Final Report on
Southeast Queensland Cloud Seeding Research Program
Sarah Tessendorf, Roelof Bruintjes, Jim Wilson, Rita Roberts, Ed
Brandes, Charlie Knight, Mike Dixon, Matt Pocernich, Kyoko Ikeda,
Courtney Arnold, Duncan Axisa, Bill Hall, Evelyn Freney, Peter R.
Buseck, Erin Towler, Peter May, Justin Peter, Louise Wilson, Michael
Manton, Steve Siems, Scott Collis, Harald Richter, Acacia Pepler, Ian
Craig, Roger Stone, Stuart Piketh, Roelof Burger, and Don Collins
Research Applications Laboratory
National Center for Atmospheric Research
2009-10-30
Report prepared by NCAR for the
State Government of Queensland
This research is in response to requirements and funding by the State Government of
Queensland. The views expressed are those of the authors and do not necessarily represent
the official policy or position of the Queensland Government.
1
Table of contents
EXECUTIVE SUMMARY .............................................................................................................7
REPORT OVERVIEW ....................................................................................................................9
1
Introduction ............................................................................................................................18
1.1
Project Objectives ......................................................................................................... 19
1.2
Personnel, Facilities, Training ...................................................................................... 19
1.2.1 Personnel ....................................................................................................................19
1.2.2 Airborne measurements .............................................................................................21
1.2.3 Radar measurements ..................................................................................................23
1.2.4 Disdrometer measurements ........................................................................................35
2
Summary of Operations .........................................................................................................37
2.1
Flight Operations .......................................................................................................... 38
2.1.1 Dual-aircraft operations .............................................................................................42
2.1.2 Randomized seeding operations ................................................................................43
3
2.2
Radar Operations .......................................................................................................... 45
2.3
Disdrometer Operations ................................................................................................ 46
Weather and Climate..............................................................................................................50
3.1
Introduction ................................................................................................................... 50
3.1.1 Overview of results ....................................................................................................50
3.2
Atmospheric Environment ............................................................................................ 52
2
3.2.1 Cluster analysis of sounding data ..............................................................................52
3.2.2 The seven-cluster case ...............................................................................................53
3.2.3 Contribution of cluster regimes to annual rainfall .....................................................56
3.2.4 Summary ....................................................................................................................58
3.3
Historical Precipitation Patterns and Relationships with Global-Scale Features ......... 59
3.3.1 Background ................................................................................................................59
3.3.2 Analysis for SEQ .......................................................................................................60
3.4
Radar Climatology ........................................................................................................ 67
3.4.1 Data sets .....................................................................................................................68
3.4.2 Synoptic regimes ........................................................................................................69
3.4.3 Statistics of Southeast Queensland storms .................................................................77
3.4.4 Conclusions ................................................................................................................87
3.5
Lightning Climatology .................................................................................................. 88
3.6
Numerical Modeling ..................................................................................................... 92
3.6.1 Air mass origins and back trajectories .......................................................................92
3.7
Mesoscale Features that Affect Cloud Development.................................................... 97
3.7.1 Impact of local environment and terrain ....................................................................97
3.7.2 High impact severe weather events..........................................................................100
4
Aerosol and Microphysics Studies.......................................................................................109
4.1
Introduction ................................................................................................................. 109
3
4.1.1 Overview of results ..................................................................................................109
4.1.2 Impacts of aerosols on precipitation ........................................................................111
4.2
Aerosols in the Atmosphere and the Impacts on Droplet Size Distributions and
Precipitation Development...................................................................................................... 112
4.2.1 General aerosol characterization ..............................................................................112
4.2.2 Chemical composition of aerosol.............................................................................119
4.2.3 Summary of aerosol conditions ...............................................................................132
4.3
Observed Microphysical Characteristics of Clouds in Southeast Queensland ........... 133
4.3.1 Precipitation development in clouds ........................................................................133
4.3.2 Cloud droplet spectra characteristics and evolution in different types of cloud
systems and impacts for seeding ..........................................................................................135
4.3.3 Primary and secondary ice production .....................................................................139
5
Cloud Seeding Studies and Assessment ..............................................................................153
5.1
Introduction ................................................................................................................. 153
5.1.1 Overview of results ..................................................................................................153
5.1.2 Background on hygroscopic seeding .......................................................................154
5.1.3 Flare particle characterization ..................................................................................157
5.2
Randomized Seeding Experiment ............................................................................... 162
5.2.1 Background on randomized experiments.................................................................162
5.2.2 Exploratory versus confirmatory experiments .........................................................162
5.2.3 Randomized design ..................................................................................................162
5.2.4 Significance and p-values ........................................................................................163
4
5.2.5 TITAN data ..............................................................................................................163
5.2.6 General statistics and summary cases ......................................................................164
5.2.7 Further statistical analyses .......................................................................................168
5.2.8 Preliminary calculations of the potential of cloud seeding to enhance water
resources ..............................................................................................................................185
5.2.9 Conclusions from randomized seeding cases ..........................................................193
5.3
Seeding Case Studies .................................................................................................. 194
5.3.1 Polarimetric radar analysis .......................................................................................194
5.3.2 Simultaneous microphysical measurements ............................................................205
5.3.3 Preliminary dual-Doppler studies ............................................................................206
6
Summary and Recommendations ........................................................................................212
6.1
Recommendations ....................................................................................................... 215
References ....................................................................................................................................218
Acknowledgements ......................................................................................................................224
7
Appendix A—Data Quality .................................................................................................226
7.1
Aircraft Instrument Intercomparisons ......................................................................... 226
7.2
Radar and Disdrometer Measurements ....................................................................... 231
7.2.1 Gamma drop size distribution model .......................................................................231
7.2.2 Disdrometer comparison ..........................................................................................232
7.2.3 CP2 calibration.........................................................................................................239
8
Appendix B—Analysis Methods .........................................................................................245
5
8.1
Statistical Methods for the Randomized Seeding Trial .............................................. 245
8.1.1 Survival analysis for evaluating the duration of targets ...........................................245
8.2
Derived Microphysical Relationships ......................................................................... 246
8.2.1 Disdrometer-derived relationships ...........................................................................246
8.2.2 Quantitative rainfall estimation algorithms .............................................................248
6
EXECUTIVE SUMMARY
During the two seasons of the Southeast Queensland (SEQ) Cloud Seeding Research Program
(CSRP), which took place between December 2007-March 2008 and November 2008-February
2009, physical measurements related to cloud and precipitation processes were collected, and
statistical cloud seeding trials were conducted in a randomized seeding experiment. Findings
from this program inform the recommendations presented herein.
Current Assessment of Cloud Seeding Technologies in SEQ
1. The results from the randomized seeding experiment in the SEQ CSRP can be
summarized as follows:
a. The statistical analysis results show similar tendencies (although based on a
relatively small sample) to hygroscopic seeding experiments conducted in Mexico
and South Africa where seeded clouds tended to rain over a longer time and larger
area than unseeded clouds
b. The infrastructure in SEQ allows for further focused research that could confirm
the statistical results, as well as improve our understanding of the physical basis
of the results from the current and previous cloud seeding studies
c. As a next step, a study should be conducted to determine the scope and cost of an
experimental program and to conduct a cost-benefit study on the potential
additional water resources provided by cloud seeding compared to other
alternatives. Initial estimates are encouraging and motivate more in-depth study.
2. Any future application of cloud seeding as a technology for enhancing water resources in
SEQ should be informed by the results of the physical measurements:
a. Glaciogenic seeding (with silver iodide) would not be advisable in clouds in the
coastal regions of SEQ
b. Hygroscopic seeding is recommended for the convective clouds in the region
c. Operations should be focused on the period of October to February (when the
clouds most suitable for cloud seeding with hygroscopic flares are more likely to
occur)
d. Hygroscopic seeding may be even more effective in convective clouds with
colder cloud bases, such as those observed earlier in the summer season, as well
as possibly further inland
e. Neither glaciogenic nor hygroscopic seeding would be effective in the deep
summer widespread stratiform systems with weak embedded convection that
sometimes traverse the region. These systems are naturally highly efficient at
producing precipitation.
3. A scientific research component should be included in any further work to gain further
confidence in the results and to improve understanding of hygroscopic seeding processes.
The randomized statistical results do indicate a tendency that seeding increased
precipitation in convective clouds in the region; however, it is recommended that further
7
randomization experiments be included in any future operations. These randomized
cases will need careful analysis using high quality radar, preferably with dualpolarization capability.
4. Seeding aircraft used in any future work should be instrumented to measure temperature,
pressure, humidity, three-dimensional winds, cloud droplet and particle spectra, and
aerosol spectra.
Enhanced Data Management and Access
5. The data collected as part of the Queensland program is a unique data set that is of
interest to many stakeholders with weather and climate applications and will be a longterm resource (detailed data sets such as this are rare and are used for decades). Therefore
a public-accessible, searchable data archive should be established for users to easily gain
access to this data set and preserve it for future use. Papers and publications developed
using this information should also be made readily available through this repository.
6. Key stakeholders should be made aware of the data arising from this project as this data
set will have applications beyond cloud seeding, including:
a. Improved operational quantitative rainfall estimation and forecasting
b. Improved water resource/hydrological management tools (beyond cloud seeding)
c. Improving understanding of severe weather events (e.g., ―The Gap storm‖) and
development of high impact weather detection and forecasting
d. Air quality and pollution monitoring and characterization.
Further Analysis
7. Detailed studies have been conducted on the meteorology of SEQ. These studies have
been conducted independently by several scientists and institutions and it is important
that this work be synthesized and the results published.
8. Numerical parcel, cloud-resolving, and mesoscale models together with the data collected
should be used to improve understanding of precipitation processes, evaluate the seeding
effects on precipitation, and refine the seeding methods for any future operational and/or
research program.
9. This was the first time that a dual-polarization radar was available to a project of this kind
anywhere in the world. Insufficient time was available in the contract period to conduct
an in-depth analysis of this data as new analysis methods are required. The data can
provide vital new insights in seeding effects on clouds and should be further analyzed. It
is recommended that support for further analysis of this data be provided:
a. To utilize the dual-polarization radar data to improve our understanding of natural
and seeded precipitation formation and study changes in cloud microphysics and
dynamics from seeding
b. To validate the hygroscopic seeding conceptual model (that seeding modifies the
initial drizzle and raindrop spectra aloft in clouds).
8
REPORT OVERVIEW
Water shortages have occurred in the Southeast Queensland region due to several years
of reduced rainfall and increasing population. As population in this region continues to increase,
water management will continue to be a vital concern. As a response, the Queensland
Government has invested resources into studying the feasibility of precipitation enhancement via
cloud seeding. A very important part of this feasibility study is to obtain high-quality
measurements of atmospheric conditions within the region as they pertain to cloud and
precipitation processes. Aerosol and microphysical measurements, in particular, can help
determine if seeding could be beneficial and also help determine what the optimal seeding
method would be with regard to its potential for enhancing precipitation in local clouds.
The potential for such man-made increases in precipitation is strongly dependent on the
natural microphysics and dynamics of the clouds that are being seeded (in this case microphysics
means the size and concentration of water droplets and ice inside clouds). These factors can
differ significantly from one geographical region to another, as well as during and between
seasons in the same region. In some instances, clouds may not be suitable for seeding, or the
frequency of occurrence of suitable clouds may be too low to warrant the investment in a cloud
seeding program. Both factors need to be evaluated from a climatological perspective. It is
therefore important to conduct preliminary studies on the microphysics and dynamics of the
naturally forming clouds prior to commencing a larger, operational experiment. It is also
important to conduct hydrological studies relating rainfall with river flows and reservoir levels,
and to determine hydrological regions where reservoir catchments are most efficient. Seeding
could then be optimized by preferentially targeting the most efficient watersheds.
If shown to be feasible, the cloud seeding technique(s) should then be evaluated using a
randomization procedure to demonstrate statistically that the seeding method works, and to
quantify the increases achieved. This approach is similar, for example, to what is commonly
done in medical trials with a new drug. Such a confirmatory randomized statistical experiment
would become the second phase of a future program.
Scientists from the National Center for Atmospheric Research (NCAR), the South
African Weather Service (SAWS), and Weather Modification Inc. (WMI), in collaboration with
the Australian Bureau of Meteorology (BoM), Monash University, the University of Southern
Queensland, the Centre for Australian Weather and Climate Research (CAWCR), CSIRO, and
MIPD Pty Ltd, conducted the feasibility study for rainfall enhancement via cloud seeding during
the summer rainfall regime in Southeast Queensland in Australia. The feasibility study included
a range of measurement systems, some using novel technologies. The most unique aspect of this
study was the availability of a dual-polarization, dual-wavelength Doppler weather research
radar (called CP2). This is the first cloud seeding experiment that had a dual-polarization, dualwavelength radar available in addition to dual-Doppler radar coverage. This is significant in that
it is the first time the potential exists to observe the evolution of microphysical precipitation
processes, such as precipitation particle type, number, and size, within a seeded cloud while at
the same time observing the air flow patterns within the cloud.
9
Typically the evaluation of cloud seeding experiments has relied on statistical evaluation
of the storm water mass, duration, size, and radar reflectivity. These parameters are generally
derived from weather radar observations, yet because of the very large natural variability in
storm characteristics and the local meteorological environment that can dwarf any cloud seeding
effects, a large sample size of randomized seeded and unseeded cases is required to obtain
statistical significance. Even then there is great uncertainty in the true understanding of physical
mechanisms that were responsible for any seeding effect the statistical analysis might have
suggested. This project was undertaken in the hopes that through the use of physical
measurements such as aircraft and a dual-polarization, dual-wavelength radar, precipitation
microphysics processes could be more directly observed making it possible to understand cause
and effect seeding relationships, thus not having to solely rely on statistical means that often
generate controversy.
Analysis efforts for the Queensland Cloud Seeding Research Program (CSRP) were
therefore focused on three major areas for the greater Brisbane region: understanding the weather
and climate, characterizing the atmospheric aerosol and cloud microphysics, and assessing the
impact of cloud seeding to enhance rainfall. These research efforts have produced a great
number of results to date, although there are still some areas that require further analysis. The
data sets collected in the two field seasons are vast and unprecedented for a cloud seeding
research project, and thus many varied research efforts can continue to utilize the Queensland
CSRP data sets well into the future.
The first effort in this feasilibility study is to understand the basic regional climatology of
precipitation, including the weather patterns and conditions that drive convection, in order to put
the cloud seeding and precipitation process analyses into context, as well as to determine the
frequency of clouds that may be suitable for seeding within the region.
Key questions to be answered by the climatology studies are:

How do the large-scale weather patterns and thermodynamic environment
influence convective activity and precipitation production in the region? What
conditions are more favorable to producing convective precipitation, and what
conditions yield the most rainfall?

What is the frequency and location of convection within the region that may be
amenable to hygroscopic seeding?
Since the primary goal of this project is to ascertain if cloud seeding is a feasible means
for enhancing rainfall in the Southeast Queensland region, analyses to study the effects of cloud
seeding are paramount. While the effects of seeding are often mostly based on randomized
seeding statistical analyses, it is also important to gain a good physical understanding of seeding
effects to be able to explain and support the statistics. Therefore, in order to fully understand the
effects of cloud seeding, a working knowledge of the natural precipitation processes in the region
is vital, including the environmental conditions that influence cloud microphysics (such as subcloud aerosol particles). The results presented herein provide an overview of the natural aerosol
and cloud microphysical processes in Southeast Queensland clouds.
10
Key questions to be answered by the aerosol and microphysics studies are:

What are the naturally occurring aerosol and droplet size spectra? How do these
microphysical characteristics affect precipitation processes, such as warm rain
formation?

What are the conditions in the mixed-phase regions of storms in Southeast
Queensland? How much liquid water is present in these regions, and how does it
relate to the precipitation efficiency in these clouds? Could they be made more
efficient from hygroscopic seeding?
The statistical results presented in this report are solely exploratory because no a priori
statistical design was conducted for this program due to the lack of knowledge of natural cloud
and precipitation processes in clouds at the start of the Queensland program. However, the
statistical analysis provides a first glance at potential effects of seeding and offers guidance for
important physical analysis of the data to interpret the statistical results. For comparison, a
statistical randomized seeding experiment very similar to those in South Africa and Mexico was
conducted.
Key questions to be answered by the cloud seeding assessment studies are:

What microphysical effects does seeding with hygroscopic material at cloud base
have on clouds in Southeast Queensland?

What are new methods that can be developed to study the physical effects of
seeding, especially those that utilize advanced radar and measurement
technologies?
The following is a summary of key results derived from the analysis of data collected
during the 2007–2008 and 2008–2009 seasons in Southeast Queensland. The results are broken
down into three categories: weather and climate, aerosol and microphysics studies, and cloud
seeding studies and assessment. Each category is a crucial component to understanding the
feasibility of seeding in Southeast Queensland.
Weather and Climate
Key findings:
1. The wet season in Southeast Queensland is broadly between October and March, with
November–February receiving the most rainfall and having the most atmospheric
moisture.
2. The synoptic cluster analyses and statistical tests have shown that Southeast Queensland
can be divided into ‗wet‘ and ‗dry‘ weather regimes, with the ‗wet‘ regimes occurring
most in summer and the ‗dry‘ regimes more in winter. This result is in itself not
surprising, as this fact is already well known to the local population and forecasters, but
this analysis has further quantified these regimes nonetheless.
11
a. The ‗wet‘ regimes are responsible for the majority of the region‘s rainfall and
include the northwesterly and ‗moist‘ southeasterly, easterly, and westerly
regimes (from the seven-cluster analysis), or the northwesterly and (when
unstable) the southeasterly regimes (from the three-cluster analysis). The
northwesterly regime contributes greatly to the total annual rainfall despite
occurring less than 10% of the time.
b. The ‗dry‘ regimes are predominantly the other southeasterly clusters (which is
also the most common regime occurring year-round yet contributes very little to
annual rainfall in the seven-cluster case), and the southwesterly regime, which is
most common in winter (in both the seven- and three-cluster cases).
3. The southeasterly regime(s) tends to be less unstable than the northwesterly regime and
can have a trade wind inversion below the freezing level, resulting in shallow convection
with rain formed via warm rain processes and only light rain. Given the tendency for
higher instability and thus deeper convection in the northwesterly regime, this regime
was also overrepresented in the aircraft measurements (as we tended to fly on days with
convection, especially focusing on deep convection).
4. Synoptic scale features (such as east coast lows) are generally the source of rain during
the winter months. Variation throughout the wet season is also attributable to synoptic
scale disturbances, as well as mesoscale features.
5. The Southern Oscillation Index (SOI) appears to be positively correlated with
precipitation rate, and more so with precipitable water (a measure of atmospheric water
vapor), such that positive SOI (La Niña) corresponds to wetter conditions. A tailored
index using sea surface temperature (SST) off the east coast of Australia showed an even
better positive correlation with precipitable water (higher SSTs relates to higher
precipitable water in the Brisbane region). These indices can be used for predicting
potential seasonal precipitation, as well as to assist in understanding historical trends in
precipitation.
6. The radar climatology indicated that cell volume, area and height all followed truncated
lognormal distributions. The same result was found when the three-cluster synoptic
regimes were considered individually. Nonetheless, the largest 20% of cells were found
to be responsible for roughly 67% of the fractional areal coverage of precipitation.
7. Student‘s t-tests applied to examine the relationship between the three clusters found that
the precipitation area was significantly larger during the winter regime than during either
of the summer regimes and cell-top height was significantly lower during the winter
months than during the summer.
8. Four regimes of surface wind flow conditions were revealed by HYSPLIT back trajectory
modeling. These regimes relate to surface aerosol conditions and were classified into two
maritime regimes (northeasterly and east/southeasterly), and two continental regimes
(northwesterly and west/southwesterly).
12
9. Based on the CG lightning activity alone, it suggests that the mixed-phase microphysics
of clouds farther west is more vigorous than for those closer to the coast, where the CSRP
domain is presently located. Cloud seeding with hygroscopic or glaciogenic flares may be
more effective in that region.
Aerosol and Microphysics Studies
Key findings:
1. Of the four HYSPLIT regimes, the northeasterly regime, on average, had the highest
fraction of aerosol that served as CCN at the 0.3% supersaturation, while the westerly
regime had the lowest. These results indicate that a higher fraction of the maritime flow
aerosol served as CCN (at 0.3% supersaturation) than from the continental flow (NWly,
Wly) days, especially so for the westerly regime.
2. The maritime flow (NEly and Ely) regimes have a more robust coarse mode in their
aerosol size distributions, especially of particles >5 µm, compared to the continental
(NWly and Wly) regimes. The Ely regime tends to have the weakest fine mode, and thus
the overall cleanest conditions. This is expected given the flow in this regime is coming
straight in from the ocean. The NWly regime often has a high number of fine particles
and/or very few coarse particles. The results show, however, that coarse mode particles
in small concentrations were observed in all four regimes, although they were more
common in higher concentrations in the NEly and Ely regimes.
3. The CCN supersaturation cycle measurements also reiterate that the Ely regime is the
cleanest regime overall with the lowest mean CCN concentrations at all measured
supersaturations. There was a general tendency for the NEly, NWly, and Wly regimes to
have similar CCN levels at each supersaturation, all of which are roughly twice as high as
for the Ely regime.
4. The fine mode fraction of aerosols is dominated by sulfates and most of the CCN active
particles are ammonium sulfate. In one case, which corresponded to an Ely regime day,
the fine modes were dominated by small sea salt particles. The intermediate and coarse
mode aerosols are a mix of dust and other particles (aluminosilicates). In the other two
cases analyzed, there was no indication from any of the size fractions that the aerosol
samples were of marine origin, and appeared more continental, possibly having an
anthropogenic source (NEly regime day with back trajectory that passed over the city of
Brisbane) or biomass burning source (NWly regime day). The NEly day, however, did
have some sodium-bearing particles that may be aged sea salt.
5. The average maximum drop concentrations are similar for all regimes, however, ranging
between 500 and 650/cc, and the results show that on average, the Wly and NWly
regimes have the highest drop concentrations and smaller mean diameters, compared to
the maritime NEly and Ely regimes.
6. The NEly and Ely regimes have the highest concentrations of drops greater than 20 µm,
which can indicate a greater likelihood that collision and coalescence will be active in
13
clouds in this regime. This is also possibly related to the fact that both of these regimes
had a strong coarse mode of aerosol particles.
7. The tendencies in the results show that the mean diameter and concentration of large
drops (>20 µm) in each regime are larger and/or higher in seeded clouds than in the
natural (non-seeded) clouds. This is consistent with the first step of the hygroscopic
seeding conceptual model and is also the same trend observed in the season one results.
8. During the early part of the season, when cloud bases are generally higher, coalescence
could be delayed, ice multiplication may not occur, and thus first ice will only form at
temperatures colder than −10o C. Hygroscopic seeding may be more effective in these
clouds by providing earlier coalescence and possibly the onset of a more efficient ice
process.
9. Clouds in Southeast Queensland generally develop precipitation initially via the ―warm
rain‖ process that then results in a more efficient mixed-phase process in deeper
convective systems that extend above the freezing level. Seeding with hygroscopic flares
could potentially enhance the ―warm rain‖ process, but glaciogenic seeding would not be
advised in these conditions.
10. These results would also explain the lack of intense lightning storms in the coastal
regions as observed during the field efforts, since liquid water content in the mixed-phase
region (necessary for storm electrification) is often rapidly depleted due to efficient
mixed-phase processes and ice multiplication.
11. Deep stratiform systems are occasionally observed in the region, yet our observations
indicate that the natural precipitation processes are very efficient in these systems. Thus,
neither hygroscopic or glaciogenic seeding may have any effect and glaciogenic seeding
may actually have a negative effect.
Cloud Seeding Studies and Assessment
Key findings:
1. The statistical results indicate that seeded clouds tend to produce more rain than unseeded
clouds due to longer lifetimes and producing rain over a larger area. The results show the
same tendencies as were observed in previous experiments in South Africa and Mexico
that used the same hygroscopic seeding techniques.
2. The sample size has a strong impact on the statistical results. Efforts have been made to
use appropriate analysis techniques to interpret and understand the data and results. The
sample size at this stage does not contain enough samples to provide an unambiguous
statistical result. Therefore, future operations should focus on increasing the randomized
sample size or attempt to design (a priori) a new confirmatory randomized experiment.
3. The statistical analysis indicates that the seeded cases tend to live longer than the
unseeded cases after seeding decision time, indicating a potential enhancement in storm
duration from hygroscopic seeding, as shown by the decreased hazard rate for seeded
14
storms. These results are similar to those found in the South African and Mexican
experiments, but must be viewed with caution, as they are not statistically significant at
the 95% level due to the small sample size (see below).
4. The statistical analysis indicates that the difference between the height of the maximum
reflectivity and the reflectivity centroid in the seeded cases seemed to increase 10 to 20
minutes after seeding commenced. This result remained consistent regardless of the
selection criteria, and is also consistent with the results of Mather et al. (1997).
5. A general tendency for seeded drop size distributions to have larger mean diameters and
higher large (>20 µm) drop concentrations in seeded clouds compared to non-seeded
clouds was observed in the season one dual-aircraft measurements, as well as in the
season two microphysical measurements. These observations suggest that the first step in
the hygroscopic seeding conceptual model is occurring. Droplet concentrations in the
Queensland clouds are very similar to those observed in South Africa and Mexico.
6. New techniques to study seeded versus non-seeded cells are being developed, to extend
the analysis beyond the traditional statistical analysis that uses reflectivity only (via
TITAN), by utilizing dual-polarization and dual-Doppler data from the CP2 radar. These
methods are in their infancy, but are valuable tools to help understand the physics of any
potential seeding effect.
7. Preliminary analyses indicate that suitable targets for seeding are major contributors to
overall seasonal rainfall. Assuming a 20% increase in precipitation by seeding, this could
result in upwards of a 5% increase in ground-measured rainfall per catchment.
8. Although the initial statistical results from this limited data set seem to behave in a
similar manner as was found for the South African and Mexican experiments, the current
results will have to be interpreted with caution because they are not yet statistically
significant and there is a possibility that the results could still be due to chance.
Recommendations:
The findings presented above inform the following recommendations:
1. The results from the randomized seeding experiment in the SEQ CSRP can be
summarized as follows:
a. The statistical analysis results show similar tendencies (although based on a
relatively small sample) to hygroscopic seeding experiments conducted in Mexico
and South Africa where seeded clouds tended to rain over a longer time and larger
area than unseeded clouds
b. The infrastructure in SEQ allows for further focused research that could confirm
the statistical results, as well as improve our understanding of the physical basis
of the results from the current and previous cloud seeding studies
c. As a next step, a study should be conducted to determine the scope and cost of an
experimental program and to conduct a cost-benefit study on the potential
15
2.
3.
4.
5.
6.
additional water resources provided by cloud seeding compared to other
alternatives.
Any future application of cloud seeding as a technology for enhancing water resources in
SEQ should be informed by the results of the physical measurements:
a. Glaciogenic seeding (with silver iodide) would not be advisable in clouds in the
coastal regions of SEQ
b. Hygroscopic seeding is recommended for the convective clouds in the region,
c. Operations should be focused on the period of October to February (when the
clouds most suitable for cloud seeding with hygroscopic flares are more likely to
occur)
d. Hygroscopic seeding may be even more effective in convective clouds with
colder cloud bases, such as those observed earlier in the summer season, as well
as possibly further inland
e. Neither glaciogenic nor hygroscopic seeding would be effective in the deep
summer widespread stratiform systems with weak embedded convection that
sometimes traverse the region. These systems are naturally highly efficient at
producing precipitation.
A scientific research component should be included in any further work to gain further
confidence in the results and to improve understanding of hygroscopic seeding processes.
The randomized statistical results do indicate a tendency that seeding increased
precipitation in convective clouds in the region; however it is recommended that further
randomization experiments be included in any future operations. These randomized
cases will need careful analysis using high quality radar, preferably with dualpolarization capability.
Seeding aircraft used in any future work should be instrumented to measure temperature,
pressure, humidity, three-dimensional winds, cloud droplet and particle spectra, and
aerosol spectra.
The data collected as part of the Queensland program is a unique data set that is of
interest to many stakeholders with weather and climate applications and will be a longterm resource (detailed data sets such as this are rare and are used for decades). Therefore
a public-accessible, searchable data archive should be established for users to easily gain
access to this data set and preserve it for future use. Papers and publications developed
using this information should also be made readily available through this repository.
Key stakeholders should be made aware of the data arising from this project as this data
set will have applications beyond cloud seeding, including:
a. Improved operational quantitative rainfall estimation and forecasting
b. Improved water resource management tools (beyond cloud seeding)
c. Improving understanding of severe weather events (e.g., ―The Gap storm‖) and
development of high impact weather detection and forecasting
d. Air quality and pollution monitoring and characterization.
16
7. Detailed studies have been conducted on the meteorology of SEQ. These studies have
been conducted independently by several scientists and institutions and it is important
that this work be synthesized and the results published.
8. Numerical parcel, cloud-resolving, and mesoscale models together with the data collected
should be used to improve understanding of precipitation processes, evaluate the seeding
effects on precipitation, and refine the seeding methods for any future operational and/or
research program.
9. This was the first time that a dual-polarization radar was available to a project of this kind
anywhere in the world. Insufficient time was available in the contract period to conduct
an in-depth analysis of this data as new analysis methods are required. The data can
provide vital new insights in seeding effects on clouds and should be further analyzed. It
is recommended that support for further analysis of this data be provided:
a. To utilize the dual-polarization radar data to improve our understanding of natural
and seeded precipitation formation and study changes in cloud microphysics and
dynamics from seeding
b. To validate the hygroscopic seeding conceptual model (that seeding modifies the
initial drizzle and raindrop spectra aloft in clouds).
17
1 Introduction
Many regions of Australia have been experiencing a severe drought over the past few
years. In response, the Queensland government decided to explore the potential for cloud seeding
to enhance rainfall. The potential for such man-made increases in rainfall using cloud seeding is
strongly dependent on the natural microphysics (i.e., the size and concentration of water droplets
and ice particles inside clouds) and dynamics (i.e., the forces affecting air motions in and around
clouds) of the clouds that are being seeded. Microphysics is in turn dependent on background
aerosol levels, because it is the aerosol particles that attract water vapor to form cloud droplets,
and in cold clouds, ice particles. Furthermore, the types and concentrations of aerosol particles
can be influenced by trace gases (i.e., air pollution). Given these dependencies, the microphysics
of clouds can differ significantly from one geographical region to another, and even between
seasons in the same region. In some instances, clouds may not be suitable for seeding, or the
frequency of occurrence of suitable clouds may be too low to warrant the investment in a cloud
seeding program. Both factors need to be evaluated in a climatological sense or at least they
should be evaluated over a sufficient period of time to account for natural variations. This
requires conducting preliminary studies on atmospheric aerosols, pollution levels, and cloud
microphysics and dynamics prior to commencing a large cloud seeding effort. If the targeted
measurements and additional data show sufficient evidence that clouds would be positively
affected by cloud seeding, the cloud seeding technique(s) should then be evaluated using a
randomization procedure to statistically demonstrate that the seeding method is effective and
measurable. This approach is similar, for example, to what is commonly done in medical trials
with a new drug. Such a seeding feasibility study was conducted in the Southeast Queensland
region December 2007–March 2008 and again from October 2008–February 2009. In both
seasons of the field effort a preliminary exploratory randomized study was also performed.
Scientists from the National Center for Atmospheric Research (NCAR) the South African
Weather Service (SAWS), and Weather Modification Inc. (WMI), in collaboration with the
Australian Bureau of Meteorology (BoM), Monash University, the University of Southern
Queensland, the Centre for Australian Weather and Climate Research (CAWCR), CSIRO, and
MIPD Pty Ltd, conducted the feasibility study for rainfall enhancement via cloud seeding during
the summer rainfall regime in Southeast Queensland in Australia. The feasibility study included
a range of measurement systems, some using novel technologies. The most unique aspect of this
study was the availability of a dual-polarization, dual-wavelength Doppler weather research
radar (called CP2). This type of advanced weather radar has been identified in a report (National
Academy of Sciences, 2003) by the U.S. National Research Council (NRC) of the National
Academy of Sciences (NAS) as essential to address some of the outstanding questions related to
the assessment of rainfall enhancement. The CP2 was acquired by the BoM as part of a joint
agreement with NCAR that continued previous joint research activities. The considerable radar
expertise within CAWCR and NCAR is a major factor contributing to this program. In addition
to the CP2 radar, the collaborative study entailed all necessary aspects of a cloud seeding project,
including: airborne chemistry, aerosol, and cloud physics measurements; assessment and
enhancement of weather radar capabilities (additional hardware and software for collection of
quantitative data); and analyses of field data. These efforts build on the experience gained from
programs in other parts of the world.
18
This report describes the results of the scientifically based Cloud Seeding Research
Program (CSRP) conducted to evaluate the feasibility for using cloud seeding technology,
specifically hygroscopic seeding, to enhance rainfall. The following chapters include
preliminary results from climatology studies, aerosol and microphysical process studies, and
cloud seeding assessment studies.
1.1 Project Objectives
The scientific objectives for the initial feasibility studies were to make preliminary assessments
of:
1) the climatological characteristics of precipitation and in particular the frequency of
clouds suitable for seeding
2) the approaches necessary to obtain robust estimates of precipitation amount and retrieve
microphysical properties of the clouds
3) the effect of cloud seeding on storm microphysics and dynamics, including precipitation
particle types, number and size of precipitation particles and horizontal and vertical air
motions, to determine if there is evidence of increased secondary convective storm
initiation and/or precipitation enhancement from cloud seeding.
1.2 Personnel, Facilities, Training
1.2.1 Personnel
1.2.1.1 Operations Director
In both seasons, Roelof Bruintjes and Sarah Tessendorf acted as the primary Operations
Directors, with brief periods of leave covered by Mike Dixon. The Operations Director worked
from the Operations Center, managing and directing the day-to-day operations and personnel.
Decisions were made after assessment of the daily forecast, including calling for aircraft
operations (coordinating with the personnel at the Archerfield hangar), guiding the aircraft to
suitable areas/clouds for investigation and seeding, and ensuring that proper data collection was
taking place. The Operations Director worked closely with all personnel, particularly those at the
Archerfield hangar and CP2 Operations Center.
1.2.1.2 Radar operations
The Operations Director oversaw radar operations; however, at least one radar scientist
was on site to handle the responsibility of changing scanning strategies according to the mission
objectives. The radar scientist was also responsible for monitoring the development of
clouds/storms, as well as the progress of the current radar objective/mission. The radar scientists
during both seasons included Justin Peter, Jim Wilson, Ed Brandes, Rita Roberts, Mike Dixon,
Peter May, and Acacia Pepler. Alan Seed also helped with radar operations in season one, and
Eric Nelson and Louise Wilson helped in season two. The radar hardware and software were
regularly maintained by Ken Glasson, the CP2 radar engineer. This included periodic
19
maintenance, calibration checks, general troubleshooting, and radar data archival. Mike Dixon
and Phil Purdam led the software development and implementation efforts.
1.2.1.3 Forecaster
In season one, Justin Peter provided the daily forecasts, and in season two several BoM
staff served terms as the project forecaster (including Scott Collis, Tony Bannister, Harald
Richter, and Peter May). The daily forecasts resulted in designating the flight plan options for
the day. The plan for the day was communicated by the Operations Director to the pilots and
crew at the Archerfield hangar by around 10 AM each day at the daily weather/flight briefing
and plans for flights were decided at this time. Changes in meteorological conditions were
monitored throughout the day and occasionally the plan was adjusted to reflect significant
changes. Besides the daily forecast, outlooks for the following 2–3 days were issued by the
forecaster to assist in staffing decisions for future operations. Another important function of the
forecaster was to archive weather data, such as upper air soundings, synoptic charts, satellite
imagery, and local meteorological observations.
1.2.1.4 Pilots
Pilots experienced in cloud seeding operations are critical to the successful
implementation of the operations plan. For the SAWS ―SEEDA1‖ research aircraft, Ret
Orsmond, a pilot with a long history of cloud seeding operations, was available to the project for
carrying out research flight operations in season one, along with his crew of pilots (Hans Kruger,
Gary Wiggins, and John Hingst). Gary Wiggins and John Hingst were the pilots for SEEDA1 in
season two. In season one, the MIPD/WMI pilots conducted the primary cloud seeding
operations, and included pilots experienced in cloud seeding operations from WMI (Jim Carr)
and from MIPD (Neville O‘Donnell, Paul Brady, and Greg Choma). Pilots were also responsible
for taking in-flight notes and submitting their flight notes/reports to the Data Managers on a daily
basis.
1.2.1.5 Aircraft and instrument maintenance
Periodic calibration checks, aircraft instrument maintenance, and general troubleshooting
of aircraft instrumentation were performed by the instrument technician. Steve Edwards served
in this role in both seasons, with Nico Mienie also filling in this role in season two. Aircraft
maintenance was monitored and performed on the SEEDA1 aircraft by Harold McGarry (both
seasons) and Rafe Zerby for WXMOD in season one. An instrument scientist flew on each
research flight, to operate and monitor instrument function during flight. These personnel
included Stephen Broccardo, Xolile Ncipha, and Shaazia Bhailall, and in season two Jared
Lodder and Nicola Walton also flew as flight scientists.
1.2.1.6 Data Managers
Data management was distributed among all the personnel, under the direction of the
Operations Director. An important part of this effort involved regular reporting of the instrument
and data status to the Operations Director and throughout the project staff. Ian Craig and Li
Fitzmaurice were the primary contacts for the gathering and archiving of project data in season
20
one and Ian Craig and all flight scientists assisted with this effort in season two. Aircraft
operations report writing duties were shared by Ian Craig, Martin van Nierop, Nico Kroese,
Karel de Waal, and Stuart Piketh in season one, and led by Ian Craig in season two. Radar report
writing duties were shared by the Radar Scientists and Operations Director. Roelof Burger,
Duncan Axisa, Bruce Woodcock (Aventech), and Bill Dawson (DMT) provided assistance with
aircraft data processing and quality checks.
1.2.2 Airborne measurements
Two aircraft were used during the first season of the project: one was primarily a research
aircraft, but also could serve as a secondary seeding aircraft if conditions warranted; the second
aircraft was the primary seeding aircraft. In season two, the research aircraft also served as the
seeding aircraft for a single aircraft operation. The research aircraft was the South African
Weather Service (SAWS) Aerocommander (ZS-JRA; call sign SEEDA1; Figure 1.1b). It carried
flare racks on each wing (10 burn-in-place hygroscopic or silver iodide flare capacity each), and
had a full suite of atmospheric instrumentation (see section 1.2.2.1). The season one primary
seeding aircraft was the Weather Modification Inc (WMI)/MIPD Piper Cheyenne II (N747RE;
call sign WXMOD; Figure 1.1a). It carried flare racks on each wing (12 burn-in-place
hygroscopic or silver iodide flare capacity each) and an undercarriage ejectable silver iodide
flare rack (306 flare capacity).
a)
b)
Figure 1.1. Photo of the a) WXMOD seeding aircraft, and b) SEEDA1 research/seeding
aircraft (a: Courtesy MIPD/WMI, b: Courtesy Scott Collis).
1.2.2.1 Instrumentation
The research aircraft had a suite of instruments capable of taking trace gas, aerosol, and
microphysical measurements in the Brisbane region, and in treated or untreated clouds (see Table
1.1 for list of instrumentation on board in each season). These instruments were monitored by an
in-flight scientist, and maintained by an instrument technician to ensure all instruments were
functioning properly.
21
Table 1.1. List of instrumentation on SEEDA1 (ZS-JRA Aerocommander).
Instrument
Purpose/Comment
Range
State Variables
Season
Rosemount Temperature,
Static and Dynamic Pressure,
and GPS parameters
Temperature, pressure, altitude, TAS, and
location – recorded on telemetry box and on
the data system (SAWS)
multiple
Both
Edgetech Dew point sensor
Moisture content (NCAR)
−40° to 60° C
2
Vaisala Temperature and
Relative Humidity
Secondary temperature and moisture content
(SAWS)
−50° to 50° C,
0–100%
Both
AIMMS-20 probe
Temperature, relative humidity, pressure,
three-dimensional wind components
multiple
Both
Cloud Physics
PMS FSSP
Cloud droplet spectra (SAWS)
0.5–47 m
1
DMT SPP-100 FSSP
Cloud droplet spectra (SAWS)
0.5–47 m
Both
PMS 2D-C
Small precipitation particle size, concentration
and shape (SAWS)
Large precipitation particle size, concentration
and shape (SAWS)
Small precipitation particle size, concentration
and shape (NCAR; part of CAPS probe listed
below)
Large precipitation particle size, concentration
and shape (NCAR)
25–800 m
1
200–6400 m
1
25–1550 m
Both
100–6200 m
2
PMS Hot-wire (King) Liquid
Water Content (LWC) Probe
Liquid water content (SAWS)
0.01–3 g m-3
Both
DMT CAPS probe
Aerosol through precipitation size
spectrometer; LWC; CIP; static and dynamic
pressure; temperature (NCAR)
multiple
Both
PMS 2D-P
DMT CIP
DMT PIP
Aerosols
DMT CCN Counter
Cloud condensation nuclei concentration and
spectra (WITS)
Depends on
Supersaturation
Both
Texas A&M DMA
0.01 to 1 m
Both
PCASP
Fine mode aerosol spectra and concentration
(NCAR)
Aerosol concentration and spectra (WITS)
0.1 to 3 m
1
DMT SPP-200 PCASP
Aerosol concentration and spectra (WITS)
0.1 to 3 m
2
ASU Aerosol Particle
Sampler
Aerosol chemical composition (NCAR)
N/A
Both
TECO SO2 (43c)
Sulfur dioxide (WITS)
0–100 ppm
Both
TECO O3 (49i)
Ozone (WITS)
0–200 ppm
Both
TECO NOy (42c)
Nitrogen oxides (NCAR)
0–100 ppm
Both
TECO CO (48c)
Carbon monoxide (WITS)
0–10,000 ppm
1
N/A
Both
Trace Gases
Cloud and Situation Imagery
Digital still camera
To show development of clouds and treatment
situations for historical purposes
22
1.2.2.2 Data quality
Data quality checks were performed in the field on a regular basis to assess instrument
performance and diagnose instrument maintenance needs. The full suite of instruments in these
two field seasons provided multiple measurements of key microphysical fields and thus allowed
for intercomparisons of these measurements to assess instrument performance and data quality.
More detailed data quality analysis has been completed post-field and show very good data
quality, especially in season two. The details of these data quality studies and instrument
intercomparisons are presented in Section 7.1 (Appendix A).
1.2.3 Radar measurements
The Australian Bureau of Meteorology (BoM) runs a large network of weather radars,
predominantly around the coast of Australia. A number of these radars cover the area around
Brisbane (see Table 1.2).
Table 1.2. BoM radars in the Brisbane region.
Site
Latitude Longitude Type
(deg)
(deg)
Wavelength
Scan
interval
Grafton
29.620 S
152.970 E
WSR 74S
10 cm
10 min
Moree
29.500 S
149.850 E
WF100C
5 cm
10 min
Mt Stapylton
27.718 S
153.240 E
Gematronik Doppler
10 cm
6 min
Marburg
27.61 S
152.54 E
EEC WSR 74S
10 cm
10 min
Mt Kanigan
25.957 S
152.577 E
EEC DWSR 8502S
10 cm
10 min
Some of the BoM radars scan a full volume once every 10 minutes, while others do this
once every six minutes. In order to get more frequent data for this project, the BoM modified the
scanning strategy for the Mount Stapylton radar to once every 6 minutes. This was important for
dual-Doppler coordination with the CP2 research radar.
The CP2 radar, originally developed and owned by NCAR, was obtained by the BoM in
2007 and set in Redbank Plains to the southeast of Brisbane (Figure 1.2). CP2 is actually two
co-located radars, the main radar being an S-band (10 cm) unit and the smaller radar being an Xband (3 cm) unit. The X-band antenna piggybacks on the main S-band antenna (see Figure 1.2.),
and is designed to view the same sample volume as the S-band radar. The technical
characteristics of CP2 are described by Bringi and Hendry (1990). The CP2 facility is supported
by BoM personnel who visit the site regularly for routine maintenance and repairs.
The CP2 system was subjected to major refurbishment work at NCAR. As part of the
joint activities of the Queensland CSRP, all drive gearboxes were refurbished, a new transmitter
focus coil assembly power supply was installed, along with a new ceramic thyatron with
associated solid state trigger drive circuitry. Following system acceptance tests, CP2 was
23
shipped to Australia for installation with spare modules and components for all updated systems.
A modern digital receiver and signal processing system is employed based on the NCAR PIRAQ
III signal processing unit along with a new antenna control and data display system.
Figure 1.2. CP2 site infrastructure at Redbank Plains: Antenna and pedestal are within an
inflated radome mounted over housing for office, storage and transceiver (left); CP2
S-band and X-band antennae (right).
1.2.3.1 Data collection and display software
The data from each BoM radar is stored in the RAPIC (RAdar PICture) data format. Each
RAPIC file contains data from a single radar volume. The RAPIC data is transmitted from each
radar site to a number of central BoM data servers. The BoM set up software to copy RAPIC
data from the radars listed in Table 1.2 from the Brisbane office to the CP2 radar operations
center. NCAR software running at CP2 converts the RAPIC data into NCAR Meteorological
Data Volume (MDV) files. The data from the CP2 radars is obtained using the NCAR PC-based
Integrated Radar Acquisition (PIRAQ) system. This PIRAQ system produces radar time series
data that is transmitted to computers running the NCAR TITAN software system, which then
computes the radar fields from the time series and stores the data in MDV files.
1.2.3.1.1 Storm tracking and display software
The TITAN system includes the capability to automatically identify and track storms in
three-dimensional radar data, such as the merged mosaic. These storm tracks can then be
displayed on a special purpose TITAN display called Rview. For a full description of the
TITAN storm tracking and analysis capabilities, see:
http://www.rap.ucar.edu/projects/titan/home/index.php
Rview is actually a pair of displays. The main Rview display shows the radar data grid,
both in plan view and in a vertical section through the 3-D radar data. Superimposed on these
data are the identified storm shapes, including the current, past, and future (or forecast) storm
shapes depending on the operational mode of the display (Figure 1.3). A secondary display,
24
TimeHist (for Time History) shows the time history of tracks selected on the main Rview display
(Figure 1.4).
Figure 1.3. Rview display, showing storms in the southern dual-Doppler lobe (yellow
circles). Orange line shows aircraft flight track for SEEDA1. Yellow numbers over the
storm area show the maximum radar reflectivity. Cyan lines outline the storm shape at the
current time.
a)
b)
c)
Figure 1.4. TimeHist window showing a) the time history of storm: volume (gray), area
(cyan), radar-estimated precipitation flux (green), radar-estimated mass (magenta) and
vertically integrated liquid (VIL) (yellow); b) the time-height profile of storm reflectivity.
The yellow line shows the storm top as estimated by the 18 dBZ contour. The numbers
indicate the height of the maximum reflectivity over time; and c) various estimates of hail
severity over time.
25
1.2.3.1.2 CIDD display software
The Cartesian Interactive Data Display (CIDD) is the main user display for the radar
system. The images showing the merged radar product (see section 1.2.3.2.1) were all produced
by CIDD. Although CIDD cannot show storm track data with the same detail as Rview, it is a
more versatile display that allows the user to overlay products such as aircraft tracks, surface
station data and lightning data on top of radar grids (Figure 1.5). CIDD also has the ability to
show high-resolution Range-Height-Indicator (RHI) displays of radar data taken by scanning
vertically through a storm rather than by the more normal horizontal scanning mode (Figure 1.6).
Figure 1.5. CIDD-generated image showing various products overlaid on the radar data:
aircraft track (orange line), surface temperature and dew point (red/gray text), surface
wind (green winds barbs), segment along which a vertical cross section or RHI is being
displayed (thick yellow line), and location of RHI scans (white tick marks).
26
a)
b)
Figure 1.6. CIDD-generated Range-Height-Indicator (RHI) display showing a detailed
vertical slice through a storm: (a) reflectivity and (b) Doppler radial velocity.
1.2.3.2 Enhancements
1.2.3.2.1 Merged radar images
In order to get the ‗big picture‘ about the weather in the Brisbane area, data from the
individual operational radars were combined into a regional ‗radar mosaic‘. Of the five
operational radars, four were operated on a 10-minute cycle, and one (Mt. Stapylton) was
operated on a six-minute cycle. Therefore it was decided to create a mosaic once every 10
minutes. To create the mosaic, the data from each radar was interpolated from its respective
radar space grid to a common Cartesian grid, and the merged reflectivity field was then
calculated by taking the maximum reflectivity at each grid point.
Figure 1.7 shows the extent of the data in the merged radar grid for heights of 1.0 km
MSL and 1.75 km MSL respectively. Because of earth curvature effects, radars are not able to
‗see‘ weather close to the ground at longer ranges. Therefore, full coverage is only obtained at
some finite height above the ground, in this case 1.75 km. Figure 1.8 shows an example of the
merged radar coverage of storms that occurred on 2008/02/05 just before 07:00 UTC. The storms
at that time were quite widespread and were detected by all five operational radars. Figure 1.9
shows the storms as seen by each of the five radars individually. The merged data is a
three-dimensional grid, which covers an area of 600 km by 600 km centered on the CP2 radar
site. The lowest plane in the grid is at 1 km MSL and the top plane is at 10.5 km MSL. The
resolution of the grid is 0.75 km in each of the three coordinate directions (x, y, z).
27
a)
b)
Figure 1.7. Coverage of the merged BoM radars, at a height of a) 1.0 km MSL and b) 1.75
km MSL.
Figure 1.8. Example of merged radar mosaic from 2008/02/05 at 06:56:08 UTC.
28
a)
b)
d)
c)
e)
Figure 1.9. Example of radar coverage from a) Mt Kanigan, b) Marburg, c) Mt Stapylton,
d) Grafton, and e) Moree on 2008/02/05 at 06:56:08 UTC that contributed to the merged
mosaic shown in Figure 1.8.
1.2.3.2.2 Dual-Doppler capabilities
In addition to measurements of reflectivity (related to precipitation content of a system)
Doppler weather radars measure the precipitation radial velocity. Using a single Doppler radar it
is difficult to resolve storm circulations and upward motions. However, by combining data from
two nearby Doppler radars, in the case of the CSRP field program the CP2 research radar and the
Bureau of Meteorology Mt Stapylton operational radar, it is possible to reconstruct the threedimensional flow field. A traditional way of determining the area where retrievals can
dependably be carried out is to map out the area where the angle of intersection between the
beams of radar varies between 150 and 30 degrees. This is known as the dual-Doppler lobes
(shown for CSRP in Figure 1.10).
Pioneer work on dual-Doppler wind retrievals is documented in Ray et al. (1980) and
later enhancements in Scialom and Lemaitre (1990). Our system is based on that described in
Protat and Zawadzki (1999). The algorithm used is similar to a three-dimensional variational
assimilation scheme used to initialize numerical weather prediction models.
29
Figure 1.10. The dual-Doppler lobes of the CP2 and Mt Stapylton radars overlaid on the
CSRP domain.
The algorithm minimizes the difference between the radial projection of the current guess
for the three-dimensional flow field and that measured by the two Doppler radars. Since the
projection of the vertical component of wind velocity onto the viewing angle of the radar beams
is small (for most cases) owing to the scanning strategy generally employed in radar meteorology,
the measured radial velocities are insensitive to vertical motions. This requires the use of an
additional constraining model to allow for the retrieval of vertical motions. In our case we use
the anelastic mass continuity equation as well as a smoothing model. The smoothing model
ensures that noise in the radial measurement does not overwhelm the calculation used to apply
the continuity constraint. The quality of the retrieval is limited by random and systematic errors
in the radial velocity data as well adequate capturing of the storm dynamics (i.e. ensuring that
storm top and base are captured in the scan by both radars to facilitate the vertical integration of
the constraining equation).
As an illustration of the dual-Doppler potential of the CSRP data set, Figure 1.11 shows a
retrieval for a shallow (warm rain processes) precipitating system to the south of the CP2 radar.
The left-hand panel shows an overhead view of a horizontal cross-section through the system at
an altitude of 1500 m. The color contours show the radar reflectivity. The vectors show the zonal
and meridional components of the flow while the overlaid contours show updraft intensity. The
right-hand panel shows a vertical cross-section through one of the precipitating cells at a constant
latitude (depicted by the dashed blue line in the left panel), and the vectors now depict the zonal
30
and vertical wind components. The figure shows a line of precipitation with pockets of enhanced
reflectivity coincident with small 1 m s-1 updrafts.
Figure 1.11: The left-hand panel shows an overhead slice at 1500m through a shallow
(warm rain) precipitating system at 03:00 UTC on 2 February 2008. The filled contours
represent reflectivity (proportional to precipitation content) while the arrows depict the
zonal and meridional wind components and the overlaid contours show updraft strength.
The right-hand panel is the same except a constant latitude vertical slice and the arrows
depict the zonal and vertical components.
As a contrast, Figure 1.12 shows a developing deep convective case from later on the
same day as Figure 1.11 (2 February 2008). In this case the storm has grown deeper to above the
freezing level (at approximately 3000 m) and is exhibiting higher reflectivities and updrafts
> 5 m s-1. The storm has a similar layout to that depicted in Figure 1.11 with a line of
precipitation along a convergent zone with areas of enhanced reflectivity returns located next to
an updraft. Further examples will be presented in Section 3.7 with an analysis of a high impact
thunderstorm.
31
Figure 1.12. The same as Figure 1.11 except for a deep convective system later (07:12 UTC)
the same day (2 February 2008).
There are several proposed applications of dual-Doppler analyses to the CSRP data set:




