____________________________________________
The evolution of convective storms initiated
by an isolated mountain range
____________________________________________
Brett Soderholm
Department of Atmospheric and Oceanic Sciences
McGill University, Montreal, Canada
April 2013
A thesis submitted to McGill University in partial fulfillment of the requirements
of the degree of Master of Atmospheric and Oceanic Sciences
© Brett Soderholm 2013
Contents
Abstract
3
Sommaire
4
Acknowledgements
5
List of figures by chapter
6
List of tables by chapter
8
Chapter 1: Introduction
10
1.1: Overview
10
1.2: Requirements for convection initiation
11
1.3: Convective storm organization and evolution
12
1.4: Mountain convection initiation mechanisms
14
1.4.1: Mechanically-forced lifting
1.4.2: Thermally-forced lifting
14
16
1.5: Motivation for the current project
Chapter 2: Methodology
18
23
2.1: Obtaining radar data
23
2.2: Storm tracking
24
2.2.1: Region of interest
2.2.2: Our algorithm
2.2.3: Cell splitting/merging, and other issues
25
25
28
2.3: Event classification
30
2.4: Estimating maximum precipitation
32
1
2.5: Obtaining Radiosonde Data
34
2.6: Obtaining Analysis Data
35
2.7: Model set-up and initialization
36
Chapter 3: Observational Analysis
39
3.1: Description of observed events
3.1.1: The STSD event of July 2nd, 2012
3.1.2: The STLD event of June 30th, 2012
3.1.3: The LTLD event of August 27th, 2011
3.2: Data source(s)
39
40
42
44
46
3.2.1: Wind comparison
3.2.2: CAPE/|CIN|comparison
46
51
3.3: Background-wind differences between categories
54
3.3.1: Differences in vertical wind shear
57
3.4: Thermodynamic analysis
58
3.4.1: Average thermodynamic profiles
3.4.2: Differences in CAPE/|CIN|
Chapter 4: Numerical Simulations
58
60
62
4.1: Motivation
62
4.2: Experimental setup
63
4.3: Results
66
4.3.1: Control
4.3.2: Changes to mid-level wind shear
4.3.3: Changes to low-level wind direction
4.4: Discussion
66
68
71
76
2
Chapter 5: Conclusions and future work
79
5.1: Conclusions
79
5.2: Future work
81
References
83
Appendix A: List of observed events
90
3
Abstract
While significant attention has been given to understanding the initiation
mechanisms of convective storms over mountainous terrain, far less has been
given to the factors controlling their subsequent evolution. Here we perform an
observational and numerical investigation of the evolution of convective storms
initiated by the Black Hills mountains of South Dakota. These Hills are
preferentially located within the United States to access moist, unstable air, and
are thus a local hot-spot for convection initiation. Applying a convective-cell
tracking algorithm to 53 observed events initiated by the Black Hills revealed
three types of storm evolution: short-lived, short-track cells; long-lived, shorttrack cells; and long-lived, long-track cells. Analysis of the background wind
profiles during each event revealed modest differences amongst the storm types,
which were tested using quasi-idealized, convection-permitting numerical
simulations. The track-lengths and durations of convective cells produced in
these simulations were consistent with those in our observed events,
demonstrating that these modest differences in background wind profile could
indeed largely explain a convective storm’s evolution.
4
Sommaire
Bien que beaucoup d’attention a été accordée à la compréhension des
mécanismes d’initiation des tempêtes convectives sur un terrain montagneux,
beaucoup moins a été mis sur les facteurs qui contrôlent leur évolution ultérieure.
Ici, nous effectuons une enquête observationelle et numérique des tempêtes
initiées par les montagnes Black Hills du South Dakota. Ces montagnes initient
beaucoup de convection grace à leur accès à l’air humide et instable.
L’application d’un algorithme suivant les pistes de 53 tempêtes initiés par les
Black Hills a révélé trois types d’évolution: tempête de courte-piste, courte-durée;
tempête de courte-piste, longue-durée; et tempête de longue-piste, longue-durée.
Un analysis des profils de vents pendant chaque événement a révélé de légères
differences entre les types de tempêtes, qui ont été testés par des simulations
numériques. Les longueurs de piste et les durées des tempêtes produites dans ces
simulations étaient similaires à celles des événements observés, ce qui démontre
que ces légères différences dans les profils de vent pourraient en effet expliquer
l’évolution de chaque type de tempête.
5
Acknowledgements
First and foremost, I would like to thank Professor Daniel Kirshbaum for
his tireless support and supervision throughout this project. His attention to detail
and encouraging attitude were invaluable assets that greatly assisted the entire
research process from start to finish. In addition to providing editorial help and
guidance with the writing of this thesis, he is responsible for initializing and
running the numerical simulations presented in Chapter 4.
I would also like to thank my office mates Mathieu Plante and David
Themens and for their willingness to brainstorm ideas with me and troubleshoot
solutions to any coding issues I happened to encounter throughout this project.
Finally, I wish to thank all my friends and family who believed in me and
convinced me to pursue this degree. Without their support and encouragement, I
would not have been able to make it this far. Thanks guys!