Use of retrieved winds to constrain a parcel model to study aerosol uptake in precipitating
systems (both background and flare produced)
To study the dynamical evolution (e.g., updraft intensity with time) of seeded and
unseeded clouds
To investigate the link between updraft intensity and microphysical behavior in clouds
(for example how strong an updraft do you need for hail and graupel formation)
To investigate high impact weather events that occurred during CSRP (for example the
destructive wind event at The Gap on 16 November 2008).
1.2.3.2.3 Rainfall products and particle identification (PID)
Rain rate and hydrometeor particle identification is calculated using the polarimetric
radar data from CP2 in real time using CIDD. Rain rate is calculated in CIDD by a variety of
relationships that are outlined in Section 8.2.1. Figure 1.13 shows examples of the rainfall maps
generated for 16 November 2008 at 06:30 UTC using four of the different calculation methods.
32
a)
b)
c)
d)
Figure 1.13. Rain rate on 16 November 06:30 UTC calculated with (a) hybrid approach,
(b) Kdp and ZDR, (c) ZH and ZDR, and (d) ZDR.
The particle identification (PID) is calculated in CIDD using a fuzzy logic method
(described in Vivekanandan et al. 1999), with the polarimetric radar variables and an input
temperature profile.. The current algorithm in CIDD classifies radar return signals into
17 categories (see Table 1.3). Figure 1.14 illustrates the CIDD PID display for multiple
elevations during The Gap storm on 16 November 2008 (see Section 3.7 for more detailed
analysis). These hydrometeor classifications can be used to determine changes in precipitation
characteristics.
33
Table 1.3. Particle IDs and corresponding color in CIDD.
PID
Clutter
2nd trip echoes
Insects
Supercooled liquid water
Irregular ice crystals
Ice crystals
Wet snow
Dry snow
Graupel/rain
Graupel/small hail
Rain/hail
Hail
Heavy rain
Moderate rain
Light rain
Drizzle
Cloud drops
a)
Color
Magenta
Gray
White
Pink
Lavender
Violet
Blue
Cyan
Dark Green
Green
Light Green
Yellow
Red
Brown
Orange
Tan
Light Gray
b)
Figure 1.14. CIDD PID display for The Gap storm on 16 November 2008; images taken at
06:30 UTC at elevation angles of (a) 1 and (b) 6 degrees. The display makes it easy to
identify the transition from varying degrees of rainfall (orange, red) to increasing
quantities of hail and graupel (yellow, shades of green). See Table 1.3 for color legend.
34
1.2.3.2.4 Quantitative precipitation estimation algorithm development
Polarimetric radar is now widely recognized as a tool for measuring the variability in
rainfall drop-size distribution and improving estimates of the resultant rain rate (e.g. Ryzhkov et
al. 2005, Bringi and Chandrasekar, 2001). However, accuracy varies as a function of rain rate,
with some polarimetric estimates exhibiting high error at low rain rates. Consequently, various
methods have arisen to optimize such estimates, typically using either decision-tree logic to
determine which algorithm to use in specific conditions (e.g. Ryzhkov et al., 2005) or highly
complex algorithms based on drop size distribution characteristics (e.g. Bringi et al. 2004,
Brandes et al. 2004).
An investigation of polarimetric estimation of rainfall in Queensland, Australia, with a
view to improving estimation methods, is presented in Section 8.2.2 (Appendix B). This is
investigated by first using 2DVD disdrometer data and scattering calculations to tune standard
algorithms to the location of interest, before using this data to investigate a number of methods
of combining algorithms in both a traditional decision-tree logic manner, and two methods of
utilizing theoretical error characteristics to combine algorithms using robust weightings. Results
are presented with some comparison to relationships such as the Marshall-Palmer R-Z relation
and the Ryzhkov et al. (2005) Oklahoma study.
1.2.3.3 Data quality
A description of the CP2 calibration analysis is presented in Section 7.2.3 (Appendix A).
1.2.4 Disdrometer measurements
Raindrop measurements were made with disdrometer(s) installed at the Willawong
Animal Shelter (62.4° and 16.4 km from the CP2 radar). The latitude and longitude of the site
were 27º 35.998′ and 153º 00.525′, respectively. In season one, a two-dimensional video
disdrometer, owned and operated by NCAR, was deployed to the Willawong site. Three
disdrometers were available for season two (a 2DVD and an impact disdrometer owned by the
Australian Bureau of Meteorology (BoM) and a Particle Video Imager developed by the U.S.
National Space and Aeronautics Administration (NASA).
The 2DVD disdrometers were manufactured by Joanneum Research at the Institute of
Applied Systems Technology in Graz, Austria. The instrument consists of two horizontally
pointing line-scan cameras whose beams are separated in the vertical by approximately seven
mm. (A detailed technical description is given by Kruger and Krajewski (2002).) Measurements
are made for a common 10 × 10 cm area. Recorded information for each raindrop includes
orthogonal silhouette images, equivalent volume diameter, oblateness, and terminal velocity.
Horizontal resolution is approximately 0.15 mm. Vertical resolution depends on particle
terminal velocity and is 0.1–0.2 mm for raindrops. The instrument is calibrated by dropping
pellets of known size into the device. Accurate determination of hydrometeor characteristics can
be made for particles with dimensions greater than about 0.5 mm.
The impact disdrometer (RD-80) is manufactured by Distromet LTD of Basel,
Switzerland. The sensor head, which has a circular surface area of 50 cm2, is displaced
downward when struck by a raindrop. The displacement causes a voltage, which induces a
35
restoring voltage, related to the drop diameter, that repositions the sensor head. (For a technical
description, see Joss and Waldvogel (1967) or Sheppard and Joe (1994).) The Joss-Waldvogel
(RD-80) impact disdrometer has long been the standard for raindrop distribution measurements.
Raindrops are sorted into 20 size categories, which for the BoM instruments have mean
diameters 0.36 to 5.36 mm. A shortcoming with the RD-80 is that during instrument restoration
drops are not recorded. To account for this ―dead time‖ loss, an adjustment proposed by
Sheppard and Joe (1994) has been applied. The adjustment, which attempts to correct for
―ringing‖ of the styrofoam cone when struck by a raindrop, increases the number count for a
particular size category according to the number of impacts in adjacent size categories. Counts
for size categories with no detected drops are not adjusted. Instrument calibration is performed
by the manufacturer.
A detailed technical description of the NASA Particle Video Imager (PVI) and an
evaluation for raindrop measurements are not available. (Previous applications have been for
snowflake imaging in snow storms.) Instrument characteristics are summarized by Newman et al.
(2009). The instrument has a single light source and a viewing area of 3.2 × 2.4 cm. Calibration
consists of inserting a graduated ruler in the field of view.
A comparison of the disdrometer measurements is provided in Section 7.2.2
(Appendix A). The disdrometer observations are being used to fine-tune the CP2 hardware
calibration, establish drop size distribution (DSD) characteristics of stratiform and convective
rains and radar-derived microphysical relationships (see Section 8.2, Appendix B), and to
develop procedures for monitoring drop size distributions in seeded and unseeded clouds with
polarimetric radar (see Section 5.3.1).
a)
b)
Figure 1.15. Photos of the a) NCAR 2DVD in season one, and b) one-dimensional NASA
disdrometer (camera and light source) with the Joss-Waldvogel impact disdrometer (on the
ground in the line of sight of the NASA disdrometer) in season two (BoM 2DVD is not
shown).
36
2 Summary of Operations
The first season of the Queensland CSRP took place between November 2007 and March
2008 (with a holiday break from 20 December 2007–14 January 2008), and the second season
took place between October 2008 and February 2009 (with a holiday break from 17 December
2008–9 January 2009) in the Brisbane region. Aircraft-based research operations began in
earnest on 12 January 2008 in season one, and 4 November 2008 in season two (see Table 2.1
and Table 2.2 for a summary of the project operations). Aircraft operations were based out of
Archerfield Airport, where daily weather and flight planning briefings were held for the pilots.
During flights, operations were coordinated via radio communications between the pilots and the
Operations Director at the CP2 radar facility and operations center (see map in Figure 2.1).
Table 2.1. Summary of all season one CSRP operations. Date is in DD/MM/YY format.
Seeding types were hygroscopic or silver iodide (glaciogenic), if seeding was done.
Operations
Date
Seeding type
Number
Number
Dual-
of Flights
of Flights
Doppler
WxMOD
SEEDA 1
Operations
Number
Dual-
of Flights
Doppler
WxMOD
SEEDA 1
Targets?
Convection Type
Date
Targets?
Convection Type
12/01/08
none
0
1
no
shallow/warm
17/02/08
none
1
0
yes
shallow/warm
14/01/08
none
0
1
no
shallow/warm
20/02/08
hygroscopic
2
0
yes
shallow/warm
15/01/08
none
0
1
yes
shallow/warm
24/02/08
none
1
0
no
shallow/warm
16/01/08
none
0
1
no
shallow/warm
25/02/08
hygro/silv.iod.
2
0
yes
deep/mixed phase
17/01/08
none
0
1
no
shallow/warm
28/02/08
none
0
1
no
shallow/warm
18/01/08
none
0
1
no
29/02/08
hygroscopic
1
1
yes
shallow/warm
21/01/08
none
2
2
no
01/03/08
none
0
1
yes
22/01/08
none
2
0
no
03/03/08
hygroscopic
2
1
yes
23/01/08
hygroscopic
1
1
no
shallow/warm
04/03/08
none
1
0
no
24/01/08
hygroscopic
1
0
no
shallow/warm
05/03/08
none
0
2
no
shallow/warm
26/01/08
hygroscopic
1
1
no
shallow/warm
06/03/08
hygroscopic
1
0
yes
shallow/warm
27/01/08
hygroscopic
1
2
no
deep/mixed phase
07/03/08
none
0
1
no
29/01/08
hygroscopic
1
1
yes
deep/mixed phase
09/03/08
hygroscopic
1
2
no
31/01/08
hygroscopic
1
2
no
deep/mixed phase
10/03/08
hygroscopic
1
1
yes
shallow/warm
01/02/08
hygroscopic
1
1
yes
deep/mixed phase
11/03/08
none
0
2
no
shallow/warm
02/02/08
hygroscopic
2
1
yes
both
13/03/08
hygroscopic
0
1
no
06/02/08
none
2
3
yes
both
15/03/08
hygroscopic
2
1
yes
shallow/warm
07/02/08
hygroscopic
1
1
no
both
16/03/08
none
0
1
no
shallow/warm
08/02/08
none
1
1
yes
shallow/warm
18/03/08
none
0
1
yes
shallow/warm
09/02/08
hygroscopic
1
1
no
shallow/warm
19/03/08
hygroscopic
1
1
yes
shallow/warm
11/02/08
silver iodide
4
1
yes
deep/mixed phase
20/03/08
hygroscopic
1
1
yes
shallow/warm
13/02/08
hygroscopic
1
0
yes
shallow/warm
22/03/08
hygroscopic
1
0
yes
14/02/08
none
0
1
no
shallow/warm
24/03/08
hygroscopic
1
1
no
16/02/08
hygroscopic
1
1
yes
shallow/warm
26/03/08
hygroscopic
1
1
yes
deep/mixed phase
27/03/08
hygroscopic
2
0
no
deep/mixed phase
28/03/08
none
1
1
no
deep/mixed phase
none
deep/mixed phase
none
37
Seeding type
Number
of Flights
both
shallow/warm
none
none
both
none
both
both
Table 2.2. Summary of all season two CSRP operations. Date is in DD/MM/YY format.
Seeding type was only hygroscopic, if seeding was done.
DualOperations
Number of
Doppler
DualOperations
Number
Doppler
of Flights
0
2
1
0
1
1
Targets?
no
no
no
no
no
no
Convection Type
shallow/warm
both
shallow/warm
none
deep/mixed phase
deep/mixed phase
Date
04/11/2008
05/11/2008
06/11/2008
07/11/2008
08/11/2008
11/11/2008
Seeding type
none
none
none
none
hygroscopic
none
Flights
1
0
1
1
1
2
Targets?
no
no
no
no
yes
no
Convection Type
deep/mixed phase
none
shallow/warm
deep/mixed phase
deep/mixed phase
shallow/warm
Date
09/01/2009
12/01/2009
13/01/2009
14/01/2009
15/01/2009
17/01/2009
Seeding type
none
none
none
none
none
hygroscopic
12/11/2008
13/11/2008
16/11/2008
17/11/2008
18/11/2008
19/11/2008
20/11/2008
21/11/2008
22/11/2008
23/11/2008
25/11/2008
26/11/2008
27/11/2008
28/11/2008
29/11/2008
30/11/2008
03/12/2008
04/12/2008
05/12/2008
06/12/2008
none
hygroscopic
none
none
none
none
none
none
none
none
none
none
none
none
none
none
none
hygroscopic
none
hygroscopic
0
2
2
0
0
1
1
0
1
1
1
1
1
0
1
0
1
1
1
1
no
yes
yes
no
no
no
no
yes
no
no
yes
no
no
no
no
no
no
no
yes
no
shallow/warm
shallow/warm
deep/mixed phase
none
deep/mixed phase
both
deep/mixed phase
none
both
shallow/warm
deep/mixed phase
both
both
none
shallow/warm
none
both
both
shallow/warm
deep/mixed phase
19/01/2009
20/01/2009
22/01/2009
23/01/2009
24/01/2009
25/01/2009
26/01/2009
27/01/2009
29/01/2009
30/01/2009
02/02/2009
03/02/2009
04/02/2009
05/02/2009
06/02/2009
08/02/2009
09/02/2009
10/02/2009
11/02/2009
12/02/2009
none
none
hygroscopic
hygroscopic
hygroscopic
hygroscopic
none
hygroscopic
none
none
none
hygroscopic
none
none
none
none
none
none
none
none
0
1
2
2
1
2
1
2
0
0
0
1
0
0
0
0
0
0
0
0
no
no
no
no
no
yes
yes
yes
no
no
no
no
no
no
no
no
no
no
no
no
none
none
deep/mixed phase
deep/mixed phase
deep/mixed phase
deep/mixed phase
deep/mixed phase
deep/mixed phase
shallow/warm
shallow/warm
shallow/warm
shallow/warm
both
none
none
none
none
none
none
none
07/12/2008
09/12/2008
11/12/2008
12/12/2008
13/12/2008
15/12/2008
16/12/2008
hygroscopic
hygroscopic
none
none
none
none
none
2
1
2
0
0
0
0
no
yes
no
no
no
no
no
deep/mixed phase
shallow/warm
shallow/warm
none
none
none
none
13/02/2009
14/02/2009
15/02/2009
16/02/2009
17/02/2009
18/02/2009
20/02/2009
21/02/2009
22/02/2009
24/02/2009
none
hygroscopic
hygroscopic
hygroscopic
hygroscopic
hygroscopic
hygroscopic
hygroscopic
hygroscopic
none
1
1
1
1
2
2
1
2
1
1
no
no
yes
yes
no
no
yes
no
no
no
deep/mixed phase
deep/mixed phase
deep/mixed phase
deep/mixed phase
deep/mixed phase
deep/mixed phase
deep/mixed phase
both
both
none
2.1 Flight Operations
Between both the 2007–2008 and the 2008–2009 seasons, there are a total of 108 flight
operation days with 164 total flights. Of the total flights, there were 142 research flights (where
research flights exclude test, calibration, or recurrency flights). In season one, 49 research flights
were flown by the SEEDA1 aircraft and 39 by WXMOD, while in season two, all operations
were conducted by the SEEDA1 aircraft and they flew 54 research flights (see Table 2.3). These
flights comprised 386 total flight hours between both seasons and both aircraft. In each season,
SEEDA1 flew 150 hours (for a total of 300 hours), and in season one WXMOD flew 86 hours.
The first season flight operations ran between December 2007 and March 2008, and in season
two between November 2008 and February 2009. A map of all flight tracks for each aircraft in
season one is shown in Figure 2.2 and Figure 2.3, and for the SEEDA1 aircraft in season two in
Figure 2.4.
38
Figure 2.1. Map of Southeast Queensland water catchments (red) and location of primary
radars used during the CSRP (green circles). Radars from west-to-east are Marburg, CP2,
and Mt Stapylton.
Table 2.3. Number of research flights per month for both aircraft and both seasons.
Season One
SEEDA1
WXMOD
Dec-07
2
0
Jan-08
14
6
Feb-08
13
16
Season Two
Mar-08
20
17
Sub-total
49
39
39
Nov-08
17
n/a
Dec-08
9
n/a
Jan-09
14
n/a
Feb-09
14
n/a
Sub-total
54
n/a
TOTAL
103
39
Figure 2.2. Map of all flight tracks for WXMOD (in season one).
Figure 2.3. Map of all SEEDA1 flight tracks (in season one).
40
Figure 2.4. Map of all SEEDA1 flight tracks (in season two).
In order to see the distribution of microphysical measurements obtained by the SEEDA1
research aircraft, the SEEDA1 flights have been broken into segments of constant altitude
relative to cloud base, and the frequency distribution of these segments can be seen in Figure 2.5.
Out of the total flight segments in each season, there were relatively more cloud base aerosol
measurements in season two, while season one operations were more dominated by warm cloud
penetrations above the cloud base level and warmer than the freezing level. Season two had
relatively more penetrations in the freezing and mixed-phase levels, rather than focusing on the
warm cloud region. This was partially due to the type of convection that occurred predominantly
in each season (more deep convection in season two). Furthermore, since SEEDA1 also
performed cloud seeding in season two, it resulted in more flight time spent at cloud base
seeding, whereas in season one SEEDA1 spent more time in cloud above WXMOD who was
seeding at cloud base.
41
Figure 2.5. Frequency of SEEDA1 flight heights relative to cloud base (CB) for both
seasons. The percentage is the fraction of all flights for each season that fell into the given
height range: Subcloud = any height below cloud base, CB Aerosol = at but just below
cloud base (out of cloud) for aerosol and CCN measurements, CB + 1kft Pen. = cloud base
penetrations made around 1000 ft above cloud base, CB+1kft to 2deg = warm cloud
penetrations above the initial cloud base penetration yet warmer than the freezing level,
0 +/− 2deg = penetrations taken within 2 degrees Celsius of the freezing level, −2 to −8 deg
= penetrations taken between −2 and −8 degrees C, Colder than −8 deg = penetrations
taken at temperatures less than −8 degrees C.
2.1.1 Dual-aircraft operations
In season one, there were two aircraft in the field operations: one for seeding (WXMOD),
and one for research measurements (as well as seeding if conditions warranted it; SEEDA1).
Thus, we attempted to collect in situ microphysical measurements in clouds by SEEDA1 that
were being seeded at the base by WXMOD to document the changes in microphysical droplet
spectra that occur as a result of hygroscopic seeding. However, there are several caveats to this
procedure, such as flight restrictions on aircraft separation (SEEDA1 always had to fly a
minimum of 2000 ft above WXMOD) and it is not trivial to determine if the research aircraft
flew through a plume of seeding material. Nonetheless, several of these flights were conducted
(summarized here in Table 2.4) and the preliminary results from those flights are presented in
Section 5.3.2.
42
Table 2.4. Summary of all dual-aircraft operations where WXMOD seeded a cloud and
SEEDA1 took cloud penetration measurements of the seeded cloud.
Date
Time SEEDING
SEEDA 1
Time Penetrations
Occurred (UTC)
Penetrations
Occurred (UTC)
1/26/08
5:00 - 5:06
5
4:59 - 5:12
1/26/08
5:11 - 5:36
9
5:18 - 5:42
1/27/08
5:24 - 5:45
6
5:30 - 6:06
1/27/08
6:36 - 6:45
6
6:27 - 6:45
1/29/08
4:54 - 5:06
2
5:00 - 5:06
1/29/08
5:06 - 5:48
5
5:06 - 5:54
1/29/08
5:54 - 6:12
4
6:00 - 6:12
1/29/08
6:20 - 6:30
5
6:18 - 6:30
2/1/08
5:24 - 5:54
14
5:12 - 6:18
2/2/08
5:48 - 6:60
8
5:30 - 6:12
2/2/08
6:54 - 7:06
10
6:56 - 7:18
2/9/08
3:30 - 3:48
13
3:30 - 4:06
2/16/08
3:41 - 4:18
16
3:30 - 4:23
3/6/08
3:30 - 3:42
17
3:30 - 4:00
3/9/08
4:00 - 4:18
35
3:42 - 5:12
3/10/08
2:12 - 2:42
7
2:12 - 2:42
3/10/08
2:54 - 3:18
10
3:00 - 3:30
3/15/08
1:36 - 2:06
10
1:48 - 2:24
3/19/08
3:00 - 3:24
3
3:06 - 3:54
3/19/08
4:18 - 4:42
6
4:18 - 4:48
3/15/08
1:36 - 2:06
10
1:48 - 2:24
3/20/08
2:24 - 3:06
12
2:12 - 3:12
3/24/08
5:42 - 5:54
6
5:36 - 6:00
3/24/08
6:12 - 6:24
8
6:18 - 6:30
3/24/08
6:30 - 6:42
6
6:36 - 6:48
2.1.2 Randomized seeding operations
62 randomized cases (#1–62) were declared in season one and 65 randomized cases were
declared in season two (cases #63–127) and are detailed in Table 2.5. All season one (two)
randomized cases were declared and seeded by the WXMOD (SEEDA1) aircraft. The
randomized cases were optimally declared within 100 km of the CP2 radar when the pilots
detected a rain-free, uniform, and dark cloud base with updraft of roughly at least 200 ft min-1.
A map of the location for all randomized cases declared between the two seasons is shown in
Figure 2.6.
43
Table 2.5. Summary of all declared randomized cases in both seasons of the CSRP. Cases
1–62 were declared in season one, and 63–127 in season two.
Seed
Seed
Seed
Case No.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Case Date
20080202
20080202
20080202
20080202
20080209
20080209
20080213
20080213
20080213
20080213
20080213
20080213
20080220
20080220
20080220
20080220
20080220
20080303
20080303
20080303
(y/n)
y
n
y
n
y
n
y
y
n
y
n
y
n
y
y
y
n
y
y
n
Case No.
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
Case Date
20080324
20080324
20080324
20080324
20080324
20080324
20080324
20080324
20080324
20080326
20080326
20080327
20080327
20080327
20080327
20080328
20080328
20080328
20080328
20080328
(y/n)
y
n
n
y
y
n
n
y
n
y
y
n
n
n
y
y
n
n
n
y
Case No.
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
Case Date
20090123
20090123
20090123
20090123
20090124
20090124
20090125
20090125
20090125
20090125
20090126
20090127
20090127
20090127
20090127
20090127
20090127
20090203
20090203
20090203
(y/n)
n
y
n
n
y
y
n
y
n
y
n
y
y
n
n
y
y
n
y
y
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
20080303
20080306
20080306
20080306
20080309
20080309
20080309
20080310
20080310
20080315
20080315
20080315
20080315
20080315
20080322
20080322
20080322
20080322
20080322
20080324
20080324
20080324
n
y
n
y
n
n
y
y
n
n
y
y
n
n
y
y
y
n
y
n
n
n
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
20081108
20081113
20081113
20081125
20081204
20081204
20081206
20081206
20081206
20081207
20081207
20081209
20081209
20081209
20081211
20090117
20090122
20090122
20090122
20090122
20090122
20090123
y
n
y
n
n
y
n
y
n
y
y
n
n
y
n
y
y
y
n
y
y
y
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
20090215
20090215
20090215
20090215
20090216
20090216
20090217
20090217
20090217
20090218
20090218
20090218
20090220
20090220
20090220
20090221
20090221
20090221
20090221
20090222
20090222
20090222
20090222
n
n
y
y
n
n
y
y
n
y
n
n
y
n
n
y
y
n
y
n
y
n
y
44
Figure 2.6. Map of the locations of all randomized seeding cases declared in season one (red)
and season two (blue). Mt Stapylton (pink) and CP2 (cyan) radar locations are overlaid
along with the Archerfield airport (green; see legend).
2.2 Radar Operations
The CP2 radar scanning strategies were designed to meet three main objectives: (1)
obtain adequate statistics of rainfall in southeast Queensland storms, (2) obtain sufficient
microphysical characteristics of the storms in support of in situ observations, and (3) gather
sufficient observations of the time for initiation of precipitation in both seeded and unseeded
storms and document the microphysical evolution of the droplet size distribution in these storms.
Since objective (1) is complemented by the network of five radars (Mt Stapylton, Mt Kanigan,
Moree, Marburg, and Grafton – see Table 1.2) and the general mode of operation of CP2 is
volumetric scanning, objective (1) should be adequately addressed. Objective (2) requires
detailed remote observations of storms and lends itself to narrow sectors, with as small a vertical
resolution as possible, however, objective (3) requires that sectors be as large as possible to
maximize the probability of sampling storms from their initiation through the onset of
precipitation. Both objectives (2) and (3) are complementary to explain any effects which
seeding may have on the microphysical evolution of the drop size distribution (DSD), however,
they are opposite with respect to the radar scanning strategy which must be employed. The
following scanning strategies represent a compromise to maximize objectives (2) and (3).
45
When unattended, CP2 operated in volume-scan mode synchronized with Mt Stapylton,
producing a volume of Plan Position Indicator (PPI) sweeps once every six minutes. A full
volume of PPI sweeps (each sweep covering the full 360-degree circle) would include
10 elevation angle tilts (0.5, 1.0, 1.7, 2.4, 3.2, 4.7, 6.5, 9.1, 12.8 and 17.8 degrees). During
operations, CP2 was operated in a PPI sector-scan mode, in which a portion of the 360-degree
circle is scanned, or Range-Height Indicator (RHI) mode, in which the antenna scans vertically
through a storm at a constant azimuth angle, and run in a sequence that covers a sector of
azimuth angles. The PPI sectors and RHI scan sequences first included a full 360-degree base
scan at an elevation angle of 1.0 degrees, and then each sector was designed to take only 3
minutes, so that two PPI sectors, two RHI sequences, or one PPI sector and one RHI sequence
could be completed in time for CP2 and Mt Stapylton to sync every six minutes. The PPI sector
scanning strategies allowed for smaller sectors with more vertical elevation angle tilts (for higher
vertical resolution or the ability to top deep or close storms) or larger sectors with less elevation
angle tilts. When seeding operations were underway, the CP2 radar scans followed the targeted
cell for at least 20 minutes after seeding ended, or if the cell was within the dual-Doppler lobes
for an additional hour after seeding ended (see Table 2.1 and Table 2.2 for a list of days with
dual-Doppler targets).
Other scanning strategies employed by the CP2 radar included vertically pointing scan
sequences for calibrating the radar data, and disdrometer scan sequences. The disdrometer scans
were a set of low-level (0.5 and 1.0 degree elevation) small sector scans for comparing radarderived rainfall products with ground-based disdrometer measurements. These scanning
strategies were employed when no other flight missions were taking place, and when there was
rain over the CP2 radar (disdrometer site) for the vertically pointing scans (disdrometer
sequence).
2.3 Disdrometer Operations
The disdrometers described in Section 1.2.4 were operated continuously throughout their
respective field seasons, and the data collected by each is summarized in Table 2.6 for season
one, and Table 2.7 for season two. The 2DVD used in season two was not installed until 22
November 2009, and thus was unable to collect data prior to that. The RD-80 and 2DVD used in
season two also had some outage periods, as seen in Table 2.7 by the periods of no data for one
of the two instruments.
46
Table 2.6. Summary of the 2D video disdrometer data collected during the QCSRP season
one. Time (HHMM-HHMM, UTC) is the starting and ending time of the data (events are
not continuous between these start and end times and not all objects may be raindrops),
Max dBZ is the maximum radar reflectivity determined from 1-min raindrop size
distributions. Max RR (mm/h) is the maximum rainfall rate determined from 1-min
raindrop size distributions. Max D (mm) is the maximum raindrop recorded. Total count
is the total number of objects recorded between the starting and ending times (some objects
may not be raindrops).
Date
Time
Max. dBZ
Max. RR
Max D
(UTC)
(dBZ)
(mm/h)
(mm)
20080113
2130-2359
35.18
24.19
2.17
20080114
0000-2359
40.86
30.89
3.87
20080115
0000-0510
41.05
34.57
3.57
20080116
0000-1610
28.32
3.85
20080117
0200-1530
36.43
20080118
0000-2359
26.89
(yyyymmdd)
Total count
Time
Max. dBZ
Max. RR
Max D
(yyyymmdd)
Date
(UTC)
(dBZ)
(mm/h)
(mm)
Total count
27285
20080218
0000-2359
46.65
49.01
4.59
160579
257651
20080219
0000-1200
24.54
3.83
1.82
21949
125396
20080220
0400-0500
15.72
0.36
1.09
555
1.73
18018
20080224
1600-2359
29.61
1.02
4.57
3575
20.66
2.7
90911
20080225
0000-2359
34.19
3.31
3.67
66560
2.9
1.76
8494
20080226
0000-1730
47.22
50.39
4.67
219900
20080119
0530-2359
25.44
2.41
2.47
6666
20080227
1630-2230
22.72
0.46
2.39
1257
20080120
0500-1945
29.23
1.85
3.18
24079
20080228
0500-2359
29.67
1.4
3.09
10209
20080121
0100-1810
28.66
3.11
2.92
19049
20080229
0300-0700
4.38
0.03
1.29
322
20080128
1450-2359
36.94
22.34
2.64
22999
20080303
0400-0700
8.91
0.08
1.48
183
20080129
0000-2359
46.59
41.75
5.32
262062
20080305
0200-0900
25.72
1.4
2.46
3278
20080130
0000-1800
49.37
95.65
4.79
107597
20080306
0000-0100
8.75
0.11
0.86
142
20080201
2100-2200
2.88
0.03
2.33
94
20080308
1000-2100
32.95
9.78
2.84
16882
20080203
0000-2359
38.83
22.24
3.35
248868
20080309
0000-2359
42.56
16.32
4.71
29171
20080204
0100-2359
45.4
45.37
4.6
215258
20080310
0000-0800
34.26
11.96
2.36
6539
20080206
0000-1130
54.64
99.13
6.62
75241
20080314
0730-2130
12.12
0.07
1.41
120
20080208
0800-1600
56.29
116.82
7.14
28357
20080315
0000-2300
29.29
3.18
2.3
2890
20080210
2100-2359
26.37
3.76
3.1
3411
20080319
0000-2359
55.37
136.77
6.46
86396
20080211
0000-2359
38.84
26.94
3.16
255184
20080320
0000-0700
41.27
9.09
3.85
5976
20080212
0000-1530
44.95
59.48
3.66
456164
20080321
1300-2359
35.76
8.42
4.14
21068
20080213
0400-1600
39.36
18.18
3.7
24567
20080322
0100-1600
38.69
6.68
3.21
9677
20080214
0000-0700
23.97
3.55
1.38
14176
20080324
0100-1230
54.27
104.7
6.08
34869
20080215
0230-2359
37.34
27
2.45
61041
20080326
0900-2359
40.81
10.17
3.82
76497
20080216
0030-2359
33.04
7.35
2.52
137508
20080327
0000-0200
35.56
15.17
3.28
39668
20080217
0000-2359
36.88
24.73
2.8
100377
47
Table 2.7. Same as Table 2.6, except for part one of season two (operations until the
December holiday break), and for the BoM 2D video disdrometer and Joss-Waldvogel
impact disdrometer.
Joss-Waldvogel Impact Disdrometer (RD-80)
Date
(yyyymmdd)
Time
Max. dBZ
Max. RR
Max D
(UTC)
(dBZ)
(mm/h)
(mm)
2-D Video Disdrometer
Total count
20081022
0501-0724
-5.27
0.01
0.66
30
20081023
0122-2244
-9.19
0.01
1.12
40
20081024
0053-2210
-2.92
0.01
0.77
36
20081025
0335-1823
20.28
0.26
2.26
890
20081026
0024-2354
-9.29
0.01
0.77
31
20081027
0027-0028
55.15
31.15
5.37
58
20081029
0102-2225
35.93
8.88
2.87
3728
20081030
0127-2350
-5.98
0.01
0.91
48
20081031
0237-0401
55.50
34.45
5.37
473
20081101
0947-2210
9.34
0.10
1.51
2434
20081102
1516-2213
19.26
3.31
0.91
200346
20081103
0129-1852
34.79
3.88
3.54
5860
20081106
0554-2157
-5.48
0.01
0.91
87
20081107
0109-2204
20.37
0.42
2.26
3374
20081108
0245-1021
53.16
64.57
5.37
40536
20081109
0020-2323
8.83
0.06
1.12
168
20081110
0855-2314
9.50
0.24
0.91
2945
2478
20081111
0031-1101
20.47
1.51
1.12
20081120
0302-1214
-5.49
0.01
4.35
181
20081125
1316-1317
46.22
16.36
5.37
4326
Time
Max. dBZ
Max. RR
Max D
(UTC)
(dBZ)
(mm/h)
(mm)
Total count
0800-2315
45.97
19.11
4.93
134174
20081126
0115-2359
51.22
74.54
4.95
283061
20081127
0000-1445
30.35
2.65
2.28
17401
20081129
0452-2343
51.15
81.63
5.55
63405
20081203
0840-1914
44.58
16.79
5.03
11144
20081205
0100-2000
38.23
10.82
4.34
87317
20081206
0314-1935
54.40
96.66
6.41
310581
20081207
0100-2400
48.60
10.21
7.34
4607
20081208
0000-1400
30.34
0.84
3.45
7098
20081209
0130-0300
52.32
114.03
5.55
148162
48
Table 2.8. Same as Table 2.7, except for part two of season two (January 8, 2009 through
the end of season two).
Joss-Waldvogel Impact Disdrometer (RD-80)
Date
(yyyymmdd)
Time
Max. dBZ
Max. RR
Max D
(UTC)
(dBZ)
(mm/h)
(mm)
2-D Video Disdrometer
Total count
20090108
0722-2316
21.17
1.46
1.33
10852
20090109
0105-2305
-3.12
0.02
2.87
214
20090110
0331-1912
17.26
0.81
1.12
845
20090111
0047-2109
1.67
0.04
0.91
539
20090112
0022-2359
3.03
0.04
0.91
955
20090113
0000-2320
0.85
0.04
0.91
401
20090114
0101-2307
-8.41
0.01
1.33
63
20090115
0351-1433
-10.21
0.00
0.91
37
20090116
0056-2242
-0.67
0.03
1.12
1626
20090117
0033-2314
-1.43
0.02
2.58
254
20090118
0110-2344
-7.72
0.01
0.66
69
20090119
0544-2353
6.17
0.07
0.91
94
20090120
0137-0202
-0.49
0.03
0.77
121
20090122
1358-2359
46.68
50.78
3.92
196781
20090123
0000-1953
15.11
0.13
1.91
1329
20090124
2037-2359
34.06
7.91
2.26
6880
20090125
0000-2232
28.81
2.35
2.58
20090126
0022-0741
35.09
3.38
3.92
20090127
0107-2329
34.66
6.58
20090128
0049-2338
38.91
20090129
0001-1938
20090130
20090131
Time
Max. dBZ
Max. RR
Max D
(UTC)
(dBZ)
(mm/h)
(mm)
Total count
1400-2000
21.88
1.36
1.29
6616
1100-1130
19.23
0.90
1.17
1056
2145-2200
5.28
0.05
0.88
138
0030-0050
19.23
0.16
2.26
116
1900-2200
28.40
1.35
2.35
2234
0635-2400
45.96
42.73
4.52
346589
0000-0020
10.30
0.07
1.22
9747
0000-2120
40.37
17.11
3.67
133815
8201
0022-2350
47.11
42.97
4.15
67758
2.26
9103
0000-2345
35.13
6.64
3.06
17269
22.90
2.58
25443
0104-2400
42.23
36.30
4.52
49145
50.47
64.00
5.37
105033
0023-2326
49.92
59.01
5.46
129002
0107-2241
46.53
23.79
4.35
12801
0216-1242
46.65
29.46
4.52
19521
0011-2347
42.88
21.12
3.92
11123
0052-2305
42.49
20.31
4.84
12962
20090201
0001-2344
47.09
46.07
4.35
38426
1102-2345
47.61
50.88
4.35
56044
20090202
0042-2356
31.73
6.19
3.20
37176
0045-2333
29.40
5.04
2.17
31137
20090203
0016-1456
55.15
31.15
5.37
27434
0035-2204
37.75
11.13
3.47
34744
20090205
0017-2338
5.80
0.05
1.12
450
20090206
0038-2230
-2.04
0.02
1.51
133
20090207
0245-2358
-4.32
0.01
0.91
116
20090208
0049-2238
3.54
0.06
0.91
209
20090209
0631-1949
2.77
0.03
1.12
95
20090210
0120-2314
21.54
0.40
2.26
5019
0811-1346
21.88
0.42
2.46
7118
20090211
0519-1229
48.70
32.26
4.86
18630
1158-1257
20090212
1049-2359
46.20
58.39
3.92
542210
0456-2359
46.17
58.99
3.64
833935
20090213
0000-0034
36.00
9.59
2.87
12450
0000-2359
42.66
29.72
4.02
697867
20090214
0100-1924
29.52
3.61
3.20
27143
0000-0929
28.80
3.01
3.33
72437
20090215
0030-0542
-6.87
0.01
1.12
136
20090216
1016-1601
3.73
0.10
0.77
0744-0817
51.62
114.47
5.59
66732
20090217
0240-2247
42.55
13.82
4.86
103844
0201-0429
7.63
0.05
1.04
271
0114-2223
52.83
136.64
5.37
103890
0742-1845
51.26
85.76
4.74
95469
20090220
0007-2119
50.44
71.64
5.37
100053
0600-1751
49.96
63.74
5.27
176013
20090221
0003-2345
32.19
2.11
3.54
5434
1230-2110
32.55
1.71
3.59
7625
20090222
0029-2324
43.75
15.06
4.35
5651
0030-1052
42.10
12.78
4.96
9067
20090223
0007-2338
-5.62
0.01
1.67
51
20090224
0037-2251
7.42
0.06
1.33
946
20090225
0050-0109
-10.24
0.01
0.91
106
20090121
334
20090218
20090219
49
193
3 Weather and Climate
3.1 Introduction
In order to put the seeding and precipitation analyses into context, the climatology of
atmospheric conditions, as they relate to precipitation and microphysics, must be well understood
within the region. This chapter describes several climatological aspects of the southeast
Queensland region. Section 3.2 describes a synoptic clustering method used to group the
synoptic weather patterns observed in southeast Queensland into common types and characterize
the associated rainfall that occurs in each synoptic weather regime. Section 3.3 explains how the
two operational seasons (2007–2008, and 2008–2009) fit into the context of the previous 60
years, in terms of precipitation and the El Niño Southern Oscillation cycle. Section 3.4 utilizes
five years of Marburg radar data to study the precipitation patterns over the localized CSRP
domain. Section 3.5 utilizes the Australian ground-based lightning network to illustrate lightning
patterns across the broader Queensland region. Lightning is a direct result of microphysical
properties in convection, and thus can add extra insight into what microphysical conditions are
like across the region. Section 3.6 uses the HYSPLIT trajectory model to establish air mass
sources and common parcel wind trajectory regimes to help characterize the aircraft
measurements. These results will be used to help complete the cloud base aerosol and
microphysical conditions analysis that is presented in Section 4.2.1. The cloud base conditions
climatology will be used to help understand which days and locations in the domain are most
suitable for cloud seeding. And finally, Section 3.7 provides a mesoscale analysis of a severe
weather event observed during season two of the Queensland CSRP (―The Gap‖ storm). This is
just one example of auxiliary analyses that can be conducted using the CSRP data set that is of
interest to weather forecasters and risk managers.
3.1.1 Overview of results
The following conclusions are drawn from the detailed analyses presented in this chapter:
1. The wet season in Southeast Queensland is broadly between October and March, with
November–February receiving the most rainfall and having the most atmospheric
moisture.
2. The synoptic cluster analyses and statistical tests have shown that Southeast Queensland
can be divided into ‗wet‘ and ‗dry‘ weather regimes, with the ‗wet‘ regimes occurring
most in summer and the ‗dry‘ regimes more in winter. This result is in itself not
surprising, as this fact is already well known to the local population and forecasters, but
this analysis has further quantified these regimes nonetheless.
a. The ‗wet‘ regimes are responsible for the majority of the region‘s rainfall and
include the northwesterly and ‗moist‘ southeasterly, easterly, and westerly
regimes (from the seven-cluster analysis), or the northwesterly and (when
unstable) the southeasterly regimes (from the three-cluster analysis). The
50
northwesterly regime contributes greatly to the total annual rainfall despite
occurring less than 10% of the time.
b. The ‗dry‘ regimes are predominantly the other southeasterly clusters (which is
also the most common regime occurring year-round yet contributes very little to
annual rainfall in the seven-cluster case), and the southwesterly regime, which is
most common in winter (in both the seven- and three-cluster cases).
3. The southeasterly regime(s) tends to be less unstable than the northwesterly regime and
can have a trade wind inversion below the freezing level, resulting in shallow convection
with rain formed via warm rain processes and only light rain. Given the tendency for
higher instability and thus deeper convection in the northwesterly regime, this regime
was also overrepresented in the aircraft measurements (as we tended to fly on days with
convection, especially focusing on deep convection).
4. Synoptic scale features (such as east coast lows) are generally the source of rain during
the winter months. Variation throughout the wet season is also attributable to synoptic
scale disturbances, as well as mesoscale features.
5. The Southern Oscillation Index (SOI) appears to be positively correlated with
precipitation rate, and more so with precipitable water (a measure of atmospheric water
vapor), such that positive SOI (La Niña) corresponds to wetter conditions. A tailored
index using sea surface temperature (SST) off the east coast of Australia showed an even
better positive correlation with precipitable water (higher SSTs relates to higher
precipitable water in the Brisbane region). These indices can be used for predicting
potential seasonal precipitation, as well as to assist in understanding historical trends in
precipitation.
6. The radar climatology indicated that cell volume, area and height all followed truncated
lognormal distributions. The same result was found when each of the three-cluster
synoptic regimes was considered individually. Nonetheless, the largest 20% of cells were
found to be responsible for roughly 67% of the fractional areal coverage of precipitation.
7. Student‘s t-tests applied to examine the relationship between the three clusters found that
the precipitation area was significantly larger during the winter regime than during either
of the summer regimes and cell-top height was significantly lower during the winter
months than during the summer. Furthermore, the two summer regimes do not separate
statistically when stratified by precipitation area.
8. Four regimes of surface wind flow conditions were revealed by HYSPLIT back trajectory
modeling. These regimes relate to surface aerosol conditions and were classified into two
maritime regimes (northeasterly and east/southeasterly), and two continental regimes
(northwesterly and west/southwesterly).
9. Based on the CG lightning activity alone, it suggests that the mixed-phase microphysics
of clouds farther west is more vigorous than for those closer to the coast, where the CSRP
domain is presently located. Cloud seeding with hygroscopic or glaciogenic flares may be
more effective in that region.
51
3.2 Atmospheric Environment
The seasonal weather of southeastern Queensland (SEQ) can be described by a wet and a
dry season, with the highest rainfall occuring during the summer half of the year. The main water
catchments are located north of Brisbane on the Brisbane River. Significant rainfall over the
main water catchments of Somerset and Wivenhoe tends to be limited to the summer months
from October through to March, with monthly falls over 100 mm only in the period of November
to February.
The reason for the strong gradient in rainfall away from the coast is explained at least in
part by the topography of the region. Along all of the eastern coast of Australia, the terrain rises
to the ridges of the Great Dividing Range, with a maximum elevation of about 1000 m in the
south of the region. The areas of maximum monthly rainfall correspond to the areas of steepest
topography. The monthly rainfall tends to decrease to the west of the ridges around 152° E.
3.2.1 Cluster analysis of sounding data
In order to understand the rainfall variability of SEQ, it is useful to consider the impact of
the different weather patterns or synoptic regimes on the regional rainfall. Comparison of
atmospheric profiles will reveal relationships between precipitation and thermal structure.
Nuijens et al. (2009) compared composite profiles of equivalent potential temperature, relative
humidity and zonal wind speed. Conner and Bonnell (1998) show that radiosonde data can be
used to classify rainfall regimes along the Queensland coast, and a similar approach is taken here.
Cluster analysis was performed on 5773 out of a possible 6939 days of radiosonde data
from Brisbane Airport obtained from the University of Wyoming spanning the period
01/01/1990–31/12/2008. Days were excluded due to missing data. The 00Z (10 AM local time)
radiosonde flight was used in all cases.
This encompasses several El Niño/La Niña events. The 1990s were predominantly
negative values of the Southern Oscillation Index (SOI), which indicated El Nino periods, with a
particularly strong El Niño in 1997–1998. However, this time period is not long enough to
capture decadal variability.
The K-means clustering algorithm (Hartigan and Wong, 1979) was used on normalized
data. To ensure stable clusters the clustering algorithm used 100 random starts. The clustering
was performed on seven atmospheric variables, which were derived from the radiosonde flight
data. The air mass variables chosen for inclusion were inversion height, total-totals, 850 hPa
winds, wind shear between 850 mb and 500 mb, moisture flux and total water calculated from
the sounding data at Brisbane Airport. The wind vector (U, V) at 850 hPa provides information
on the basic dynamical flow, particularly whether the flow is onshore (easterly) or offshore
(westerly). Positive values of U are winds from the west; positive values of V are winds from the
south. The shear variable indicates whether raining systems are likely to be susceptible to
shearing forces, and the total-totals (T-T) variable is a simple measure of the stability of the
atmosphere. The water flux vector (QU, QV) is expected to be important in determining the
amount of moisture available for rainfall in the region, and the total water (TW) is a measure of
the water in the atmosphere at a given time.
52
Commonly, the decision of how many clusters describes the greatest part of the variance
in the data is made by looking at the within sum of squares error. It is found that the goodness of
fit increases only marginally as the number of clusters increases from three or four. However
there is valuable detail to be found when moving to higher numbers of clusters. The sevencluster case was chosen as it shows clearly the seasonality of regimes and in particular describes
a high rainfall, low frequency regime.
The physical characteristics of the seven-cluster case will be described. The mean sea
level pressure analyses and soundings corresponding to the cluster centroids are presented and
will be discussed. The centroid of each cluster is the mean of all points within that cluster. The
day closest to centroid was found by minimizing the square of the difference between each
variable and the cluster centre. The annual distribution of regimes will be examined using bar
plots of the monthly breakdown of each cluster. In addition, the contribution of each regime to
annual rainfall will be quantified and discussed.
3.2.2 The seven-cluster case
The seven-cluster case describes three separate southeasterly regimes, three westerly
regimes and an easterly regime. Figure 3.1 shows the environment for each regime. The vertical
profiles and synoptic environment for each cluster centre are shown in Figure 3.2 and Figure 3.3.
The seasonal cycle of each regime is shown in Figure 3.4.
The dominant regime is southeasterly (SE; black in figures). This regime is characterized
by a southeasterly moisture flux and moderate values of T-T and shear. This regime accounts for
24% of all days and occurs throughout the year.
A ‗moist‘ southeasterly trade regime (SE ‗moist‘; green in figures) accounts for 16% of
all days. Consideration of the plot of annual variability (Figure 3.4) reveals that this regime
occurs most frequently during the late summer (beginning of the year) although it still accounts
for approximately 10% of days during the winter months. There is significant southeasterly
moisture flux and values of atmospheric moisture and T-T are moderate. A key feature of the
sounding is the trade inversion at about 800 mb and high moisture up to approximately 500 mb.
The main feature of the MSLP for this regime is the position of the Queensland inland trough,
which can be associated with the advection of moisture into the region, leading to significant
rainfalls.
A ‗dry‘ southeasterly regime (SE ‗dry‘; cyan in figures) accounts for 13% of all days and
occurs most frequently during the winter half of the year. This regime has a southwesterly
component to the moisture flux and has the highest value of shear. The sounding shows a drying
out of the atmosphere above 950 mb.
The wintertime is also described by a southwesterly regime (SW; purple in figures). This
regime has a westerly moisture flux, high values of shear and the atmospheric profile is very dry
above 800 mb. This regime accounts for 11% of all days. Typically this half of the year has low
rainfall; however heavy, widespread falls can occur due to intense extratropical lows that form
off the coast (e.g., east coast lows).
53
An easterly regime (E; red in figures) accounts for 13% of all days. There is a small
northerly component to both the wind and predominantly easterly moisture flux, along with high
values of total water. This regime occurs during the summer half of the year. When moisture is
advected into the region this scenario can lead to deep convection. Analysis of storm locations
reveals that storms form on the ranges and advect east (see Section 3.4; Peter et al. 2009).
The yellow bars in the figures represent a northwesterly (NW) regime that accounts for
6% of all days, with a slight preference for the summer half of the year. This regime has very
strong northeasterly components to the moisture flux and inspection of the atmospheric profile
remains reveals a very moist column. The values for T-T are the highest of all regimes.
A westerly regime (W; blue in figures) has a strong westerly component to the moisture
flux. This regime accounts for 16% of all days and occurs throughout the year with a preference
for early summer.
Figure 3.1. Scaled variables for cluster centres. The colours correspond to individual
clusters: SE (black); SW (purple); SE 'moist' (green); NW (yellow); SE 'dry' (cyan); E
(red); W (blue). The percentage of total days in each cluster is shown in the % column.
54
Figure 3.2. Atmospheric profiles corresponding to the centre of each cluster, clockwise
from top left: SE (black); SW (purple); SE 'moist' (green); NW (yellow); SE 'dry' (cyan); E
(red); W (blue). Courtesy of the University of Wyoming.
Figure 3.3. Mean sea level pressure analysis charts corresponding to the centre of each
cluster, clockwise from top left: SE (black); SW (purple); SE 'moist' (green); NW (yellow);
SE 'dry' (cyan); E (red); W (blue). Courtesy of the Bureau of Meteorology.
55
Figure 3.4. Frequency of occurrence clusters for each month. SE (black); SW (purple); SE
'moist' (green); NW (yellow); SE 'dry' (cyan); E (red); W (blue).
3.2.3 Contribution of cluster regimes to annual rainfall
To investigate the contribution of each regime to monthly rainfall we look at the total
rainfall over the period sorted according to cluster. Each day in the period can be described by
one of the regimes described above.
The initial analysis has been carried out on the SEQ region, which is taken to be the area
east of 151° E and lying between 29° S and 26° S. Daily rainfall records for this region were
obtained through the Bureau of Meteorology. The data was extracted for the period
01/01/1995–01/06/2008 (4754 days) for all months of the year, equating to 768 sites in the target
region. These were matched to the days used in the cluster analysis. A total of 4168 days were
used in this section of the analysis.
The mean daily rainfall across all sites in the target area was calculated and then the total
sum was found for each month according to cluster. The results are shown in Figure 3.5. The
results agree with the mean monthly rainfall for the period 1994–2009 from the Bureau of
Meteorology (Figure 3.6).
56
Figure 3.5. Contribution of clusters to annual rainfall. The contribution of each cluster to
the total is represented by the color: SE (black); SW (purple); SE 'moist' (green); NW
(yellow); SE 'dry' (cyan); E (red); W (blue). The total is for all years in the period 19952008 inclusive.
Figure 3.6. Mean monthly rainfall totals calculated for the period 1949-2000. Courtesy of
the Bureau of Meteorology.
The seasonal rainfall cycle is as expected with low monthly totals in the winter months
and high totals in the summer months. May and June should be noted as having above average
rainfall over the 13-year period. Examination of the monthly rainfall totals for each year (not
shown here) reveals that these high totals are the result of isolated east coast low events.
57
Major contributions to monthly rainfall are made by the easterly regime (red in figures)
during the summer half of the year. This regime is responsible for the majority of the rainfall
during the month of May, which is somewhat surprising. Taking a closer look at the monthly
rainfall totals for each year reveals that May 1996 recorded more than 300 mm of rainfall. This
is triple the long-term average of 100 mm, and still more than double the average taken over the
last 15 years (125 mm) (see Figure 3.6). These heavy falls occurred during the first week and
were associated with a long-lived east coast low.
The northwesterly regime (yellow in figures) makes important contributions to rainfall
during the summer half of the year, despite only occurring 6% of the time. This regime is
responsible for the majority of the rainfall during the month of June. Inspection of June rainfall
for each year reveals that in 2005 the monthly total was more than 150 mm, more than double
the long-term average June rainfall of about 75 mm (see Figure 3.6). There was a cold outbreak
towards the end of the month that brought heavy rainfall and there were reports of snow.
The ‗moist‘ southeasterly trade regime (green in figures) contributes a moderate amount
of rainfall during all months of the year. The westerly regime (blue in figures) also makes
sizeable contributions during the summer half of the year, with its influence weakening over the
winter months.
The dominant synoptic regime for the region, the southeasterly regime (black in figures),
does not contribute significantly to the total rainfall in any month. There does not appear to be a
lot of difference between the mean sea level pressure analyses for this regime and the ‗moist‘
southeasterly regime, however when the atmospheric profiles are considered the difference in
moisture is evident. This is a ‗no-rain‘ regime most likely composed of dry days where little or
no precipitation was recorded.
The winter regimes do not make any sizeable contributions to rainfall in any month. The
winter southwesterly regime (purple in figures) does not make any significant contributions to
monthly rainfall. The ‗dry‘ southeasterly regime (cyan in figures) only makes small noteworthy
contributions to total monthly rainfall during June, July, and August.
3.2.4 Summary
It is apparent that the rainfall is largely limited to the coastal strip, with maxima near
regions with steep terrain. The main rainfall regime is from November through February, with
peak falls tending to be in February. As such, the cluster regimes for the Southeast Queensland
region can be divided into ‗wet‘ and ‗dry‘ scenarios.
The ‗wet‘ regimes are responsible for the majority of the region‘s rainfall, and include the
southeasterly ‗moist‘, northwesterly, easterly and westerly clusters. The ‗moist‘ southeasterly
regime and the northwesterly regime can produce deep convection and are major contributors to
the region‘s annual rainfall. The easterly and westerly regimes also are major contributors to
annual rainfall. Southeasterly wind regimes are commonly associated with trade wind or
‗stream‘ showers and coastal trade wind cumulus. When there is significant moisture such for the
‗moist‘ southeasterly regime these systems can bring significant rainfall to the region. The
position of the Queensland inland trough plays an important role during the wet season over
58
southeastern Queensland. Eastward movement of the trough causes a shift in wind direction from
southeasterly to northwesterly. This brings warm moist air into the region leading to a potential
increase in convective activity. The northwesterly regime can produce deep convection and
contributes greatly to the total annual rainfall despite occurring less than 10% of the time.
The dry scenarios typically occur in the most common southeasterly cluster, which
occurs year round, as well as the southeasterly ‗dry‘, and the southwesterly clusters that are more
prominent in the wintertime. The winter regimes are generally ‗dry‘ and do not make notable
contributions to annual rainfall in any month. The southeasterly regime that describes the largest
number of days also does not make a notable contribution to annual rainfall.
However, there are also significant rainfall events during the dry season, such as intense
subtropical cyclones (east coast lows) that bring sustained strong winds and intense rainfall to the
region. In general, however, the winter season is dry. The greater part of the ‗no-rain‘ days are
described by the dominant southeasterly regime.
3.3 Historical Precipitation Patterns and Relationships with Global-Scale
Features
3.3.1 Background
Quayle (1929) first suggested that the risk of drought in eastern Australia is increased
during an El Niño phase, especially during both the onset phases and mature phases of an event.
McBride and Nicholls (1983) provided comprehensive knowledge based on correlation analysis
of relationships between the Southern Oscillation Index (SOI) and eastern Australian rainfall,
while Stone and Auliciems (1992) provided information related to the non-linearity of
rainfall-Southern Oscillation relationships and the value of a ‗phase‘ analysis to delineate
additional information regarding relationships between the El Niño/Southern Oscillation (ENSO)
and rainfall and temperature for this region. Cai et al. (2001) suggested that the SOI and
northeast Australian rainfall were highly correlated throughout the 20th century (r ≈ 0.7) except
during the period 1930–1945.
In Australia, including southern Queensland, climate analyses and forecast outputs have
mostly been based on either large-scale SST patterns (also related to ENSO) or the use of
SOI phases (Stone et al., 1996) based on principal component and cluster analysis to identify lagrelationships between changes in the SOI and patterns of rainfall probabilities. SST relationships
and subsequent development of forecasts using principal component analysis and discriminant
analysis of SST-rainfall relationships have been developed as forecasting systems by the
Australian Bureau of Meteorology (Drosdowsky and Chambers, 2001), although recent analysis
has demonstrated that this type of approach is slightly less useful in eastern Australia and
southern Queensland than use of either the SOI phases or Niño 4 SST anomalies (Murphy and
Ribbe 2004, Fawcett and Stone 2009).
Indeed, Murphy and Ribbe (2004) show that relationships between the Niño 4 region of
the equatorial Pacific and rainfall were strong over much of eastern Australia. They show that
―the highest correlations are more spatially extensive between rainfall and an index based on the
59
Niño 4 region, and therefore rainfall in southern Qld as a whole is more closely related to
SST variations in the central rather than the eastern Pacific‖ (Murphy and Ribbe, 2004). They
also point out that the Niño 4 index and the SOI have very similar rainfall correlations in SEQ in
more recent decades.
3.3.2 Analysis for SEQ
The historic precipitation was characterized by analyzing both NCEP/NCAR reanalysis
data (Kalnay et al. 1996), which can be obtained from the NOAA-CIRES Climate Diagnostics
Center website (http://www.cdc.noaa.gov), and data from the Australian Government Bureau of
Meteorology (http://www.bom.gov.au/cgi-bin/silo/cli_var/area_timeseries.pl). The reanalysis
project utilizes a frozen state-of-the-art data assimilation system and observational database to
produce a long-term, standardized record of atmospheric data on a global grid. For this analysis,
two precipitation output fields were used: precipitable water (a measure of atmospheric water
vapor) and precipitation rate. Each output field has been classified by its relative influence from
observational data and the model. Precipitable water is a measure of the amount of moisture
available in the atmosphere, and is of a variable class that is directly affected by observational
data, although the model also has a very strong influence on the value. On the other hand,
precipitation rate is derived solely from the model fields forced by the data assimilation. Hence,
caution should be used when interpreting results, as they may not be consistent with
observational data from a particular location. Nevertheless, the precipitation fields provide
valuable insight in assessing historic climate, while eliminating any perceived climate changes
due to particular data collection systems (Kalnay et al. 1996). The Bureau of Meteorology (BoM)
offers a one degree gridded dataset, which is based on observed monthly precipitation amounts
and is specific to the Australian land surface.
The Brisbane/Toowoomba region was approximately defined to correspond to the study
area. The area examined was approximately between 26° to 29° S and 151° to 154° E, though
slight discrepancies occurred due to differences in the underlying grids (see Table 3.1). Monthly
data was obtained from January of 1948 through August 2009.
Table 3.1. Latitude and longitude ranges for precipitation variables used in analysis.
Data Source
Variable (units)
Latitude Range
Longitude Range
Reanalysis
Precipitation (mm/day)
26.7–28.6 S
151.0–152.9 E
Reanalysis
Precipitable Water (kg/m2)
26.3–28.9 S
151.3–153.8 E
BoM (Bureau)
Precipitation (mm/month*)
26.5–28.5 S
151.5–153.5 E
* To compare with the reanalysis precipitation, this was converted to mm/day by dividing by number of days per
month.
From the reanalysis and BoM data, precipitation rate and monthly precipitation total
patterns (Figure 3.7) and precipitable water (Figure 3.8) over the Brisbane/Toowoomba region
reveal that there is a wet season spanning from October through March (ONDJFM, Figure 3.7).
The seasonal patterns of precipitation are shown as box plots, in which the box represents the
60
25th and 75th percentile, the whiskers show the 5th and 95th percentiles, points are values
outside this range, and the horizontal line represents the median. Comparing the BoM and
reanalysis precipitation rates for the wet months (ONDJFM) of the year (Figure 3.9, left) shows
that there is a strong correspondence between the datasets (r=0.63), though there is considerable
scatter around the one-to-one line. Further, as expected, there is a strong correlation between the
reanalysis precipitation rate and precipitable water (Figure 3.9, right).
Figure 3.7. Precipitation rate (top) and monthly precipitation totals (bottom) for the
reanalysis (left) and BoM (right) data.
61
Figure 3.8. Monthly precipitable water for the reanalysis data.
Figure 3.9. Precipitation rate for the BoM and reanalysis (left) and reanalysis precipitable
water versus precipitation rate (right) for the wet months (ONDJFM). Black line shows
one-to-one ratio and r value is linear correlation.
To characterize the year-to-year variability, precipitation was examined for total
accumulation in all the wet months (ONDJFM, Figure 3.10) and for the average seasonal
precipitation rate for December, January, and February (DJF, Figure 3.11). From year-to-year,
there is considerable variability in terms of precipitation, as well as differences between the BoM
and reanalysis datasets. Since the season spans from October to March, the years correspond to
March of each season. For all descriptive statistics examined, the BoM‘s dataset yields higher
values than the reanalysis (Table 3.2). This is not surprising, given the fact that the BoM dataset
is strongly influenced by direct observations. On the other hand, the reanalysis data is derived
62
from an underlying model, which may be less able to capture some of the large precipitation
events observed.
Bureau
Reanalysis
1200
Season (ONDJFM) Total Precipitaiton (mm)
1000
800
600
400
200
0
1949
1959
1969
1979
1989
1999
2009
Year
Figure 3.10. Total wet season (ONDJFM) precipitation amounts from 1949–2009.
Bureau
Reanalysis
Season (DJF) Average Precipitation Rate (mm/day)
10
9
8
7
6
5
4
3
2
1
0
1949
1959
1969
1979
1989
1999
Year
Figure 3.11. Average seasonal (DJF) precipitation rate from 1949–2009.
63
2009
Table 3.2. Descriptive statistics for total and average precipitation for the Bureau and
reanalysis.
ONDJFM Monthly Precipitation
Totals (mm)
DJF Average
Rate (mm/day)
Precipitation
BoM
Reanalysis
BoM
Reanalysis
Mean
664
604
4.2
4.0
Std. Dev
168
162
1.4
1.2
Max (Year)
1084 (1956)
966 (1984)
9.2 (1971)
6.7 (1953)
Min (Year)
335 (1969)
153 (1969)
2.0 (2002)
1.5 (1969)
To explore potential sources of the wet season variability, relationships with standardized
climate indices can be examined. The Southern Oscillation Index (SOI) is an index number
computed by comparing air pressure at sea level between Darwin and Tahiti. In general, during
El Niño, the index becomes negative, and is characterized by warming of the central and eastern
Pacific Ocean. During La Niña, the reverse happens and cooling takes place in these same
waters. In the northern and eastern parts of Australia, positive SOI indices (or La Nina years) are
generally associated with wet weather. The scatter plot of the concurrent average DJF
precipitation rate from the reanalysis and the Bureau with standardized SOI shows moderate
positive associations (Figure 3.12), which is higher for the Bureau (r=0.42) than for the
reanalysis (r=0.29). A local smoother (Loader 1999) is also overlaid (black line). Similar
associations are seen by examining the relationship between average DJFM precipitable water
and the concurrent and preceding SOI (Figure 3.13). The concurrent season graph for
precipitable water shows that besides a few negative SOI index outliers, the relationship is
strongly positive once SOI reaches zero. The linear correlation value is 0.41.
64
Figure 3.12. Concurrent average seasonal SOI index versus the precipitation rate of the
reanalysis (left) and BoM (right). Black line is local smoother and r value is the linear
correlation.
Figure 3.13. Average seasonal concurrent (left) and preceding (right) SOI index versus the
DJFM precipitable water of the reanalysis. Black line is local smoother and r value is the
linear correlation.
Rather than using standard climate indices such as the SOI, tailored indices can be
developed to examine correlations of large-scale climate information with local-scale climate
(e.g., Grantz et al. 2005). For instance, Figure 3.14 shows the spatial correlation between the
DJFM precipitable water for the Brisbane/Toowoomba area and sea surface temperatures (SSTs)
in the waters surrounding Australia for the preceding and concurrent seasons. The box with the
white dashed lines shows the area of highest correlation for the period of record. The plot of
SST averages from each boxed area against DJFM precipitable water shows a positive
relationship (Figure 3.15), with higher correlations (i.e., r=0.50 and r=0.43) than resulted from
65
analysis with the SOI. Hence, for particular regions, additional insight regarding climate
variability may be gained from exploring tailored indices.
Figure 3.14. Correlation between DJFM precipitable water for the Brisbane/Toowoomba
region and the average seasonal preceding (top) and concurrent (bottom) SSTs. Dashed
white box represents area of highest correlation (top: lat. −20.0 to −31.4, long. 163.1 to
174.4; bottom: lat. −29.5 to −41.0, long: 153.8 to 166.9).
66
Figure 3.15. Average seasonal concurrent (left) and preceding (right) SSTs versus the
DJFM precipitable water, from the reanalysis. Black line is local smoother and r value is
the linear correlation.
3.4 Radar Climatology
This study evaluates the mean characteristics of clouds observed by radar, utilizing
objective classification of storm properties and is also presented in Peter et al. (2009). We report
similar cloud characteristics as observed by Houze and Cheng (1977) and Potts et al. (2000).
Their studies focused on short periods during summer, while our analysis examines cells during
all seasons of the period 2000–2006. Furthermore, this analysis builds on the work of Houze and
Cheng (1977) and Potts et al. (2000) as we also examine how the cell statistics vary with
synoptic situation and season.
The statistics that we examine are precipitation area, cell volume and cell-top height. The
size distribution (both horizontal extent and height) of clouds is required for parameterization of
the effects of clouds in general circulation models (GCMs) for two reasons. First, the cloud size
distribution is needed to describe the scattering of incoming solar radiation and secondly because
the parameterization of convection is dependent on cloud size. Many convective
parameterization schemes are based upon mass flux, which requires cloud area and cloud updraft
as input (e.g. Arakawa and Schubert, 1974). There have been several observational studies which
have attempted to quantify the cloud size distribution and some discrepancy has arisen on the
functional form which the size distribution takes. Plank (1969) studied cumulus cloud fields in
Florida and found an exponential fit to the data, while later studies found a truncated lognormal
distribution best described the cloud size (Lopez, 1976, 1977). Most recently, observational
studies (Cahalan and Joseph, 1989) and modeling studies (Neggers et al., 2003) have found a
power law best describes the cloud size distribution.
67
Other studies of storms in Australia have investigated extreme events (e.g. Abbs and
McKinnes, 2004; Matthews and Geerts, 1995), because these events are responsible for the most
damage to property and also for loss of life. Since this paper is intended to provide a background
climatology we do not focus solely on extreme events but conduct our analysis on all observed
cells. Analyzing all cells also provides a description of a range of cloud systems and will enable
the representativeness of systems chosen for future QCSRP case studies to be placed in context.
Furthermore, statistical analysis of all cells will enable comparison of observations with results
from cloud resolving model simulations to assist in the development of better cumulus
parameterization schemes (e.g. Neggers et al. 2003; Plant, 2009).
The data sets employed are described in Section 3.4.1. In Section 3.4.2 we conduct a
k-means clustering of variables derived from sounding observations obtained twice daily at
Brisbane airport, from which we determine the major synoptic regimes prevalent in SEQ. This
method is similar to that presented in Section 3.2, however instead presents a 3-cluster case for
use in the subsequent radar analysis. In Section 3.4.3 we evaluate statistics of cells determined
from an objective storm classification methodology (Dixon and Weiner, 1993) for the entire
observational period and then extend the analysis to examine how the statistics vary under the
differing synoptic regimes determined from the clustering analysis. Section 3.4.4 summarizes the
main conclusions of this study.
3.4.1 Data sets
3.4.1.1 Marburg radar
The radar data set used in this study was obtained from the Australian Bureau of
Meteorology‘s (BoM) Marburg radar (27.61 S, 152.54 E). It is a WSR74S/14 radar, which
transmits 10-cm (S-band) wavelength electromagnetic radiation. It was installed in 1993 and has
been taking near-continuous measurements of rainfall west of Brisbane since. The data set covers
the period January 2000–July 2006. The Marburg radar collects a three-dimensional volume of
data every 10 minutes. Each volume consists of 15 tilts ranging in elevation angles from 0.5 to
32 degrees.
The Marburg radar data is analyzed using the objective classification method of Dixon
and Weiner (1993) (TITAN). TITAN is used operationally by BoM forecasters to enable
identification of storms and allow short-term forecasting of storm tracks (Potts, 2005). The polar
data which is output from the Marburg radar is transformed to a Cartesian coordinate system.
TITAN then defines a cell as a contiguous volume of pixels that exceed a prescribed reflectivity
and size threshold. For the present analysis, these thresholds are 35 dBZ and 30 km 2 for
reflectivity and size, respectively. The 35 dBZ threshold was considered to be sufficient for this
climatology as manual observations indicate that even shallow cumulus readily produce echoes
in excess of this value. Where a cell split or merger occurs, TITAN retains two numbers for cell
identification; a simple track number where the separate elements are tracked and a complex
track number where the cell is considered a single entity combining splits or merger components.
For this analysis we use the complex track number to avoid ambiguity with regard to cell size.
For example, when we analyze cell height, the maximum height (of the 35 dBZ echo) is the
68
height attained at any time throughout the cell lifetime regardless of whether the maximum
occurs before or after a cell split or merger.
3.4.1.2 Sounding data
The second data set is the historical record of radiosonde measurements at Brisbane
airport over the period 1995 to April 2008. The Australian Bureau of Meteorology obtains
soundings twice daily (00Z and 12Z) from balloon releases. The environment for a day is
generally computed from the 00Z sonde. However, if that sounding is not available then the
relevant 12Z sounding is used. The analysis has been carried out on observations only at the
standard levels from 1000 hPa to 20 hPa. The standard levels are 1000, 925, 850, 700, 500, 400,
300, 250, 200, 150, 100, 70, 50, 30 and 20 hPa. The wind observations were generally not
recorded before 2000, and so we limit the radiosonde data to the period after 1999.
3.4.2 Synoptic regimes
3.4.2.1 K-means clustering of sounding data
The climate in Southeast Queensland (SEQ) can be broadly described as consisting of
two seasons, a ―dry‖ and a ―wet‖. The dry season occurs during the Austral winter months
(roughly April–September). SEQ, however, is in the subtropics and therefore has quite a few
departures from the typical wet-dry season scenario prevalent further north. For instance, during
the dry season, midlatitude cyclones to the south of SEQ can penetrate far enough into the
subtropics to bring occasional but appreciable rainfall to the region. Also during the dry season,
intense sub-tropical cyclones (known locally as east coast lows) form, which can bring sustained
strong winds and intense rainfall. The wet season weather is modulated predominantly by the
position of the subtropical ridge and the Queensland inland trough. The Queensland inland
trough usually resides over the west of Queensland, however, also moves east towards SEQ
mostly as a consequence of the position of the subtropical ridge. As the subtropical ridge
oscillates east-west off the east coast of Australia, the winds shift between southeasterly to
northwesterly. Severe thunderstorms often occur during the wet season. Operational
meteorologists in SEQ generally associate southeasterly winds with ―stream‖ or trade wind
showers, while the increased moisture and heat fluxes associated with northwesterly winds bring
an increased chance of severe convective activity.
The most common scenario encountered during the wet season is one in which
southeasterly winds prevail. It occurs when a high pressure cell is located off the east coast of
Australia and results in trade wind cumulus and showers along the coast. If there is sufficient
moisture flux, these showers may penetrate further inland to the plains; however, they tend to
dissipate once they encounter the range of mountains to the west of Brisbane. The second
scenario occurs when the surface winds at Brisbane are mainly northerly. Since the northerly
winds advect warm and moist air, this scenario can often result in deep convection, especially
during unstable conditions. The storms are noted to form on the ranges and advect east with the
steering level winds. Northerly winds are also present when the tropical monsoonal trough
migrates south, which will often result in widespread deep stratiform rain with embedded
convection. The third scenario, associated with the dry (winter) season, has winds which are
69
generally from the southwest and a mainly continental origin with low moisture content. The
winter season is typically dry, however, synoptic scale forcing can result in widespread
precipitation events even during this time of year. One of the major contributors to wintertime
precipitation are east coast lows which are intense extratropical lows which form close to the
south or central Queensland coast (Hopkins and Holland 1987).
It is desirable to quantify the synoptic regimes, which have been qualitatively identified
by operational meteorologists. Several attempts have been made at using an objective
classification method, including studies by Abbs and McKinnes (2004) who analyzed mean sealevel pressure charts and Connor and Bonnell (1998) who analyzed radiosonde data. The study
of Abbs and McKinnes (2004) was intended to describe synoptic situations under which extreme
weather events prevail. The intent of this paper is not to describe extreme events, but rather, the
statistical properties of storm events. Here we take a similar approach to Connor and Bonnel
(1998) and conduct a k-means clustering analysis (Hartigan and Wong, 1979) on variables
derived from soundings obtained at Brisbane airport. The variables chosen to represent the
synoptic environment are listed in Table 3.3 and are calculated from the standard level sounding
data.
Table 3.3. Variables used to classify the synoptic regimes of Southeast Queensland.
Variable
U
Description
Westerly wind at 850 hPa
V
Southerly wind at 850 hPa
Shear
RMS wind shear between 500 and 850 hPa
Total-totals (TT)
T(850)-2*T(500)-T(850) where T and T are the temperature
and dew point, respectively.
QU
Westerly water flux up to 250 hPa
QV
Southerly water flux up to 250 hPa
TW
Total water up to 250 hPa
The seasonal cycle of the clustering variables is shown in Figure 3.16. The variable with
the most pronounced cycle is total atmospheric water (TW), followed closely by atmospheric
stability (TT). In particular both TW and TT are lower in the winter months; that is, the
atmosphere is generally drier and more stable, however, TT values approaching those of the
summer months may still occur during winter. The largest values of atmospheric moisture occur
during the summer months (November–February). The variables used in the cluster analysis
clearly distinguish the subtropical climate characterized by a ―wet‖ (summer) and ―dry‖ (winter)
season.
For the purposes of this paper only three clusters will be considered as this number is
sufficient to illustrate how the statistical properties of the cells vary with synoptic situation and
season. The relationships between the variables chosen for the cluster analysis when three
clusters are considered are shown in Figure 3.17.
70
It is instructive to examine the seasonal cycle of the three regimes so that we may
determine during which time of year they are most likely to occur. This is shown in Figure 3.18.
It is clear that the regime with low moisture, high instability and southwesterly winds (black
circles) is a predominantly winter regime and the other regimes occur mainly during the summer
months. From Figure 3.18, it is evident that the synoptic situations corresponding to each cluster
can occur at any time of the year, particularly the northwesterly and southeasterly regimes. Even
the regime identified as occurring mainly during winter has events which occur during the
summer months. It must therefore be kept in mind that the naming convention we have adopted
is one which best describes the characteristics of the cluster as a whole, however, departures are
to be expected, primarily due to the statistical nature of the clustering algorithm.
Finally, the clustering analysis indicates that there are three predominant regimes in SEQ.
These are: (1) a winter regime with low TW and TT, and mainly southwesterly winds (black
circles), (2) a trade wind regime where the winds are southeasterly (green circles), and (3) a
regime with predominantly northwesterly winds (red circles). While both the northwesterly and
southeasterly trade wind regimes have large moisture fluxes, the northwesterly regime is slightly
more unstable than the regime with winds from the southeast. It is apparent that the clustering
analysis (when considering three clusters) has extracted the synoptic situations often considered
by operational forecasters.
Figure 3.16. Seasonal variation of environmental variables for the period 2000 to 2008.
Values for 2008 are colored in grey.
71
Figure 3.17. Relationships between environmental variables in Table 3.3 with three clusters.
The three clusters correspond to the following synoptic regimes: (1) a winter regime (black
circles), (2) a northwesterly regime (red circles) and (3) a southeasterly (trade) regime
(green circles).
Figure 3.18. Seasonal cycle of total water for the synoptic regimes determined by k-means
clustering analysis; black is the winter regime, red is the northwesterly regime and green is
the southeasterly trade wind regime.
72
3.4.2.2 Characteristics of the synoptic regimes
In this subsection we present some of the basic radar-retrieved morphology and
microphysical descriptors of the storms present during each of the synoptic regimes determined
in the previous subsection. The images shown were obtained with the BoM‘s CP2 radar. CP2 is a
dual-polarized, dual-wavelength (10 cm and 3 cm wavelength) Doppler radar and is located to
the east of the Marburg radar. CP2 has only been operational since November 2007, so is
unsuitable for the climatology analysis which we will present in later sections, however, due to
its capabilities it provides us with much better understanding of microphysical and dynamical
processes occurring within cells.
3.4.2.2.1 Southwest (winter) regime
The weather during the winter months in SEQ is characterized by predominantly dry and
stable conditions. However, heavy precipitation events do occur during these months. Synoptic
scale features are the major generators of rain during the winter months. One such feature is the
presence of upper level troughs and lows. The strong instability associated with cold air aloft
leads to the development of deep systems which can produce showers or thunderstorms. Local
forecasters have noted that the presence of high moisture content in the lower levels is not a
necessary requirement for the development of deep convection associated with upper lows.
Another system, which can bring widespread rainfall to SEQ is the presence of East Coast lows
(Hopkins and Holland 1987). These intense extratropical lows form close to the south or central
Queensland coast and, although they can form at any time of year are most common in autumn
or early winter. They form when thermal lows in the middle troposphere (500–700 hPa) coincide
with upper divergence caused by the subtropical jet.
3.4.2.2.2 Northwest regime
During the Austral summer months, a high-pressure ridge is generally present near the
east coast of Australia. As the ridge migrates eastward, due to the passage of midlatitude
cyclones south of the Australian continent, the high will generally ridge along (or offshore of)
the eastern Australian coast. In such cases, the winds in SEQ tend to northwesterly–northeasterly.
The northerly flow advects warm and moist air into the region and, as the high moves east, the
trade wind inversion erodes providing conditions conducive for isolated thunderstorm
development. Operational meteorologists make the common observation that storms often form
over the elevated terrain and will subsequently propagate towards the coast (due to the upper
level steering winds) where they may dissipate or intensify depending on the stability profile in
the low-lying coastal region. Synoptic scale disturbances can intensify thunderstorm activity
during this scenario: these include (but are not limited to), the location of the Queensland inland
trough which usually runs north-south through central Queensland, the presence of upper level
cold pools and the passage of mid-latitude cyclones, the northern edge of which will sometimes
penetrate into SEQ. Mesoscale disturbances, especially the sea breeze which is a common
phenomenon in Brisbane in the afternoon during the summer months, can also contribute to
intensification (or dissipation) of storms located in the plains.
An example of a typical vertical profile obtained when the winds are northerly (cluster 2)
is shown in Figure 3.19. The sounding was obtained on 22 January 2009 at 00Z (10 AM local
73
time). The dashed curve represents the moist adiabatic profile of a surface parcel that has been
lifted above the lifting condensation level and indicates that such an environment promotes the
development of thunderstorms, which will reach well above the freezing level. On this particular
day, the convection was triggered due to topographic influences and the resultant thunderstorms
were of moderate intensity.
Figure 3.20a shows a PPI taken at 04:18 UTC (14:18 local time) on 22 January 2009,
while Figure 3.20b is a synthetic RHI obtained by reconstructing successive PPIs through the
azimuth indicated by the red line in Figure 3.20a. Several storms are evident, especially in the
western portion of the domain over the elevated terrain. The RHI indicates that cloud top in the
storms extends to 6 km, which is well above the freezing level (4.7 km). A moderate reflectivity
approaching 50 dBZ was reached in the storm. The lower panels show the PPI and RHI images
after fuzzy logic particle identification algorithms (Vivekanandan et al. 1999) were applied to the
data. Strong signatures of the presence of rain and graupel particles are evident throughout the
storm and, to a lesser extent, signatures of the presence of hail. The storms remained isolated on
this day, which was probably due to a combination of factors: (1) the dry environment through
the mid-levels, which will decrease the buoyancy of cloudy parcels as environmental air is
entrained, and (2) a lack of synoptic or mesoscale features discussed above to promote extra
vertical development. When these additional factors are present, the storms will often merge,
resulting in long-lived thunderstorms or squall lines whose tops may reach the tropopause and
contain significant hail and damaging winds.
3.4.2.2.3 Southeast (trade) regime
Southeast winds and associated trade wind cumulus (known locally as ―stream‖ showers)
are a very common occurrence during the summer months in SEQ. The typical vertical profile of
a southeast regime is shown in Figure 3.21 and shows a relatively moist unstable air mass capped
by a strong temperature (subsidence) inversion. The height of the inversion varies due to several
factors but is typically in the range 2–3 km with a moist layer extending to just below 2 km. A
wind shear is evident across the trade inversion, along with significant drying above the
inversion. The orientation of the surface winds is noted to contribute to the ability of the showers
to penetrate inland; if the winds are predominantly from the south and parallel to the coast then
the showers are usually confined to the coastal regions, however if the wind has more of an
easterly component and is aligned perpendicular to the coast, then the showers will penetrate
further inland.
The height of the freezing level during the summer months is usually from 3–4 km, above
the height of the trade inversion. As a result, rain produced in such showers is formed via the
warm rain process. This is evident in the hydrometeor classification shown in Figure 3.22. It can
be seen that despite the absence of ice processes involved in precipitation production, high
reflectivities (in this case about 40 dBZ, however during the field campaign reflectivities up to
50 dBZ were noted). The trade showers generally follow a diurnal pattern peaking in shower
activity during the night and early morning and subsiding during the day. However, synoptic
forcing, such as upper level troughs may provide enough upper-level instability to maintain
shower activity throughout the day.
74
Figure 3.19. Skew-T plot for 22 January 2009 illustrating sounding conditions typical of the
northwesterly (cluster 2) regime.
Figure 3.20. PPI and RHI images of reflectivity and particle identification obtained on 22
January 2009 illustrating typical observed clouds during the northwest regime (cluster 2).
75
Figure 3.21. Skew-T plot for 12 November 2008 illustrating sounding conditions typical of
the southeasterly (cluster 3) regime.
Figure 3.22. PPI and RHI images of reflectivity and hydrometeor classification obtained on
12 November 2008 illustrating typical observed clouds during the southeast regime
(cluster 3).
76
3.4.3 Statistics of Southeast Queensland storms
3.4.3.1 Cell location frequency
The Marburg radar is located southwest of Brisbane in the elevated terrain which bounds
Brisbane to the west (see Figure 3.23a). The spatial distribution of cell location frequency for all
35 dBZ cells during the complete observational period is shown in Figure 3.23b. The cell
frequency has been normalized by the maximum number of cells registered and is therefore in
the range 0–1. It shows the location in which a cell was registered at any time throughout its
history, and will therefore indicate information about cell movement. There is a maximum in cell
location through the ranges to the west and south of Marburg radar. Figure 3.23c shows the cell
location for the first occurrence of a cell on a particular day; it is clear that cells exhibit a
propensity to form over the higher terrain. Figure 3.23d shows the cell location frequency for the
first occurrence of a complex track, and is therefore independent of the motion of the cell. It
resembles Figure 3.23a, suggesting that many cells which are registered do not travel distances
greater than the resolution of the histogram spacing (approximately 17.3 km) throughout their
lifetime. Similar maxima in observed cell frequency over ranges were obtained by Matthews and
Geerts (1995) and Potts et al. (2000) for cells in the Sydney region.
The observed decrease in cell frequency with range is partially due to beam geometry: at
distances greater than about 200 km from the radar, the beam height is greater than 4 km, which
is approximately the height of the freezing level, so many of the shallow trade wind cumuli
which are confined below this level will not be registered by the TITAN software. Moreover the
width of the beam increases with distance from the radar, and many cells may be smoothed over
a large beam volume. At close distances cells may not be topped by the radar beam. Despite
these caveats, it is clear that many cells have their origin over the ocean and near the coast. The
problems associated with beam broadening and beam height are more clearly illustrated in
Figure 3.24, which shows a scatter plot of the maximum reflectivity observed as a function of
distance from the Marburg radar. The centre curve shows the median of maximum reflectivity
while the upper and lower curves represent one standard deviation from the median. For the
continuing analysis, only observations in the range 20–130 km are considered from hereon. The
total number of cells measured for the observational period was about 26 600 and although some
numbers reported (e.g. mean, median, etc.) may change when including the complete domain,
the main conclusions were found to be independent of the domain size.
We now examine the cell location frequency for the different synoptic regimes
determined by the clustering analysis of Section 3.4.2. Figure 3.25 shows the spatial frequency
when three synoptic regimes are considered. Again, there is a clear indication that cells have
their genesis over the elevated topography to the west of the coast. However, the spatial
frequency for cluster 1 (winter regime) also shows that many cells occur near the coast to the
south of Brisbane. The northwesterly and trade wind regimes also have maxima in cell location
frequency along the ranges and the coast.
77
Figure 3.23. Spatial frequency of all storms observed by the Marburg radar for the period
2000–2006: (a) is a topographic map of SEQ showing the location of the Marburg radar.
Elevation contours are separated by 100 m intervals, (b) shows the spatial occurrence
frequency distribution for all storms at all times during which they were registered as a
TITAN echo, (c) for the first time a storm was registered on a particular day, and (d) for
the first unique occurrence of a storm’s complex track number.
Figure 3.24. The variation of the median of the maximum reflectivity measured in storms
from the Marburg radar is shown in the middle curve. The upper and lower curves
represent one standard deviation from the median of the maximum reflectivity.
78
Figure 3.25. Spatial occurrence frequency of storms when three synoptic clusters are
considered. The clusters numbers correspond to the following synoptic regimes: (1) winter
regime, (2) northwesterly regime, and (3) southeasterly trade wind regime.
3.4.3.2 Precipitation area
The areal extent of precipitation was examined. Previous studies have shown that this
property is described by a truncated lognormal distribution (e.g., Lopez 1976, 1977; Potts et al.,
2000). The frequency histogram of precipitation area is shown in Figure 3.26a and shows that
small cells dominate the number frequency of cells. We now examine the frequency distribution
of precipitation area to determine if it is described by a lognormal distribution. Figure 3.26b is a
histogram of the natural logarithm of the precipitation area. If the areal extent of precipitation
area is distributed in a lognormal manner, then it follows that its logarithm should be distributed
normally. The normal distribution described by the mean and the standard deviation of the
observations has also been superimposed on the observations. There are departures from
normality with the observations tending to be skewed towards smaller precipitation area. To
examine this in more detail consider Figure 3.26c, which shows a plot of the quantiles of the
logarithm of the observed precipitation areas against theoretical quantile values for a normal
distribution (a q-q plot). If the logarithm of the precipitation area is distributed normally (i.e., if
the precipitation area is distributed log normally) then the q-q plot should lie along the straight
line also plotted. There are departures at both small and large values of the precipitation area,
however, the departures at large values are minimal. The region that does lie on the straight line
corresponds approximately to cell areas in the approximate range 12–250 km2. Also plotted is
the empirical cumulative distribution function (ecdf) of the observations compared with expected
values from a normal distribution. Again, we have obtained another qualitative description that
precipitation area of cells is well described by a lognormal distribution function.
79
Figure 3.26. Frequency distribution of 35 dBZ precipitation area (km) for all storms.
To quantitatively address whether a lognormal distribution describes precipitation area in
the measured cells there are numerous tests that can be applied. We have chosen to implement
the Kolmogorov-Smirnov (KS) test, which is a test for normality applied to the ecdf. Since it is a
test for normality we applied it to the ecdf of the natural logarithm of precipitation area. The
results are summarized in Table 3.4. Since the p-value is greater than 0.05 when all synoptic
clusters are considered together and separately we conclude that precipitation area is distributed
log normally.
Table 3.4. Kolmogorov-Smirnov test for normality summary statistics.
Cell property
Synoptic cluster
All
Precipitation area 0.59
Cell volume
0.98
Cell height
0.87
1
0.82
0.43
0.79
2
0.87
0.99
0.68
3
0.72
0.99
0.99
We have examined similar descriptors of the distribution of the logarithm of precipitation
area for cells observed during the each of the synoptic clusters described in Section 3.4.2. For
brevity, we refer to only q-q plots for each of the clusters. Figure 3.27 shows q-q plots of the
precipitation area for all observed cells and for each of the synoptic clusters. Similar to the
observations for all observed cells, the precipitation area for each of the synoptic clusters is also
distributed log normally. Again, significant departures are seen for small precipitation areas, but
cell areas up to at least 250 km are distributed log normally and the departures up to cell sizes of
approximately 3000 km are minimal. The KS test (Table 3.4) also confirms that precipitation
area is distributed log normally.
80
Figure 3.27. Normal q-q plots of the logarithm of precipitation area for all storms and for
the synoptic clusters.
3.4.3.3 Cell volume
The histogram of cell volume for all 10-minute samples was computed and is shown in
Figure 3.28a. A total of 26 227 storms were examined in this study. It is seen that the majority of
cells have a volume of less than 250 km. The geometric mean (a measure of the central tendency)
of storm volume was about 120 km. The logarithm of storm volume and the normal fit computed
from the mean and standard deviation is shown in Figure 3.28b. As for precipitation area, it is
seen that cell volume is skewed to smaller sizes. The q-q plot indicates that cell volume is
distributed log normally over the range 55–700 km (4–6.5 on the q-q plot in Figure 3.29), which
accounts for approximately 90% of cells. The departures for large cells have been shown by
Lopez (1977) to be characteristic of a truncated lognormal distribution due to environmental
limits placed on the maximum size of cells. Houze and Cheng (1977) also pointed out that very
large cells are underestimated in size when they fall outside the domain of the radar. The KS test
values (Table 3.4) again confirm that cell volume is distributed log normally.
81
Figure 3.28. Frequency distribution of 35 dBZ storm volume (km) for all storms.
Figure 3.29. Normal q-q plots of the logarithm of storm volume for all storms and for the
three synoptic clusters.
3.4.3.4 Cell-top height
The frequency distribution of cell-top height for all cells is shown in Figure 3.30a. Here,
storm-top height is defined as the maximum height of the 35 dBZ echo achieved by a storm
during each 10-minute sample. As for precipitation area and cell volume the frequency histogram
is skewed towards lower heights, the geometric mean being 6.5 km. About a quarter of cells are
confined below 4 km, which is the approximate height of the freezing level during the summer
82
months. Examination of the frequency histogram of the logarithm of cell-top height and the
associated q-q plot indicates that cell top height is distributed log normally in the approximate
range 2–9.5 km. The departures for small sizes are most likely a consequence of the lower limit
of 30 km specified for the analysis by the TITAN software, while departures for large cells are
due to environmental limits (for instance, the height of the tropopause) which constrain the
height of cells. The KS test values (Table 3.4) again confirm that storm-top height is distributed
log normally.
Figure 3.31 shows q-q plots for all cells and for each of the synoptic regimes. During
winter (cluster 1), 25% of cells are below 3.75 km, while for the northwesterly and southeasterly
regimes, 25% of cells have tops lower than 5.25 km and 4.75 km, respectively. During the trade
wind scenario, the freezing level is located at about 4 km indicating that the warm rain process is
important for the production of precipitation in many cells during this regime. As for
precipitation area and cell volume, cell-top height is distributed log normally over a couple of
orders of magnitude. The q-q plot for cluster 3 (the trade regime) indicates that cell-top height is
distributed log normally in the range 1.5–16 km. This corresponds roughly to all altitudes in the
range between the typical heights of the trade wind inversion and tropopause.
Figure 3.30. Frequency distribution of 35 dBZ storm height (km) for all storms.
83
Figure 3.31. Normal q-q plots of the logarithm of storm height for all storms and for the
three synoptic clusters.
3.4.3.5 Intercomparison between clusters
Some summary statistics of the cell properties examined are shown in Table 3.5. The top
value in each row represents the mean and standard deviation while the lower value represents
the geometric mean. The geometric mean is quoted as it represents a measure of scale for a
lognormal distribution (Potts et al., 2000). Lopez (1977) suggested that cell properties were
distributed in a lognormal manner because cloud growth is a stochastic process whereby growth
is a random multiplicative proportion of the current size of the cloud. For processes that change
multiplicatively, the appropriate measure of the central tendency is the geometric mean rather
than the arithmetic mean.
Despite the precipitation area for all three regimes exhibiting a truncated lognormal
distribution it is useful to examine the interrelationship of cell properties between each regime.
To achieve this we apply a Student‘s t-test to examine whether the mean of precipitation area is
statistically different between the regimes. The results of the t-test statistic analysis are shown in
Table 3.6. In all cases, the critical value for the t-test is about 1.96 (at 5% confidence limit for a
two-sided test). Only the t-test applied to clusters 2 and 3 suggests that they exhibit a mean in
precipitation area drawn from the same sample. Recalling that cluster 1 corresponds to the winter
regime, this suggests that the characteristics of precipitation area for the winter regime differ
from the northwesterly and trade regimes.
Stated in another manner, the mean precipitation area observed during winter was
significantly higher than during the summer months. The t-test has therefore quantitatively
demonstrated the difference between the wet and dry seasons. We also note that the t-test statistic
values are sensitive to changes in the choice of the bounds of distance from Marburg radar
chosen. However, the conclusions regarding the differences between the precipitation area in the
different synoptic clusters are not. The t-test has also been applied for cell volume and cell-top
84
height, the results of which are also summarized in Table 3.6. The t-test statistic when applied to
cell-top height also indicated that the winter regime is distinct from the two summer regimes,
with a significantly lower cell-top height. This is not surprising since (1) the cluster analysis
indicated that the winter regime was more stable than either of the summer regimes, and (2) the
height of the tropopause will be lower during winter months which will in turn lead to lower celltop heights. Interestingly, when the t-test statistic is applied to cell volume, the winter regime
exhibits more similarity with the southeasterly trade regime.
Figure 3.32 shows frequency histograms of storm-top height when all storms are
considered and for each of the synoptic clusters. The winter time regime (cluster 1) is skewed
towards smaller size, while the two summer regimes display an increased probability for the
occurrence of slightly larger storms. The trade wind regime (cluster 3) displays the largest
probability of deep storm occurrence, however, it is apparent that deep storms can form at any
time of the year. The geometric (and arithmetic) mean of storm-top height was very similar for
each of the synoptic clusters (see Table 3.5), which is due to the statistics being dominated by
smaller storms. As was mentioned in the introduction, the statistics capture the gross features of
storms present during each synoptic regime, however, they fail to capture the influence of the
larger storms which make up the tails of the distributions.
Table 3.5. Cell-property summary statistics for the three synoptic clusters. The arithmetic
mean and standard deviation are shown in the top of each row. The value in parentheses
below is the geometric mean.
Cell property
Synoptic cluster
All
Precipitation area 101±186
(46)
Cell volume
339±799
(121)
Cell height
7.4±4.0
(6.5)
1
113±215
(47)
351±844
(120)
6.9±4.1
(6.5)
2
94±175
(51)
291±670
(116)
7.4±3.6
(6.0)
3
97±175
(43)
367±857
(126)
7.7±4.3
(6.7)
Table 3.6. Cell-property t-test statistics to examine the inter-relationship between clusters.
In all cases the t-test critical value is 1.96.
Cell property
Statistic
Precipitation area t-value
p-value
Cell volume
t-value
p-value
Cell height
t-value
p-value
Synoptic cluster
1 and 2
2.59
0.001
2.11
0.03
3.16
0.002
85
1 and 3
2.25
0.02
0.54
0.59
4.6
<0.001
2 and 3
0.54
0.59
3.05
0.002
1.8
0.07
Figure 3.32. Frequency histograms of storm-top height for all storms and for each of the
synoptic clusters.
3.4.3.6 The contribution to precipitation by differing-sized storms
From the preceding analysis of the statistics of cell volume and precipitation area, it is
clear that frequency histograms are skewed to smaller cells, i.e., most cells are likely to be
shallow, short-lived cells. We now attempt to address the question: what is the relative
contribution to the total precipitation of the different sized cells? Figure 3.33 shows the
cumulative frequency of areal coverage of precipitation area versus the cumulative frequency of
precipitation area ranked by diminishing size. Curves for all cells and also for the individual
clusters are shown. It shows that the largest 20% of cells are responsible for approximately 67%
of precipitation. A virtually identical result was obtained by Lopez (1976) for observations of
shallow cumulus in the northwestern Atlantic. The curve for the northwesterly regime (cluster 2)
matches that for all cells, while the winter and trade regimes (clusters 1 and 3, respectively) show
a slightly increased fractional areal coverage, however, this is only on the order of a few percent.
Given that the preceding analysis of precipitation area indicated that the winter regime was quite
distinct from the summer (northwesterly and trade) regimes, it is not clear why the same
similarities are evident in the fractional areal coverage of precipitation.
86
Figure 3.33. Cumulative frequency distribution of fractional areal coverage of
precipitation area. Curves are shown for all storms during the period 2000–2006 and for
the individual clusters.
3.4.4 Conclusions
This section investigates the statistical properties of cells in Southeast Queensland. The
data used for the analysis were obtained from radiosonde observations of the vertical structure of
the atmosphere and measurements of reflectivity from a ground-based 10-cm wavelength radar.
The soundings provided observations of wind direction, wind shear, stability (in this case,
total-totals), moisture fluxes and total integrated water. These variables were then analyzed with
a k-means clustering algorithm to determine three predominant synoptic regimes for the SEQ
region. The main result of the clustering analysis was:
• Three distinct synoptic regimes were identified. The distinctions were most obvious in the
moisture and stability fields. The regimes classifications were: (1) a dry, stable regime
with mainly westerly winds, which occurred during the winter months (April–September);
(2) a moist, unstable regime with northwesterly winds; and (3) a moist, unstable regime
with southeasterly winds. Regimes 2 and 3 occurred mainly during the summer months
(October–March).
An objective method of storm identification and classification was applied to the radar
observations and various statistics of the cells analyzed for the period January 2000–July 2006.
The statistics examined included cell volume, area, height and the cumulative areal coverage of
precipitation. Furthermore, the statistics were also investigated for each of the individual regimes
determined from the cluster analysis.
87
Despite clear differences in the moisture and stability characteristics of the synoptic
regimes the gross statistics of cells appear to vary little between the different regimes. It was
shown that the mean in the precipitation area of the winter regime was statistically larger from
that of the northwesterly or trade regimes and that the cell-top height was significantly lower.
However, the same distinction was not evident in the areal coverage of precipitation: all regimes
indicated that the largest 20% of cells were responsible for about 67% of the fractional areal
coverage of precipitation.
The main conclusions obtained from the analysis of the radar data were:
• When the entire period was considered it was found that cell volume, area and height all
followed truncated lognormal distributions. The same result was found when each of the
synoptic regimes was considered individually.
• Student‘s t-test applied to examine the relationship between the clusters found, (1) the
precipitation area was significantly larger during the winter regime than during either of
the summer regimes, (2) cell-top height was significantly lower during the winter months
than during the summer, (3) the summer regimes do not separate statistically when
stratified by precipitation.
Not surprisingly, the cluster analysis and following statistical tests have identified the
distinction between the summer (―wet‖) and winter (―dry‖) seasons, known to the local
population and forecasters alike.
That various cell properties are distributed log normally is an interesting result. However,
since the same conclusion appears true for all of the synoptic regimes, it would seem that these
properties are not dependent so much on the vertical structure of the atmosphere. Raymond
(1997) suggested that the lognormal distribution of cell properties is a consequence of the
interaction of the subcloud layer turbulence and humidity fields with the cloud layer at all scales.
Further studies are required to examine the fundamental processes controlling the emergence of
lognormal distributions for the cell properties presented in this study.
3.5 Lightning Climatology
Lightning is a tool that can be used to help assess convective-storm microphysics.
Lightning activity is related to the dynamic vigor of a storm, but also depends on the presence of
both ice and supercooled liquid water at temperatures colder than -10° C for non-inductive
charging processes to be effective. Thus, the lightning activity in a storm (or lack thereof) can be
used to indicate the nature of the microphysics in the mixed-phase region of clouds. The total
(intracloud plus cloud-to-ground) lightning behavior is the most useful to assess these
characteristics of clouds, given that thunderstorms often produce most of their lightning in cloud
(i.e., intracloud lightning). Cloud-to-ground (CG) lightning occurs in most thunderstorms, but
not all. However, instruments to measure total lightning are not widespread at the present,
whereas CG lightning detection networks are more commonplace. In Australia, the GPATS
88
(Global Position and Tracking Systems Pty. Ltd1) CG lightning detection network observed CG
lightning in both seasons of the CSRP, and a climatology of this CG lightning behavior is
illustrated in Figure 3.34 and Figure 3.35.
It is clear in these maps that most of the CG activity in these months occurs farther west
of the CSRP domain (200 km radius from CP2). The only month with notable CG lightning
activity in the CSRP domain was December 2008 in season two (Figure 3.35c). Thus, based on
the CG lightning activity alone, it suggests that the mixed-phase microphysics of clouds farther
west are more vigorous than those closer to the coast, where the CSRP domain is presently
located.
1
http://www.gpats.com.au
89
a)
b)
c)
d)
e)
Figure 3.34. Maps of monthly total GPATS CG lightning counts from CSRP season one
for a) November 2007, b) December 2007, c) January 2008, d) February 2008, and
e) March 2008. The map origin is the CP2 radar and the white radial lines extend to a
200 km radius from CP2.
90
a)
b)
c)
d)
e)
Figure 3.35. Maps of monthly total GPATS CG lightning counts from CSRP season two for
a) October 2008, b) November 2008, c) December 2008, d) January 2009, and e) February
2009. The map origin is the CP2 radar and the white radial lines extend to a 200 km radius
from CP2.
91
3.6 Numerical Modeling
3.6.1 Air mass origins and back trajectories
In order to help understand the natural rain formation processes in the domain, we need to
characterize the ambient aerosol conditions. We have employed the use of a back trajectory
model to determine the air mass origin and path taken to the measurement location, based on the
assumption that these factors (air mass origin and path) have an effect on the measured aerosol
conditions.
Back trajectories were calculated using the Hybrid Single-Particle Lagrangian Integrated
Trajectory (HYSPLIT) model (Draxler and Hess 1998; information available online at the
National Oceanic and Atmospheric Administration [NOAA] Air Resources Laboratory [ARL]
website, http://www.ready.noaa.gov/ready/hysplit4.html). We used the Global Data Assimilation
System (GDAS) archived data, with a temporal resolution of three hours and gridded to 1 degree
x 1 degree in latitude and longitude. The GDAS data set is the only one available that covers the
CSRP project domain for the entire duration of the measurements; however, comparisons
between trajectories calculated using GDAS and other data sets for the same region in past years
yield highly similar results (Figure 3.36).
Figure 3.36. Comparison of GDAS, National Centers for Environmental Prediction (NCEP)
final global analyses (FNL), and NCEP/NCAR reanalysis meteorological datasets and
resulting back trajectories (on 22 February 2005).
HYSPLIT offers three methods for calculating vertical motion along back-trajectories:
isobaric flow, isentropic flow, or three-dimensional flow. Isobaric solutions have been found to
be less realistic, while three-dimensional trajectories surpass both of the other methods in
accuracy when vertical wind data is available (Martin et al. 1990, Draxler 1996, Stohl 1998),
therefore we used three-dimensional flow for our calculations.
We ran the model for a 24-hour duration backward in time from the time of our
measurements. The HYSPLIT trajectory model has an error of roughly 15–30% of the total
92
travel distance (http://www.arl.noaa.gov/faq_hg11.php), thus we chose 24 hours in order to
minimize travel distance (and therefore error) while providing a useful amount of synoptic
information. Trajectories of less than 24 hours ignore any possible diurnal cycle effects on air
mass trajectory. We also observed anything less than 24 hours to be an insufficient number of
time steps to identify regime borders or aerosol trends.
To get a representative view of the common trajectories over the entire season, the model
was run daily between November 2008 and February 2009, starting hourly from
02:00–06:00 UTC, and at starting heights of 500, 1000, and 1500 m above ground level (AGL),
covering the majority of flight operation time and observed cloud base heights in season two.
The trajectories were then labeled according to residence time in predefined regions (Figure
3.37):
1)
2)
3)
4)
Ocean north of Brisbane (hereafter, northeasterly, NE)
Ocean south of Brisbane (hereafter, east/southeasterly, E/SE)
Land north of Brisbane (hereafter, northwesterly, NW)
Land south of Brisbane (hereafter, west/southwesterly, W/SW)
Figure 3.37. Map of the 0.5 deg x 0.5 deg grid used in sector analysis, illustrating four
sectors: 1) ocean north of Brisbane in cyan, 2) ocean south of Brisbane in green, 3) land
north of Brisbane in red, and 4) land south of Brisbane in black.
All of the time steps along each 24-hour trajectory are assigned to the sector within which
it falls and then the number of time steps within each sector is tracked for each trajectory. Other
studies making use of sector analysis have ranged between taking the sector within which a
trajectory spends the majority of its time (Toledano et al. 2009), or the majority of time as in the
93
previous example but weighting the final 20% of the trajectory more heavily (Hopkins et al.
2002), or requiring 100% of the trajectory to fall within a given sector to qualify as characteristic
of that regime (Dayan and Levy 2005). While the 100% requirement certainly eliminates
uncertainty due to air masses coming from or traversing multiple surface conditions/regions, it is
an unreasonable condition given our project‘s location on the coast. We also found that when a
trajectory spent time in more than two sectors, the sector in which the trajectory spent the
majority of its time may not represent the air mass character. For example, if a trajectory spent
15 hours over the ocean, split between the north and south ocean sectors for eight and seven
hours, respectively, and then nine hours over one land sector, the trajectory would be classified
as the land sector, even though its source and nature was more maritime. Thus, we first
determine if each trajectory spent the majority of its time in an ocean or land sector, then we
distinguish which of the two ocean (or land) sectors that it spent the most time within. Therefore,
in the example above, the trajectory would be classified as the north ocean sector.
The sector regime breakdown showed that the majority of trajectories in the
November 2008–February 2009 season fell into the east/southeast (E/SE) region (Figure 3.38a,
Table 3.7). The next most common trajectory sectors were the west/southwest (W/SW) and
northwest (NW), while the least common regime was the northeasterly (NE) trajectory sector.
The exception to this was for starting heights of 1500 m, in which the W/SW sector was most
common.
The methods above were then applied to specific flight times when the aircraft flew just
below cloud base. Taking cloud base measurements was a standard procedure on every flight, in
order to build up a quasi-climatology of the cloud base conditions in the region. The percentage
of aircraft measurements in each sector roughly reflects the relative frequency of these regimes
in the overall November-February season (for 500 and 1000 m starting heights; Figure 3.39,
Table 3.7). Cloud base heights (for the cloud base aircraft measurements) ranged between
400–2200 m, but the majority were between 800–1500 m (not shown). This sector breakdown,
however, illustrates a somewhat disproportionate percentage of aircraft measurements in the
E/SE and NW regimes compared to what was observed over the entire season, especially when
you consider the number of separate days represented by the aircraft measurements for each
regime (Table 3.7). In other words, some days had multiple cloud base measurements within the
same regime, thus inflating the percentages of total aircraft measurements per regime, and vice
versa. However, the NW regime was often an active convective regime and thus a typical flight
operations day, while the southeast trade wind regime often inhibited (deep) convection and was
not a preferred flight operations day. As such, note that most of the trajectories in the
E/SE sector are more easterly than southeasterly, and thus we have dubbed this region the
easterly (E) region in future discussion. Similarly for the W/SW regime, most of the aircraft
measurements in this regime are more westerly (W), and thus in future discussion of the aircraft
measurements (Section 4.2.1), we refer to this regime as W.
94
Figure 3.38. Maps of all back trajectories between November 2008–February 2009 starting
at 500 m (top), 1000 m (middle), and 1500 m (bottom). The northeasterly (NE) regime is in
cyan, east/southeast (E/SE) regime in green, northwest (NW) in red, and west/southwest
(W/SW) in black.
95
Figure 3.39. Map of aircraft cloud base measurement back trajectories, color-coded by
regime. The northeasterly (NE) regime is in cyan, east/southeast (E/SE) regime in green,
northwest (NW) in red, and west/southwest (W/SW) in black.
Table 3.7. Percentage of total November 2008–February 2009 back trajectories at each
starting height (500 m, 1000 m, or 1500 m) and percentage of all cloud base aircraft
measurement back trajectories that were classified in each regime shown in Figure 3.38
and Figure 3.39. The number of separate days represented by the aircraft measurements
for each regime is listed in parentheses (out of 35 total days, thus some days represented
multiple regimes). The highest percentage in each scenario is highlighted in bold.
NE
E/SE
NW
W/SW
Nov 2008–Feb 2009 500 m
6.6%
47.8%
23.5%
22.1%
Nov 2008–Feb 2009 1000 m
9.9%
44.4%
20.2%
25.6%
Nov 2008–Feb 2009 1500 m
8.3%
36.1%
13.8%
41.8%
15.7% (5)
35.0% (12)
28.6% (13)
20.7% (15)
Aircraft measurements
In addition to tracing the path of the air mass, HYSPLIT can output several other
variables. Therefore, every flight time trajectory also has 24 hours of history for precipitation,
terrain height, pressure, temperature, downward solar radiation flux, relative humidity, and
mixed layer depth. These other variables can be used to provide a more complete
characterization of each regime, as well as to explain variances in aerosol concentrations. These
regimes will be utilized in Section 4.2.1 to help characterize the aerosol measurements in
southeast Queensland.
96
3.7 Mesoscale Features that Affect Cloud Development
3.7.1 Impact of local environment and terrain
To understand some of the factors that could be affecting the natural variability of the
clouds in the Brisbane area, CP2 radar echoes rendered as circles have been overlaid onto plots
of Brisbane terrain to examine where the clouds form in relation to the local surface and steering
level winds and relative to the complex terrain. Figure 3.40 and Figure 3.41 show the diurnal
variability of cloud development on 22 and 27 January 2009, respectively. The two cells that
will be discussed in Section 5.3.1.1.1 are indicated in these figures with red annotation. The
steering level flow was from the north at 10 kts on 22 January and surface winds were generally
from the N–NE. The steering level winds were from the southeast at 15 kts on 29 January and
surface winds were generally from the SE.
During the early afternoon of 22 January, cloud development was oriented in N–S cloud
streets located above horizontal convective rolls in the boundary layer (Figure 3.42). The
majority of >35 dBZ cell development occurred on the upslope sides of the higher terrain. The
ridge located to the N–NW of CP2 was the preferred hotspot for new convection throughout the
afternoon due to the favorable northeasterly component of the surface winds combined with
slightly higher surface moisture values and a surge in the wind speed evident in the radar radial
velocities that began at 01:30 UTC and propagated southwestward (Figure 3.40). Cell ―b‖
discussed in Section 5.3.1.1.1 formed along the eastern side of this ridge. The northerly steering
level flow favored storm longevity while clouds remained on the higher terrain but storms
generally dissipated as they moved southward off the ridge. The northeasterly surge or region of
enhanced surface convergence initiated >35 dBZ storms on the upslope (eastern) sides of
elevated terrain for two hours as it moved through the domain. No significant storms developed
in the surrounding low-lying areas.
The storms on 27 January formed initially along the coast during the morning hours and
moved inland as bands or waves of convection (Figure 3.42). This pattern of convection is
indicative of mesoscale dynamics influencing the orientation of convection in contrast with
22 January where convection was more locally driven. Storms were located primarily over the
higher terrain. Subtle shifts of the surface winds during the afternoon, i.e., those that shifted
from southeast to easterly, played a role in the subsequent formation of convection in the
southwestern portion of the domain, where cell ―a‖, described in Section 5.3.1.1.1, formed.
Radar echoes were observed to form and evolve quickly. This is evident in Figure 5.34b and
discussed in Section 5.3.1.1.1. As with 22 January, these storms formed on the upslope side of
the terrain.
Without question the local wind, terrain, and atmospheric stability are important factors
in initial development of clouds and subsequent storms. Furthermore, differences from one day
to the next also have an impact on the storm morphology, as shown here and discussed in later
sections, and impact the potential effectiveness of cloud seeding. Information on the local
environment will be used in conjunction with radar analysis of Z and ZDR profiles to further
characterize the type of cloud development and environment most suitable for enhancing storm
precipitation processes through hygroscopic seeding. Additionally, careful examination of
several days may indicate that for the Brisbane area, the preferred location for seeding clouds
97
should be on the windward (upslope) sides of the terrain to maximize hygroscopic seeding in the
strongest updrafts within the clouds.
Figure 3.40. CP-2 radar echoes (colored circles) on 22 January 2009 overlaid onto the
terrain map for the Brisbane area. White (blue) circles represent the location of echoes of
0–15 dBZ (20–50 dBZ) intensities. Yellow circles represent echoes chosen at random for
conducting average and volumetric profiles of reflectivity and Zdr for comparison with
seeded clouds. Clouds chosen for randomized seeding are represented by red (seeded) or
orange (unseeded) circles. Echoes in close proximity to seeded cells (―sister cells‖) are
represented by green circles and green and yellow striped circles; the latter convention is
for those cells that have had additional computations done, including profiles of maximum
Do. Location of aircraft aerosol/CCN sampling is shown by the blue ovals. Surface station
data and wind barbs are overlaid. The leading edge of a surge of northeasterly flow that
spreads southeastward during the early afternoon is indicated by the curved, dashed line
and black arrow.
98
Figure 3.41. Same as Figure 3.40, but for 27 January 2009.
99
a)
c)
b)
d)
Figure 3.42. Environmental conditions on 22 and 27 January 2009: a) radial velocity on
22 January; b) visible satellite imagery for 22 January; c) CP2 reflectivity on 27 January;
d) radial velocity image for 27 January. Yellow polylines indicate bands of convection.
3.7.2 High impact severe weather events
The highest impact convective weather event in southeast Queensland during the CSRP
occurred on 16 November 2008. There were many severe thunderstorms in the area during the
afternoon including supercells. Some storms featured hook echoes and one storm produced
extreme surface winds with the CP2 radar measuring radial velocities of about 50 m s-1
(~180 km/h). This storm, ―The Gap‖ storm, produced significant damage at ―The Gap‖, a
Brisbane suburb west of the Brisbane CBD, with power cuts to almost 230,000 homes and
tragically one fatality.
3.7.2.1 National Thunderstorm Forecast Guidance System (NTFGS) forecast
Forecasters at the Australian Bureau of Meteorology commonly use a model-and
ingredients-based approach to convective forecasting known as the National Thunderstorm
Forecast Guidance System (NTFGS). This tool takes numerical weather forecast fields from the
Australian operational high-resolution model and uses a rule-based system to outline areas of
potential storminess including individual convective hazard classes (Figure 3.43). This objective
100
tool outlined high potential for damaging winds, but there was insufficient shear in the model
forecasts for supercell thunderstorms to be indicated. The area outlined for large hail was also
too confined to the south, demonstrating that the numerically based convective guidance for the
event suffered from underestimating the severity of the storms.
3.7.2.2 Convective environment
The surface conditions (shown in Figure 3.44) associated with the severe storms in
southeast Queensland were marked by a west-northwest to east-southeast running inland trough
interacting with a frontal system moving northward along the coast. Deep convection first
initiated near the intersection between these two boundaries over elevated terrain (almost 1 km
above sea level) in the northeastern corner of New South Wales.
3.7.2.3 Sounding
The 00Z (10 AM Brisbane time) morning radiosonde ascent (Figure 3.46) showed
approximately 16 g/kg of deep (~700 m) boundary layer moisture with the precipitable water at
~37 mm (climatologically a high value for November in Brisbane). The morning surface parcel
showed a convective available potential energy of ~2000 J/kg and a convective temperature of
33° C. Note that there was a significant low-level inversion (cap) present in this trace, so that
convective initiation was delayed until significant heating over elevated terrain in the presence of
boundaries managed to break this cap.
3.7.2.4 Observed storm activity
Convection first initiated about 01:00 UTC over elevated terrain in New South Wales
(NSW), with the first Queensland cell going up southwest of the CSRP domain around
02:00 UTC. These cells developed rapidly with radar echo tops extending to the tropopause and
satellite imagery (not shown) clearly showing cloud tops overshooting the tropopause. The
initial NSW cells were generally steered towards the coast, with a leftward propagating cell after
02:00 UTC going through several cycles of new updraft formations to its north along its gust
front. Around 03:50 UTC, such an updraft regeneration cycle finally resulted in the formation of
a high precipitation (HP) supercell by 04:20 UTC featuring very strong outflow on its western
flank or rear flank (Figure 3.45). The cold pool from this cell, including its predecessors,
combined with the frontal passage from the south, initiated and amplified a new round of severe
storms after 04:00 UTC west of the CP2 radar. These new cells organized into two discrete HP
supercells by ~06:20 UTC quickly moved across the area west of the Brisbane CBD. Severe
thunderstorm warnings were issued by the Queensland Regional Forecast Office (RFC) for these
storms, identifying them as supercell thunderstorms.
The synoptic conditions for these severe storms were well anticipated by the RFC, with
the likelihood of severe weather being forecast several days out. Of course, this did not include
the extreme nature of this event.
101
Figure 3.43. Output from the 00:00 UTC (16 November 2008) National Thunderstorm
Forecast Guidance System (NTFGS) algorithm based on the Australian mesoscale
numerical model output. This outlines the high potential for heavy rain and damaging
winds associated with thunderstorms (Mills and Colquhoun, 1998).
102
Figure 3.44. Subjective surface analysis of the 01:00 UTC/16 November 2008 surface dew
point and boundaries as well as the 00:00 UTC/16 November 2008 winds at the 500 hPa
level for the Australian state of Queensland. Apparent are the higher surface dew points in
the coastal areas of SEQ in the presence of lifting mechanisms to the south and west of
Brisbane. Due to backed surface winds in the coastal areas combined with stronger flow
aloft, deep layer shear was maximized in the southeast corner of the state.
103
Figure 3.45. TITAN tracks up to 06:48 UTC/16 November 2008, the time of the destructive
surface winds at ―The Gap‖. Evident are the earlier storm crossing the state border into
Queensland which became supercellular west of the Gold Coast (―GOLD‖) and the new
round of severe convection to the west of the original storm.
104
Figure 3.46. The morning (00Z/16 November 2008) radiosonde ascent from Brisbane
airport.
3.7.2.5 Dual-Doppler results
The Gap storm passed through a dual-Doppler lobe of the Mt Stapylton and CP2 Doppler
radars at the peak of its life cycle. Dual-Doppler syntheses (see Section 1.2.3.2.2), focused on
The Gap storm about 20 minutes prior to the onset of damaging winds, show a very strong
updraft (upward vertical velocities >20 m s-1) and a well defined bounded weak echo region and
echo overhang (Figure 3.47a). A persistent storm-scale circulation of supercell strength allowed
the individual cell to persist for some extended time and also for the recirculation of falling ice to
produce large hail. Hail greater than 4 cm in diameter was reported from the nearby cells and the
polarimetric microphysical classifications (Vivekanandan et al. 1999) show significant areas of
large hail (not shown). This rotating updraft produces a significant local circulation with the
cyclonic rotation at 4 km above the ground clearly visible in Figure 3.47a. This rotation
corresponds to a relative vorticity of ~ 4x10-3 s−1. However, this rotation remained limited to the
mid-levels in this particular storm, without ever extending to near ground level. A possible
explanation for the absence of low level genesis of storm-scale rotation might be the significant
temperature dew point spread in the pre-storm environment (typical surface parcels of
33o C/17o C) which allowed the formation of relatively cold, non-buoyant outflows (Markowski
et al. 2002).
The situation at the time of the destructive surface winds (06:48 UTC, Figure 3.47b)
shows an intense downdraft with an amplitude > 20 m s-1, associated with very high reflectivity
values (>60 dBZ) near the surface. The high reflectivity values indicate the presence of large
hail near the surface. There is also a high likelihood that a very high precipitation (liquid and ice)
loading with attendant evaporation and sublimation in relatively dry environmental air
contributed to the severity of the downdraft. Given the strongest surface winds were located
105
close to the midlevel circulation center of the storm, storm-scale pressure gradients might have
dynamically contributed to a further strengthening of the surface winds. However, the damaging
surface winds did not seem to be associated with any noteworthy low-level circulation, but
seemed to be the result of a dynamically enhanced intense wet microburst.
In terms of the life cycle of The Gap storm, Figure 3.48 indicates that the most intense
surface flow was preceded by a descending maximum reflectivity core and large hail (> 70 dBZ)
reaching the surface. The figure further strengthens the hypothesis that a large descending hail
core is linked to the destructive surface winds at ―The Gap‖.
This event is being analyzed further to better understand the dynamics of the storm.
Severe events such as this covered by dual-Doppler and polarimetric radar are relatively rare.
Analysis of such events is crucial for improvements in severe weather forecasting, detection, and
warning.
106
a)
b)
Figure 3.47. a) Dual-Doppler reconstruction of mid-level circulation at 06:30 UTC showing
the reflectivity and the wind at a height of 4 km as well as an east-west cross-section
through the main updraft and b) near surface reflectivity and wind field at 06:48 UTC
showing the divergent burst of wind through The Gap. An east-west cross-section shows
the intense downdraft with an amplitude > 20 ms-1 corresponding to very high reflectivity
near the surface.
107
Figure 3.48. Lowest height of a 70 dBZ+ return from the Mt Stapylton radar near Brisbane
on 16 November 2008 (blue curve) and the height of the strongest returns (red curve) and
the value of those returns (green curve) highlighting the storm collapse. The strongest
winds recorded near ―The Gap‖ occurred around 06:48Z.
108
4 Aerosol and Microphysics Studies
4.1 Introduction
In order to best understand how seeding might affect precipitation in the region, it is
crucial to understand the natural precipitation formation processes that occur in the clouds that
will be seeded. Thus, this chapter describes the microphysical conditions observed in the region,
including subcloud/cloud base aerosol conditions that will affect the initial drop sizes in the
cloud and the subsequent microphysical processes that occur. Section 4.2 describes the cloud
base aerosol measurements from all cloud base measurements collected, including the chemical
composition of the measured aerosol for three cases of cloud base measurements. The chemical
composition influences the ability of the aerosol to serve as nuclei for cloud droplets, and thus is
an important aspect of the analysis to understand aerosol influences on cloud microphysics.
Section 4.3 discusses the in situ microphysical cloud measurements, which includes a summary
of the initial cloud base droplet spectra, and the subsequent microphysical droplet growth
evolution and ice formation for several cases. Characterizing the initial droplet spectra and
droplet growth is important to interpreting how hygroscopic seeding would be expected to affect
the subsequent droplet growth and rain formation.
4.1.1 Overview of results
The following conclusions are drawn from the detailed analyses presented in this chapter:
1. Of the four HYSPLIT regimes, the northeasterly regime, on average, had the highest
fraction of aerosol that served as cloud condensation nuclei (CCN) at the
0.3% supersaturation, while the westerly regime had the lowest. These results indicate
that a higher fraction of the maritime flow aerosol served as CCN (at 0.3%
supersaturation) than from the continental flow (NWly, Wly) days, especially so for the
westerly regime.
2. The maritime flow (NEly and Ely) regimes have a more robust coarse mode in their
aerosol size distributions, especially of particles >5 µm, compared to the continental
(NWly and Wly) regimes. The Ely regime tends to have the weakest fine mode, and thus
the overall cleanest conditions. This is expected given the flow in this regime is coming
straight in from the ocean. The NWly regime often has a high number of fine particles
and/or very few coarse particles. The results show, however, that coarse mode particles
in small concentrations were observed in all four regimes, although they were more
common in higher concentrations in the NEly and Ely regimes.
3. The CCN supersaturation cycle measurements also reiterate that the Ely regime is the
cleanest regime overall with the lowest mean CCN concentrations at all measured
supersaturations. There was a general tendency for the NEly, NWly, and Wly regimes to
have similar CCN levels at each supersaturation, all of which are roughly twice as high as
for the Ely regime.
109
4. The fine mode fraction of aerosols is dominated by sulfates and most of the CCN active
particles are ammonium sulfate. In one case, which corresponded to an Ely regime day,
the fine modes were dominated by small sea salt particles. The intermediate and coarse
mode aerosols are a mix of dust and other particles (aluminosilicates). In the other two
cases analyzed, there was no indication from any of the size fractions that the aerosol
samples were of marine origin, and appeared more continental, possibly having an
anthropogenic source (NEly regime day with back trajectory that passed over the city of
Brisbane) or biomass burning source (NWly regime day). The NEly day, however, did
have some sodium-bearing particles that may be aged sea salt.
5. The average maximum drop concentrations are similar for all regimes, however, ranging
between 500 and 650/cc, and the results show that on average, the Wly and NWly
regimes have the highest drop concentrations and smaller mean diameters, compared to
the maritime NEly and Ely regimes.
6. The NEly and Ely regimes have the highest concentrations of drops greater than 20 µm,
which can indicate a greater likelihood that collision and coalescence will be active in
clouds in this regime. This is also possibly related to the fact that both of these regimes
had a strong coarse mode of aerosol particles.
7. The tendencies in the results show that the mean diameter and concentration of large
drops (>20 µm) in each regime is larger and/or higher in seeded clouds than the natural
(non-seeded) clouds. This is consistent with the first step of the hygroscopic seeding
conceptual model and is also the same trend observed in the season one results.
8. During the early part of the season, when cloud bases are generally higher, coalescence
could be delayed, ice multiplication may not occur, and thus first ice will only form at
temperatures colder than −10o C. Hygroscopic seeding may be more effective in these
clouds by providing earlier coalescence and possibly the onset of a more efficient ice
process.
9. Clouds in Southeast Queensland generally develop precipitation initially via the ―warm
rain‖ process that then results in a more efficient mixed-phase process in deeper
convective systems that extend above the freezing level. Seeding with hygroscopic flares
could potentially enhance the ―warm rain‖ process, but glaciogenic seeding would not be
advised in these conditions.
10. These results would also explain the lack of intense lightning storms in the coastal
regions as observed during the field efforts, since liquid water content in the mixed-phase
region (necessary for storm electrification) is often rapidly depleted due to efficient
mixed-phase processes and ice multiplication.
11. Deep stratiform systems are occasionally observed in the region, yet our observations
indicate that the natural precipitation processes are very efficient in these systems. Thus,
neither hygroscopic or glaciogenic seeding may have any effect and glaciogenic seeding
may actually have a negative effect.
110
4.1.2 Impacts of aerosols on precipitation
Investigations of aerosols in the atmosphere are important for two reasons: 1) their
influence on climate and clouds, and 2) their health and environmental impacts. Our focus here is
their effect on clouds because aerosol particles affect the radiative properties and precipitation
efficiency of clouds. The background pollution levels will also impact the efficiency of rainfall
enhancement via cloud seeding with hygroscopic or glaciogenic flares.
Aerosol particles are injected into the atmosphere from either natural or anthropogenic
(human activity) sources. Aerosols also form in the atmosphere through gas-to-particle
conversion. Dry aerosol particles can be classified into three size categories, nucleation or
Aitken mode particles (radii <0.1 m), large or accumulation mode particles (0.1 < radii <1.0 m)
and giant particles (radii >1.0 m) (Pruppacher and Klett, 1998). Potential aerosol radiative
forcing and the efficiency of clouds to develop precipitation are complex functions of the size,
chemical composition and other physical properties of aerosols. An increase in the total number
concentration of particles in the accumulation mode at a given liquid water content, results in a
decrease in the mean droplet radius in the cloud and a corresponding increase in the cloud albedo
(the reflectance of radiation from the earth to space). In addition, the smaller cloud droplets will
have the effect of a less efficient precipitation process in clouds by inhibiting the warm-cloud
coalescence mechanism (cloud droplet collisions).
Different categories of aerosols affect clouds in different ways. Water-soluble aerosol
particles in the atmosphere consist mostly of sulfate derived from natural and anthropogenic
sources, and often serve as good sources of cloud condensation nuclei (CCN). Anthropogenic
sulfate in the atmosphere is produced mostly through gas-to-particle conversion of SO2 emitted
from fossil fuel combustion and metal smelting (Charlson et al., 1992). Emissions of crustal
aeolian dust material, often termed ―mineral dust‖, are caused primarily by surface winds acting
on dry soils where vegetation cover is or has become sparse (Tegen and Fung, 1995). The term
―mineral dust‖ refers to a large range of species that are highly variable in their chemical
composition and include such diverse compounds as quartz, clay, calcite, gypsum, hematite and
others. When mineral dust exists in conjunction with other pollutants such as sulfates, the dust
particles can be coated with the sulfate and become more active as CCN. This is especially
important in desert regions where both sources exist and have a likely impact on the precipitation
processes in clouds. In some instances, the coated dust particles could enhance the precipitation
efficiency of clouds, while in others they may have a negative impact.
Recent research has suggested an important role for organic aerosols as CCN (Cruz and
Pandis, 1998; Facchini et al., 1999; Liu et al., 1996; Rivera-Carpio et al., 1996; Shantz et al.,
2003; Raymond and Pandis, 2002). Organic compounds either in mixed or free state, are a major
source of atmospheric aerosol particles (Novakov and Penner, 1993). Novakov and Penner (1993)
calculated that approximately 65% of the CCN concentrations measured at 0.5% supersaturation
in the marine atmosphere were due to organic aerosols. Cruz and Pandis (1997) implied that two
hydrophilic compounds can be good CCN, while hydrophobic compounds are not good CCN.
Raymond and Pandis (2002) showed how secondary organic aerosols with low carbon numbers
and various functional groups are an excellent source of CCN, whereas longer carbon chains
with fewer functional groups tend to be inactive under typical atmospheric conditions. Organic
compounds in the atmosphere are very important; however, uncertainty about the chemical
111
speciation and the solubility of the large organic component precludes a rigorous analysis of its
contribution to nucleation activity. Recent studies in the Amazon have indicated that forests may
be an important natural source of organic nuclei and this could be important for the forested
regions of southeast Queensland as well. Guenther et al. (2006) in fact showed that eucalyptus
forests along the eastern coast of Australia are strong sources of isoprene, a volatile organic
compound that promotes the production of sulfates.
4.2 Aerosols in the Atmosphere and the Impacts on Droplet Size
Distributions and Precipitation Development
4.2.1 General aerosol characterization
Cloud base aerosol concentrations and spectra have been compiled for all cloud base
flight segments that had good DMA, PCASP (SPP-200), and FSSP (SPP-100) data (see Table
1.1 for more information on the instrumentation), and then compared for each regime defined in
Section 3.6.1 to investigate how air mass origin and path influences the cloud base aerosol
conditions at the measurement location. By characterizing the cloud base aerosol and
microphysical conditions for similar wind (trajectory) regimes, we can better understand the
natural microphysical response to cloud base aerosol, and look for statistical changes to that
response in cases which have been seeded, as well as to possibly predict when environmental and
aerosol conditions are most amenable to seeding.
CCN measurements were collected with a CCN counter that was run continuously at the
0.3% supersaturation throughout all flight operations, except under certain conditions when it
would be cycled through additional supersaturations of 0.5% and 0.8%. The conditions for the
supersaturation cycle were that the aircraft had to be flying at constant altitude, out of cloud and
usually below cloud base, for at least 11 minutes (the duration of the CCN supersaturation cycle).
As a result, regular cloud base measurements (just below cloud base and out of cloud) were
taken at the 0.3% supersaturation, but only a few measurements at the higher supersaturations
were conducted. Furthermore, for inclusion in this analysis, the measurements for each
supersaturation had to be within 10% of the actual desired supersaturation2. The median CCN
concentrations during the time that the measurements met this criteria for each supersaturation
were included in the subsequent analysis.
All aerosol and CCN data presented herein were collected at cloud base (just below and
out of cloud). The median PCASP concentrations for each cloud base segment are depicted in
Figure 4.1 and Figure 4.2 and summarized in Table 4.1. The easterly (E) regime is typically very
clean with the lowest mean aerosol concentration and variability, while the westerly (W) and
northwesterly (NW) regimes are more polluted, ranging up to nearly 2500/cc (Figure 4.2). Note
in Figure 4.2 how the northeasterly (NE) regime is highly variable, yet mimics the easterly
2
There is a time delay for the temperature plates which control supersaturation in the instrument
to heat/cool to achieve the desired supersaturation, thus resulting in unreliable measurements
until the actual supersaturation is within an allowed range of the desired supersaturation.
112
regime when it has similar low cloud bases (warmer cloud base temperatures) and resembles its
continental counterparts at higher cloud base (colder cloud base temperatures). This is possibly
due to the proximity of the measurement to the coast, whereas farther inland the cloud bases and
aerosol concentration are higher. The outlier (very low) PCASP concentration in the W regime
(see Figure 4.2) is also an outlier in its HYSPLIT trajectory (see purple E trajectory in the W
regime in Figure 4.1). This is an example of a trajectory that was seemingly misclassified by the
objective HYSPLIT regime classification described in Section 3.6.1. Otherwise, the HYSPLIT
regimes resemble similar paths and have similar aerosol characteristics.
Figure 4.1. Maps of HYSPLIT trajectories for each cloud base measurement by HYSPLIT
regime, and color-coded by the median PCASP concentration of each cloud base flight
segment.
113
Figure 4.2. Median PCASP concentration of each cloud base flight segment vs cloud base
temperature, and color-coded by HYSPLIT regime.
Table 4.1. Statistics of cloud base aerosol concentrations for each HYSPLIT regime: mean
and standard deviation of the median PCASP concentration (see Figure 4.2), and ratios of
the 0.3% supersaturation CCN to PCASP median concentrations, 0.3% supersaturation
CCN to all aerosol >0.1 µm median concentrations, and fine (0.01–1.0 µm) to coarse
(>=1.0 µm) mode and accumulation (0.1–1.0 µm) to coarse mode aerosol median
concentrations (see Figure 4.3).
NE
E
NW
W
Mean PCASP (/cc)
678
345
1005
1031
Std Dev
470.6
159.5
420.0
446.8
CCN:PCASP CCN:All >0.1µm Fine:Coarse Accum:Coarse
490.5
0.47
0.45
379.4
477.8
0.41
0.38
258.7
2084.1
0.38
0.37
1466.5
1705.7
0.27
0.27
670.0
114
a)
b)
c)
d)
Figure 4.3. Scatter plots of a) 0.3% supersaturation CCN vs PCASP concentrations,
b) 0.3% supersaturation CCN vs all aerosol >0.1 µm concentrations, c) fine mode aerosol
(0.01–1.0 µm) vs coarse mode (>=1.0 µm) concentrations, and d) accumulation mode
aerosol (0.1–1.0 µm) vs coarse mode (>=1.0 µm) concentrations color-coded by HYSPLIT
regime (see legend).
The fraction of PCASP aerosol that served as CCN at the 0.3% supersaturation was
highest on average for the NE regime and lowest for the W regime, and similar results were
found for the ratio of 0.3% supersaturation CCN to all aerosol (>0.1 µm; Table 4.1). Figure
4.3a-b illustrates these relationships for every cloud base measurement, and it is clear that in
most regimes the ratio lies roughly around 0.4, except for several points in the W regime (and
some of the NW regime) that have much lower CCN fractions. Thus, these results indicate that a
higher fraction of the maritime flow aerosol serve as CCN (at 0.3% supersaturation) than from
the continental flow (NW, W) days, especially so for the W regime. Given that aerosol size and
chemical composition both are strong factors that influence their ability to serve as CCN, and we
are comparing concentrations of the same aerosol size ranges between regimes, it is possible that
the chemical composition (or other physical properties) of the aerosol from the maritime to
continental regimes vary such that a higher fraction of the maritime flow aerosol serve as CCN at
the 0.3% supersaturation.
115
The aerosol size distributions (ASDs) were created by combining aerosol measurements
from the DMA, PCASP (SPP-200), and FSSP (SPP-100) for each cloud base segment in season
two. The ASDs for each regime are presented in Figure 4.4 with the geometric mean ASD for
each regime overlaid in thick blue. The corresponding 3-mode lognormal fits for each regime‘s
mean ASD is shown in Figure 4.5 with the statistics summarized in Table 4.2. The mean ASDs
in each regime illustrate some key differences between the regimes. A few features to note are
that the NE and E regimes have a more robust coarse mode, especially of particles >5 µm
compared to the NW and W regimes. The individual measurements are fairly noisy in the coarse
mode range, however, for all regimes. The fine mode portions of the ASDs are similarly noisy
among all regimes, although the fitted ASD statistics indicate that the E regime tends to have the
lowest fine and accumulate mode (Modes 1 and 2, respectively) concentrations, and thus the
overall cleanest conditions (Table 4.2). This is expected given the flow in this regime is coming
straight in from the ocean.
As another way to quantitatively compare the regimes‘ ASDs, we calculated the ratio of
fine mode aerosol to coarse mode (defined as all aerosol smaller or larger than 1µm,
respectively). This ratio is highest on average for the NW regime and lowest for the NE and E
regimes (Table 4.1). This indicates that the NW regime often has a high number of fine particles
and/or very few coarse particles, while the NE and E (maritime flow) regimes have the least
number of fine particles and/or a larger fraction of coarse particles (perhaps aged sea salts). The
W regime, however, was more variable and had a few outlier days with very high concentrations
of fine aerosol (Figure 4.3c, Figure 4.4). Thus, we have also studied the ratio of accumulation
mode aerosol (0.1–1 µm) to coarse mode, and found similar relationships between the two size
modes for each regime, without the outliers3. These results show that coarse mode particles in
small concentrations were observed in all four regimes, however, were more common in the NE
and E regimes (Figure 4.3d).
The CCN supersaturation cycle measurements also reiterate that the E regime is the
cleanest regime overall with the lowest mean CCN concentrations at all measured
supersaturations (Figure 4.6, Table 4.3). Given the small sample size for the other three regimes
at higher supersaturations, it is not feasible to confidently characterize their CCN supersaturation
spectra, however, the general tendencies are for the NE, NW, and W regimes to have similar
CCN levels at each supersaturation that are roughly twice as high as for the E regime. Recall
that the NE regime had the highest ratio of 0.3% CCN to aerosol (Table 4.1, Figure 4.3), thus
even though this regime had similar CCN concentrations to the two continental regimes, the
fraction of aerosol that served as CCN was more similar to that in the E (maritime) regime, thus
suggesting the chemical composition of the aerosol in this regime differs from those that result in
similar CCN concentrations in the continental regimes.
3
The outliers with very high fine aerosol measurements, seen as the W regime aerosol size
spectra in Figure 4.4 that reach nearly 105 in the smallest size bin (0.01 µm) may be days with
bad DMA data or anomalous days worthy of case study. Given the consistency of PCASP and
DMA measurements in these cases, we have confidence in the DMA data.
116
Figure 4.4. Aerosol size distributions (ASDs) for each cloud base measurement grouped by
HYSPLIT regime (note the log scales). The geometric mean ASD is overlaid in thick blue.
Figure 4.5. 3-mode lognormal fits for each regime mean ASD.
117
Table 4.2. Summary of aerosol size distribution statistics for each of the three lognormal
fit modes for each regime.
Mode 1
Mode 2
Mode 3
Diameter µm
0.022
0.132
1.53
Number Conc (/cc)
5385
1540
4.91
Standard deviation
2.54
1.62
2.27
Diameter µm
0.025
0.128
1.89
Number Conc (/cc)
3530
890
2.22
Standard deviation
2.12
1.63
2.08
Diameter µm
0.031
0.132
1.18
Number Conc (/cc)
6250
2290
1.47
Standard deviation
2.10
1.68
2.27
Diameter µm
0.031
0.126
1.22
Number Conc (/cc)
6055
2935
1.63
Standard deviation
1.76
1.69
2.33
NorthEast
East
NorthWest
West
118
Figure 4.6. Summary of season two cloud base CCN measurements at varying
supersaturation, color-coded by HYSPLIT regime as in Figure 4.2 (NE: cyan, E: green,
NW: red, W: black). The lines connect the mean CCN concentration at each
supersaturation for each regime.
Table 4.3. Summary of CCN concentrations per regime for each supersaturation level.
The mean CCN concentrations are an average of all the individual cloud base segments’
median CCN concentrations for each regime. (Note: The median concentration is taken
over the course of the time segment where the actual instrument supersaturation was
within 10% of the desired supersaturation.) The instrument was routinely run at 0.3% SS,
and only occasionally conducted a cycle through 0.5% and 0.8% also, thus it should be
noted that there are fewer samples in each mean at the higher supersaturations.
NE
E
NW
W
0.3% supersaturation
Mean Conc . # samples
(/cc)
285
22
146
40
349
41
247
27
0.5% supersaturation
Mean Conc . # samples
(/cc)
455
2
253
10
587
5
533
4
0.8% supersaturation
Mean Conc .
#
(/cc)
samples
1133
1
546
5
757
2
1098
2
4.2.2 Chemical composition of aerosol
Samples were collected using MPS-3 samplers, which contain three stages that are used to
collect particles from diameters of 0.5 to 3 microns. We will call these stages 3, 2, and 1, and
119
their 50% cut-off aerodynamic diameters are 0.05, 0.3, and 2.0 µm, respectively. Particles were
analyzed for composition using energy dispersive spectroscopy (EDS) and their hygroscopic
properties were measured using environmental transmission electron microscopy (ETEM).
Samples have been analyzed for three cases from measurements taken on 22 January,
23 January, and 26 January at cloud base (Figure 4.7, Table 4.4). The HYSPLIT back
trajectories for these three cases are illustrated in Figure 4.8, and the CCN supersaturation
spectra are presented in Figure 4.9.
a)
b)
c)
Figure 4.7. Aerosol size distribution created from the DMA, PCASP, and FSSP
measurements for a) 01:15:49–01:24:22 UTC on 22 January 2009,
b) 02:58:53–03:10:16 UTC on 23 January 2009, and c) 03:17:01–03:25:33 UTC on
26 January 2009.
Table 4.4. Summary of aerosol size distribution statistics for each of the four lognormal fit
modes for each case study.
Mode 1
Mode 2
Mode 3
Mode 4
Diameter µm
0.012
0.082
0.157
1.70
Number Conc (/cc)
12365
355
416
0.90
Standard deviation
2.030
1.364
1.513
2.20
0.0538
0.278
2.881
--
Number Conc (/cc)
3750
56.41
0.964
--
Standard deviation
2.0353
1.373
1.877
--
Diameter µm
0.025
0.152
2.4
6.0
Number Conc (/cc)
4530
242
9.25
1.2
Standard deviation
1.9
1.608
1.8
2.1
22 January 2009
23 January 2009
Diameter µm
26 January 2009
120
Figure 4.8. Map of the Brisbane region illustrating HYSPLIT back trajectories for each of
the three January 2009 cloud base aerosol filter sample case studies: 22nd (cyan; NE
regime), 23rd (red; NW regime), and the 26th (green; E regime).
Figure 4.9. CCN mean concentrations over a cycle of supersaturations (data points at 0.3,
0.5, 0.8%) at cloud base from: 01:15:00–01:25:00 UTC on 22 January 2009 (cyan; NE
regime), 02:52:07–03:10:43 UTC on 23 January 2009 (red; NW regime), and
03:17:01–03:25:33 UTC on 26 January 2009 (green; E regime). (Note: Up to 20%
supersaturation error was allowed on these cases to have data points at all three
supersaturations.)
121
4.2.2.1 22 January 2009
On 22 January samples were collected at 3937 ft from 01:15:20 to 01:31:26 UTC below
cloud base (see Figure 4.7a, Table 4.4).
Stage 3 – Samples collected on January 22 contain a greater number of particles with
diameters less than 0.01 µm (Figure 4.7a) and higher sulfur (55%) content than particles
collected on the 23rd and 26th. Two types of sulfur-bearing particles were detected. Both
occurred as either externally mixed grains or as coatings on mineral particles. The first type
shows only S and O in the EDS signal (Figure 4.10a), is sensitive to the electron beam, and is
likely NH4SO4. It deliquesces at a relative humidity (RH) of 81%, consistent with literature
values for the DRH of NH4SO4 (DRH 80%) (Figure 4.10c). The second type of sulfur-bearing
particle, making up 13% of the analyzed particles, contains Mg, O, S, and in some cases F
(Figure 4.10d). These particles give no diffraction spots or rings and thus are presumably
amorphous. No morphological changes were observed at increasing RH (Figure 4.10f).
Other species collected in this size range include carbon in the form of fractal soot
aggregates (12%) and mineral particles (22%) that commonly include O, Si, Al, K, Mg, and Na.
They are likely alkali aluminosilicates, presumably clays, feldspars, or both. Some
aluminosilicates also contain Ti and Fe.
Stage 2 – Of 100 particles analyzed 53% were uncoated aluminosilicates. A small number
(5%) of analyzed aluminosilicate-bearing particles were surrounded by a coating containing Mg,
Na, S and O, likely Mg or Na sulfate mixed with NH4SO4 (Figure 4.11, Figure 4.12a). In most
cases these coatings are spread along the bars of the TEM grid and fill the corners of the holes,
which may suggest that they were hydrated at the time of arrival on the grid (Figure 4.11a-b).
About 25% of particles contained sodium sulfate or sodium nitrate. These sodium-bearing
particles may be aged sea salt.
Grains containing Fe, Ti, and O coated with a Na, Cl, Mg-bearing material (Figure 4.11c)
were detected. This coating was beam-sensitive and amorphous with low levels of S. Grains
composed of K2SO4 and those having a core and coating of Na, Mg, O, and S were also
identified. These coatings were beam-sensitive, suggesting that they may also contain NH4SO4
mixed with Na2SO4 or MgSO4. The coatings grew hygroscopically at high RH, without
achieving a fixed deliquescence point (Figure 4.12a-c). Other grains in this size fraction show
minor morphological changes at high RH (Figure 4.12d-f, g-i). Several aerosol particles were
composed of Na, Mg, and Ca sulfates attached to aluminosilicate grains. In some cases these
aluminosilicate grains contain P, Ti, and Fe and are surrounded by carbonaceous coatings that
have small (20 to 50 nm) beam-sensitive inclusions containing O, Na, Mg, and S.
Stage 1 – Coarse-mode particles include Na2SO4, K-salts, and aluminosilicates (Figure
4.13). The K-salts grew hygroscopically at increasing RH (Figure 4.13a-c), and the Na2SO4
particles deliquesced at 83% (Figure 4.13f). Several other particles contain Na, Cl, and O and are
coated by Na and O with low levels of S. From a selection of 30 grains, only 20% showed
changes with increasing RH. In two of these cases a hydrophobic material coated hygroscopic
inclusions. The inclusion changed with increasing RH, but the coating remained unchanged and
possibly inhibited the growth of the inclusion (Figure 4.13g-i).
122
There is no indication from any of the size fractions that this aerosol sample is of marine
origin. The S-bearing particles in the fine fraction likely originated from a continental source –
they are not mixed with K or organics that would suggest they originated from biomass burning.
Figure 4.10. Two grid areas (a to c, and d to f) that contain fine particles collected on
22 January 2009 and exposed to increasing RH (b, c, and e, f) using the ETEM. The black
arrows in frames b, c, e, and f indicate that the RH was increasing when the image was
acquired. The element contents of the particles are shown by the black arrows in a and d.
Red (a to c) and yellow (d to f) arrows point to the same particles and show changes in
particle morphology as the RH was increased.
Figure 4.11. Examples of aerosol particles collected on stage 2 on 22 January. The black
arrows show the element contents of the particles.
123
Figure 4.12. Three particles (a to c, d to f, g to i) collected on stage 2 on 22 January and
exposed to increasing RH. The arrows in frames b and c indicate whether the RH was
increasing when the image was acquired. Black arrows in frames a, d, and g indicate the
element contents of the particles. White arrows show changes in particle morphology as
the RH was increased. The white dotted circle in a, b, and c indicates the area of the
particle and coating.
124
Figure 4.13. Three particles (a to c, d to f, g to i) collected on stage 1 on 22 January and
exposed to increasing RH. The black arrows in frames b, c, e, f, h, and i indicate that the
RH was increasing when the image was acquired. Black arrows in frames a, d, and g
indicate the element contents of the particles. White arrows show changes in particle
morphology as the RH was increased.
4.2.2.2 23 January 2009
The samples on 23 January 2009 were collected at 3530 ft from 02:57:04 to 03:10:30
UTC below the cloud base (see Table 4.4).
125
Stage 3 – It is likely that aerosol samples collected in the fine fraction on the 23rd
originated from a biomass-burning source. K-salts, including KCl, are the most abundant
inorganic particles in the fine fraction. Other elements detected include Na, Cl, and O with minor
Ca (Figure 4.14). Particles range in diameter from 20 nm to 1.5 µm, with most from 100 to
600 nm (Figure 4.14a). Most of the KCl grains have euhedral morphologies, but some are
rounded. Some of the rounded particles containing K and Cl are more beam sensitive than KCl
(Figure 4.14d). These are likely mixtures of KCl and NH4Cl that formed through reactions of
chlorine and NH3 emitted from the fire. Fine K-bearing particles not attributable to mineral dust
or sea salt are indicators of biomass burning. CaSO4 and K2SO4 have been observed in smoke
from African savanna and wetlands fires (Gaudichet et al. 1995).
This sample showed little growth with increasing humidity. Deliquescence occurred for a
fraction of particles at RH values of 75% and 86% (Figure 4.15), consistent with the DRH of
NaCl and KCl. These inorganic salts occur with carbonaceous material having diameters less
than 100 nm. Other grains collected on the grid contain Si, Al, O, and Fe.
Stage 2 – Particles collected on stage 2 are mainly aluminosilicates, some of which
contain F, Fe, and Ti. Particles showing only Fe and O in EDS, presumably magnetite, hematite,
or an Fe hydroxide also occur. These Fe-bearing particles are either externally mixed or attached
to S-bearing material, likely NH4SO4 (Figure 4.16g). During the ETEM experiment, the NH4SO4
deliquesced, but the Fe oxide(s) remained unchanged (Figure 4.16g-i). Other deliquescent
particles include NaCl mixed with MgSO4. K, Na, and O, likely in the form of NaNO3 or KNO3,
coat several NaCl grains; these particles may be of marine origin (Figure 4.16a, Figure 4.17a,d)
and grew hygroscopically at higher RH, with morphological changes appearing in the coating
before the entire particle dissolved (Figure 4.16a-c).
Externally mixed grains that contain F were also detected (Figure 4.17b). These particles
are euhedral and decomposed under the electron beam. F-bearing particles that can exist in the
atmosphere include NH4F, NaF, and KF. The lack of any other chemical signals in the EDS
spectra suggests that they may be NH4F, but further work is needed to verify. The major
industrial sources of F release into the environment are via phosphate production and Al
manufacture (Tsai 2007). Sixteen particles were chosen for ETEM analysis. The nine that
deliquesced had coatings that contain S or N (Figure 4.16).
Stage 1 – Most coarse particles are aluminosilicates, similar to those collected on stage 1
on January 22.
Summary - The composition, size, and morphology of the fine fraction suggest that this
aerosol sample was emitted from anthropogenic or biomass-burning sources. There is little soot,
organics, or aged K-salts (K2SO4 or KNO3) on the grid so we assume that this aerosol sample
was collected from a freshly emitted burning source. The larger size fractions contain mainly
mineral dust particles. Some of the samples collected on stages 1 and 2 were coated with
sulfates, which might indicate they experienced atmospheric processing prior to collection.
126
Figure 4.14. Examples of aerosol particles collected on stage 3 on 23 January. Image b
shows a high-magnification image of the area surrounded by the red box in a. The black
arrows show the element contents of the particles.
127
Figure 4.15. Three areas (a to c, d to f, g to i) of the TEM grid from stage 3 on 23 January
exposed to increasing RH. The black arrows in frames b, c, e, f, h, and i indicate that the
RH was increasing when the image was acquitted. Black arrows in frames a, d, and g
indicate the element contents of the particles. Blue (a, b, and c), red (d, e, and f) and yellow
(g, h, and i) arrows point to the same particles and show changes in particle morphology as
the RH is increased. A red box in images d, e, and f highlights an area where changes in
particle morphology are observed.
128
Figure 4.16. Three areas (a to c, d to f, g to i) of the TEM grid collected on stage 2 on
23 January and exposed to increasing RH. The black arrows in frames b, c, e, f, h, and i
indicate that the RH was increasing when the image was acquired. Black arrows in frames
a, d, and g indicate the element contents of the particles. White arrows show changes in
particle morphology as the RH was increased.
129
Figure 4.17. Examples of aerosol particles collected on stage 2 on 23 January. The black
arrows show the element contents of the particles.
4.2.2.3 26 January 2009
Samples were collected on 26 January 2009 at 3400 ft from 03:16:00 to 03:26:00 UTC
below cloud base (see Table 4.4).
Stage 3 – Most particles in the fine fraction are of marine origin. The fine fraction was
90% dominated by fine sea salt particles. All of these particles were freshly emitted with no or
little signs of aging. Most sea salt particles consisted of a NaCl-rich core with a Mg-rich coating.
This coating is natural and not a result of aging; only a result of the drying process of a sea salt
droplet. Grain sizes range from 0.2 to 1.5 µm, and most are between 0.2 and 0.5 µm. Most NaCl
particles are internally mixed with Mg- and Ca-sulfates, presumably from evaporation of sea salt
droplets (Figure 4.18). A small number of particles contain Al, O, Si, and S and are presumably
aluminosilicates coated with sulfate.
Stage 2 – The particles from this stage differ significantly from those in the fine fraction.
Most particles are mineral grains, with only a few sea salt grains (Figure 4.19). CaCO3 is the
most abundant particle type. Other types of particles contain Mg, S, and O or an amorphous
carbonaceous material.
130
Stage 1 – Only few grains were collected in the coarse fraction. The particles that were
analyzed contain various quantities of Fe, V, Al, Ag, and O (Figure 4.20). An ETEM experiment
was not carried out on this sample, but because of the morphology and composition of the
particles, we believe that it is unlikely they will display significant, if any, hygroscopic growth.
Figure 4.18. Examples of aerosol particles collected on stage 3 on 26 January. The black
arrows show the element contents of the particles.
131
Figure 4.19. Examples of aerosol particles collected on stage 2 on 26 January. The black
arrows show the element contents of the particles.
Figure 4.20. Examples of aerosol particles collected on stage 1 on 26 January. The black
arrows show the element contents of the particles.
4.2.3 Summary of aerosol conditions
It is clear that even within one week‘s time, the thermodynamic conditions in the
Brisbane region vary considerably, resulting in variations such as a maritime versus continental
source to the subcloud air mass and varying degrees of instability (not shown). The presence of
a subsidence inversion is a major limiting factor in the vertical development of convection, and
132
the presence of this inversion, as well as its height, varies considerably throughout the wet
season in Brisbane. On most days when it is at low levels and quite strong, flight operations
were not conducted due to the lack of convection. Thus, this set of case studies reflects more of
the unstable days when convection, and hence flight operations, occurred.
The resulting aerosol conditions also vary dramatically. The 22 January ASD had a high primary
mode concentration, with a very small modal diameter, indicating high concentrations of very
fine particles (Figure 4.7a). The trajectory over the Brisbane metro area may have influenced
this feature of the aerosol spectra on that day (Figure 4.8). As a result, the majority of those fine
particles were not easily activated as CCN, and thus CCN concentrations were not as high as
they were for the more continental case of the 23rd. The coarse mode on 23 January was at least
an order of magnitude less than that of the other cases, and its air mass had the most continental
source (Figure 4.8), and the resulting CCN concentrations were highest on this day (Figure 4.9).
The sample on 26 January had a strong coarse mode, and had the most maritime source to the air
mass (Figure 4.8). CCN concentrations on this maritime flow day were the lowest of all cases
studied (Figure 4.9). This was also the only day that contained particles originating from the
sea, but only particles collected in the fine fraction contained these sea salt-bearing particles.
The middle and coarse fraction of particles collected on 26 January only contained mineral dust
particles. It is not clear at this point why there were not larger particles containing sea salt
collected during this period. One possibility would be that the larger sea salt particles were
short-lived and just not sampled.
The chemical composition analysis illustrated that the fine fractions from all three days
(only 22, 23, and 26 January have been analyzed so far) display greater variability than the
coarser fractions. The fine fraction on 22 January was dominated by sulfate, soot, and mineral
dust, while the 23rd contained K- and Na- Cl particles mixed with carbonaceous compounds, and
the 26th was dominated by NaCl particles internally mixed with Ca- and Mg- sulfates, with a
small number of mineral dust particles. The S-bearing particles on both 22 and 23 January were
mixed with fractal soot aggregates, mineral dust, and organic carbon and are not likely to be of
marine origin unlike those collected on the 26th, which only contained NaCl particles with Caand Mg- sulfates.
Samples from stages 2 and 1 on all three days were similar with a high number of
aluminosilicates, some of which also contained Fe and Ti. Particles collected on the 22nd were
coated with organics and sulfates that indicate that the aerosol has undergone some atmospheric
processing. This was not evident for the aerosol particles collected on the 23rd and 26th.
4.3 Observed Microphysical Characteristics of Clouds in Southeast
Queensland
4.3.1 Precipitation development in clouds
Precipitation initiation and development can proceed via several physical paths (Figure
4.21). These involve various microphysical processes, which proceed simultaneously, but at
different rates, resulting in one path becoming dominant because of its greater efficiency under
given atmospheric conditions. The efficiency with which clouds produce rain at the surface
133
varies greatly. Precipitation efficiency, defined as the ratio of the rate of rain reaching the
ground to the flux of water vapor passing through cloud base (Marwitz et al., 1972), can range
from zero in non-precipitating clouds to greater than unity for short times in very intense,
convective systems (Cotton and Anthes, 1989). Some of the earliest studies showed that ordinary
thunderstorms transform less than 20% of the in-flux of water vapor into rain on the ground
(Braham, 1952). The principles of most, if not all, precipitation enhancement hypotheses are
rooted in these efficiency factors, which in general, seek to improve the effectiveness of the
precipitation evolution path. The seeding conceptual model (physical hypothesis; see Section
5.1.2) describes how this is accomplished by the seeding intervention, and specifically how the
initiation and development of precipitation in seeded clouds differs from that in unseeded clouds
and ultimately affects the dynamics of the cloud.
Figure 4.21. Various pathways by which water vapor is transformed into various types of
cloud particles and precipitation. Adapted from Houze (1993, p. 96).
Precipitation formation mechanisms can differ dramatically from one location to another,
and even at one location, depending on the meteorological setting and stage of cloud evolution.
Precipitation growth can either take place through coalescence or the ice process or a
combination of the two (Figure 4.21). In clouds with tops warmer than 0o C, precipitation can
develop only by means of the coalescence process. Clouds are further categorized as either
continental or maritime, which describes their degree of colloidal instability. However, when
134
cloud tops reach temperatures colder than 0o C, ice may develop and precipitation can form
through a different set of paths, as is depicted in Figure 4.21.
The number concentration and size spectrum of cloud droplets will influence the cloud‘s
precipitation efficiency and can also vary dramatically, depending on the CCN size distribution.
A maritime droplet spectrum will tend to consist of fewer particles but more large drops than in a
continental spectrum (Pruppacher and Klett, 1998) leading to a more efficient precipitation
process. The vertical depth of the cloud also impacts how efficient it is at precipitation
production (Cotton and Anthes, 1989). The deeper the cloud, the more likely precipitation will
form. One way of inferring the depth of the cloud is to look at the horizontal scale of the
convective core, defined by an elevated region of liquid water content and associated with the
peak vertical velocities in the cloud. Generally, a cloud with an adiabatic core greater than 2 km
is likely to form precipitation, with the likelihood increasing as the maximum liquid water
content increases. Another generality stated in Cotton and Anthes (1989) is that unless a cloud
produces liquid water contents of 0.5 g m-3, it is unlikely to precipitate.
4.3.2 Cloud droplet spectra characteristics and evolution in different types of cloud
systems and impacts for seeding
Cloud base droplet measurements were collected regularly during the season two field
program at roughly 1000 feet above cloud base (sometimes higher if terrain height restricted
flight altitude, thus we have included declared ―cloud base penetrations‖ up to 2000 feet of cloud
base in the analysis). They have also been grouped into similar wind regimes as the cloud base
aerosol measurements in Section 4.2.1. The total droplet concentrations range from 100–900/cc
(Figure 4.22). The average maximum drop concentrations are similar for all regimes, however,
ranging from 500–650/cc. The drop size distributions (DSDs) for each regime are shown in
Figure 4.23, and illustrate that the NE and E regimes are the most variable, while the NW regime
drop spectra are quite consistent. Note that some regimes (such as the W regime) have very few
samples of the initial cloud base droplet spectra, so we have less confidence in the statistics of
the spectra for these regimes. Nonetheless, these results show that on average, the W and NW
regimes have the highest drop concentrations and smaller mean diameters, compared to the
maritime NE and E regimes (Table 4.5). Furthermore, the NE and E regimes have the highest
concentrations of drops greater than 20 µm, which can indicate a greater likelihood that collision
and coalescence will be active in clouds in this regime. This is also possibly related to the fact
that both of these regimes had a strong coarse mode of aerosol particles (recall Table 4.1, Figure
4.4).
135
Figure 4.22. Maximum cloud base FSSP (SPP-100) concentration vs temperature (the
average temperature during the cloud base penetration segment), color-coded by
HYSPLIT regime.
136
Figure 4.23. Drop size distributions (DSDs) for each cloud base penetration (of natural,
non-seeded clouds) grouped by HYSPLIT regime. The mean DSD for each regime is
overlaid in thick blue.
Table 4.5. Summary of mean cloud base droplet statistics per HYSPLIT regime, including
the mean maximum and standard deviation SPP-100 FSSP concentration, the mean and
standard deviation mean drop diameter, and mean concentration of drops >20 µm (/cc).
NE
E
NW
W
Max FSSP (/cc)
513
534
646
564
Std Dev Mean Drop Diam. (µm) Std Dev Mean Conc >20
157.5
12.01
1.34
6.94
211.9
12.11
1.18
6.20
192.7
11.32
0.07
2.63
176.5
11.62
1.16
4.63
The initial cloud base DSDs have also been compiled for all seeded clouds in season two
(Figure 4.24, Table 4.6). Even though some regimes have very few samples, the tendencies of
these results show that the mean diameter and concentration of large drops (>20 µm) in each
regime is larger and/or higher in seeded clouds than the natural (non-seeded) clouds presented in
Table 4.5. This is consistent with the first step of the hygroscopic seeding conceptual model (see
137
Section 5.1.2) and is also the same trend observed in the season one results (see Section 5.3.2).
Furthermore, calculating these mean differences by separating the measurements by regimes that
should have similar natural aerosol and drop spectra characteristics, attempts to reduce some of
the natural variability and adds confidence to the observed trends. A Wilcoxon rank-sum test on
the mean diameters of the overall seeded and unseeded populations indicated a p-value of 0.018,
signifying support for the alternate hypothesis that mean diameters are larger in the seeded
population than in the unseeded. Similarly for concentration, the same test indicated a
p-value of 0.040, supporting the hypothesis that drop concentration is smaller in the seeded
population than in the unseeded. Additionally, the variance of the total seeded and unseeded
populations illustrated that seeded clouds may have broader cloud droplet spectra in accordance
with the hygroscopic seeding model but not significantly at the 95% level (p-value = 0.067).
Similar tests on differences between unseeded and seeded populations within regimes suffer
from sample size and are not statistically significant.
Figure 4.24. Same as Figure 4.23, except for seeded clouds.
138
Table 4.6. Same as Table 4.5, except for seeded clouds.
NE
E
NW
W
Max FSSP (/cc)
426
359
608
531
Std Dev Mean Drop Diam. (µm) Std Dev Mean Conc >20
8.56
131.1
12.41
1.26
8.89
102.4
13.09
1.31
12.29
133.9
12.63
1.68
6.11
130.66
12.29
1.64
4.3.3 Primary and secondary ice production
The ―warm rain‖ process was often active in the Southeast Queensland clouds, primarily
due to the warm cloud bases and thus large vertical extent of cloud warmer than 0o C (several
kilometers). However, on several occasions deep convection with a combination of warm and
mixed-phase processes was observed, although the frequency of these events was lower than for
the ―warm rain‖ processes only. The existence and efficiency of the condensation/coalescence
process also has important implications on the evolution of the mixed-phase processes.
The most important elementary ice processes that are sensitive to the cloud droplet size
distributions are riming rates and secondary ice generation. The earlier formation of drizzle or
raindrops in natural or seeded storms may provide the initial graupel embryos through the
freezing of these large drops. Larger drops freeze preferentially at temperatures below 0o C
(usually between the −5 and −10o C levels, while primary nucleation usually occurs around
−10o C (or colder) and these frozen drops provide more efficient graupel embryos than graupel
formation through the primary ice nucleation process.
The accretion of cloud droplets by ice crystals is subject to alterations due to changes in
cloud droplet size distributions and the associated changes in collection efficiencies. To the
extent that more maritime or hygroscopic seeding either increases the concentrations of the cloud
droplets at the large end of the spectrum, or shifts the whole spectrum to larger sizes, it will
increase the collection efficiencies of droplets on ice particles.
The breakup of ice crystals and the fragmentation of rime clusters on ice crystals are both
subject to changes as a consequence of broadened droplet size distributions, but not enough is
known about these processes to make more precise statements about the expected changes. The
most clearly established mechanism for secondary ice generation, the Hallett-Mossop (H-M)
rime splintering process (Hallett and Mossop 1974), is known in enough detail to consider the
possible effects of altered droplet size distributions, and this effect is quite simple. The
probability of splinter production (per collision between ice crystals and cloud droplets) is
directly proportional to the number concentration of cloud droplets with diameters exceeding
24 m. Thus, if the shift in droplet spectra, whether it be naturally in deep warm clouds or due to
hygroscopic seeding, can be well predicted then the secondary ice generation rate that might
arise in the cloud, provided all other conditions for this process are fulfilled, can be readily
estimated.
139
As in the case of riming, the generation of additional ice crystals via an accelerated H-M
process does not lead to a clearly predictable effect on precipitation efficiency. Model
experiments comparing the precipitation produced with or without the H-M process have given
somewhat divergent results, from no change in precipitation over the lifetime of the cloud to
relatively modest changes. Because of the multitude of processes influencing precipitation
efficiency, and limited by available moisture and instability, the impact of the H-M mechanism,
or of any other specific process, on the total amount of precipitation is limited. However, with
an active H-M process, glaciogenic seeding may not have any effects and could decrease rainfall
under some conditions.
In summary, with a broadened cloud droplet spectra it is expected that frozen drizzle or
raindrops will form the initial graupel particles, riming efficiencies would be increased, and that
a secondary ice process might be initiated. The biggest effect of riming would be to speed
graupel production somewhat. Only quantitative analysis can assess the likely importance of this.
However, one should expect it to be less important than any process that increases the number of
ice particles. With secondary ice generation in clouds with bases at relatively warm temperatures
(10o C and higher), the impact of hygroscopic seeding on the H-M process is likely to be small.
The droplet spectrum at the height of the –3 to –8o C temperature level (where the H-M process
is active) will already have a substantial concentration of droplets larger than the critical 24 m
size. The colder the cloud bases (<10o C), the larger the potential effect of hygroscopic seeding
will be.
The measured cloud base heights and temperatures for the two seasons when the aircraft
were flying are displayed in Figure 4.25. Although there are large variations it is clear that there
is a general tendency for cloud bases to be higher and at cooler temperatures during the early part
of the summer season and that as the summer season progresses cloud bases tend to lower and
occur at warmer temperatures. The variations are also larger in the early part of the season and
tend to become smaller as the season progresses towards March and April.
140
Figure 4.25. Cloud base heights and temperatures as a function of date during the season
(2007–2008 and 2008–2009 seasons).
The result of the lower cloud bases in the latter part of the season also then provides for a
deeper part of the cloud warmer than 0o C, as is shown in Figure 4.26. The ―warm‖ cloud depth
tends to increase from the beginning of the summer season towards the end of the season. The
depth of the ―warm‖ cloud is important because this will determine in many instances if
coalescence will be active and large drops present before the cloud top reaches temperatures
colder than 0o C. This will certainly impact the efficiency of the ice processes in the cloud and
could also affect precipitation production as discussed earlier.
141
Figure 4.26. Cloud depth warmer than 0o C as a function date for the two seasons.
During the second season of the Queensland CSRP, 21 deep convective mixed phase
clouds were studied by the aircraft and two vertical profiles were conducted in deep stratiform
cloud systems which produced widespread rain over the area. In 19 of the 21 deep convective
cases, the aircraft measured large drops in concentrations of between 1 and 10 L-1 at the 0o C
level. The cloud bases in these cases ranged from 700 to 1300 m MSL. When large drizzle or
small raindrops were present at the 0o C level, graupel tended to form around the −5o C level and
subsequently an active secondary ice process (ice multiplication) evolved. The measurements in
a typical case are displayed in Figure 4.27, which are representative of the other 18 deep
convective cases with similar results.
Figure 4.27 displays the time histories for temperature, droplet concentration, CIP
particle concentrations larger than 40 µm diameter and the cloud liquid water content as a
function of time for several penetrations at different levels through a deep convective cloud. The
maximum cloud droplet concentrations were around 530 cm-3 near cloud base, indicating a
continental type of cloud droplet concentration.
142
Figure 4.27. Time histories for temperature (o C), droplet concentration (cm-3), CIP
particle concentrations larger than 40 µm diameter (L-1), and the cloud liquid water
content (g m-3) as a function of time for several penetrations at different levels through a
deep convective cloud on 26 January 2009. A size distribution of the particles is also shown
in the figure at the −8o C level. The width of the image bars represents 1 mm and the same
scale is true for the vertical.
143
The cloud droplet spectra are also plotted for near cloud base (bottom left panel Figure
4.27). It is clear that the spectra close to cloud base exhibits a continental-like size distribution
with very few droplets larger than 25 µm in diameter.
Cloud liquid water contents and droplet concentrations also decrease above the freezing
level indicating again an active coalescence process. In addition, the particle concentrations as
measured by the CIP indicate larger particles (>50 µm diameter) are substantially increasing
above the freezing level.
Figure 4.27 also shows the associated images of the particles along with a size
distribution of the particles at the −8o C level. The images around the freezing level indicate a
large concentration of cloud droplets larger than 25 µm diameter and also a significant
concentration of drizzle drops (~200 to 500 µm diameter) indicating the existence of large drops
lofted through the freezing level that subsequently freeze into graupel particles at relatively
warm sub-freezing temperatures. The criteria for the H-M process are thus satisfied and a large
amount of ice crystals suddenly occur around the −8o C level (see CIP particle images, Figure
4.27). The cloud top temperature never reached colder than −12o C. Due to the H-M process and
the graupel formation, the cloud liquid water contents also decrease rapidly and riming and
accretion become more effective. In these situations, glaciogenic seeding may have no effect and
may even contribute to a more rapid depletion of the liquid water content and less efficient
precipitation formation process. In addition, in these cases because of a very efficient ice process
and the associated rapid depletion of cloud liquid water; very little liquid water is lifted to
temperatures colder than −10 o C. This results in generally less lightning in these types of clouds
because the microphysical processes contributing to charge buildup usually occur in the
temperature regime between −10 and −20o C.
Another case (also described more in Section 5.3.3) on 21 February 2009 exhibits the
same features (Figure 4.13) with cloud base droplet concentrations around 600 cm-3 and a
continental cloud droplet size distribution. During a penetration at the 0o C level large drops were
already evident and the images collected during a penetration around the −10o C level showed
graupel and capped columns indicating that ice multiplication was occurring. Again cloud
droplet concentrations at the 0o C level had also decreased due to the onset of coalescence in the
clouds.
144
Figure 4.28. Same as Figure 4.27, except for a deep convective cloud on 21 February 2009.
Size distributions of the particles are also shown in the figure at the −0.8o C, 3.5o C, and
3.2o C, levels (left to right).
145
All the other deep convective cases in season two (not shown here) exhibit the same
features except for two cases that were penetrated by the aircraft in November 2008 (20 and
22 November) in the early part of the season when cloud bases were generally higher (Figure
4.25). The cloud bases in these cases ranged from 1500 to 2400 m. An example of the
measurements of such a case is shown for 22 November 2008 in Figure 4.29.
In this case no large drops were evident at the 0o C level as is evident from the images.
Cloud droplet concentrations and liquid water contents remained high and close to adiabatic
values even at the penetrations colder than the 0o C level. The cloud droplet size distribution also
exhibits a continental nature with very few droplets larger than 20 µm in diameter. In this case
and the 20 November case the first ice was only detected at temperatures colder than −10o C.
146
Figure 4.29. Same as Figure 4.27, except for 22 November 2008. CIP images are shown
from −2.5o C.
147
During the second season on two days widespread rain was experienced with a deep
stratiform layer with often times weak embedded convection in these systems. These two cases
occurred on 5 December 2008 and 13 February 2009. The aircraft on both days conducted a
vertical profile through the stratiform deck from cloud base to close to the top of the layer. The
cases exhibit the same processes based on the measurements. Cloud bases on both cases were
around 900 m MSL. The measurements for the 13 February 2009 case are displayed in Figure
4.30. On both days cloud droplet concentrations varied between 150 and 300 cm-3 near cloud
base, but with a fairly wide cloud droplet size distribution (Figure 4.30) with in both cases cloud
droplets larger than 25 µm in diameter. This resulted in an active coalescence process and ice
multiplication around the −5o C level, and then aggregates of needles and some dendritic growth
at colder temperatures. Very little, if any, supercooled liquid water was evident at temperatures
colder than 0 o C. The existence of numerous needle-type ice crystals in concentrations of more
than 20 L-1 indicated that enough ice was present to provide for an efficient ice process.
Because of the large amount of ice present already and very little supercooled liquid
water, silver iodide seeding in these systems may have negative effects and would not be
effective. The case on 5 December 2008 (not shown) exhibited the same features. In addition,
with the more maritime cloud droplet size distributions near cloud base, hygroscopic seeding
would also have no effect in these cases. These results indicate that these systems are very
efficient in producing precipitation and cloud seeding would not be advisable in these situations.
148
Figure 4.30. Same as Figure 4.27, except for a deep stratiform system on 13 February 2009.
149
For the 22 cases when penetrations were made in deep convective clouds, the cloud base
cloud droplet concentrations generally varied between 160 and 700 cm-3 with more than 80% of
the cases having droplet concentrations larger than 400 cm-3 (Figure 4.31). The total peak
concentrations for the cloud base penetrations in the 22 cases as a function of standard deviation
of the cloud droplet spectra are shown in Figure 4.31. The tendency seems to be for larger
standard deviations with larger initial droplet concentrations, which seems somewhat counter
intuitive because in general one would expect (using the maritime/continental distinction) that
smaller droplet concentrations would result in broader cloud droplet spectra and thus a larger
standard deviation. However, the more polluted air with the higher droplet concentrations also
could contain more large cloud condensation nuclei (CCN). This is an important point because
in many instances classifying clouds solely on cloud droplet concentrations could be misleading,
such as the case here.
Figure 4.31. Total peak droplet concentrations (cm-3) as a function of standard deviation of
the cloud droplet spectra for the 22 cases when penetrations were conducted in deep
convection near cloud base in growing non-precipitating parts of the cloud.
150
Figure 4.32. Mean diameter of the cloud base doplet spectra as a function of standard
deviation of the cloud droplet spectra for the 22 cases when penetrations were conducted in
deep convection near cloud base in growing non-precipitating parts of the cloud.
The mean diameter as a function of standard deviation near cloud base for the same
sample as displayed in Figure 4.31 is shown in Figure 4.32. As expected there is strong
relationship between the standard deviation and the mean diameter, with the standard deviation
of the droplet spectrum increasing as the mean diameter increases. These results are consistent
with other observations around the world.
4.3.3.1 Summary
In summary, these results indicate that clouds in Southeast Queensland generally develop
precipitation initially via the ―warm rain‖ process that then results in a more efficient mixedphase process also in deeper convective systems that extend above the freezing level. Seeding
with hygroscopic flares could potentially enhance the ―warm rain‖ process, but glaciogenic
seeding would not be advised in these conditions.
During the early part of the season, when cloud bases are generally higher, coalescence
could be delayed and ice multiplication may not occur and first ice will only form at
temperatures colder than −10o C. Hygroscopic seeding may be more effective in these clouds
providing for earlier coalescence and possibly the onset of a more efficient ice process.
These results would also explain the lack of intense lightning storms in the coastal
regions as observed during the field efforts. Further inland lightning frequency was much higher
151
than in the coastal region (see Section 3.5) that may indicate that the ―warm rain‖ process is less
efficient and hygroscopic seeding may be potentially more effective farther west of the current
CSRP domain.
In deep stratiform systems that are sometimes be observed in the region, natural
precipitation processes are very efficient and neither hygroscopic or silver iodide seeding may
have any effect and silver iodide seeding may actually have a negative effect.
152
5 Cloud Seeding Studies and Assessment
5.1 Introduction
From the 2007–2008 and 2008–2009 field operations seasons in southeast Queensland it
was determined that the frequency of opportunities for glaciogenic seeding in the Brisbane
region were much fewer than those for hygroscopic seeding. Furthermore, preliminary
measurements in the mixed phase region of deep convection indicated limited liquid water
content, which is crucial for glaciogenic seeding to be efficient (see also Section 4.3.3). Thus, as
a result, hygroscopic seeding has been the focus of the seeding trials in the region.
In this chapter we present the results of the randomized hygroscopic seeding analysis
conducted in southeast Queensland (Section 5.2), as well as some case studies that have been
analyzed by polarimetric and dual-Doppler radar (Section 5.3). Note that the application of both
polarimetric and dual-Doppler radar data for cloud seeding research is unprecedented, and as
such, these methods are only now being developed for use in analyzing the Queensland CSRP
data set and are still quite preliminary. Those methods are discussed and a few examples of their
use are presented (Sections 5.3.1 and 5.3.3). Finally, we present some of the results from the
simultaneous microphysical measurements collected during seeding with the dual-aircraft
operations in season one (Section 5.3.2).
5.1.1 Overview of results
The following conclusions are drawn from the detailed analyses presented in this chapter:
1. The statistical results indicate that seeded clouds tend to produce more rain than unseeded
clouds due to longer lifetimes and producing rain over a larger area. The results show the
same tendencies as wereobserved in previous experiments in South Africa and Mexico
that used the same hygroscopic seeding techniques.
2. The sample size has a strong impact on the statistical results. Efforts have been made to
use appropriate analysis techniques to interpret and understand the data and results. The
sample size at this stage does not contain enough samples to provide an unambiguous
statistical result. Therefore, future operations should focus on increasing the randomized
sample size or attempt to design (a priori) a new confirmatory randomized experiment.
3. The statistical analysis indicates that the seeded cases tend to live longer than the
unseeded cases after seeding decision time, indicating a potential enhancement in storm
duration from hygroscopic seeding, as shown by the decreased hazard rate for seeded
storms. These results are similar to those found in the South African and Mexican
experiments, but must be viewed with caution, as they are not statistically significant at
the 95% level due to the small sample size (see below).
4. The statistical analysis indicates that the difference between the height of the maximum
reflectivity and the reflectivity centroid in the seeded cases seems to increase 10 to
153
20 minutes after seeding commenced. This result remained consistent regardless of the
selection criteria, and is also consistent with the results of Mather et al. (1997).
5. A general tendency for seeded drop size distributions to have larger mean diameters and
higher large (>20 µm) drop concentrations in seeded clouds compared to non-seeded
clouds was observed in the season one dual-aircraft measurements, as well as in the
season two microphysical measurements. These observations suggest that the first step in
the hygroscopic seeding conceptual model is occurring. Droplet concentrations in the
Queensland clouds are very similar to those observed in South Africa and Mexico.
6. New techniques to study seeded versus non-seeded cells are being developed, to extend
the analysis beyond the traditional statistical analysis that uses reflectivity only
(via TITAN), by utilizing dual-polarization and dual-Doppler data from the CP2 radar.
These methods are in their infancy, but are valuable tools to help understand the physics
of any potential seeding effect.
7. Preliminary analyses indicate that suitable targets for seeding are major contributors to
overall seasonal rainfall. Assuming a 20% increase in precipitation by seeding, this could
result in upwards of a 5% increase in ground-measured rainfall.
8. Although the initial statistical results from this limited data set seem to behave in a
similar manner as was found for the South African and Mexican experiments, the current
results will have to be interpreted with caution because they are not yet statistically
significant and there is a possibility that the results could still be due to chance.
5.1.2 Background on hygroscopic seeding
Hygroscopic seeding, as opposed to glaciogenic (cold cloud seeding with silver iodide)
seeding, is directed at promoting the coalescence of water droplets in the cloud. The intention is
to promote particle growth through coalescence and thereby improve the efficiency of the
rainfall formation process. Appropriately sized salt particles, water droplets from sprays of either
water or saline solution (Bowen, 1952; Biswas and Dennis, 1971; Cotton, 1982; Murty et al.,
2000; Silverman and Sukarnjanasat, 2000), and hygroscopic flares (Mather et al., 1997; WMO,
2000) have been used. Statistical results, observations and modeling results for large particles
(>10 m diameter) have provided some statistical evidence (Murty et al., 2000; Silverman and
Sukarnjanasat, 2000) and evidence that under certain conditions with optimal seed drop size
spectra, precipitation can be enhanced (Farley and Chen, 1975; Rokicki and Young, 1978;
Young, 1996). The hygroscopic flare particle seeding experiments have provided statistical
support for rainfall increases due to seeding based on single cloud analyses, but the physical
processes leading to these increases in precipitation are not well understood. Despite the wide
use of hygroscopic seeding, the results have been inconclusive due to a lack of physical
understanding and, in some cases, inconclusive statistical evaluations.
Table 5.1 lists examples of field experiments or operations in which hygroscopic seeding
was employed. The results from these programs that have contributed to the current state of
knowledge in weather modification are that:
154







Both the South African and Mexican experiments produced remarkably similar statistical
results in terms of the differences in radar estimated rainfall for seeded versus non-seeded
groups (Bigg, 1997; Silverman, 2000; WMO, 2000);
In the South African and Mexican experiments, reevaluation of the results showed an
increase in rain mass 30–60 minutes after seeding, significant at the 96 percent level
(α = 0.04) or higher;
Marked differences in concentrations of ice particles were found in maritime clouds (high)
versus continental clouds (low) signifying the active role of collision and coalescence in
maritime clouds compared to continental clouds (Scott and Hobbs, 1977; Cotton, 1972;
Koenig and Murray, 1976);
Freezing temperatures increased with increasing drop size because larger droplets contain
or have a higher probability of colliding with ice nuclei;
Relatively large droplets (>24 μm) played a role in ice multiplication processes, including
mechanical fracturing during melting and evaporation and ice splinter formation during
riming (Hallet and Mossop, 1974);
Delayed response in radar-derived storm properties was a possible function of seedinginduced dynamic processes beyond the classical cloud physics results that links cloud
condensation nuclei and droplet spectra to rain production (WMO, 2000); and
Hygroscopic seeding might overcome the inhibiting effects on rainfall of air pollution
(Rosenfeld et al., 2002).
Table 5.1. Examples of hygroscopic seeding experiments in precipitation enhancement
Experiment
South African experiments
Indian experiments
Thailand experiments
Mexico experiments
Reference
Mather et al., 1997
Murty et al., 2000
Silverman and Sukarnjanasat, 2000
WMO, 2000
Based on the physical chain of events in the development of precipitation in seeded and
unseeded clouds, the seeding conceptual model and hypothesis for a microphysical propagation
of the seeding effect as originally anticipated in Queensland is summarized in Table 5.2. Note
that the seeding conceptual model primarily incorporates the microphysical aspects of the
conceptual model. Based on the previous randomized seeding experimental results in South
Africa and Mexico, additional dynamic effects in terms of enhanced cloud growth due to
possible modification of the updraft/downdraft structures and the development of progeny clouds
had to be invoked to explain the statistical results. However, these aspects are as yet not well
understood. A graphical description of the seeding hypothesis is provided in Figure 5.1. Clouds
were selected by the pilots based on predetermined criteria as depicted in Figure 5.2, and
described in Section 2.1.2.
155
Table 5.2. Steps in the hygroscopic seeding conceptual model.
Hygroscopic seeding broadens the cloud droplet spectra near cloud base
Hygroscopic-flare seeding enhances the production of drizzle
Drizzle production occurs in near-adiabatic ascent of parcels (where LWC is high), resulting in
drizzle near the tops of new turrets and in downdrafts (which can be stronger)
Recirculation is important, helping drizzle enter high-LWC regions of the same turret and spread
to other turrets
If the cloud vertically extends to temperatures colder than 0C, large drizzle and raindrops would
freeze preferentially and become the primary graupel embryos
Broader cloud droplet spectra below 0C would result in higher riming rates
Combination of large drop freezing and broader cloud droplet spectra may result in secondary ice
generation
Additional loading of precipitation at lower levels in seeded clouds results in changes in
updraft/downdraft structures and modify dynamic aspects of the storm
Figure 5.1. Hygroscopic cloud seeding conceptual model.
156
Figure 5.2. Typical candidate clouds for randomized cloud seeding experiments.
5.1.3 Flare particle characterization
Several different manufacturers have started to make hygroscopic flares, following the
initial promising results from South Africa and Mexico (WMO, 2000). It is important to evaluate
the output particle spectra from these flares in order to understand the effect they will have on
the condensation/coalescence process and precipitation development in convective clouds.
Airborne measurements in Texas in 1998, designed to evaluate the output particle spectra
from the different hygroscopic flares, found that the different flares produced different size
spectra that could have significant impacts on the evolution of the warm rain process. According
to previous modeling studies, the differences in the particle size spectra could have significantly
different effects on the condensation-coalescence process. It is also important to consider the
natural background aerosol spectra in the evolution of these differences.
The majority of the hygroscopic cloud seeding flares currently in use are based on the
formula of Hindman (1978) that was developed to initiate fog for cover of military vessels.
These flares incorporate sodium and lithium salts along with potassium perchlorate, magnesium
powder and an organic binder as the fuel. When burned, the flares produce a plume of sodium,
lithium and potassium salt particles along with magnesium oxide. All of these species are
incorporated into each of the flare particles that are produced.
Modeling studies on hygroscopic seeding show that the size of the particles introduced by
the seeding flare influences the effectiveness of seeding on rainfall enhancement. Under the
157
conditions modeled, the most effective seeding occurred when the introduced CCN had a size of
approximately 1 m. Little information is available on the particle sizes generated by the flares
currently in use. A size distribution measurement of the dry particles indicates that a majority of
the particles are in the 0.2–0.4 m size range (Cooper et al. 1997). These measurements were
made using flares manufactured with the original Hindman formula. Other airborne studies and
ground-based wind tunnel studies confirm that a majority of the particles produced from the
magnesium/potassium perchlorate flares are less than 0.5 m. However, the large particle tail
that is produced by the flares is largely unknown at this stage. The modeling studies also suggest
that the use of calcium salts in the seeding material could enhance seeding effectiveness.
It is difficult to obtain measurements of the particle sizes produced by the flares in a field
environment. This generally requires two aircraft, one to generate the seeding material, and the
second to make the measurements. A further difficulty is the ability to reproduce these
measurements, since the environment changes. To study hygroscopic cloud-seeding flares, a test
facility has been designed and constructed that simulates the burning of flares from an aircraft.
The test facility has been designed to provide a reproducible environment for combustion
of flares and measurement of the resultant particles. The facility is also designed to simulate the
burning of flares on the wing of an aircraft. Taking into account the calculations of particle loss
due to the different dynamical mechanism, the flare testing facility was designed and calibrated
to provide the means for sampling the size distribution from a burning flare and to be able to
trace back to gravitational and diffusional losses if required by the final analysis. The final
objective is to be able to calculate the actual size distribution spectra from the burning flare to
understand and study the mechanisms that modify or change the size distribution.
Based on numerous tests with different formulations of flares, a new flare was developed
by Ice Crystal Engineering (ICE) that produced larger particles than the original South African
flare used both in South Africa and Mexico. Modeling studies have shown that larger particles
would be more effective in hygroscopic seeding. Figure 5.3 shows an initial test of the particle
size and volume spectra of the ICE flare, which was the type of flare used in the Queensland
program to enhance rainfall.
158
Figure 5.3. Particle data from a flare manufactured by ICE containing 70% potassium
perchlorate.
A more recent test of the ICE flare is shown in Figure 5.4. Figure 5.4a shows the time
history of the size concentrations of the combusted flare particles as a function of time while
Figure 5.4b displays the associated particle size spectra measured by the PCASP and SPP-100
particle probes.
159
a.
b.
Figure 5.4. Hygroscopic flare tests on June 23, 2004.
The concentrations in the 0.8 to 6µm diameter size range are substantially increased as
compared to previous flares including the original South African flare based on the Hindman
(1978) formula. Based on the original size spectra of the South African flare (Cooper et al. 1997)
and the current measured size spectra for the ICE flare we replicated the Cooper et al. (1997)
model simulations to investigate if the current flare enhances drizzle production compared to the
original South African flare. The results of these simulations are displayed in Figure 5.5 and
Figure 5.6.
It is clear from Figure 5.5 and Figure 5.6 that the ICE flare produces substantially more
drizzle drops at shorter times that the South African flare. After 1 minute the ICE flare already
initiates drizzle formation with concentrations reaching a maximum when the South African flare
starts producing the first drizzle size drops. In addition the drizzle also results in a greater
coalescence process forming rain, and the transformation to rainwater is also much faster than
for the original South African flare. Initially the ICE flare produces two orders of magnitude
more drizzle water content than the South African flare.
160
Figure 5.5. a) Mass distribution functions as in Cooper et al. (1997) for 20, 600, 1200 and
1800 seconds after passage through cloud base for the unseeded case. b) Mass distribution
functions as in the unseeded case but for Cooper et al. flare (solid line) and the ICE 70%
(dotted line).
Figure 5.6. Changes in the distribution of the condensate between cloud droplets (<40 μm),
drizzle droplets (40–500 μm) and rain drops (>500 μm) after droplet activation for the
background case (thin solid line), flare from Cooper et al. (medium solid line) and the ICE
70% (thick solid line).
161
5.2 Randomized Seeding Experiment
This section summarizes the results of the randomized seeding experiment. When
reading these results, one must keep in mind that this experiment is exploratory. That is, the
analyses presented here were not defined prior to collecting or analyzing the data. The
implications of this are discussed in the background section below.
For the sake of simplicity and clarity, these analyses consider the mass of the target as
estimated using TITAN to be the response. The reason for initially focusing on TITANestimated mass is that it is the primary variable discussed in reports on experiments in South
Africa, Mexico and Thailand. New and innovative analysis methods are being developed
however (see Section 5.3) and will be utilized in future analysis to make use of new technology
(i.e., CP2 polarimetric, dual-wavelength radar) and explore insightful techniques to study the
effects of seeding on cloud properties.
5.2.1 Background on randomized experiments
Because the natural variability of clouds and rainfall can be ten to hundred times larger
than the effect from seeding on rainfall the World Meteorological Organization suggested that
―Weather modification experiments have to be randomized and evaluated by statistical
methods.‖
5.2.2 Exploratory versus confirmatory experiments
Weather modification is replete with discussions and arguments centered around the
distinctions between exploratory and confirmatory experiments and analysis. These stem from
the fact that weather modification experiments typically collect vast amounts of data that is very
useful in interpreting the observed results. The tendency has been to scour this information to
find variables which identify conditions under which seeding is effective. This by itself is a
logical and responsible thing to do with the data. Problems arise when one tries to take a result
as proof that seeding is effective. In addition to the problem of specifying the analysis after
viewing the data, issues related to multiplicity are typically ignored.
To be in any sense a confirmatory experiment, the experiment and analysis need to
follow a pre-specified design and address a fixed set of hypotheses. Since this experiment did
not have a statistical experimental design because the microphysics of the natural clouds
were not known at the start of this experiment, only exploratory analyses are conducted on
the results. Results from an exploratory analysis cannot be elevated to that of a confirmatory
experiment. These results cannot be interpreted as proving the effectiveness of cloud seeding in
Queensland, but will help guide the analysis of the physical data and help in the design of a
specific statistical experiment.
5.2.3 Randomized design
Due to the large natural variability associated with natural cloud systems and rainfall in
one location, from day to day, month to month, and between seasons, rainfall enhancement
experiments have reverted to randomized cloud seeding experiments in order to be able to
162
quantify the results. Natural variability can often be a hundred times larger than the response to
seeding and thus randomized experiments are the only avenue to evaluate the cloud seeding
results.
Randomization is used to generate an approximately equal number of seeded and
unseeded cases. This is also helps to eliminate any biases that might be introduced by virtue of
the staff knowing in advance the action to be used on a particular target.
Gabriel (2000) noted that ―Randomization not only protects against bias but also provides
a probability model that allows the assessment of the chance of various outcomes under
particular assumptions regarding treatment effects. It allows the calculation of significance levels
under the null hypothesis of no effect, as well as the estimation of power under hypotheses of
effects of given magnitudes.‖
Gabriel (2000) also gave the following quote in the context of confirmatory experiments,
however, it is still relevant for the current exploratory analyses:
―An important issue is the temptation to peek at the intermediate results of
experiments and declare a significant outcome as soon as the calculations indicate
a P value of, say, 5% or less. The probability that such calculations will show a P
value less than 0.05 sometime before the experiment has run its planned course is
considerably greater than 5%, so this strategy can greatly bias the conclusions and
has been generally discouraged by statisticians.‖
5.2.4 Significance and p-values
In the context of this report the p-values that will be quoted do not provide proof that
seeding works or does not work based on the above. It rather provides a first glance at the results
and is exclusively exploratory in nature. The results quoted in this report provide an overview of
the analyses to date and provide guidance for future experiments and analysis. Furthermore,
unless otherwise stated, use of the term ―significant‖ implies at the 95% level for this report.
5.2.5 TITAN data
The statistical analyses described in the next paragraphs were based on the radar-derived
TITAN tracks of the randomized cases similar to the analysis of the South African and Mexican
experiments (WMO, 2000).
In order to compare the analyses to the South African and Mexican experiments, the set
of cases for inclusion in the statistical analysis was determined using the same reflectivity
threshold (35 dBZ) and criteria that were used in those experiments, including the same size
criteria. In summary, the following criteria were applied:



TITAN radar (Mt Stapylton radar) derived data and responses.
35 dBZ track threshold.
35 dBZ TITAN track was detectable during at least two volume scans as seen by the
Mt Stapylton radar.
163

Volume of the 35 dBZ echo less than 750 km3 at decision time to eliminate targets so
large that they could not plausibly be affected by seeding (Mather et al. 1997). This also
removed targets that were incorrectly described due to limitations in TITAN tracking (see
subsequent sections).
Based on these factors, which were set to match the criteria used for the South African
and Mexican experiments, 41 (20 seeded and 21 unseeded) of the 127 total randomized cases fit
the criteria to be included in the statistical analysis. The reason for the limited amount of cases
was because all the other cases did not satisfy the aforementioned criteria. Especially during the
first season with the maritime shallow trade-wind cumulus clouds, many of the randomized cases
never developed a 35 dBZ echo that lived long enough (>2 volume scans).
However, in order to document the effect of storm volume on the results, this set of cases
was expanded to include larger storms, such that the volume criterion was changed to include
cases with a volume of the 35 dBZ echo up to 3000 km3 (compared to 750 km3 in the analysis
set). This resulted in 48 cases, which were then partitioned into 250 km3 storm volume intervals.
This method was applied so that the effect of seeding as we increase the volume threshold could
be evaluated, which furthermore lends credit to the robustness in using 750 km3 for the volume
threshold in the statistical analysis, especially for our sample. Using this set we also look at the
effect of mergers and splits and deficiencies in utilizing only the objective TITAN storm tracking
software.
5.2.6 General statistics and summary cases
The following sections provide an overview of the preliminary randomized statistical
results. Figure 5.7 shows the time history of all the tracks identified by TITAN for the 20 seeded
(red) and 21 unseeded (black) cases that satisfied the statistical analysis criteria. It is
immediately evident that there is large natural variability in both the seeded and unseeded sets of
the data, which is clearly the case considering the logarithmic scale on the y-axis even before and
at decision time when seeding was commenced.
164
Figure 5.7. Mass of targets as a function of minutes from decision time (e.g. T10 is
10 minutes past decision time.) for seeded (red) and unseeded (black) cases. Be aware that
y-axis is a log scale making a comparison of values between tracks difficult.
In order to correct for the initial biases, Figure 5.8 shows the radar-derived mass as a
function of time, but normalized at decision time. The large natural variability both before
seeding and after seeding commenced is clearly evident. It indicates the difficulty in collecting a
homogeneous data set and emphasizes the large variability that occurs in nature and the difficulty
to predict which clouds will further grow naturally or not. While some of the cases with the
greatest growth are unseeded, this figure suggests that a majority of the cases with greatest
growth are seeded. It also suggests that many of the cases with the longest duration were seeded.
These observations will be explored more in subsequent sections.
165
Figure 5.8. Target mass normalized by mass at decision time for seeded (red) and
unseeded (black) cases.
In order to further explore the relation between growth before decision time and after
seeding commenced, Figure 5.9 displays the relation between initial growth (change DT-10 to
decision time) and maximum growth (DT10 to 45) for radar-derived precipitation flux, volume,
mass, and area of the cases. Based on Figure 5.8 there seems to be an indication that if storms
are growing before decision time that they will continue to grow after decision, and the larger the
storm the greater the possibility of further growth of the storm.
These results indicate that it is important to interpret the responses to seeding in the
context of what is happening naturally to the clouds. Making conclusions on just the time series
data and the associated results can often mask other variations that naturally occur. In addition,
outliers can also have significant effects on the results in the time series data that could tilt the
results in either direction depending on the action (seed or no-seed) that was executed (see also
re-randomization results presented in Sections 5.2.7.4 and 5.2.7.5). Based on the results
summarized in Figure 5.9, the area of the 35 dBZ echo seems to have the highest correlation with
the later growth of the storm. Intuitively this would be expected because a larger more defined
storm would have potentially already established itself dynamically and provides for a candidate
to continue to grow. However, this could also point to a deficiency in how the TITAN analyses
treat cell mergers, by then including the clouds that merged from the start of the analyses (see
discussion on the effects of storm volume in Section 5.2.7.5). For the smaller sized storms there
is more variability and there is no clear indication if initial size changes will result in the same
after seeding. In general there seems to be more seeded cases that follow the suggestion that the
initial growth will result in future growth or maintain the storm, but the differences are not
significant at this stage.
166
Figure 5.9. Relation between initial growth (change DT-10 to decision time) and maximum
growth (DT+10 to 45 minutes).
Based on the South African and Mexican results that found that the primary response to
seeding was to enhance the duration of the seeded storms (Mather et al. 1997, WMO 2000),
Figure 5.10 displays the number of active storms as a function of time after seeding. Although
from this plot there may be some indication that more seeded storms lived longer it does not
seem significant based on this figure. There were more seeded storms that lived longer than
45 minutes than unseeded storms, but at earlier times there does not seem to be much difference
between the groups. There seems to be an indication that more seeded storms tend to dissipate
early on in their lifetime. Additional statistical methods have been employed to study duration
effects, and are presented in Section 5.2.7.3.
167
Figure 5.10. Duration as defined by the presence/absence of a TITAN value for mass. Note
values of 45 minutes are censored in that it is unknown how long targets existed.
5.2.7 Further statistical analyses
5.2.7.1 Wilcoxon test comparing mean mass recorded DT10 – DT45
The Wilcoxon test was used to test for differences in radar-derived mean mass (WMO,
2000) between the seeded and unseeded clouds. The results are presented in the form of box and
whisker plots displayed in Figure 5.11 from ten minutes after seeding to 45 minutes after seeding.
It shows that the seeded storms seem to be larger than the unseeded storms although outliers are
evident in both cases. With no assumptions made about the distribution of the two sets of data
(seeded and unseeded values), the Wilcoxon test evaluates the hypothesis that one group tends to
produce larger values than the other group. Slightly stronger assumptions are needed to make
statements about differences in group medians. Here, the Wilcoxon test indicated a p-value
of .0322 indicating support for the alternative hypothesis that seeded clouds tended to have
greater maximum mass than unseeded clouds. However, these results have to be interpreted with
caution because if there was an initial bias in the time before or at seeding times towards the
seeded storms these corrections or adjustments will have to be taken into account also.
168
Figure 5.11. Box plot of mean mass (DT10 - DT45).
5.2.7.2 Wilcoxon test on residuals from linear model (which use initial conditions to predict
mean mass in interval from DT10 – DT45)
In order to take into account varied initial conditions, the Wilcoxon test was also applied
to the residuals from the median regression model, which used the mass at decision time to
predict the radar-derived maximum mass during the period ten to 45 minutes after seeding.
The results are shown in Figure 5.12. A variety of statistical models were explored to
describe this relation. Transforming the variables using the logarithm produced a good fit but
required that variables with the value of zero be censored (recall Figure 5.10: note five unseeded
and three seeded clouds no longer had TITAN values for mass at DT10). While there is a lot of
scatter in the relation between the initial and maximum value, the median regression (i.e.,
quantile regression) suggests that initial mass was a significant variable in predicting the final
mass. Median regression is less affected by outliers than a linear model. Taking into account
initial conditions, again seeded targets tended to produce a larger maximum mass. It is also clear
that for larger storm masses there is a possibility that seeding will further increase the mass of
the storm with most of the red dots (seeding cases) lying above the one-to-one line.
169
Figure 5.12. Linear median regression model fit to mass at DT vs. max mass DT10 - DT45.
This includes 41 cases: 20 seeded and 21 unseeded (Set2 data).
5.2.7.3 Survival analysis for duration
In both the South African and Mexican results it seemed that the major effect of seeding
was to prolong the lifetime of the clouds (WMO, 2000). Thus, it is important to study the effect
of lifetime in more depth with the Queensland data and so using the cases selected for the
analyses discussed in the previous paragraphs, a survival analysis was conducted. Table 5.3
provides the data for the survival analyses.
Table 5.3. Lifetime analysis for duration of unseeded and seeded targets.
Unseeded
Time
from DT
0
5
10
15
20
25
30
35
40
45
# at risk
21
19
19
16
15
11
8
8
4
2
# end
2
3
0
1
4
3
4
0
2
0
Survival
Rate
0.90
0.76
0.76
0.71
0.52
0.38
0.19
0.19
0.10
0.10
Seeded
Hazard
Rate
0.10
0.16
0.00
0.06
0.27
0.27
0.50
0.00
0.50
0.00
170
# at risk
20
20
17
17
17
15
11
11
8
7
# end
0
3
0
0
2
4
0
3
1
0
Survival
Rate
1.00
0.85
0.85
0.85
0.75
0.55
0.55
0.40
0.35
0.35
Hazard
Rate
0.00
0.15
0.00
0.00
0.12
0.27
0.00
0.27
0.13
0.00
The analysis was done for each volume scan with the survival rate being the percentage
of radar derived storm tracks (seeded and unseeded) still alive and the hazard rate being the
percentage of storms in the seeded and unseeded categories that dissipated (see Section 8.1.1 for
more details). It is clear that the seeded storms had a lower hazard rate and larger survival rate at
all time periods after ten minutes and that at 45 minutes after seeding 35% of the seeded storms
still were tracked while only 10% of the unseeded storms remained active. These results are
presented in graphical form in Figure 5.13. Note that the hazard rate for targets in the final time
period is zero. Since this is the final time period it cannot be determined whether targets
dissipate or live longer. The subsequent analysis takes this artifact of the data into account.
These results seem to be consistent with the South African and Mexican results, but with only a
limited data set there is still a possibility that these results were obtained by chance.
Figure 5.13. Hazard rate for unseeded and seeded targets. The hazard rate indicates the
likelihood a target will disappear in a given interval.
Several generalized linear models were used to study the effect of initial growth, mass,
and action (seeded versus unseeded) on the hazard rate. The key result from this analysis was
that initial conditions (mass at decision time) are a significant factor in predicting the duration of
target. Including the expected effects from the initial conditions, the seeding action remained a
significant effect (p-value = 0.02) and seeded storms had a significantly lower hazard rate.
A lower hazard rate is equivalent to a longer duration.
171
5.2.7.4 Re-randomization statistics summary
This section describes the results from the re-randomization tests similar to the South
African and Mexican experiment analysis for 41 cases that satisfied the criteria that were used
for the previous experiments. It is important to note that this is a very small sample set and by
changing the criteria widely different results can be obtained, thus at present these results are just
preliminary. Furthermore, it should be noted that in this section (and the following section) the
p-values are used to describe the confidence in the alternative hypothesis (that there are
differences between the two groups), to follow the convention of Mather et al. (1997). This
convention uses a higher p-value (x) to indicate higher confidence that there is a difference
between the two groups (traditional statistics would normally use 1-x such that a lower p-value
is a sign of confidence the two groups are different).
The tendencies for the radar-derived rain mass, area, precipitation flux, and integrated
precipitation mass (see Figure 5.14) are very similar to what was found in the South African and
Mexican experiments. At a first glance one could easily interpret these results that seeding does
have an effect (more rain mass and larger area), but in some re-analyses (see following sections)
it was found that with such a small sample set, including additional cases or excluding some
cases by changing some of the criteria changed the results dramatically. The p-values should also
be taken with caution because of multiplicity effects and the small sample size. However, the
results seem to indicate that the tendencies seem to be in the same direction as was found in the
South African and Mexican experiments. It is also important to note that the statistical tests for
the differences between seeded and unseeded cases for the same variables over the total period of
the analyses indicated that none of the differences are significant with a p-value of 0.68 for mass
and 0.78 for area. This means that although the time evolution of the clouds seems to indicate
some responses in a global sense, the responses are still not statistically significant.
The only significant difference at the 5% level between the seeded and unseeded clouds
in the re-randomization tests was for the duration of the clouds after seeding, with the seeded
clouds living significantly longer than the unseeded clouds with a p-value of 0.96. This is also a
similar result as to what was found in the South African and Mexican results. The difference in
the area time integral (total rain over lifetime after seeding) between the seeded and unseeded
clouds had a p-value of 0.89. However, this was mostly due to the fact that seeded clouds seem
to live longer than the unseeded clouds.
172
a)
b)
c)
d)
Figure 5.14. Radar-derived mass (a) and area (b), precipitation flux (c), and integrated
precipitation mass (d) of the 35 dBZ echo as a function of time from 15 minutes prior to
seeding decision time to 45 minutes after decision time for seeded (red) and unseeded (blue)
cases. The numbers on the plots provide the p-values of the differences between the seeded
and unseeded clouds.
In order to test the consistency of these results in the framework of the previous
experiments, Figure 5.16 displays the height of the maximum reflectivity and the difference
between the height of the maximum reflectivity and the height of the centroid as a function of
time from the start of seeding. The figure only includes the TITAN tracks (seeded and unseeded
storms) that were still living (zero values due to storms dissipating were treated as missing
values in this analysis as opposed to the analysis described in the previous figures); Figure 5.15
shows how many storms were included at each time step. These variables were also found to be
significant in the South African and Mexican experiments (Mather et al. 1997 and WMO, 2000).
173
Figure 5.15. Number of cases included in analysis at each time step.
The trends are again similar and seem to indicate a potential seeding effect, but the
caveats as mentioned above pertain to these results also. However, in a general sense the trends
seem to be in same direction as was observed in the South African and Mexican experiments.
The trend between the difference of the height of the maximum reflectivity and the height of the
centroid between 10 and 20 minutes was also observed and mentioned specifically by Mather et
al. (1997).
a)
b)
Figure 5.16. (a) Height of the maximum reflectivity and (b) the height of maximum
reflectivity minus height of centroid as a function of time [as in Figure 5.14 except that
zeros are accounted for as missing values and the analyses are limited to 30 minutes after
seeding because the sample becomes too small (<10 cases) after 30 minutes].
In summary, these results seem to indicate similar trends in the seeding effects with
hygroscopic flares as were found in the South African and Mexican experiments. However, in
the current case our sample size is too small to make any conclusive statements about the seeding
effects.
174
5.2.7.5 Effects of storm volume on re-randomization results
In this section we explore the effects of sample size, biases and deficiencies in the
objective TITAN tracking analysis in terms of mergers and splits to evaluate the robustness and
consistency of the results.
The results described in the previous paragraphs, indicating that the seeded storms lived
longer and produce more precipitation, are primarily dominated by the seeded storms living
longer because when storms dissipate the zero values are included in the statistical analysis. To
further explore this result and determine if other variables showed a response also, Figure 5.17
displays a comparison between the seeded and unseeded storms in the statistical analysis for
maximum reflectivity (dBZ) and area for those storms that still had an 35 dBZ echo at any
specific time interval and treating the zero values (storms that dissipated) as missing values. We
limit our analyses for the period up to 30 minutes after seeding because for time periods after 30
minutes the sample size becomes too small to make any meaningful conclusions (fewer than ten
cases altogether). The results for the maximum reflectivity values indicate that after ten minutes
the reflectivity in the seeded storms was only slightly higher than for the unseeded storms.
Although this is only a small difference in reflectivity it could still impact precipitation because
reflectivity is represented on a logarithmic scale. However, although the intensity of precipitation
might have been slightly increased the results indicate that this was not a major response. A more
pronounced result was evident in the differences between the area of the seeded and unseeded
clouds, with the area in the seeded clouds being larger than the unseeded clouds from about ten
minutes after seeding and remaining larger for the rest of the analysis period up to 30 minutes
after seeding. This could be consistent with the seeding hypothesis that the formation of drizzle
by hygroscopic seeding would mix through larger parts of the clouds providing for a larger area
affected by seeding.
a)
b)
Figure 5.17. Maximum reflectivity (dBZ) (a) and area (b) for seeded (red) and unseeded
(blue) storms as a function of ―decision time‖ when seeding was conducted.
It is clear from Figure 5.17 that there are small non-significant differences between the
seeded and unseeded storms for maximum reflectivity. The maximum reflectivity provides an
indication for the intensity of rainfall. This is consistent with the earlier results from Mexico and
South Africa. The only variable that did show more pronounced differences between the seeded
and unseeded clouds for those clouds that still had an echo was the area of the 35 dBZ echo
175
(Figure 5.17b). Starting ten minutes after seeding began the area of the unseeded clouds seem to
be larger than the unseeded clouds. These results should only be interpreted as tendencies
because the sample size is very small. However, it seems that apart from seeding potentially
increasing the lifetime of the clouds, the area may also be increased as a result of seeding. These
results also are consistent with what was found in the South African and Mexican experiments. It
is important to consider these points as we discuss the results in the following paragraphs.
One of the major obstacles in the statistical analysis of rainfall enhancement experiments
such as the Queensland program has always been the effect of initial biases, but especially of
outliers (large storms) that could easily overwhelm and dominate the statistical results. In
addition, the effects of merging or splitting storms can influence and complicate the analysis
substantially. This includes deficiencies in the TITAN objective tracking system on how it
handles the mergers and splits in the current analysis. Currently when storms that are included in
the randomized sample are merging with other storms during the period from decision time to
45 minutes after decision time, these merging storms are included in the analysis from the time
that the merging storms existed. This can complicate and make the analysis meaningless because
these storms could be separated from the initial randomized case by more than 20 km at decision
time and therefore one would expect no effect from seeding on those storms. Figure 5.18
provides an example of such a case. The TITAN track that was seeded at decision time is marked
with an X but all the TITAN tracks to the north of the randomized case were also included in the
analysis because these tracks all merged with the randomized case after 30 minutes (Figure 5.18).
This is clearly not meaningful because the TITAN tracks to the north of the storms were in no
manner affected by the seeding of the initial TITAN track. These kinds of mergers then provide
for very large storms that dominate the results. The reason for the storms merging is primarily
because they were growing against the uplift of terrain in that area. Several such cases were
observed in the data set. Mather et al. (1997) also identified this problem (line-storms in their
case) and therefore instituted the criterion that only TITAN tracks with a volume less than
750 km3 at decision time were included in the randomized sample set analysis.
176
a)
b)
c)
d)
e)
f)
g)
h)
Figure 5.18. Case 67 at a) decision time (DT) and seeding (X), b) DT +18 minutes,
c) DT +24 minutes, d) DT +30 minutes, e) DT +36 minutes, f) DT +42 minutes,
g) DT +48 minutes, h) DT +54 minutes.
177
In order to study the effects of such large storms on the analysis we stratified the storms
by volume. Figure 5.19 provides a frequency histogram of TITAN tracks (storms) as a function
of volume of the storms. In addition, a frequency diagram showing the seeded, unseeded and
total tracks as a function of volume is displayed in Figure 5.19. It is clear that most of the tracks
have volumes less than 1000 km3 and that the seeded and unseeded tracks are nearly equally
represented in this sample. Between 1000 and 2000 km3 more seeded tracks are added while
nearly no unseeded tracks were added. However, a lot more unseeded tracks are added once the
volume is larger than 2000 km3 while only few seeded clouds are added. This bias in not having
equal representation of seeded and unseeded tracks added at larger volumes does impact the
analysis as will become evident in the following paragraphs.
Figure 5.19. Frequency histogram of TITAN tracks >35 dBZ, and normalized frequency
distribution of seeded, unseeded and total tracks as a function of volume of tracks.
An example of the type of TITAN tracks that are included in the analysis (with
volume < 750 km3) at decision time are shown in Figure 5.20. It is clear that this is a more
typical track with one single, easily identifiable storm without the mergers and splits shown in
Figure 5.18.
To study the biases associated with not equally adding larger storms in the seeded and
unseeded categories, Figure 5.21 through Figure 5.24 show the analysis from the rerandomization tests for differences between seeded and unseeded clouds for area (Figure 5.21),
mass (Figure 5.22), precipitation flux (Figure 5.23), and area time integral (ATI) of precipitation
(Figure 5.24) as a function of time from decision time. In addition, the analyses are conducted
by stratifying the results by volume of the storms starting with storms with a volume less than
250 km3 and then repeating the analysis by adding storms every 250 km3 up to maximum of
3000 km3. In this way it is anticipated that we could identify the effects of larger storms as they
are added to the sample in each category (seeded versus unseeded).
178
It is clear from Figure 5.21 through Figure 5.24 that for the analysis up to a volume of
1000 km3 the results are very consistent with seeded storms having larger areas, mass, and
precipitation fluxes than the unseeded storms without any initial and major biases.
However, as the analyses are conducted for storms between 1000 and 2000 km3 it is clear
that the biases are favoring the seeded storms because as we discussed earlier (Figure 5.19) we
are adding many more seeded cases than unseeded cases in this category. Because the storms in
this category (1000 to 2000 km3) are larger they start to dominate the results. Initial biases now
also favor the seeded storms. For storms larger than 2000 km3 the results now start to favor the
unseeded storms because many more unseeded than seeded storms (Figure 5.19) are now added
to the sample. The added unseeded storms are now dominating the results and also the initial
biases are now in favor of the unseeded storms.
These analyses indicate that it is extremely important to take these effects into account
when interpreting the results. In addition, it is important to note that for most storms larger than
1000 km3 the effects of mergers and splits in storm tracks and the associated deficiencies in the
TITAN tracking software in tracking these storms induces unrealistic storm tracks into the
results. Based on these analyses it is clear that the focus of the analysis and interpreting should
focus on the storms which are less than 1000 km3 in volume because they are best treated by the
automated analysis. This is consistent with the findings of Mather et al. (1997).
179
a)
b)
c)
d)
e)
f)
g)
h)
Figure 5.20. Case 90 at a) decision time (DT) and seeding (X), b) DT +18 minutes,
c) DT +24 minutes, d) DT +30 minutes, e) DT +36 minutes, f) DT +42 minutes,
g) DT +48 minutes, h) DT + 54 minutes.
180
Figure 5.21. Area (km2) of seeded (red) and unseeded (blue) storms and adding storms in
250 km3 intervals and redoing the re-randomization analyses as a function of time from
―decision time‖ (time of seeding). The p-values for the differences are also noted on the
plots. In addition at the bottom the p-values and the differences in area between seeded and
unseeded storms as a function volume are shown with a frequency plot showing the seeded
and unseeded storms in each volume category.
181
Figure 5.22. Same as Figure 5.21 but for mass (ktons).
182
Figure 5.23. Same as Figure 5.21 but for precipitation flux (m3/s).
183
Figure 5.24. Same as Figure 5.21 but for area time integral (ATI) (km2/h).
184
5.2.8 Preliminary calculations of the potential of cloud seeding to enhance water
resources
5.2.8.1 Representativeness of the randomized cases compared to climatology
To be able to interpret the results from the randomized seeding experiments, it is also
important to determine how representative this sample of storms is of the overall population of
storms that occurred over the past two seasons in the SEQ coastal region. For this purpose we
have analyzed all the storms using the TITAN radar tracking software and compared the
properties of the overall population of storms with the properties of the case storms which met
the 35 dBZ threshold and had a minimum duration of 900 s. The total number of storms
identified by TITAN during the last two summer seasons from October to March totaled 44808
as identified by the Mt Staplyton radar and are compared to the 50 randomized seeding cases that
met the above criteria. The duration of the storms is shown in Figure 5.25. It is clear that the
randomized cases were all biased toward the storms that generally lived longer; however, it must
be noted that the randomized cases include both seeded and unseeded cases. Although it was
found that seeding may have affected the lifetime of the storms, the unseeded cases also tended
to generally have longer lifetimes, indicating that the selection of storms was biased towards the
larger storms.
a)
b)
Figure 5.25. a) Histogram of percentage of storm duration as a function of time and
b) cumulative distribution function of storm duration.
This is further confirmed when we compare the precipitation area of the storms between
the overall population and the randomized cases (Figure 5.26). It is clear that more than 50% of
the randomized cases were in the top 9% of the overall population for the two seasons. These
facts are important because it indicates that the storms forming part of the randomized cases are
also the important storms contributing to the annual rainfall in the region. In previous studies
utilizing the same methods in the South African experiment (Mather et al., 1997), it was found
that in the monthly number of TITAN storm tracks affecting the target area:
•
15% of the TITAN tracks produced more rain mass than the 1st quartile control group in
terms of rain mass
185
•
10% produced more rain mass than the 2nd quartile control group in terms of rain mass
•
4.3% produced more rain mass than the 3rd quartile control group in terms of rain mass
In order to affect the annual rainfall it is extremely important to target the larger storms because
they contribute the most to the annual rainfall.
Figure 5.26. Histogram of precipitation area as defined by the 35 dBZ echo displayed as a
percentage as a function of area.
The comparison between the overall population and the randomized cases for
precipitation flux is shown in Figure 5.27. This figure indicates that the 75% of the randomized
cases fell in the top 25% of the overall population, confirming again that the randomized case
selections were biased towards the larger and more intense storms that generally contribute more
to the annual rainfall. Future analyses should focus on using these results together with the
results from the randomized experiments to more accurately determine the contribution to annual
rainfall, and if a certain increase in precipitation is achieved as indicated by randomized
experiments, what the cost-benefit ratio of such a program would be for increasing water
resources compared to other alternatives.
Determining the cost-benefit ratio and translating the individual storm effects into area
effects and impacts on runoff, groundwater, and agriculture is crucial to the funding
organizations. Preliminary theoretical studies in South Africa by the hydrological community
have projected a ~25% increase in annual run-off in typical catchments if the annual rainfall
could be increased by 7% (Görgens and Roosenboom, 1990). If attainable, this would result in
186
additional water at about 1/5 the cost of the cheapest alternative in South Africa. It is important
to note that this study was based on a hypothetical increase of 7% in area-wide rainfall.
Figure 5.27. Histogram plot of percentage of storms and cumulative distributions for
precipitation flux.
187
5.2.8.2 Preliminary rainfall enhancement potential
This section provides an estimate of the capability of an operational hygroscopic seeding
program in SEQ to enhance rainfall. It is based on climatological rain gauge and radar data.
The radar data serves to provide an estimate of the typical number and size of storms prevalent in
a catchment region of SEQ (for instance Wivenhoe) while the rain gauge data provides
ground-based measurements of rainfall which serve to provide a baseline measurement of
rainfall in the absence of seeding. It should be noted that this is a ―rough estimate,‖ however,
should be regarded as a ―best possible outcome‖ should an operational cloud seeding program be
initiated in SEQ.
Figure 5.28 shows the location of rain gauges throughout the whole of the CSRP domain,
and a time series of rain gauge data throughout the whole year for the period for which rain
gauge data is available (1995–2008) is shown in Figure 5.29. The accumulated rainfall has been
scaled from whole CSRP domain to a domain of 50x50 km, a size representative of a catchment
area. The scaling has been obtained assuming that rainfall is evenly distributed throughout the
domain.
Figure 5.28. Rain gauge locations in the CSRP domain.
188
Figure 5.29. Time series of total accumulated rainfall within a 50x50 km catchment for the
period 1995–2008.
If cloud seeding were to be instigated on an operational basis, it would be confined to the
summer months, which we define as the period October–March, inclusive. Figure 5.30 shows the
time series of rain gauge data measured during the summer season.
189
Figure 5.30. Time series of summer rainfall within a 50x50 km domain for the summer
period (October-March)
For the purposes of an operational cloud seeding program, we will assume the same
criteria for echoes which formed the basis of the statistical analysis presented throughout this
chapter and also for the statistical analyses of the South African and Mexican cloud seeding
experiments. These criteria are that an echo registers as a TITAN track (reflectivity threshold of
35 dBZ; minimum volume of 30 km3) and that the storm volume is less than 750 km3. Note that
the analysis presented in the previous section used a different set of radar data and did not have a
minimum volume and thus there are differences in reported total numbers of cells.
The total number of TITAN cells registered throughout the period 1995–2006 was 26 227
registered over 1081 days. Of these storm cells, 23 838 (or approximately 91%) satisfied the
criteria that the storm volume be less than 750 km3. When only the summer period is selected,
this number reduces to 18 898 or approximately 72% of storms. These storms occur on 584
unique days within the 2000–2006 period (2557 days), or about 23% of the total number of days.
If only the summer periods are considered (1177 days), then approximately 50% of days are
suitable for seeding. We will term these 584 unique days ―seedable days.‖
After application of the maximum storm volume threshold of 750 km3, 91% of the total
rain-gauge measured rain is accounted for. When only summer months are included, 80% of
storms are accounted for. The implication is that while storms of the appropriate size and
190
temporal occurrence manifest themselves 23% of the time (when considering a whole year), they
account for 80% of the measured precipitation on the ground.
Figure 5.31 is a time series of the number of storms on any given day. The total number
of TITAN cells registered on a given day is shown in black. The green circles represent the
unique occurrences of a TITAN track. The mean number of storms on any given day is 1.3±2.5
storms per day. The frequency histogram of the number of TITAN cells occurring during the
summer months and with a volume less than 750 km3 is shown in Figure 5.32. It can be seen that
rarely do more than five TITAN tracks appear within a catchment area on any given day. For our
current purposes, we will assume that a seeding aircraft can seed up to 10 distinct storms on a
given day. This is reasonable given that the average flight hours available to a light twin-engine
aircraft is on the order of four hours, and 15 minutes is required to burn three sets of flares
typically used in hygroscopic seeding. Therefore, the majority of the time, the seeding aircraft
should be able to target all cells within a catchment, assuming that they all occur within flying
hours.
Figure 5.31. Time series of the number of seedable storms occurring. The black dots
correspond to all TITAN tracks. The green dots correspond to only unique occurrences of
a storm's complex track number.
191
Figure 5.32. Frequency histogram of the number of unique seedable storms in a catchment
region 50x50 km.
We are now in a position to make some calculations of the expected enhancement to
precipitation in a catchment region. We assume that seeding increases a storm‘s
ground-measured precipitation by 20%. Assuming that 10 storms can be seeded on any particular
day, 55% of the total summer rain comes from potential seeding targets. This equates to 39% of
the total rain during a whole year coming from seedable clouds. If we increase the amount of
precipitation from seedable storms by 20%, then this percentage increases to about 44%.
Therefore, seeding has the potential to increase the total rainfall in a catchment from 39% to 44%,
an increase of 5%.
Figure 5.33 is a time series of the total rain-gauge-measured precipitation for the period
1995–2006. The above calculations have been obtained where we have corresponding radar data
(i.e., the period 2000–2006). It can be seen that there has been a general decreasing trend of
rainfall during the 2000–2006 period compared to 1995–2006, so it could be argued that the
above value will increase during periods of increased rainfall. The main limiting factor is the
number of clouds able to be seeded. Recall that on average only 1–3 are available for seeding
and we have made our calculations assuming that up to 10 clouds per day are able to be seeded
(if they exist); thus it would take a substantial increase in rainfall from seedable clouds during
wetter periods to increase the rainfall enhancement by much more than the 5% estimate provided
above.
192
Figure 5.33. Time series of total rain-gauge-measured rainfall for the period 1995–2008.
The seasonal cycle is shown in the second panel and the trend in the third panel down. A
decreasing rainfall trend is seen to occur from 2000 onward.
5.2.9 Conclusions from randomized seeding cases
Although the initial statistical results from this limited data set seem to behave in a
similar manner as was found for the South African and Mexican experiments, such that seeded
clouds tended to rain over a longer time and larger area than unseeded clouds, the current results
will have to be interpreted with caution. Initial biases and the small sample size could still result
in a given result by chance rather than due to seeding. Furthermore, a more detailed cost-benefit
analysis should be conducted to further assess the impact and utility of cloud seeding as a
method for augmenting water resources in the region.
193
5.3 Seeding Case Studies
5.3.1 Polarimetric radar analysis
In this section we investigate if cloud seeding had an effect on the differential radar
reflectivity (ZDR) measurements made by CP2. ZDR is related to the size distribution of the
raindrops (Bringi et al. 1986, Wakimoto and Bringi 1988, Brandes et al. 2004) and thus can be
used to estimate the drop sizes. Since larger drops fall with their horizontal axis longer than their
vertical axis, ZDR values tend to be larger for bigger drops. Assuming cloud seeding alters the
natural drop size distribution (Mather et al. 1997) monitoring ZDR measurements within seeded
and unseeded clouds may provide a means for determining if the rain drop size distribution is
being modified.
5.3.1.1 Possible CP2 analysis techniques
Two radar responses observable with CP2 are expected to be especially sensitive to the
development of warm rain, and hence to seeding effects. First, if the natural clouds are very
continental, with about 1000 droplets per cubic centimeter in undiluted ascending parcels, then
the earliest radar echo from rain at low reflectivity (Z) values (less than 20 dBZ) should have
rather high ZDR (greater than around 1dB, often several dB). This means that most of the cloud
water is in very small droplets that only seldom collide and coalesce. The earliest rainfall then
comes from a few giant particles that start coalescing early and form large drops while most of
the cloud water remains invisible to the radar. Such clouds are expected to be particularly
susceptible to hygroscopic seeding, producing more of the larger drops. Maritime cumulus have
been shown not to have this radar response (i.e., low Z with high ZDR) because they have lower
droplet concentrations, the natural coalescence process is faster, and the initial, low values of Z
correspond with low values of ZDR, showing small-droplet drizzle with few large drops. As a
result, the first thing to study is the echo growth of Z and ZDR together.
The second radar response to cloud microphysical processes is the time to the
development of precipitation. Here, the two-wavelength combination of CP2 is a key. At 10 cm
(S-band), the radar reflectivity needs to be above about 5 dBZ (occasionally as high as 10 dBZ)
before one can be sure that it is caused by drizzle or rain. This is because Bragg scattering, from
turbulent mixing inside the clouds, produces this magnitude of a radar echo. However, at
X-band (3 cm), this threshold is 20 dB lower – when the reflectivity is above −15 to −10 dBZ, it
can be relied upon to be from water drops. Thus the S-band radar echo can be used to estimate
cloud lifetime, while the X-band can be used to estimate a time when precipitation starts forming.
Thus if hygroscopic seeding is done early enough, there is potential here for seeing its effect on
radar, both through the time to radar evidence of precipitation and the early comparison of Z and
ZDR.
5.3.1.1.1 Preliminary results
Only one thorough study of the Z alongside ZDR growth has been performed in maritime,
trade wind cumulus during the Rain In Clouds over the Ocean (RICO) experiment (Knight et al.
2008), and the dual-wavelength approach has not been employed anywhere before. The
194
Queensland data allow a first look at these two approaches, not only for detecting seeding effects
but also for documenting the natural variability as it may be associated with changes in the
natural aerosol and with differences in cloud dynamics, such as over land versus over the ocean.
Before looking to compare seeded and unseeded cases, we need to develop knowledge of the
natural variability of the cloud behavior. So far there is a distinct difference on average between
the Queensland clouds and those from RICO that were entirely over the ocean (Knight et al.
2008), such that the clouds in the Queensland area are more continental.
So far, eight cases have been analyzed on 22 January 2009 and seven on 27 January 2009,
covering entire lifetimes of convective ―cells‖ when possible, but always including the echo
growth stages. Primarily, the radar reflectivity factor values (Z) at 10- and 3-cm wavelengths
(S and X) and the differential reflectivity (ZDR) at 10 cm have been used. The values studied are
not maximum values but ―total‖ values within single sweeps and with single cloud volume scans
(see Knight et al. 2008 for details on these methods). It is important that these are not average
values, but the numbers one would get if single sweeps or single volumes correspond to single
radar pulse volumes. That is the only valid way of estimating what is going on at the larger
scales. ―Cloud‖ from the radar perspective is defined as greater than 0 dBZ on S Band.
To illustrate the variability observed in the analysis thus far, two very different cases
from Queensland are presented: one from 22 January 2009 (see cell labeled ‗b‘ in Figure 3.40c)
and one from 27 January 2009 (see cell labeled ‗a‘ in Figure 3.41b). The case on 27 January is a
very simple single-cell, growing and decaying, illustrating rapid increase and decrease of both Z
and ZDR pretty much in unison (Figure 5.34b). The big drops grow and fall out quickly, leaving
essentially zero ZDR for scans 5-9. The slight negative offset of ZDR essentially indicates 0 dB.
This case is notable for starting obvious coalescence growth very quickly. The case on
22 January is more complicated and longer-lasting, and there is no evident coalescence growth
for about 20 minutes after the cloud becomes identifiable on radar (Figure 5.34-Figure 5.36a).
Its updraft evidently does not die as suddenly as the one in the 27 January case.
In general, the variability of cloud behavior on each of these two individual days is great
enough that it cannot be determined yet if the two days are distinctly different in any general
aspects of their radar cloud histories. Additional analysis is needed to further assess the natural
variability of precipitation processes in the region. Nonetheless, it seems clear so far that these
clouds studied in Queensland are more continental than those studied in RICO.
195
a)
b)
Figure 5.34. Total cloud volume reflectivity (dBZ) versus total volume differential
reflectivity (dB) for a) 22 January 2009 and b) 27 January 2009. Each individual volume
scan is numbered in sequence. In b), growing phase scans are red, and decaying phase
scans are blue. Of the two curves, the one on the left represents the Marshall-Palmer
distribution, and the one on the right shows the Z vs ZDR relation for a range of drop sizes
at concentration of 1 m-3 (see Appendix of Knight et al. 2008).
a)
b)
Figure 5.35. Total area reflectivity (dBZ) versus total area differential reflectivity (dB) for
a) 22 January 2009 and b) 27 January 2009. Each point represents an individual sweep
(total area value). The two curves are the same as in Figure 5.34.
196
a)
b)
Figure 5.36. Time series plots of reflectivity factor (dBZ; horizontal reflectivity as red and
vertical reflectivity as cyan using left axis), differential reflectivity (dB; blue using right
axis), and height of the echo (km; green using right axis) for a) 22 January 2009 and
b) 27 January 2009. [Note only horizontal reflectivity is plotted for 22 January 2009]
5.3.1.2 “Sister cell” analysis procedure
For this analysis we chose to select convective clouds that were new and growing and not
part of an existing cluster of precipitating clouds. In addition, we selected cases that appeared on
radar mostly as single cells and where the seeding started before the cloud reached its maximum
radar reflectivity. By choosing single cell storms containing primarily one updraft, we felt it
maximized the chance to observe any seeding modifications by reducing the likelihood that rain
drops from nearby updrafts would mask events in the seeded updraft. Utilizing the data from
both seasons, 24 single cell type storms have been selected for analysis. Once a cell was selected,
a polygon was manually drawn at each radar elevation around the cell of interest within which
volumetric radar statistics could be calculated.
Figure 5.37 shows the draw tool that was used for this purpose. A polygon was drawn
around the cell for each radar scanned elevation angle throughout the evolution of the radar cell.
Within the polygon, 17 values were computed including the mean horizontal reflectivity (ZH),
mean ZDR, median ZDR, echo area greater than a specified ZH threshold, and estimates of the
median size drop diameter (derived from equations presented in Section 8.2.1). The analysis
presented here examines the mean ZH and mean ZDR. These values were computed as follows:
Mean ZH = ∑ ZH /N where the summation is over all radar gates within the polygon
when ZH ≥ 10 dBZ, and
Mean ZDR = ∑ (ZH)i / (ZV)i where i is each radar gate within the polygon when ZH ≥ 10
dBZ and HV is ≥ 0.95.
197
Figure 5.38 is an example of a time height profile of the mean radar reflectivity (ZH, dBZ)
for cell #9 (i.e., randomized case #3) on 2 February 2008. The individual values in the plot are
the mean ZH for each polygon at the selected radar elevation angle and time for which the
polygon was drawn. A dual-Doppler analysis of this case is also presented in Figure 1.11.
Figure 5.37. Example of polygon drawing capability (yellow lines). Orange arrow is the
overlaid aircraft track.
198
Figure 5.38. Example of mean ZH (dBZ) values within each polygon after registering to a
time-height grid. Example is for cell #9 on 02 February 2008. Time is the abscissa and
height is the ordinate.
5.3.1.2.1 Preliminary results
Figure 5.39 shows the analyzed field of mean ZH from Figure 5.38 overlaid with the
mean ZDR analyzed field. Figure 5.39 shows a common feature that was often observed – high
ZDR is located near the axis of maximum mean reflectivity. The tendency was for higher ZDR
values to be located before the maximum reflectivity. This feature will be apparent in subsequent
figures.
Figure 5.40 and Figure 5.41 show plots of mean ZDR versus mean ZH through the lifetime
of 7 cells that all initiated within 24 minutes of each other in the south dual-Doppler lobe of CP2
on 2 Feb 2008 (see figure caption for interpretation of the plots). The plots trace the evolution of
ZH and ZDR through the cells‘ lifetime. The individual points are about 3 min apart. Figure 5.42
shows the location and cell number at a central time for the cells that are shown in Figure 5.40
and Figure 5.41. In 5 of the 7 cells, the mean ZDR increases as mean ZH increases, which is
generally what is expected if more intense rain has larger drops. Also, for the same reflectivity,
the mean ZDR tends to be larger (bigger drops) during the growth stage than during the
dissipation stage. The seeded case (C09) had a trace similar to cells C02, C03 and C07. Seeding
began only a couple minutes before the storm reached its maximum reflectivity; thus it could be
argued it is not a particularly good case for examining seeding effects on a growing cloud. Cell
C06 is very different from the others in that it initially has relatively large mean ZDR values. The
larger sample of cases showed this to occur numerous times. Probably the most striking
observation is the large amount of variability from cell to cell, particularly when it is considered
that all the cells in Figure 5.40 and Figure 5.41 initiated within 24 minutes and in close proximity
to each other.
199
Figure 5.39. Time-height of mean differential reflectivity (ZDR) overlaid on mean radar
reflectivity (Z). This is the same case as Figure 5.38. Note seeding began just before
maximum reflectivity.
200
Figure 5.40. Plots of mean ZDR versus mean ZH for four single cell storms on
2 February 2008. The values are for a height of ~1.5 km above ground. Each point
represents a different radar scan time from the beginning to the end of the cell. The red
indicates the growing phase and the blue the decaying phase. Thus the first red point is the
beginning of the cell and the last blue point is the end of the cell, where the beginning and
end are defined as a cell mean ZH of 10 dBZ. The yellow points indicate times the cell was
being seeded. See Figure 5.42 for the location and time of the cells. The cells are numbered
as C0X. No seed means the cell was selected for seeding, but the randomized decision was
―don’t seed‖. Sister means a nearby (unseeded and not a randomized case) cell.
201
Figure 5.41. Same as Figure 5.40 except for three other cells on 2 February 2008.
202
Figure 5.42. Radar reflectivity image from CP2 on 2 February 2008. The white crosses
indicate the initiation location and cell number of the nine storms that initiated within a
22-minute period. Plots of mean ZDR versus mean ZH for seven of these nine cases are
shown in Figure 5.40 and Figure 5.41.
5.3.1.2.2 Future analysis
The purpose of Figure 5.43 and Figure 5.44 is to provide examples of other single cell
cases. In all these cases the mean ZDR tended to be relatively large initially. The question as to
whether the seeding modified the natural rain drop size distribution will require more extensive
study. In some occasions there is a suggestion the seeding may have altered the drop size
distribution, but in other cases the opposite. The large natural variability in the evolution of the
drop size distribution greatly complicates determining the effects of seeding. Future efforts will
utilize dual-Doppler analyses to make certain the cell in question was actually seeded. The
analysis shown here was done at 1.5 km height; other heights will also be examined. In those
cases where the seeding was done very earlier in the growing stage of the cell it is desirable to
examine ZDR data for reflectivity less than 10 dBZ. However, because of data quality issues for
ZDR at very low reflectivity, special care will be required to ensure data quality.
203
Figure 5.43. Same as Figure 5.40 except for other days as noted.
204
Figure 5.44. Same as Figure 5.40 except for other days as noted.
5.3.2 Simultaneous microphysical measurements
The dual-aircraft measurements from season one, in which SEEDA1 collected in situ
measurements in clouds which WXMOD targeted for seeding, have been compiled in a similar
manner as the cloud base DSDs from season two (recall Section 4.3.2). All penetrations by
SEEDA1 were taken roughly 2000 ft above cloud base (where WXMOD was seeding), a
separation distance mandated by air traffic control. In addition to seeding cases, we have
compiled the unseeded drop measurements within 2000 ft of cloud base from season one for
comparison with the seeded cloud drop spectra. A few caveats to this analysis are that it is not
certain that droplet measurements in seeded clouds were measured in a portion of the cloud
affected by seeding material, and furthermore, these results have not been separated by wind and
aerosol regimes given the small overall sample size (would result in some regimes with no data),
and thus natural variability from varying environmental and aerosol conditions could influence
the observed DSDs also.
The DSDs for unseeded and seeded cloud base penetrations are illustrated in Figure 5.45.
As observed in the season two DSDs, the mean diameter and concentration of large particles
(>20 µm) are larger and/or higher for seeded clouds than unseeded clouds (Table 5.4). However,
it must be noted that the seeded cloud sample is very small (see Figure 5.45), and thus the results
may be overrepresented by measurements from regimes that have naturally larger, broader
205
spectra. Nonetheless, the trends are similar to that observed for each HYSPLIT regime in season
two (recall Section 4.3.2), and the differences in mean diameters are again statistically significant
(p-value = 0.027), although these results must be interpreted with care as no natural variability
(e.g., a regime-based bias towards larger diameters) is taken into account.
Figure 5.45. Drop size distributions (DSDs) for each cloud base penetration (of natural,
non-seeded clouds) vs seeded clouds (that were being seeded simultaneously with the
measurement) from season one.
Table 5.4. Summary of mean cloud base droplet statistics for Figure 5.45, including the
mean maximum and standard deviation SPP-100 FSSP concentration (/cc), and the mean
and standard deviation mean drop diameter (µm), and the mean concentration >20 µm
(/cc).
Unseeded
Seeded
Max Conc
843
661
Std Dev
439.5
86.9
Mean Diam
8.43
10.7
Std Dev
2.24
2.25
Mean Conc >20
2.39
9.62
5.3.3 Preliminary dual-Doppler studies
The ensemble of observations made by aircraft, polarimetric and Doppler radars, and
raindrop disdrometers during the Queensland cloud seeding project affords a unique opportunity
for studying the details of precipitation development within storms, especially related to
understanding the effects of cloud seeding. In this section we present preliminary analysis of an
(unseeded) storm complex that occurred on 21 February 2009. This case includes simultaneous
radar and aircraft measurements of a mixed-phase storm (recall Section 4.3.3, Figure 4.28). This
analysis is presented to document some of the capabilities we can explore for studying the
dynamics and microphysics of seeded and unseeded clouds for future cloud seeding analysis.
206
The computational procedure, an enhancement of that described by Brandes (1976),
utilizes upward and downward integrations of the continuity equation to reduce error propagation
in the vertical velocity calculation. Figure 5.46 shows portions of a dual-Doppler analysis using
measurements from the CP2 and Mt Staplyton radars. The reconstructed horizontal wind field,
in storm-relative reference frame, is shown by vectors. Thin contours show estimated vertical
drafts. Radar reflectivity (in dBZ) is contoured in blue.
The storm complex propagated roughly toward the north-northwest at about 5 m s–1.
Maximum storm top was 8–9 km. Figure 5.46a (1 km elevation) reveals that storm updrafts
were fed by air from the north and north-northeast. Air exited updrafts toward the southeast
(Figure 5.46b; 6 km). The reflectivity field is elongated in this direction, and all cells lean
southeastward with height as well. The reflectivity maximum >50 dBZ at x = −4, y = −30 km
(1 km) represents an older storm cell that is raining out. Small reflectivity maxima seen at 6 km
(x = −6, y = −28 km and x = −8.5, y = −34.5 km) are newly developing cells whose precipitation
has not yet lowered to 1 km. A vertical cross-section through the storm complex (Figure 5.47)
reveals low-level precipitation within the 50 dBZ core has fallen from a sloping updraft. Peak
computed updrafts in the cross-section are 7.3 m s–1. Newer cells at 6 km associate with
extensions of the updraft.
The green curved line in Figure 5.46b depicts the SEEDA1 aircraft track for the period
05:49 to 05:53 UTC. The mean aircraft height was very close to 6 km. Recorded temperatures
varied from –9.5oC outside the complex to –6.2oC in the updraft of the cell at x = −6, y = −28 km.
Observed hydrometeors in the northern and southern developing cells are presented in Figure
5.48 and Figure 5.49, respectively. Mixed-phase conditions are evident in both cells.
(Supercooled liquid water was detected by other aircraft probes and implied by wind shield
camera images showing ice buildup, see Section 4.3.3.) Drops with diameters as large as 3 mm
were observed in both cells. The presence of large drops is confirmed by differential reflectivity
measurements >3 dB that are paired with reflectivity measurements greater than 40 dBZ (not
shown).
Using CP2 polarimetric measurements, along with Eq. (15), and Eq. (17) (in Section
8.2.1) drop maximum diameters Dmax and median volume diameters D0 were estimated for the
1-km level (Figure 5.50). [Alternately, storm properties could be examined using differential
reflectivity and/or radar reflectivity measurements. However, for many purposes physical
properties of the DSD are preferable.] Maximum drop diameters of 6 mm are suggested by the
differential reflectivity measurements. The maximum drop estimates are consistent with that
obtained from radar reflectivity [Eq. (14)]. Largest drops are displaced slightly northward from
the 50 dBZ reflectivity maximum. This probably results from size sorting along the leading edge
of the storm complex in which large drops fall faster than small drops. In general, maximum
drop diameters are smaller in trailing portions of the storm. Because the analysis in Figure 5.50
is based on the ZDR measurement alone, the estimated D0 field is similar. A small region with
D0 = 3.5 mm is indicated. Along the storm‘s southern (trailing) edge, D0s tend to be less than
1 mm. This is also a consequence of size sorting as small drops from declining cells are the last
to arrive at the surface.
207
a)
b)
Figure 5.46. Dual-Doppler analyses at a) 1 and b) 6 km elevation AGL for a storm complex
observed at 05:49 UTC on 21 February 2009. Horizontal winds are indicated by red
vectors. Updraft (downdraft) regions are shown by solid (dashed) red contours. Blue
contours are radar reflectivity (dBZ). The heavy black line in (a) is the location of the
cross-section shown in Figure 5.47. The SEEDA1 track is shown by a green line in (b).
Green dots indicate aircraft positions at 1-min intervals.
208
Figure 5.47. Storm-relative wind flow, as in Figure 5.46, except for a vertical cross-section
through storm on 21 February 2009. See Figure 5.46 for location.
209
Figure 5.48. SEEDA1 images obtained with the PIP probe between 05:55:06 and
05:51:13 UTC in the northern developing cell.
Figure 5.49. As in Figure 5.48, except for particles detected from 05:51:14 to 05:51:16 UTC
in the southern cell.
210
a)
b)
Figure 5.50. Radar-estimated a) maximum drop (Dmax) diameter and b) median volume
diameter (D0) computed from CP2 measurements of differential reflectivity. The analysis
is for 1 km elevation and the domain in Figure 5.46.
If cloud seeding is effective, storms must be altered in a quantifiable way. Due to time
constraints our evaluation to date has been restricted to summary metrics that describe the
general behavior of seeded and unseeded storms. Many storms were within the dual-Doppler
lobes formed by the CP2 and Mt Staplyton radars. Detailed polarimetric and dual-Doppler radar
and aircraft-based analyses of these storms will allow trajectories of seeding material to be
determined and detailed evaluation of the storms‘ microphysical response.
Dual-Doppler analysis has also been performed for one volume scan of a seeded warm,
shallow convective cloud (randomized case #3, or cell #9 in Section 5.3.1) and a nonrandomized seeded, deeper convective cloud on 2 February 2008 (see Figure 1.11 and Figure
1.12). These analyses are preliminary so far, and will require more time to complete and utilize
for cloud seeding research such as that stated in Section 1.2.3.2.2:
1) Use of retrieved winds to constrain a parcel model to study aerosol uptake in
precipitating systems (both background and flare produced), and
2) To study the dynamical evolution (e.g., updraft intensity with time) of seeded and
unseeded clouds.
These types of analyses take considerable time and care to assure quality wind retrievals,
especially for multiple radar volumes per case and multiple cases, and thus use of these
techniques on many cases has not been feasible in the analysis to date. Nonetheless, these are
examples of the unprecedented and innovative cloud seeding research that can be done with this
grand data set.
211
6 Summary and Recommendations
Two seasons of field operations were conducted for the Queensland Cloud Seeding
Research Program (CSRP), with the first season taking place between December 2007-March
2008, and the second season between November 2008 and February 2009. A total of 108 flight
operation days with 164 total flights took place between the two seasons. Of the total flights,
there were 142 research flights: 49 research flights were flown by the SEEDA1 aircraft and 39
by WXMOD in season one, and 54 research flights by SEEDA1 in season two. These flights
comprised 386 total flight hours between both seasons and both aircraft. In each season,
SEEDA1 flew 150 hours (for a total of 300 hours), and in season one WXMOD flew 86 hours.
62 randomized cases (#1-62) were declared in season one and 65 randomized cases were
declared in season two (cases #63-127).
Analysis efforts for the Queensland CSRP were focused on three major areas for the
greater Brisbane region: understanding the weather and climate, characterizing the atmospheric
aerosol and cloud microphysics, and assessing the impact of cloud seeding to enhance rainfall.
These research efforts have produced a great number of results to date, although there are still
some areas that still require further analysis. The data sets collected in the two field seasons are
vast and unprecedented for a cloud seeding research project, and thus many varied research
efforts can continue to utilize the Queensland CSRP data sets. A summary of the key findings
from the analysis efforts to date is presented below, and is followed by a set of recommendations
for future operational and research efforts.
The wet season in Southeast Queensland occurs broadly from October–March, with
November-February receiving the most rainfall and having the most atmospheric moisture. Two
types of synoptic cluster analysis were presented: a seven-cluster case and a three-cluster case.
Both have shown that Southeast Queensland can be divided into ‗wet‘ and ‗dry‘ weather regimes,
with the ‗wet‘ regimes occurring most in summer and the ‗dry‘ regimes more in winter. This
result is in itself not surprising, as this fact is already well known to the local population and
forecasters, but this analysis has further quantified these regimes nonetheless.
The ‗wet‘ regimes were responsible for the majority of the region‘s rainfall and include
the northwesterly and ‗moist‘ southeasterly, easterly, and westerly regimes (from the sevencluster analysis), or the northwesterly and (when unstable) the southeasterly regimes (from the
three-cluster analysis). The northwesterly regime contributed greatly to the total annual rainfall
despite occurring less than 10% of the time. The ‗dry‘ regimes were predominantly the other
southeasterly clusters (which was also the most common regime occurring year-round yet
contributes very little to annual rainfall in the seven-cluster case), and the southwesterly regime,
which was most common in winter (in both the seven- and three-cluster cases). The
southeasterly regime(s) tended to be less unstable than the northwesterly regime and can have a
trade wind inversion below the freezing level, resulting in shallow convection with rain formed
via warm rain processes only. Given the tendency for higher instability and thus deeper
convection in the northwesterly regime, this regime was also overrepresented in the aircraft
measurements (as we tended to fly on days with convection, especially focusing on deep
convection). Synoptic scale features (such as east coast lows) were generally the source of rain
212
during the winter months. Variation throughout the wet season was also attributable to synoptic
scale disturbances, as well as mesoscale features.
The Southern Oscillation Index (SOI) appeared to be positively correlated with
precipitation rate, and more so with precipitable water (a measure of atmospheric water vapor),
such that positive SOI (La Nina) corresponds to wetter conditions. A tailored index using sea
surface temperature (SST) off the east coast of Australia showed an even better positive
correlation with precipitable water (higher SSTs relates to higher precipitable water in the
Brisbane region). These indices can be used for predicting potential seasonal precipitation, as
well as to assist in understanding historical trends in precipitation.
The radar climatology indicated that cell volume, area and height all followed truncated
lognormal distributions. The same result was found when each of the three-cluster synoptic
regimes were considered individually. Nonetheless, the largest 20% of cells were found to be
responsible for roughly 67% of the fractional areal coverage of precipitation. Student‘s t-tests
applied to examine the relationship between the three clusters found that the precipitation area
was significantly larger during the winter regime than during either of the summer regimes and
cell-top height was significantly lower during the winter months than during the summer.
Four regimes of surface wind flow conditions were revealed by HYSPLIT back trajectory
modeling. These regimes relate to surface aerosol conditions and were classified into two
maritime regimes (northeasterly and east/southeasterly; hereafter NEly and Ely), and two
continental regimes (northwesterly and west/southwesterly; hereafter NWly and Wly). Of the
four HYSPLIT regimes, the northeasterly regime, on average, had the highest fraction of aerosol
that served as CCN at the 0.3% supersaturation, while the westerly regime had the lowest. These
results indicate that a higher fraction of the maritime flow aerosol served as CCN (at 0.3%
supersaturation) than from the continental flow (NWly, Wly) days.
The maritime flow (NEly and Ely) regimes had a more robust coarse mode in their
aerosol size distributions, especially of particles >5 µm, compared to the continental (NWly and
Wly) regimes. The Ely regime tended to have the weakest fine mode, and thus the overall
cleanest conditions. This is expected given the flow in this regime is coming straight in from the
ocean. The NWly regime often had a high number of fine particles and/or very few coarse
particles. The results showed, however, that coarse mode particles in small concentrations were
observed in all four regimes, although they were more common in higher concentrations in the
NEly and Ely regimes. The CCN measurements also reiterated that the Ely regime is the
cleanest regime overall with the lowest mean CCN concentrations at all measured
supersaturations. There was a general tendency for the NEly, NWly, and Wly regimes to have
similar CCN levels at each supersaturation, all of which were roughly twice as high as for the
Ely regime.
Three aerosol filter samples were analyzed from three different days, each representing a
different HYSPLIT regime (NEly, NWly, and Ely). The fine mode fraction of aerosols was
mostly dominated by sulfates and most of the CCN active particles were ammonium sulfate,
however in one (Ely, maritime flow) case, the fine modes were dominated by small sea salt
particles. The intermediate and coarse mode aerosols, on this day however, were mostly mineral
dust. In the other two cases analyzed, there was no indication from any of the size fractions that
213
the aerosol samples were of marine origin, and appeared more continental, possibly having an
anthropogenic source (NEly day with a back trajectory that passed over the city of Brisbane) or
biomass burning source (NWly day). The NEly day, however, did have some sodium-bearing
particles in the intermediate size fraction that may have been aged sea salt. Otherwise, the coarse
mode was usually a mix of mineral dust and other particles (aluminosilicates).
The average maximum drop concentrations at roughly 1000–2000 ft above cloud base
were similar for all HYSPLIT regimes, however, and range between 500 and 650/cc. The results
show that on average, the Wly and NWly regimes had the highest cloud base drop concentrations
and smaller mean diameters, compared to the maritime NEly and Ely regimes. The NEly and
Ely regimes had the highest concentrations of drops greater than 20 µm, on the other hand,
which can indicate a greater likelihood that collision and coalescence will be active in clouds in
those regimes. This was also possibly related to the fact that both of these regimes had a strong
coarse mode of aerosol particles.
The cloud base droplet spectra were combined for all natural (unseeded) clouds in each
HYSPLIT regime, as well as for seeded clouds in each regime. The tendencies from these
results showed that the mean diameter and concentration of large drops (>20 µm) in each regime
was larger and/or higher in seeded clouds than the natural (non-seeded) clouds. This is
consistent with the first step of the hygroscopic seeding conceptual model and this trend was also
observed in the season one cloud base droplet spectra results.
During the early part of the summer season, when cloud bases are generally higher, our
results indicated that coalescence could be delayed, ice multiplication may not occur, and thus
first ice will only form at temperatures colder than −10o C. Hygroscopic seeding may be more
effective in these clouds by providing earlier coalescence and possibly the onset of a more
efficient ice process. Otherwise, clouds in Southeast Queensland generally develop precipitation
initially via the ―warm rain‖ process that then results in a more efficient mixed-phase process in
deeper convective systems that extend above the freezing level. Seeding with hygroscopic flares
could potentially enhance the ―warm rain‖ process, but glaciogenic seeding would not be advised
in these conditions. These results would also explain the lack of intense lightning storms in the
coastal regions as observed during the field efforts, since liquid water content in the mixed-phase
region (necessary for storm electrification) was often rapidly depleted due to the efficient mixedphase processes and ice multiplication. Based on the CG lightning activity alone, it suggested
that the mixed-phase microphysics of clouds farther west were more vigorous than those closer
to the coast, where the CSRP domain was located. Cloud seeding with hygroscopic or
glaciogenic flares may be more effective in that region.
Deep stratiform systems were occasionally observed in the region, yet our observations
indicate that the natural precipitation processes are very efficient in these systems. Thus, neither
hygroscopic or glaciogenic seeding may have any effect and glaciogenic seeding may actually
have a negative effect.
The randomized seeding statistical results indicated that seeded clouds tended to produce
more rain than unseeded clouds due to longer lifetimes and by producing rain over a larger area.
The results showed the same tendencies that were observed in previous experiments in South
Africa and Mexico, which used the same hygroscopic seeding techniques. Nonetheless, the
214
sample size had a strong impact on the statistical results. Efforts were made to use appropriate
analysis techniques to interpret and understand the data and results. Unfortunately, the sample
size at this stage does not contain enough samples to provide an unambiguous statistical result.
Therefore, future operations should focus on increasing the randomized sample size or attempt to
design (a priori) a new confirmatory randomized experiment.
That said, the statistical analysis indicated that the seeded cases tended to live longer than
the unseeded cases after seeding decision time, indicating a potential enhancement in storm
duration from hygroscopic seeding, as was shown by the decreased hazard rate for seeded storms.
These results are similar to those found in the South African and Mexican experiments, but must
be viewed with caution, as they are not statistically significant at the 95% level due to the small
sample size. The statistical analysis also indicated that the difference between the height of the
maximum reflectivity and the reflectivity centroid in the seeded cases seemed to increase 10 to
20 minutes after seeding commenced. This result remained consistent regardless of the selection
criteria, and is also consistent with the results of Mather et al. (1997).
The aircraft measurements showed a general tendency for the seeded cloud base drop size
distributions to have larger mean diameters and higher large (>20 µm) drop concentrations
compared to unseeded clouds in both the season one dual-aircraft measurements, as well as in the
season two microphysical measurements. These observations suggest that the first step in the
hygroscopic seeding conceptual model was occurring. Moreover, droplet concentrations in the
observed Queensland clouds were very similar to those observed in South Africa and Mexico.
Although the initial statistical results from this limited data set seem to behave in a similar
manner as was found for the South African and Mexican experiments, the current results will
have to be interpreted with caution because they are not yet statistically significant and there is a
possibility that the results could still be due to chance.
Finally, new techniques to study seeded versus unseeded cells have been developed in
order to enhance the analysis beyond the traditional statistical analysis that uses reflectivity only
(via TITAN), by utilizing dual-polarization and dual-Doppler data from the CP2 radar. These
methods are in their infancy and still need further development, but are unique tools to help
understand the physics of any potential seeding effect. It will be extremely valuable to continue
refining these tools and to utilize them for future analysis of the effects of hygroscopic seeding.
6.1 Recommendations
The following recommendations are concluded from the findings of this report:
1. The results from the randomized seeding experiment in the SEQ CSRP can be
summarized as follows:
a. The statistical analysis results show similar tendencies (although based on a
relatively small sample) to hygroscopic seeding experiments conducted in Mexico
and South Africa where seeded clouds tended to rain over a longer time and larger
area than unseeded clouds
215
2.
3.
4.
5.
6.
b. The infrastructure in SEQ allows for further focused research that could confirm
the statistical results, as well as improve our understanding of the physical basis
of the results from the current and previous cloud seeding studies
c. As a next step, a study should be conducted to determine the scope and cost of an
experimental program and to conduct a cost-benefit study on the potential
additional water resources provided by cloud seeding compared to other
alternatives. Initial estimates are encouraging and motivate more in-depth study.
Any future application of cloud seeding as a technology for enhancing water resources in
SEQ should be informed by the results of the physical measurements:
a. Glaciogenic seeding (with silver iodide) would not be advisable in clouds in the
coastal regions of SEQ
b. Hygroscopic seeding is recommended for the convective clouds in the region,
c. Operations should be focused on the period of October to February (when the
clouds most suitable for cloud seeding with hygroscopic flares are more likely to
occur)
d. Hygroscopic seeding may be even more effective in convective clouds with
colder cloud bases, such as those observed earlier in the summer season, as well
as possibly further inland
e. Neither glaciogenic nor hygroscopic seeding would be effective in deep summer
widespread stratiform systems with weak embedded convection that sometimes
traverse the region. These systems are naturally highly efficient at producing
precipitation.
A scientific research component should be included in any further work to gain further
confidence in the results and to improve understanding of hygroscopic seeding processes.
The randomized statistical results do indicate a tendency that seeding increased
precipitation in convective clouds in the region; however, it is recommended that further
randomization experiments be included in any future operations. These randomized
cases will need careful analysis using high quality radar, preferably with dualpolarization capability.
Seeding aircraft used in any future work should be instrumented to measure temperature,
pressure, humidity, three-dimensional winds, cloud droplet and particle spectra, and
aerosol spectra.
The data collected as part of the Queensland program is a unique data set that is of
interest to many stakeholders with weather and climate applications and will be a longterm resource (detailed data sets such as this are rare and are used for decades). Therefore
a public-accessible, searchable data archive should be established for users to easily gain
access to this data set and preserve it for future use. Papers and publications developed
using this information should also be made readily available through this repository.
Key stakeholders should be made aware of the data arising from this project as this data
set will have applications beyond cloud seeding, including:
216
a. Improved operational quantitative rainfall estimation and forecasting
b. Improved water resource management tools (beyond cloud seeding)
c. Improving understanding of severe weather events (e.g., ―The Gap storm‖) and
development of high impact weather detection and forecasting
d. Air quality and pollution monitoring and characterization.
7. Detailed studies have been conducted on the meteorology of SEQ. These studies have
been conducted independently by several scientists and institutions and it is important
that this work be synthesized and the results published.
8. Numerical parcel, cloud-resolving, and mesoscale models together with the data collected
should be used to improve understanding of precipitation processes, evaluate the seeding
effects on precipitation, and refine the seeding methods for any future operational and/or
research program.
9. This was the first time that a dual-polarization radar was available to a project of this kind
anywhere in the world. Insufficient time was available in the contract period to conduct
an in-depth analysis of this data as new analysis methods are required. The data can
provide vital new insights in seeding effects on clouds and should be further analyzed. It
is recommended that support for further analysis of this data be provided:
a. To utilize the dual-polarization radar data to improve our understanding of natural
and seeded precipitation formation and study changes in cloud microphysics and
dynamics from seeding
b. To validate the hygroscopic seeding conceptual model (that seeding modifies the
initial drizzle and raindrop spectra aloft in clouds).
217
References
Abbs, D. and K. McKinnes, 2004: The impact of climate change in extreme rainfall and coastal
sea levels over south-east Queensland, Part 1. Report for Gold Coast City Council,
CSIRO.
Andsanger, K., Beard, K.V., Laird, N.F. (1999) Laboratory Measurements of Axis Ratios for
Large Raindrops, J. Atmos. Sci., 56, 2673–2683.
Arakawa, A. and W. Schubert, 1974: Interaction of a cumulus cloud ensemble with the largescale environment. Part 1. J. Atmos. Sci., 31, 674–701.
Beard, K. V., Chuang, C. (1987) A new model for the equilibrium shape of raindrops. J. Atmos.
Sci., 44, 1509–1524.
Bigg, E. K. 1997. An independent evaluation of a South African hygroscopic cloud seeding
experiment, 1991–1995. Atmos. Res. 43, 111–27.
Biswas, K.R., and A.S. Dennis, 1971: Formation of rain shower by salt seeding. J. Appl. Meteor.,
10, 780–784.
Bowen, E. G. 1952. A new method of stimulating convective clouds to produce rain and hail. Q.
J. Roy. Meteorol. Soc. 78:37–45.
Braham, R.R. Jr., 1952: The water and energy budgets of the thunderstorm and their relation to
thunderstorm development. J. Meteor., 9, 227–242.
Brandes, E. A., 1976: Flow in severe thunderstorms observed by dual-Doppler radar. Mon. Wea.
Rev., 105, 113–120.
Brandes, E. A., G. Zhang, and J. Vivekanandan, 2002: Experiments in rainfall estimation with a
polarimetric radar in a subtropical environment. J. Appl. Meteor., 41, 674–685.
Brandes, E. A., G. Zhang, and J. Vivekanandan, 2005: Corrigendum. J. Appl. Meteor., 44,
461–475.
Brandes, E. A., G. Zhang and J. Vivekanandan, 2004: Comparison of polarimetric radar drop
size distribution retrieval algorithms, J. Atmos. Oceanic Technol., 21, 584–598.
Bringi, V. N., R. M. Rasmussen and J. Vivekanandan, 1986: Multiparameter radar measurements
in Colorado convective storms, Part I: Graupel melting studies. J. Atmos. Sci., 43,
2545–2562.
Bringi, V., and A. Hendry, 1990: Technology of Polarization Diversity Radars for Meteorology.
Radar Meteorology, David Atlas, Ed., Amer. Meteor. Soc., 153–190.
Bringi, V. N., and V. Chandrasekar, 2001: Polarimetric Doppler Weather Radar. Principles and
Applications. Cambridge University Press, 636 pp.
Bringi, V. N., Tang, T., Chandrasekar, V. (2004) Evaluation of a New Polarimetrically Based
Z-R Algorithm., J. Atmos. Oceanic Technol., 21, 612–623.
Cahalan, R. and J. Joseph, 1989: Fractal statistics of cloud fields. Mon. Wea. Rev., 117, 261–272.
Cai, W., P.H. Whetton, and A.B. Pittock, 2001: Fluctuations in the relationship between ENSO
and northeast Australian rainfall. Climate Dynamics, 17, 421-432.
Charlson R.J., S.E. Schwartz, J.M. Hales, R.D. Cess, J.A. Coakley, J.E. Hansen and
D.J. Hofmann, 1992: Climate forcing by anthropogenic aerosols, Science, 255, 422–430.
Connor, G.J., and M. Bonnell, 1998: Air mass and dynamic parameters affecting trade wind
precipitation on the northeast Queensland tropical coast. Inter. J. Climatol., 18,
1357–1372.
Cooper, W.A., R.T. Bruintjes, and G.K. Mather, 1997: Some calculations pertaining to
hygroscopic seeding with flares. J. Appl. Meteor., 36, 1449–1469.
218
Cotton, W. R., 1972: Numerical simulation of precipitation development in supercooled cumuli.
Part II. Mon. Weather Rev., 11, 764–84.
Cotton, W. R., 1982: The modification of warm clouds—a review. Bull. Amer. Meteor. Soc., 63,
146–160.
Cotton, W. R., and R.A. Anthes, 1989: Storm and Cloud Dynamics. Academic Press, San Diego,
CA. International Geophysics Series, vol. 44.
Cruz, C. N. and S. N. Pandis, 1997: A study of the ability of pure secondary organic aerosol to
act as cloud condensation nuclei. Atmos. Environ., 31, 2205–2214.
Cruz, C. and S. Pandis, 1998: The effect of organic coatings on the cloud condensation nuclei
activation of inorganic atmospheric aerosol. J. Geophys. Res., 103, 13,111–13,123.
Dayan, U., and I. Levy, 2005: The Influence of Meteorological Conditions and Atmospheric
Circulation Types on PM10 and Visibility in Tel Aviv. J. Appl. Meteor., 44, 606–619.
Dixon, M. and G. Weiner, 1993: TITAN: Thunderstorm identification, tracking, analysis, and
nowcasting — a radar based methodology. J. Atmos. Sci. Technol., 10, 785–797.
Draxler, R.R., 1996: Boundary layer isentropic and kinematic trajectories during the August
1993 North Atlantic Regional Experiment Intensive. J. Geophys. Res., 101,
29 255–29 268.
Draxler, R.R., and G. D. Hess, 1998: An overview of the HYSPLIT_4 modeling system for
trajectories, dispersion, and deposition. Aust. Meteor. Mag., 47, 295–308.
Drosdowsky, W., and L.E. Chambers, 2001: Near-global sea surface temperature anomalies as
predictors of Australian seasonal rainfall. J. Climate, 14, 1677–1687.
Evans, K.F., D. O‘Connor, P. Zmarzly, and R.P. Lawson, 2006: In Situ Cloud Sensing with
Multiple Scattering Lidar: Design and Validation of an Airborne Sensor. J. Atmos.
Oceanic Technol., 23, 1068–1081.
Facchini, M. C., M. Mircea, S. Fuzzi, and R. J. Charlson, 1999: Cloud albedo enhancement by
surface-active organic solutes in growing droplets, Nature, 401, 257–259.
Farley, R. D., and C. S. Chen, 1975: A detailed microphysical simulation of hygroscopic
seeding on the warm rain process. J. Appl. Meteor., 14, 718–733.
Fawcett, R.J.B., and R.C. Stone, 2009: A comparison of two seasonal rainfall forecasting
systems for Australia. Australian Meteorological Magazine (accepted).
Gabriel, K.R., 2000: "Parallels between Statistical Issues in Medical and Meteorological
Experimentation", Journal of Applied Meterology, 39, 1822–1836.
Gaudichet, A., Echalar, F., Chatenet, B., Quisefit, J. P., Malingre, G., Cachier, H., Buatmenard,
P., Artaxo, P. and Maenhaut, W., 1995: Trace-Elements in Tropical African Savanna
Biomass Burning Aerosols. Journal of Atmospheric Chemistry 22:19–39.
Görgens, A.H. and A. Roosenboom, 1990: Potential impacts of rainfall stimulation in South
Africa: A Research Planning Study. NSI Report No. 1665/5028, Ninham Shand Inc.,
PO Box 1347 Cape Town, South Africa.
Gorgucci, E., G. Scarchilli, and V. Chandrasekar, 1999: A procedure to calibrate multiparameter
weather radar using the properties of the rain medium. IEEE Trans. Geosci. Remote
Sensing, 37, 269–276.
Grantz, K., B. Rajagopalan, M. Clark, and E. Zagona, 2005: A technique for incorporating largescale climate information in basin-scale ensemble streamflow forecasts, Water Resour.
Res., 41, W10410.
Guenther, A., T.Karl, P. Harley, C. Wiedinmyer, P. I. Palmer, C. Geron, 2006: Estimates of
global terrestrial isoprene emissions using MEGAN (Model of Emissions of Gases and
Aerosols from Nature). Atmospheric Chemistry and Physics, 6, 3181–3210.
219
Hallet, J., and S.C., Mossop, 1974: Production of secondary ice crystals during the riming
process. Nature, 249, 26–28.
Hartigan, J.A. and M.A. Wong, 1979: A K-means clustering algorithm. Applied Stats., 28,
100–108.
Hindman, E.E., 1978: Water droplets fogs formed from pyrotechnically generated condensation
nuclei. J. Wea. Modif., 10, 77–96.
Hopkins, J.R., C.J. Barnett, A.C. Lewis, P.W. Seakins, 2002: Measurements of C6–C8
hydrocarbons at a UK rural site during January 1999. Site evaluation and correlations
between species, J. Environ. Monit., 5, 14–20.
Hopkins, L. and G. Holland, 1987: Australian heavy-rain days and associated east coast cyclones:
1958–92. J. Climate, 10, 621–635.
Houze, R.A. Jr., 1993: Cloud Dynamics, Academic Press, Inc., Orlando, FL, 573 pp.
Houze, R. and C. Cheng, 1977: Radar characteristics of tropical convection observed during
GATE: mean properties and trends over the summer season. Mon. Wea. Rev., 105,
964–980.
Joss, J., and A. Waldvogel, 1967: Ein spectrograph für Niederschagstropfen mit automatisher
Auswertung (A spectrograph for the automatic analysis of raindrops). Pure Appl.
Geophys., 68, 240–246.
Kalnay, E., and co-authors, 1996: The NCEP/NCAR 40-year reanalysis project, Bull. Am.
Meteorol. Soc., 77, 437–471.
Koenig, L. R., and F. W. Murray. 1976. Ice-bearing cumulus cloud evolution: Numerical
simulation and general comparison against observations. J. Appl. Meteor. 7, 747–62.
Kruger, A., and W. F. Krajewski, 2002: Two-dimensional video disdrometer: A description.
J. Atmos. Oceanic Technol., 19, 602–617.
Liu, P. S. K., W. R. Leaich, C. M. Banic, and S. M. Li, 1996: Aerosol observations at Chebogue
Point during the 1993 North Atlantic Regional Experiment: Relationships among cloud
condensation nuclei, size distribution and chemistry, J. Geophys. Res., 101,
28,971–28,990.
Loader, C., 1999: Local Regression and Likelihood, Statistics and Computing, 290 pp., Springer,
New York.
Lopez, R., 1976: Radar characteristics of the cloud population of tropical disturbances in the
northwest Atlantic. Mon. Wea. Rev., 104, 268–283.
Lopez, R., 1977: The lognormal distribution and cumulus cloud populations. Mon. Wea. Rev.,
105, 865–872.
Markowski, P.M., Jerry M. Straka, and Erik N. Rasmussen, 2002: Direct Surface
Thermodynamic Observations within the Rear-Flank Downdrafts of Nontornadic and
Tornadic Supercells, Mon. Wea. Rev., 130, 1692–1721
Martin, D., G. Bergametti, and B. Strauss, 1990: On the use of the synoptic vertical velocity in
trajectory model: Validation by geo-chemical tracers. Atmos. Environ., 24, 2059–2069.
Marwitz, J. D., A. H. Auer, Jr. and D. L. Veal, 1972: Locating the organized updraft on severe
thunderstorms. J. Appl. Meteor., 11, 236–238.
Mather, G. K., D. E. Terblanche, f. e. Steffens and L. Fletcher, 1997: Results of the South
African cloud-seeding experiments using hygroscopic flares, J. Appl. Meteor., 36,
1433–1447.
Matthews, C. and B. Geerts, 1995: Characteristic thunderstorm distribution in the Sydney area.
Aust. Meteor. Mag., 44, 127–138.
220
McBride, J.L., and N. Nicholls, 1983: Seasonal relationships between Australian rainfall and the
Southern Oscillation. Mon. Wea. Rev., 111, 1998–2004.
Mills, G.A. and J. R. Colquhoun, 1998: Objective prediction of severe thunderstorm
environments: Preliminary results linking a decision tree with an operational regional
NWP model, Wea. Forecast., 13, 1078–1092.
Murphy, B.F. and J. Ribbe, 2004: Variability of southeast Queensland rainfall and its predictors.
International Journal of Climatology, 24 (6), 703–721.
Murty, A.S.R., and co-authors, 2000: 11-year warm cloud seeding experiment in Maharashtra
state, India. J. Wea. Mod., 32, 10–20.
Neggers, R., H. Jonker, and A. Siebesma, 2003: Size statistics of cumulus cloud populations in
large-eddy simulations. J. Atmos. Sci., 60, 1060–1074.
Newman, A. J., P. A. Kucera, and L. F. Bliven, 2009: Presenting the snowflake video imager.
J. Atmos. Oceanic Technol., 26, 167–179.
Novakov, T. and Penner, J. E., 1993: Large contribution of organic aerosols to cloudcondensation nuclei concentrations. Nature, 365, 823–826.
Nuijens, L., B. Stevens, and A. Siebesma, 2009: The environment of precipitating shallow
cumulus convection. J. Atmos. Sci., 66, 1962–1979.
Peter, J.R., M.J. Manton, L. Wilson, S.M. Collis, P.T. May, and R. Potts, 2009: Radar-derived
characteristics of storms in South East Queensland. J. Appl. Meteor. Climatol., in review.
Plank, V., 1969: The size distribution of cumulus clouds in representative Florida populations.
J. Appl. Meteorol., 8, 46–67.
Plant, R., 2009: Statistical properties of cloud lifecycles in cloud-resolving models. Atmos. Chem.
Phys., 9, 2195–2205.
Potts, R., 2005: The application of TITAN for thunderstorm nowcast operations. Hydrological
applications of weather and climate modeling: Extended abstracts of presentations at the
seventeenth BMRC modeling workshop, 3–6 October 2005, BMRC Research Report
No 111., Bureau of Meteorology.
Potts, R., T. Keenan, and P. May, 2000: Radar characteristics of storms in the Sydney area. Mon.
Wea. Rev., 128, 3308–3319.
Protat, A. and I. Zawadzki, 1999: A Variational Method for Real-Time Retrieval of ThreeDimensional Wind Field from Multiple-Doppler Bistatic Radar Network Data. Journal of
Atmospheric and Oceanic Technology, 16, 432–449.
Pruppacher, H. R., and J. D. Klett, 1998: The Microphysics of Clouds and Precipitation, 954 pp.,
Kluwer Academic Press, Amsterdam, 1998.
Quayle, E.T., 1929: Long range rainfall forecasting from tropical (Darwin) air pressures.
Proceedings of the Royal Society of Victoria, 41, 160–164.
Ray, P. S., C. L. Ziegler, W. Bumgarner, and R. J. Serafin, 1980: Single- and Multiple-Doppler
Radar Observations of Tornadic Storms. Monthly Weather Review, 108, 1607–1625.
Raymond, W., 1997: A theoretical evaluation of the relevance of lognormal distributions for the
moisture flux and wind components. Mon. Wea. Rev., 125, 3018–3023.
Raymond T. M, Pandis S. N., 2002: Cloud activation of single-component organic aerosol
particles. J. Geophys. Res., 107, doi:10.1029/2002JD002159.
Rivera-Carpio, C. A., C. E. Corrigan, T. Novakov, J. E. Penner, C. F. Rogers, and J. C. Chow,
1996: Derivation of contributions of sulphate and carbonaceous aerosols to cloud
condensation nuclei from mass size distributions. J. Geophys. Res., 101, 19 483–19 493.
221
Rogers, D.C., J. Hallet, C. Twohy, A. Schanot, G. Vidaurre, J. Jensen, J. Stith, (2006, July):
Comparison of LWC Measurements on the NCAR C-130 in AIRS-2. 12th Conference on
Cloud Physics. American Meteorological Society, Madison, WI.
Rokicki, M. L., and K. C. Young, 1978: The initiation of precipitation in updrafts. J. Appl.
Meteor., 17, 745–754.
Rosenfeld D., R. Lahav, A. P. Khain, M. Pinsky. 2002. The role of sea-spray in cleansing air
pollution over ocean via cloud processes. Science, 297, 1667–70.
Ryzhkov, A.V., S. E. Giangrande, V. M. Melnikov, and T. J. Schuur, 2005: Calibration issues of
dual-polarization radar measurements. J. Atmos. Oceanic Technol, 22, 1138–1155.
Scialom, G. and Y. Lemaître, 1990: A New Analysis for the Retrieval of Three-Dimensional
Mesoscale Wind Fields from Multiple Doppler Radar. Journal of Atmospheric and
Oceanic Technology, 7, 640–665.
Scott, B. C., and P. V. Hobbs. 1977. A theoretical study of the evolution of mixed-phase cumulus
clouds. J. Atmos. Sci. 34:812–26.
Shantz, N.C., W.R. Leaitch, and P.F. Caffrey, 2003: Effect of organics of low solubility on the
growth rate of cloud droplets. J. Geophys. Res., 108, doi:10.1029/2002JD002540.
Sheppard, B. E., and P. I. Joe, 1994: Comparison of raindrop size distribution measurements by a
Joss–Waldvogel disdrometer, a PMS 2DG spectrometer, and a POSS Doppler radar.
J. Atmos. Oceanic Technol., 11, 874–887.
Silverman, B. A., 2000: An independent statistical reevaluation of the South African hygroscopic
flare seeding experiment. J. Appl. Meteorol. 39(8):1373–78.
Silverman, B.A., and W. Sukarnjanaset, 2000: Results of the Thailand warm-cloud hygroscopic
particle seeding experiment. J. Appl. Meteor., 39, 1160–1175.
Stohl, A., 1998: Computation, accuracy and applications of trajectories: A review and
bibliography. Atmos. Environ., 32, 947–966.
Stone, R.C., and A. Auliciems, 1992: SOI phase relationships with rainfall in eastern Australia.
International Journal of Climatology, 12, 625–636.
Stone, R.C., G.L. Hamme, and T. Marcussen, 1996: Prediction of global rainfall probabilities
using phases of the Southern Oscillation Index. Nature, 384, 252–255.
Tegen, I., and I. Fung, 1995: Contribution to the atmospheric mineral aerosol load from land
surface modification. J. Geophys. Res., 100, D9, 18 707–18 726.
Tessendorf, S., and coauthors, 2008: Interim report on Southeast Queensland Cloud Seeding
Research Program. NCAR, Boulder, CO, 193 pp.
Tessendorf, S., and coauthors, 2009: Interim report on Southeast Queensland Cloud Seeding
Research Program Season Two. NCAR, Boulder, CO, 112 pp.
Toledano, C., V.E. Cachorro, A.M. de Frutos, B. Torres, A. Berjón, M. Sorribas, and R.S. Stone,
1998: Air mass Classification and Analysis of Aerosol Types at El Arenosillo (Spain). J.
Appl. Meteor. Climatol., 48, 962–981.
Tsai, W. T., 2007: The Decomposition Products of Sulfur Hexafluoride (Sf6): Reviews of
Environmental and Health Risk Analysis. Journal of Fluorine Chemistry 128:1345–1352.
Vivekanandan, J., D. Zrnic, S. Ellis, R. Oye, A. Ryzhkov, and J. Straka, 1999: Cloud
microphysics retrieval using S-band dual-polarization radar measurements. Bull. Amer.
Meteor.. Soc., 80, 381–388.
Vivekanandan, J., B. E. Martner, M. K. Politovich, and G. Zhang, 1999a: Retrieval of
atmospheric liquid and ice characteristics using dual-wavelength radar observations.
IEEE Trans. Geosci. Remote Sens., 37, 2325–2334.
222
Wakimoto, R. M. and V. N. Bringi, 1988: Dual-polarization observations of microbursts
associated with intense convection: The 20 July storm during the MIST Project. Mon.
Wea Rev., 116, 1521–1539.
WMO (World Meteorological Organization), 2000: Report on the WMO International workshop
on Hygroscopic seeding: Experimental results, physical processes and research needs.
Geneva: WMO.
Young, K. C., 1996; Weather modification – a theoretician‘s viewpoint. Bull. Amer. Meteor. Soc.,
77, 2701–10.
Zmarzly, P.M and R.P. Lawson. 2000: An Optical Extinctionmeter for Cloud Radiation
Measurements and Planetary Exploration. Final Report submitted to NASA Goddard
Space Flight Center in fulfillment of Contract NAS5-98032, September 2000, 131 pp.
223
Acknowledgements
This project is a collaborative effort by several institutions, and sponsored by the
Queensland Government through the Environmental Protection Agency‘s Queensland Climate
Change Centre of Excellence (QCCCE). The institutions involved in this project included the
National Center for Atmospheric Research (NCAR), the Australian Bureau of Meteorology
(BoM), the Centre for Australian Weather and Climate Research (CAWCR), Monash University,
the University of Southern Queensland (USQ), Australia's Commonwealth Scientific and
Industrial Research Organization (CSIRO), the University of Witwatersrand (WITS), the South
African Weather Service (SAWS), Orsmond Aviation, Weather Modification Inc. (WMI), MIPD
Pty Ltd, Arizona State University (ASU), and Texas A&M University.
The success of the program is the result of a combination of efforts by several people
who must all be acknowledged. First off, we wish to acknowledge Dave Mcrae (QCCCE) for
providing excellent program management and logistics support for both seasons of the project.
Lynne Turner (QCCCE), Roger Stone (USQ), Michael Manton (Monash), Peter May (CAWCR),
Steve Siems (Monash), Tom Keenan (CAWCR), and Deborah Abbs (CSIRO) have provided
scientific guidance to the project on the Science Advisory Group. The CP2 radar was maintained
by Ken Glasson (BoM) and other BoM staff. Ken also oversaw the operation of the
disdrometers deployed at the Willawong Animal shelter (who kindly lent us space for the
instruments and computers) and he ensured all radar and disdrometer data was regularly backed
up. The forecasting efforts for this project were vital to planning operations, and thus we greatly
thank the forecasters (provided by the BoM/CAWCR): Scott Collis, Harald Richter, Tony
Bannister, and Justin Peter. Special thanks to Scott Collis for setting up the archive of weather
data and providing us with special access to internal BoM products, as well as for forecasting
remotely during the project extension time period. The aircraft operations for SEEDA1 were
maintained by a team of staff from Orsmond Aviation and/or contracted through SAWS/WITS,
and included Steve Edwards and Nico Meinie for instrument maintenance; Harry McGarry for
aircraft maintenance; Ret Orsmond, Hans Kruger, Gary Wiggins, and John Hingst as pilots;
Stuart Piketh, Shaazia Bhailall, Nicola Walton, Jared Lodder, Xolile Ncipha, and Stephen
Broccardo for instrument operation and data quality control. The aircraft operations for
WXMOD were maintained by a team of staff from WMI and MIPD, and included Jim Carr, Nev
O‘Donnell, Paul Brady, and Greg Choma as pilots, Rafe Zerby and Mike Clancy for aircraft
maintenance, and Noel O‘Doherty and Lindsay Bishop for logistics management. Ian Craig
(USQ) also provided data management and data quality control support for the aircraft
operations, and Li Fitzmaurice (QCCCE) and Martin van Nierop, Nico Kroese, and Karel de
Waal (SAWS) helped with this in season one. Justin Peter (BoM/Monash) provided radar
operations support, forecasting assistance, and scientific analysis in the field and Louise Wilson
(Monash) and Acacia Pepler (University of New South Wales) provided radar operations
assistance and scientific analysis in the field. NCAR staff (Mike Dixon, Jim Wilson, Ed Brandes,
Rita Roberts, Eric Nelson) provided radar operations support and software development. In
particular, we thank Mike Dixon for setting up the radar displays and upgrading display features
for use in the field program. MIPD oversaw the flare management and safety logistics in both
seasons.
224
Outside of the field program, data processing efforts have been undertaken and should be
acknowledged as well. Roelof Burger (WITS) has been the primary person working on the
aircraft data processing and data quality control efforts. The Texas A&M staff has processed the
DMA data. Bruce Woodcock (Aventech) has worked with WITS and NCAR to process the
AIMMS data and provided extensive support during the field project. Peter Buseck and Evelyn
Freney (ASU) processed and analyzed the aerosol filter samples. Paul Kucera (NCAR) and
Andy Newman have processed and quality controlled the NASA disdrometer data and Kyoko
Ikeda (NCAR) has processed the NCAR (season one) and CAWCR (season two) disdrometer
data. Shelly Knight and Nancy Rehak (NCAR) have provided data management and software
assistance at NCAR, and Lara Ziady (NCAR) was the web designer for the RAL Queensland
CSRP website (that was also used in the field for daily report submissions in season two).
225
7 Appendix A—Data Quality
7.1 Aircraft Instrument Intercomparisons
In season one, one flight was coordinated with the 00Z weather balloon release from the
Brisbane airport to compare measurements. This sounding flight shows fairly good agreement
between the aircraft and balloon measurements (Figure 7.1). The agreement was best for the
Rosemount temperature probe (though there is a slight deviation aloft, likely due to different
balloon/aircraft paths). Although it captured general trends compared to the balloon profile, the
dew point measurement (from the Vaisala instrument) in season one was less accurate, especially
at the start of the balloon ascent when measurements should have been most comparable, and
thus warranted purchasing a new dew point instrument for the second field season.
Figure 7.1. Comparison between balloon and aircraft measurements of air temperature
and dew point temperature on 9 March 2008.
Three sounding comparison flights were conducted in season two, two of which are
depicted in Figure 7.2 (the third was conducted early in the season when some instruments were
still being tested). The dew point temperature compares very well for the 20 January sounding,
but the balloon temperature was warmer than that measured on the aircraft, especially in the
strength of the inversion above ~800 hPa; however, the balloon temperature on this day may
have been suspect due to a late release of the balloon (it may have sat in the sun) given the
unrealistic superadiabatic lapse rate measured by the balloon near the surface. Nonetheless, the
temperature and dew point temperature comparisons are very good in the 24 February profile
(Figure 7.2b). Both of these two sounding comparison flights had AIMMS wind measurements
(Figure 7.2), which were noisier than those measured by the balloon, but the general trend agrees
with the weather balloon remarkably well.
226
a)
b)
Figure 7.2. Comparison between aircraft and weather balloon measurements of air
temperature, dew point temperature, and wind speed and direction for a) 20 January 2009
and b) 24 February 2009.
The curves in Figure 7.3 show two examples of aerosol size distributions (ASDs) from
season two. This illustrates how the DMA, SPP-200, and SPP-100 compare to one another for
season two. The probes seem consistent in their overlap regions, but both the SPP-200 and
SPP-100 seem to suffer slightly in their smallest size bins; thus, any size distribution analysis
where all three measurements are available utilizes the DMA in the DMA/SPP-200 overlap
range and the SPP-200 in the SPP-200/100 overlap range. Unfortunately, the performance and
data collected by the DMA and PCASP was irregular in season one (the FSSP, however, had
regular measurements), such that we are unable to construct complete ASDs from season one.
a)
b)
Figure 7.3. Comparison of DMA, SPP-200, and SPP-100 aerosol size spectra for a) 17
January 2009, and b) 25 January 2009.
227
Many of the probes used in the project are capable of calculating liquid water content.
The King and the CAPS liquid waters are both direct hotwire measurements, while the
FSSP/SPP-100 is derived from measured drop spectra. Both hotwire measurements need to be
corrected for the amount of cooling experienced in dry air; to do this, an average value was
calculated when out of cloud (as determined by FSSP/SPP-100 total concentrations) and
subtracted from the total raw value. A comparison of the measurements from the hotwires and
the derived liquid water from the FSSP/SPP-100 is shown in Figure 7.4. The season one data
indicates close agreement between the King hotwire and FSSP measurements, with the FSSP
slightly underestimating the King. The linear fits imply that both hotwires reasonably agree with
one another in season two, but the SPP-100 typically underestimates liquid water compared to
either hot wire. This trend of spectral-derived liquid water underestimating hotwire
measurements has been noted in several other projects (Evans et al., 2006; Zmarzly and Lawson,
2000, Rogers et al., 2006).
a)
b)
Figure 7.4. Scatter and linear fit of hotwire and derived liquid water contents for a) season
one and b) season two (King hotwire in blue, CAPS hotwire in green). The one-to-one
comparison is shown in black.
During season one, SEEDA1 was equipped with both the PMS FSSP and DMT SPP-100
FSSP during March 2008. As Table 1.1 shows, these both measure cloud droplet spectra between
0.5 and 47 µm. The two instruments operate on the same principles of light scattering; however
the SPP-100 has superior high-speed circuitry. As Figure 7.5 illustrates, the FSSP typically
measured slightly higher concentrations than the SPP-100. Figure 7.6a-d shows a number of drop
size distributions from both instruments, which also show that the SPP-100 measured lower
concentrations, and it also appears that while the shape of the distribution is similar, the SPP-100
shows a shift towards larger diameters as compared to the FSSP. Despite differences in spectral
measurements, the two instruments‘ derived liquid water contents are highly similar (Figure
7.5b).
228
a)
b)
Figure 7.5. Scatter plot comparison of SPP-100 and FSSP a) total concentrations and b)
derived liquid water content for March 2008; black line is 1:1 and blue line is the linear fit.
a)
b)
c)
d)
Figure 7.6. Drop size distributions for SPP-100 (red) and FSSP (blue) at the maximum
drop concentration on (a) 6 March, (b) 13 March, (c) 15 March, and (d) 24 March 2008.
229
Measurements for the 2DC, 2DP, and the CIP optical array probes from season one have
been compared. The data quality from the CIP probe was quite good while the 2DC and 2DP
probes were often wet, which resulted in unacceptable data. This wetting problem was severe for
the 2DP probe. Nonetheless, there are some days with good data, and an example spectrum from
season one is shown in Figure 7.7.
Figure 7.7. Comparison of SPP-100, 2DC, 2DP and CIP data in season one on 2 February
2008.
During season two, the CIP probe continued to provide very good data. The curves in
Figure 7.8 shows two examples of drop size distributions measured in a cloud during a cloud
penetration from season two. This plot demonstrates how the three drop-measuring probes
(SPP-100, CIP, and PIP) performed compared to one another, and the agreement across the size
ranges attests to the quality of the data that was collected.
230
a)
b)
Figure 7.8. Comparison of SPP-100 (FSSP), CIP, and PIP drop spectra in season two from
a) 22 January 2009, and b) 21 February 2009.
7.2 Radar and Disdrometer Measurements
Dual-polarization radars typically transmit horizontally and vertically polarized waves
and receive backscattered signals. Because illuminated particles are not spherical, their radar
backscattering cross sections are not the same for the two polarizations. The electromagnetic
waves are subject to scatter, differential attenuation, differential phase shifts, and depolarization.
Changes in returned signals yield information regarding particle size, shape, orientation, and
thermodynamic phase.
Interpretation of polarimetric measurements is facilitated by knowing the properties of
hydrometeors and how radar measurements respond to them. This is readily determined by
computing radar variables for disdrometer-observed drop populations. This section describes the
computational procedures for determining drop distributions from the disdrometer
instrumentation described in Section 1.2.4. The various instruments are compared and calibrated.
Disdrometer measurements provide an independent method for verifying the CP2 radar
calibration.
7.2.1 Gamma drop size distribution model
For this study, raindrops are assumed to be represented by a gamma distribution model

N ( D )  N 0 D exp(   D )
,
(1)
where N0 (mm−μ−1 m−3) is a number concentration parameter, μ is a distribution shape parameter,
Λ (mm−1) is a slope term, and D (mm) is the drop equivalent volume diameter. The three
231
governing parameters of the distribution were determined using the 2nd, 4th, and 6th moments of
the observed drop distributions.
Drop median volume diameter D0 was computed from

D0

D N ( D ) dD 
3
D m in
D m ax
3
D N ( D ) dD
D0
,
(2)
where Dmax is the diameter of the largest observed drop in the distribution and Dmin is the
smallest drop. Rain rates (mm h–1) are given by
R  6   10
4

D m ax
D m in
3
D v t ( D ) N ( D ) dD
.
(3)
Drop terminal velocities vt(D) are computed as in Brandes et al. (2002). Finally, the liquid water
content (LWC) for precipitation-size drops is given by
LW C 

D m ax
D m in
 w   10
3
3
(4)
D N ( D ) dD
6
where ρw is the density of water.
7.2.2 Disdrometer comparison
An example of a DSD sampled by the RD-80 is presented in Figure 7.9. Concentrations
are based on observed drop counts, as provided by Distromet LTD software. The correction of
Sheppard and Joe (1994) has not been applied. Fitted curves for assumed truncated exponential
and gamma distribution models are plotted. (The 3rd and 6th moments are used for fitting the
exponential model.) Derived DSD attributes are listed in Table 7.1. For the observation shown
the maximum drop detected was 5.37 mm, the median volume diameter was 2.32 mm. Note that
the distribution is flat, that is, N(D) is near constant, for drop diameters between 0.5 and 1.8 mm.
Application of the dead time adjustment (Figure 7.10) increases the number concentration
of small drops. In fact, the estimated total number of drops rises 40% from 1052 to 1480. This
makes the distribution more exponential as manifest by the decrease in the shape term. This may
or may not be a good thing depending on the true DSD shape. The adjustment lowers the
estimated D0 from 2.32 mm to 2.17 mm and increases the estimated rain rate from 58.2 mm h−1
to 63.3 mm h−1. Clearly, the adjustment is significant.
The observed DSD from the 2DVD (Figure 7.11) reveals less down-turning at small drop
sizes. This is largely due to increased numbers of detected small drops. Close inspection reveals
that the 2DVD also had somewhat higher drop concentrations in the 2 to 3.5 mm size interval.
The total number of drops detected by the instrument was 5446 (Table 7.1). The estimated rain
rate is similar to that of the adjusted RD-80 rate. This is usually true for other computed
parameters, such as ZH and ZDR, as well.
232
Time series of measurements made with the BoM 2DVD and RD-80 are shown in Figure
7.12 and Figure 7.13. Even though a dead time correction has been applied it is readily apparent
that the drop counts with the RD-80 are much lower than with the 2DVD. The primary cause for
the drop count discrepancy is the occurrence of large drops and the associated long time for
instrument restoration. There is good agreement between the characteristic drop sizes D0 and
Dmax but a considerable difference in the estimated shape parameters of gamma DSD model.
Larger values with the 2DVD indicate narrower distributions. Further, the DSD shape factors
computed from the BoM 2DVD data are so noisy it is difficult to discern patterns in the data.
A direct comparison between the NCAR and BoM 2DVDs is not possible. However, we
can compare summaries for season one and season two. Measurements collected over long time
periods and for a large number of storms should exhibit a similar climatology (Figure 7.14 and
Figure 7.15). Median volume diameters computed for the NCAR 2DVD are much smaller than
that seen by the BoM 2DVD (or the RD-80). For the NCAR 2DVD the frequency distribution
for median volume diameter peaks at about 0.6 mm, whereas the D0 distribution for the BoM
2DVD peaks at 1.0 mm. For maximum drop diameters the peaks are 1.2 and 2.0 mm,
respectively. This is a huge difference. A different mix of storms may contribute to the
discrepancy, but the primary cause is believed to lie with instrument sensitivity. Small drop
issues with the impact disdrometer artificially shift D0 and Dmax estimates to large values. Hence,
agreement between the BoM RD-80 and BoM 2DVD may not be a good thing. Clearly,
additional study of the disdrometer datasets is needed.
The NASA PVI disdrometer data takes detailed manual processing and editing for quality
control. This process is time consuming, and has not been able to be completed to date, but
future work will include a comparison of the PVI data with the BoM disdrometer data from
season two.
Table 7.1. Computed drop size distribution attributes for spectra in Figure 7.9 and Figure
7.10. RD-80 values without the dead time correction (unadjusted) and with the correction
(adjusted) are given.
RD-80
Unadjusted
RD-80
Adjusted
2DVD
Drop count (min−1)
1052
1480
5446
Rain rate (mm h−1)
58.2
63.3
63.7
Shape parameter
3.8
2.6
1.4
Slope term (mm−1)
3.0
2.7
2.3
Median
volume
diameter (mm)
2.32
2.17
2.25
233
Figure 7.9. Measurements from the BoM RD-80 (20 February 2009, 14:43 UTC). A dead
time correction has not been applied.
Figure 7.10. As in Figure 7.9, except that the dead time correction of Sheppard and Joe
(1994) has been applied.
234
Figure 7.11. As in Figure 7.9, except for the BoM 2DVD disdrometer.
235
Figure 7.12. Time series of drop size distribution characteristic computed from
observations obtained with the BoM 2DVD on 20 February 2009.
236
Figure 7.13. As in Figure 7.12, except for the BoM RD-80 disdrometer. The dead time
adjustment was applied.
237
Figure 7.14. Frequency distribution of DSD attributes for convective storms from
measurements obtained with the NCAR 2DVD during season one. (592 data points)
238
Figure 7.15. As in Figure 7.14, except for measurements obtained with the BoM 2DVD
during season two. (1359 data points)
7.2.3 CP2 calibration
Retrieval of drop size distribution information and improvement in rainfall rate
estimation with polarimetric radar measurements is possible because of the differential
reflectivity measurement. The parameter in linear units is defined as the ratio of the reflectivity
at horizontal and vertical polarization. Raindrops are flattened while falling and orient
themselves with their major axis close to horizontal. Thus, the reflectivity at horizontal
polarization is larger than at vertical polarization. As the size of drops increases, the degree of
flattening and ZDR increases. In order to estimate DSD parameters such as the median volume
diameter or maximum drop size from radar measurements, the radar must be well calibrated.
239
7.2.3.1 Radar reflectivity
The reflectivity calibration was checked by comparing radar-measured reflectivity with
values computed from disdrometer observations. Results, based on 78 comparisons for 7 storm
days, revealed a small difference. The radar measurements were 0.3 dB higher than the
disdrometer estimates. In spite of the large number of factors that could influence the
comparison this is an excellent result suggesting the radar reflectivity measurement is well
calibrated.
7.2.3.2 Differential reflectivity
Attempts to verify the differential reflectivity calibration were more challenging. The
measurement can be corrupted several ways, such as rain on the radome which is a temporary
condition that lowers the mean ZDR value over broad regions. (The squall line of
21 October 2008 is an example.) Measurements can also be contaminated through radar antenna
side lobes by ground clutter and by neighboring strong reflectivity cores. Clutter is widespread
with CP2 because of its mountain top location. Mismatches between the horizontal and vertical
radar beams can also be a problem.
One method for verifying the differential reflectivity calibration is to point the radar
antenna vertically during light precipitation (Gorgucci et al. 1999). When viewed from below,
hydrometeors have no preferred orientation and drop flattening is not detectable. Hence, the
measured ZDR at vertical incidence is expected to be 1.0 in linear units or 0 dB. The CP2
antenna was pointed vertically on numerous occasions during the field campaigns. Sample
measurement profiles presented in Figure 7.16 reveal both meteorological features and what
appears to be a radar calibration issue. 4 The radar reflectivity bright band between 4.3 and
5.0 km altitude marks the melting layer. Melting is also indicated by a lowering of the
correlation coefficient ρHV (due to the presence of mixed-phase hydrometeors and a mixture of
particle shapes) and a decrease in Doppler velocity as frozen hydrometeors are transformed to
raindrops and their fall speed increases.
The vertical trace for ZDR points to a problem. The measurement is not 0 dB as
anticipated nor is it constant. (A constant value would indicate a simple power offset between
the two polarimetric channels and would be readily corrected.) Rather, the measured ZDR varies
with height (range). Examination of the reflectivity and differential profiles reveals the
magnitude of ZDR is sensitive to the range gradient of reflectivity. For example, in the height
interval 3.5 to 3.9 km, the reflectivity gradient is approximately +10 dB km−1 and ZDR is as low
as 0.05 dB. In the layer 4.2 to 4.5 km, the reflectivity gradient is +18 dB km−1 and the minimum
ZDR is −0.03 dB. Between 4.7 and 5.1 km the gradient is −18 dB km−1 and ZDR has a maximum
value of +0.37 dB. In the layer from 5.3 to 8 km the reflectivity gradient averages −6 dB km−1
and the ZDR measurement averages +0.23 dB.
4
Large fluctuations in radar measurements within 1.5 km of the radar fit the pattern to be
described, but are within the near field of the radar and should be ignored.
240
The described relationship between reflectivity gradients and the magnitude of ZDR is
seen in all profiles. Unfortunately, the gradient problem often can be seen in radar scans at lowelevation antenna angles as well. Some sort of timing error or small range offset (probably less
than one range gate) between the two radar channels is indicated that causes features at
horizontal polarization to appear at slightly greater distances than at vertical polarization. The
problem is a contributor to the general high noise level in the CP2 ZDR measurements. Because
the ZDR bias is closely tied to the magnitude of the reflectivity gradient, it may be possible to
improve the ZDR measurements in collected datasets with a correction scheme, but that has not
yet been attempted.
To better estimate the ZDR bias a subset of the vertical pointing data, consisting of
profiles with deep layers of near constant reflectivity, was assembled. For the example shown
(Figure 7.17), the computed bias for the layer 1.4 to 4 km was 0.21 dB. The average for profiles
obtained on three days suggested a system bias of 0.21 dB, that is, the radar-measured ZDR was
too high. This differs from season one for which the radar measured ZDR was about 0.45 dB too
low.
Another method for verifying the calibration is to compare disdrometer calculations of
ZDR with radar measurements (Table 7.2).5 The radar measurements could be sensitive to the
gradient issue and ground clutter contamination. Small differences in instrument times could
also be important. For the calculations, the drop aspect ratio of Brandes et al. (2002, 2005) is
used. Results for the RD-80 and 2DVD are presented. For long-lived events (22 January and
12 February) the agreement between disdrometers is excellent, affirming the utility of
disdrometer measurements for calibration verification. The comparison suggests that the CP2
ZDR measurements are a little high in agreement with the vertical pointing radar data. The
disdrometer-estimated bias is somewhat less. The difference could be due to supposed low small
drop concentrations. Of the two ZDR bias values in Table 7.2, the larger one is preferred because
a larger storm sample is represented. Note that the mean bias shown in the table represents a
systematic hardware offset. Gradients are an added problem that can create a measurement bias
of several dB.
5
A third method for evaluating the radar calibration is to test for consistency among radar
measurements. For example, it is possible to compute the differential phase from radar
reflectivity and differential reflectivity. Because time (phase) measurements can be made quite
accurately, a difference between the computed and measured values would indicate an error in
the power measurement of reflectivity or differential reflectivity. Time did not allow for a
consistency test.
241
Figure 7.16. Vertical profiles of radar reflectivity, differential reflectivity, radial velocity,
and correlation coefficient obtained with the CP2 radar on 22 January 2009 at
06:40:52 UTC.
242
Figure 7.17. As in Figure 7.16, except for 6 December 2008 at 08:42:51 UTC.
243
Table 7.2. Comparison of ZDR as computed from disdrometer observations and measured
by the CP2 radar for season two. Results for the RD-80 and 2DVD are shown. The
difference parameter is computed as the disdrometer value minus the radar-measured
value. Only radar measurements with a radar reflectivity ≥ 20 dBZ and a correlation
coefficient ≥ 0.95 are used. The number of time-matched comparisons and apparent
needed CP2 adjustment is also given.
Date
RD-80 Radar Difference
ZDR
8-Nov
0.57
ZDR
0.84
#
2DVD
(dB)
-0.27
ZDR
Radar Difference
ZDR
#
(dB)
16
25-Nov
0.48
0.62
-0.14
17
26-Nov
0.48
0.74
-0.24
47
5-Dec
0.65
0.84
-0.19
36
6-Dec
0.70
0.78
-0.08
33
22-Jan
0.46
0.49
-0.03
30
0.47
0.49
-0.02
30
12-Feb
0.33
0.44
-0.11
71
0.34
0.45
-0.11
70
19-Feb
1.67
1.77
-0.10
5
1.12
1.20
-0.08
3
CP2 ZDR adjustment:
(weighted)
-0.11 dB
-0.13 dB
244
8 Appendix B—Analysis Methods
8.1 Statistical Methods for the Randomized Seeding Trial
8.1.1 Survival analysis for evaluating the duration of targets
Survival analysis is most commonly used in the medical community to study the survival
of members of groups subject to a treatment. A classic example is where patients with a terminal
disease are randomly selected to be given a treatment or a placebo. Some participants may die
during the course of the study. Since a study has a fixed duration, some patients may be alive at
the end of the study. Correctly handling information from the people who are alive at end of the
study is an issue. Patients may leave the study in the middle of the trial. Clearly, one cannot
consider them alive for the remainder of the experiment, but one does not wish to lose the
information present in the fact that they had the disease up until a certain point.
A fundamental assumption for survival analysis is that the event can only occur once.
When the event is death, this assumption is easy to intuit. This has the following implication in
the application of survival analysis to weather modification data. The target is assumed to have
―died‖ when the TITAN signal is lost. Scientists realize this does not mean that a target may still
be producing precipitation. This is acknowledged to be a limitation in the method. This assumes
that from the decision time until the loss of the TITAN signal, the target is considered to be
active.
Some other key concepts and assumptions needed to apply the survival analysis for weather
modification experiments include the following:




An event must be defined and that event can occur only once. As discussed in the
preceding paragraph, the event is defined as the loss of a TITAN signal.
A beginning time must be defined. In this experiment, the decision time (DT) defines the
starting time.
Data is available at fixed intervals. This affects the characteristics of the response and
requires that it be viewed as discrete and not continuous data. With TITAN data based
on radar data, information about a target is only available every five minutes.
Targets which are considered active after 45 minutes are considered censored. Further
information about a target is not available after this time.
A hazard function is the probability an individual will experience the event (i.e., die)
within a given time interval. This can also be interpreted as a conditional probability. Given that
an individual has survived until the current time period, what is the probability that an event
(death) will occur to a specific individual in the next time interval? A hazard plot identifies the
places where the risk to an individual is greatest.
A survival function is the probability an individual will survive until a given point in time.
This is a monotonically decreasing function, which by definition begins near a value of 1 and
declines towards 0.
245
8.2 Derived Microphysical Relationships
8.2.1 Disdrometer-derived relationships
In this section we present expressions for estimating drop median volume diameter,
maximum drop diameter, rainfall rate, and liquid water content. In general, relationships are
derived in terms of radar reflectivity and differential reflectivity. Phase measurements are used
sparingly because of their high noise level. Relationships are derived for convective and
stratiform rain. Discrimination by precipitation type was by examination of the radar echoes and
the time series of disdrometer observations. Convective rain rate estimators for specific
differential phase are derived for possible application in regions with ground clutter and when
hail may contaminate the power measurements. Related estimators for stratiform rain and liquid
water content are not given because the high noise level in the parameter can result in estimates
that exceed expected values for stratiform precipitation.
The distribution of ZH and ZDR, as computed with disdrometer measurements from the
NCAR disdrometer, is shown in Figure 8.1. The range in computed values sets limits for the
algorithms presented. Radar measurements that differ significantly from these distributions
should be viewed with caution. Further, extrapolation beyond the data limits is not
recommended.
Figure 8.1. Radar reflectivity plotted versus differential reflectivity for disdrometer
raindrop measurements.
246
Convective and stratiform rains exhibit different ZH/ZDR distributions. At a reflectivity of
40 dBZ, the minimum ZDR for convection is about 0.5 dB, whereas for stratiform rain the
minimum value is ~1.1 dB. The lower boundary of the distributions is better defined than the
upper boundary. A minimum ZDR value at a specific reflectivity associates with a minimum D0.
DSDs at the bottom of the distribution may result from equilibrium between coalescence and
breakup processes. Larger ZDR values (drop sizes) for stratiform rain probably arise from
aggregation just above and within the melting layer.
Until small drop issues are resolved, relations developed with the NCAR disdrometer
(used in season one) are recommended. For convective rainfall rates (in mm h–1) the suite of
polarimetric relations is
R ( Z H )  0.0396 Z H
0.679
R ( Z H , Z D R )  1.45  10
(5)
2
Z H  10
R ( K D P )  sign ( K D P )53.3 K D P
2
( 0.244 Z D R  1.28 Z D R )
(6)
0.669
R ( K D P , Z D R )  sign ( K D P )192 K D P
(7)
0.946
 3.45
Z DR
,
(8)
where ZH is the radar reflectivity at horizontal polarization, ZDR is the differential reflectivity,
and KDP is the specific differential phase. All units are linear, except in Eq. (6) where ZDR has
units of dB. Relations (5) and (6) are based on a sample of 592 1-min data points. Relations (7)
and (8) are based on a subset of the data (349 observations with KDP >0o km−1). An alternate
algorithm for radar reflectivity that gives a better match for rain rates >40 mm h−1 is
R ( Z H )  0.0262 Z H
0.670
.
(9)
For stratiform storms we have
R ( Z H )  0.0190 Z H
0.678
R ( Z H , Z D R )  1.09  10
(10)
2
Z H  10
2
( 0.414 Z D R  1.35 Z D R )
(11)
The latter expressions for stratiform rain, based on 2346 1-min disdrometer samples, are valid for
rain rates as large as 10–15 mm h−1. Relations for specific differential phase parameter have not
been constructed because the signal at S-band is weak for stratiform rain and the standard error
in the KDP calculation is large (roughly 0.10o km−1 or 10 mm h−1).
Liquid water contents (LWC, g m–3) for precipitation-size raindrops in convective storms
can be estimated with
L W C ( Z H , Z D R )  1.49  10
3
Z H  10
2
( 0.328 Z D R  1.66 Z D R )
247
.
(12)
A corresponding relation for stratiform rain is
L W C ( Z H , Z D R )  9.82  10
4
Z H  10
2
( 0.576 Z D R  1.77 Z D R )
.
(13)
The units in (12) and (13) are linear for ZH and dB for ZDR.
Relations have also been derived for estimating Dmax and D0 (mm) from ZH and ZDR. For
convective rains we derive
D m ax  0.844  0.0171 Z H  0.000278 Z H  0.0000351 Z H
(14)
D m ax  0.963  3.49 Z D R  0.842 Z D R  0.0853 Z D R
(15)
2
3
2
3
D 0  0.427  0.0166 Z H  0.000765 Z H  0.0000208 Z H
(16)
D 0  0.466  1.22 Z D R  0.523 Z D R  0.128 Z D R
(17)
2
3
2
3
.
Similarly, relations for stratiform rain are
D m ax  0.844  0.0342 Z H  0.000482 Z H  0.0000171 Z H
(18)
D m ax  0.874  3.59 Z D R  1.41 Z D R  0.201 Z D R
(19)
D 0  0.528  0.0123 Z H  0.000414 Z H
(20)
2
3
2
3
2
D 0  0.557  1.17 Z D R  0.226 Z D R
2
.
(21)
8.2.2 Quantitative rainfall estimation algorithms
For this study, CP2 polarimetric radar data was retrieved from the first scan at
0.5° elevation at 6-minute intervals for seven high-rainfall events between February 2008 and
February 2009, totaling 20 days (or parts thereof) and 665 complete half hours of radar data. A
simple quality control procedure is applied to remove data with ρHV<0.8 or σ(φDP)>10 over ten
range gates, with values of KDP>5° km-1 or ZDR>5 dB truncated. Boxcar smoothing is applied
over the four adjacent data points prior to any further analysis.
The data is supplemented by 30-minute rainfall totals from a dense tipping-bucket raingauge network (Figure 8.2), with 293 stations within the range covered by the radar,
concentrated to the east along the coastline, including ten high-resolution Automatic Weather
Source (AWS) stations which also record rainfall at 1-minute intervals. Drop size distributions
(DSDs) are also available from a 2D video disdrometer (see also Section 1.2.4) located at
63° azimuth, 16.6 km range, from which can be determined droplet characteristics including
mass-weighted median drop size (DM), number of droplets (NT) and rain rate. Simulated radar
variables were derived using T-matrix scattering calculations, assuming (i) normalized gamma
248
distributions, (ii) the oblate approximation of the Beard and Chuang (1987) shapes, (iii) with
maximum drop diameter 2.5DM, (iv) standard deviation of canting angle of 7°, (v) water
temperature of 20° C, and (vi) for an elevation of 0°. Radar data is compared with ground-based
measurements by smoothing over seven range gates and two azimuth gates, then averaged over
the five 6-minute scans to compare with half-hourly data.
0
150
330
30
100
300
60
50
270
90
240
120
210
150
180
Figure 8.2. Location of gauges used in this study relative to the CP2 radar, with the QLD
coastline marked. Red circles identify those gauges which record 1-minute data, with a
black circle indicating the disdrometer.
Error characteristics will be analyzed in terms of the mean and standard deviation of the
fractional bias (FB), and the fractional root mean squared error (FRMSE), defined below. To
minimize the impact of outliers, a ‗trimmed‘ FRMSE using only data with fractional squared
error between the 10th and 90th percentiles will also be used.
FB i 
R i  R Ti
(22)
R Ti
FRMSE 
 R i  R Ti


N i 1  R Ti
1
N




2
(23)
8.2.2.1 Radar-rainfall algorithms
A total of 18 significant rainfall events, which occurred over the disdrometer between
20 November 2008 and 19 February 2009 were used, with a total of 2299 minutes of non-zero
disdrometer rain data and derived radar variables and 97% of which were less than 35 mm hr-1.
These data were used to determine optimized constants for each the four standard polarimetric
relationships from bootstrapping regression over 1000 samples. The algorithms determined using
this method (and for season two; see Section 8.2 for season one derived relationships) are given
by:
249
Z h  200 R D
1 . 36
(24)
R D ( K DP )  44 K DP
0 .8
(25)
R D  Z h , Z dr   0 . 017 Z h
0 . 84
 4 . 47
Z dr
(26)
R D  K DP , Z dr   88 . 9 K DP Z dr
(27)
0 . 88
 2 . 51
When applied to simulated radar variables and compared with the rain rate derived from
the same DSD data, each relationship had a mean bias less than ±6% at rain rates between
1 mm hr-1 and 35 mm hr-1. The lowest biases were for multiparameter algorithms R(Zh,ZDR) and
R(Kh,ZDR), both of which had FB of +2%. However, the standard deviation of bias varied
significantly between methods, reaching 0.55 for R(Zh) due to several very high outliers, whereas
all other algorithms had standard deviations lower than 0.30. Trimmed FRMSE was also low for
all algorithms at rain rates <35 mm hr-1, decreasing slightly with increasing rain rate for
multiparameter algorithms, but increasing slightly for the R-Z relation (Figure 8.3a). In all cases
the multiparameter algorithms outperform their single parameter counterparts in the absence of
measurement error.
This disdrometer data was then used to simulate real-world radar data by simulating
expected measurement errors. This was achieved by applying to each data-point a set of
100 normally-distributed random numbers with mean zero and standard deviation defined by
theoretical calculations (as given in Bringi and Chandrasekar 2001) using CP2 characteristics
(below). This created a larger dataset of 229 900 simulated data points to use for validation.
 ( Z h )  0 . 479 Z h
(28)
 ( Z dr )  0 . 1360 Z dr
(29)
 ( K DP )  0 . 745 .km

1
(30)
Using this data, very similar error characteristics were seen to that anticipated from
literature research, with KDP-based algorithms exhibiting high error at low rain rates. R(Zh) and
R(Zh,ZDR) perform best at rain rates <40 mm hr-1, with FB only +2% for R(Zh) at these rain rates.
Similarly, where R(KDP) and R(KDP,ZDR) are most valid, at rain rates >40 mm hr-1, R(KDP) has
FB of only +3%. In both cases, the higher measurement error in multiparameter algorithms
appears to counteract the lower parameterization error seen in Figure 8.3a.
250
0.5
2
RD(Zh)
RD(KDP)
RD(Zh,Zdr)
RD(KDP,Zdr)
0.45
0.4
1.6
a)
1.4
Trimmed FRMSE
Trimmed FRMSE
0.35
0.3
0.25
0.2
1.2
1
0.8
0.15
0.6
0.1
0.4
0.05
0.2
0
RD(Zh)
RD(KDP)
RD(Zh,Zdr)
RD(KDP,Zdr)
1.8
0
5
10
15
20
25
0
30
Disdrometer Rainrate (mm.hr-1)
0
b)
10
20
30
40
50
60
70
80
Disdrometer Rainrate (mm.hr-1)
Figure 8.3. Trimmed FRMSE for disdrometer-derived algorithms using a) 1-minute
disdrometer data and b) disdrometer data with propagated theoretical errors, using bins of
5 mm/hr.
8.2.2.2 Synthetic algorithm methodology
8.2.2.2.1 Decision-tree logic method
In order to optimize rain rate estimations at all rates, polarimetric algorithms may be
combined, generally using simple decision-tree logic. Using disdrometer data, we therefore
examined FRMS error after propagation as a function of reflectivity (Z), in order to identify
which disdrometer-tuned algorithm has the lowest error at each reflectivity and thus determine
reflectivity-based decision criteria. For each algorithm we then re-performed a regression as in
Section 8.2.2.1, utilizing only the disdrometer data with reflectivities in the range at which that
algorithm will be applied. This allowed us to develop a decision-tree logic based combination
algorithm for non-hail cases given by:
R s  0 . 0144 Z h
0 . 8474
R s  0 . 0131 Z h
0 . 9148
if Zh < 25dBZ
 1 . 4238
R s  73 . 5 K DP Z dr
0 . 968
 7 . 2283
Z dr
(31)
if 25 ≤ Zh < 40dBZ
if Zh ≥ 40dBZ
(32)
(33)
8.2.2.2.2 Theoretical error weighting method
In comparison, a potentially more robust combination method is proposed, weighting the
algorithms by the inverse of the theoretical error. On an individual point basis, this involves
determining the theoretical errors of measurement σ(εM), as derived from Bringi and
Chandrasekar (2001), and the parameterization error σ(εP). Parameterization error can either be
treated as zero or represented by the FRMS error as derived from disdrometer data. The best
results were found by approximating error as a linear function of derived rain rate.
251
The combined rain rate for each data point is then estimated by summing the derived rain
rate from each disdrometer algorithm weighted by the inverse of its total error, i.e.,
4
R
a

i 1
R i , where
i
1
a
4

1

i 1
(34)
i
8.2.2.2.3 Discrete error weighting method
A third method of algorithm combination utilized the propagation of error through
disdrometer data to combine both the measurement and parameterization errors. This method
used the simulated error dataset to calculate the FRMS errors for each disdrometer algorithm in
discrete boxes as a function of one or more radar variables, where a box contained at least five
measurements. These errors were used to create a set of matrices of weighting functions in a
similar method to theoretical error weighting, which could then be applied to radar
measurements with the correct weighting factors applied dependent on the discrete ‗box‘ the
variables fell in.
The best results were found using either all variables (in boxes of 5 dBZ ZH, 0.5 mm hr-1
KDP and 0.5 dB ZDR) or just reflectivity and specific phase difference (in boxes of 5 dBZ ZH,
0.2 mm hr-1 KDP). Box sizes were chosen to optimize the range of data represented while
remaining robust to variations in real algorithm behavior with variables. However, the discrete
nature of the boxes results in a potential problem at values infrequently sampled in the
disdrometer data set, particularly at very high rain rates.
8.2.2.3 Validation against disdrometer data
These three methods were applied to both raw disdrometer and in the presence of
simulated error, and compared with the Ryzhkov et al. (2005) decision-tree method, the Bringi et
al. (2004) DSD-derived complex method, and the Bringi (unpublished) CP2 code.
Using raw disdrometer data at rates less than 35 mm hr-1, the decision-tree logic method based
on disdrometer data significantly outperformed both disdrometer algorithms and the similar
Ryzhkov and Bringi methods, with a mean bias of −0.2% and a standard deviation of 0.15. Using
the Ryzhkov or Bringi decision-tree steps optimized to disdrometer data provided similar results
to our method, but no improvement (not shown). However, the more complex DSD-tuned
method of Bringi et al. (2004) performs fairly well, although still worse than our weighted
methods. Both the methods of weighting combination performed similarly at these rates, with
low biases (< ±1% for all cases), but significantly higher FRMSE than the decision-tree method
(Figure 8.4a).
Under simulated error conditions, combinations optimized to disdrometer data continue to have
lower errors than the Ryzhkov or Bringi methods at most rain rates (Figure 8.4b). However, the
FRMS error for the decision-tree method increased substantially, with a mean bias of +50% and
high susceptibility to outliers using untrimmed FRMSE (not shown), performing generally
similar to or worse than the Bringi complex DSD-based method. The method weighted by
theoretical error also exhibits rapidly increasing errors at low rain rates, with neither
outperforming the disdrometer-derived R(Zh) relationship at rain rates <25 mm hr-1.
252
In comparison, the method using discrete weightings exhibits consistently low trimmed FRMS
error at all rain rates, with biases remaining below ~±10% after propagation for all combinations
except (ZDR,KDP).
0.9
2
Decision-tree method
Theory-weighted method
Discrete weighting method
Ryzhkov (2005) Method
Bringi CP2 Method
Bringi complex
0.8
0.7
1.6
1.4
Trimmed FRMSE
Trimmed FRMSE
0.6
0.5
0.4
0.3
1.2
1
0.8
0.6
0.2
0.4
0.1
0
Decision-tree method
Theory-weighted method
Discrete weighting method
Ryzhkov (2005) Method
Bringi CP2 Method
Bringi complex
1.8
0.2
0
5
10
15
20
25
a)
0
30
Disdrometer Rainrate (mm.hr-1)
0
10
20
30
40
50
60
70
80
Disdrometer Rainrate (mm.hr-1)
b)
Figure 8.4. Trimmed FRMSE for combination algorithms using a) 1-minute disdrometer
data and b) disdrometer data with propagated theoretical errors, using bins of 5 mm/hr.
8.2.2.4 Validation against CP2 data
8.2.2.4.1 Disdrometer-tuned algorithms
Our disdrometer-tuned algorithms were then applied to the filtered radar data and
compared with the Bureau of Meteorology gauge network for all rain rates exceeding 1 mm hr-1,
below which tipping bucket gauges have poor resolution. These were compared to the standard
Marshall-Palmer R-Z relation, and the polarimetric algorithms found most accurate in a study by
Ryzhkov et al. (2005) (Table 8.1).
Table 8.1. Mean fractional bias (FB) and trimmed fractional root mean squared error
(FRMSE) where R ≥1 mm hr-1 for disdrometer-tuned algorithms and literature algorithms
of similar types.
Mean FB where R≥1mm hr-1
Disdrometer
Literature
Trimmed FRMSE where R≥1mm hr-1
-1
R≥1mm.hr
Disdrometer
Literature
R(Zh)
-0.08
-0.30
0.60
0.59
R(KDP)
0.94
2.1
1.1
2.2
R(Zh,Zdr)
-0.001
0.80
0.59
1.0
R(KDP,Zdr)
1.6
0.19
1.8
0.66
The fractional mean bias for the disdrometer-tuned R(Zh) remained very low, with −8%
compared to −30% for the Marshall-Palmer method, indicating an improved representation via
tuning. However, the standard deviation of FB was significantly higher for the disdrometer-tuned
algorithm, 1.4 compared to 0.8, indicating a strong influence of high outliers. This resulted in
253
significantly worse FRMSE for the disdrometer-tuned method at all rain rates, but fairly similar
values for trimmed FRMSE (Figure 8.4a), with Marshall-Palmer only superior at rain rates less
than 10mm.hr-1.
In comparison, the disdrometer-tuned R(KDP) and R(Zh,Zdr) algorithms were found to perform
significantly better than their literature counterparts, with FRMSE approximately halved and low
bias, nearly nil for R(Zh,Zdr). However, the Ryzhkov et al. (2005) R(Zh,Zdr) algorithm has
lower trimmed FRMSE than the disdrometer-tuned version at rates between 5 and 20 mm hr-1
(Figure 8.5a). The disdrometer-tuned R(KDP,Zdr) appears to have significantly higher bias than
its literature counterpart, belying this trend; however, at rates exceeding 10 mm hr-1 this
exhibits significantly lower FRMSE than its literature counterpart (Figure 8.5b), and a mean
FB of just −9%.
1
1
RD(Zh)
Marshall-Palmer R(Zh)
RD(Zh,Zdr)
Ryzhkov R10(Zh,Zdr)
0.9
0.8
0.8
a)
0.7
Trimmed FRMSE
Trimmed FRMSE
0.7
0.6
0.5
0.4
0.6
0.5
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
RD(KDP)
Ryzhkov R4(KDP)
RD(KDP,Zdr)
Ryzhkov R14(KDP,Zdr)
0.9
0
10
20
30
40
-1
Gauge Rainrate (mm.hr )
50
0
60
b)
0
10
20
30
40
50
60
Gauge Rainrate (mm.hr-1)
Figure 8.5. Trimmed FRMSE error using radar and gauge data for both disdrometer and
literature algorithms by type: a) Zh-based algorithms b) KDP-based algorithms, with dotted
lines representing the literature version.
8.2.2.4.2 Combination algorithms
We also applied all of the combination algorithms specified in Section 8.2.2.2 to the same
CP2 data. Of the decision-tree methods, the lowest overall error was found for that tuned against
disdrometer data, so this will be used to represent this methodology. However, tuning the Bringi
CP2 methodology to disdrometer data outperformed our method at rates >= 30 mm hr-1 by ~10%,
although performing worst of all the algorithms at lower rain rates, indicating some combination
of these may be viable. The Bringi et al. (2004) complex methodology will also be considered as
an alternate methodology.
Using all rain rates in excess of 1 mm hr-1, the decision-tree combination method has a
very low fractional bias of +3%, significantly better than the bias of +24% for the theory
weighted method using linear regression, which was the best of these methods, or the Bringi et al.
(2004) method. However, the decision-tree method is also strongly affected by outliers, with
σ(FB) of 1.45, compared to 1.25 for the theory weighted method, leading to significantly higher
FRMSE at all rain rates. Of the discretely weighted algorithms, the lowest FB was found for the
Z-K combination, at −4%. However, the lowest standard deviation was found for the three254
dimensional combination at just 1.00, leading to very similar values for FRMSE of ~1 but
potentially different error characteristics.
However, with the removal of outliers using trimmed FRMSE the errors were fairly
consistent between methods, all having trimmed FRMSE within ±0.02 of 0.55 (Table 8.2). This
is too insignificant to indicate superiority of any method, indicating that the predominant
difference between methods is in their sensitivity to outliers, with the discrete weighted method
showing significantly better untrimmed FRMSE. Trimming also appeared to change the mean
bias by ~−20% for all algorithms, indicating that the majority of outliers were due to
overestimation in each case, with the smallest bias now +4% for the theory weighted method.
The smallest trimmed bias is +0% for the Bringi et al. (2004) method; however, this method has
a significantly higher trimmed and untrimmed FRMSE. This suggests that although this method
should represent a better representation of actual drop size characteristics, it appears more
sensitive to measurement error than our weighted methods.
Table 8.2. Errors using radar data (where R ≥ 1 mm hr-1) for two combination algorithms.
Untrimmed
Trimmed
FB
FRMSE
FB
FRMSE
Decision-tree
0.03
1.4
-0.16
0.57
Weighted
0.24
1.3
0.04
0.54
Discrete ZK
-0.04
1.1
-0.20
0.55
Discrete All
-0.09
1.0
-0.24
0.55
Bringi 2004
0.22
1.6
+0
0.67
The theoretically weighted method has the lowest trimmed FRMSE at all rain rates
greater than 10 mm hr-1, but very poor estimation at low rain rates, probably due to
overrepresentation of KDP-based algorithms at these ranges (Figure 8.6). This is closely followed
by the discretely weighted method using Zh and KDP, while error in the decision-tree method
increases with rain rate. However, the variation between the two weighting methodologies is low,
and all three methods have lower trimmed FRMSE than the Bringi et al. (2004) methodology and
all disdrometer algorithms for R>=1 mm hr-1.
255
1
Decision-tree method
Theory-weighted method
Discrete weighting method - all
Discrete weighting method - Z H, KDP only
Bringi (2004) complex
0.9
0.8
Trimmed FRMSE
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
10
20
30
40
50
60
Gauge Rainrate (mm.hr-1)
Figure 8.6. Trimmed FRMSE using radar and gauge data for four methods of synthetic
algorithm combination.
8.2.2.5 Discussion and conclusion
An investigation of a number of methods of rainfall determination from polarimetric
radar data has reiterated the known error characteristics of various polarimetric radar-rainfall
relations as a function of rain rate. Parameterization of algorithms to disdrometer data for a
specific location has been shown to generally produce superior results to algorithms derived
from literature, due to the strong variability of rainfall drop size characteristics in various areas.
However, the presence of measurement and comparison errors means this is not necessarily
constant at all rain rates.
Several methods of combining algorithms to create a more robust estimation have also
been attempted. All of these have slightly lower trimmed FRMSE than individual polarimetric
algorithms or two decision tree methods derived from other studies. However, accuracy is poor
at very low rain rates, indicating a continued need for simple Z-R relations at these rates.
Nonetheless, using weighted averages of algorithms is found to consistently outperform
decision-tree logic for both trimmed and untrimmed error, although it is yet unclear whether the
discrete or theoretical weightings are most accurate. These algorithms also outperform a more
complex DSD-based algorithm derived by Bringi et al. (2004), to a greater extent with radar data
than disdrometer-derived, indicating this may be more susceptible than measurement error.
However, this may be a consequence of the applicability of assumed DSD relationships to this
area and warrants further study. These results indicate that more robustly weighted algorithm
combinations may be a superior method for rainfall estimation at rates exceeding 5 mm hr-1.
256