6
List of figures by chapter
Chapter 1:
Figure 1: Probability of observing a radar echo > 40 dBZ
as a function of solar time of day
21
Figure 2: Frequency of SPC storm reports over a portion
of the United States
22
Chapter 2:
Figure 3: Analysis box over the Black Hills
26
Figure 4: Implementing terrain into the Bryan cloud model
37
Chapter 3:
Figure 5: Distribution of UTC hours in which the eventdefining cell initiated and dissipated
39
Figure 6: Composite radar reflectivity and storm tracks
atop terrain for the STSD event of July 2nd, 2012
41
Figure 7: Composite radar reflectivity and storm tracks
atop terrain for the STLD event of June 30th, 2012
43
Figure 8: Composite radar reflectivity and storm tracks
atop terrain for the LTLD event of August 27th, 2011
45
Figure 9: RAP/NAM/NARR wind roses at 250, 500, and
800 hPa for all STSD events
48
Figure 10: RAP/NAM/NARR wind roses at 250, 500, and
800 hPa for all STLD events
49
Figure 11: RAP/NAM/NARR wind roses at 250, 500, and
800 hPa for all LTLD events
50
7
Figure 12: Distribution of CAPE values per storm category
52
Figure 13: Distribution of |CIN| values per storm category
53
Figure 14: Wind barbs for all events in each category in
the upper-, mid-, and lower-levels
55
Figure 15: Average thermodynamic profiles for all events
per storm category
59
Chapter 4:
Figure 16: Average thermodynamic profile used in all
simulation cases
64
Figure 17: Wind profiles used in each simulation
64
Figure 18: Composite radar reflectivity and storm tracks
atop terrain for Case 1
67
Figure 19: Composite radar reflectivity and storm tracks
atop terrain for Case 2
68
Figure 20: Composite radar reflectivity and storm tracks
atop terrain for Case 3
70
Figure 21: Composite radar reflectivity and storm tracks
atop terrain for Case 4
71
Figure 22: Composite radar reflectivity and storm tracks
atop terrain for Case 5
72
Figure 23: Composite radar reflectivity and storm tracks
atop terrain for Case 6
74
Figure 24: Composite radar reflectivity and storm tracks
atop terrain for Case 7
75
8
List of tables by chapter
Chapter 2:
Table 1: Event classification
30
Chapter 3:
Table 2: Average deviation from RAP data for NAM
and NARR in STSD events
48
Table 3: Average deviation from RAP data for NAM
and NARR in STLD events
49
Table 4: Average deviation from RAP data for NAM
and NARR in LTLD events
50
Table 5: Values of vertical wind shear per storm
category between various levels
57
Table 6: Values of vertical speed shear per storm
category between various levels
57
Chapter 4:
Table 7: Description of the simulation cases
66
Table 8: Summary of storm tracking results for all
cases
76
9
Chapter 1: Introduction
1.1: Overview
Convective storms are responsible for major hazards and control the largescale atmospheric circulation (e.g., Maddox et al., 1978; Richter and Rasch,
2008). These storms are characterized by nonlinear, interacting processes over a
wide range of scales, rendering them difficult to fully understand. Despite
intensive study, aspects of these storms remain poorly understood and numerically
simulated, which leads to errors in operational weather forecasts (Zhang et al.,
2003; Hohenegger and Schär, 2007). Further study is thus required to address
these deficiencies in understanding and prediction. One way this can be done is
by analyzing convective storms that develop in mountainous regions. Mountains
serve as hot-spots for convection initiation (e.g., Goudenhoofdt and Delobbe,
2012) and can serve as excellent natural laboratories to build this understanding
(Raymond and Wilkening, 1980). Although the initiation of mountain convection
has received intensive study in recent years, very little attention has been devoted
to analyzing the subsequent evolution of mountain storms. This understanding is
of critical importance for life and property in mountainous regions, which are
often subject to severe weather (e.g., flash-flooding, hail, and strong winds). In
this study, we focus on deep convection over the Black Hills of South Dakota: a
region ideally located in the Great Plains region of the United States, but which
has received minimal previous attention in the literature.
10
1.2: Requirements for convection initiation
Here we briefly review the necessary conditions for moist convection
initiation to occur. One such condition is that the atmosphere be conditionally
unstable (Bluestein, 1993). An atmosphere is defined as conditionally unstable
when the temperature lapse rate of an atmospheric layer lies between the dry and
moist adiabatic lapse rates (Banta, 1990). In this situation, the potential exists for
an air parcel - if forced upwards through the layer - to become positively buoyant,
where it can continue to ascend vertically under the power of its buoyancy (Banta,
1990). To realize this instability, the air parcel must first ascend dry adiabatically
to its lifting condensation level (LCL; Wilde et al., 1985). At this point, the parcel
begins to release latent heat through condensation, which causes its lapse rate to
become moist adiabatic (Banta, 1990). Given that the lapse rate of a conditionally
unstable layer of air is larger than a moist adiabat, the parcel eventually becomes
warmer than its environment if lifted sufficiently high (Markowski and
Richardson, 2010). The height at which the temperature of the air parcel equals
that of the environment is the level of free convection (LFC; Bluestein, 1993).
Air parcels lifted beyond this level will become warmer and lighter than the
surrounding environment, and thus ascend freely (Banta, 1990). An air parcel will
continue to ascend until it becomes colder than its environment, which occurs
when the parcel reaches its equilibrium level (EL; Bluestein, 1993). For deep
convection, this frequently lies at or above the tropopause, which allows clouds to
extend from a short distance above the surface up to 10-12 km (Banta, 1990).
11
Another necessary condition for convection initiation is the presence of
sufficient moisture (Hagen et al., 2011). In order for an air parcel to reach its
LFC, the moisture content must be sufficient for the air parcel to saturate (Wilde
et al., 1985). In addition, a lifting mechanism is required to drive parcels upward
to their LFC. Possible lifting mechanisms include frontal circulations, low-level
horizontal convergence zones, boundary layer thermals, and orography (Banta,
1990; Wilson and Megenhardt, 1997; Houze Jr., 2012). While specific orographic
initiation mechanisms will be discussed in detail in Section 1.4, it is first
necessary to discuss how convective storms organize and evolve.
1.3: Convective storm organization and evolution
After an air parcel reaches its LFC, three modes of convective storms can
be produced: single (ordinary) cells, multi-cells, and supercells (Bluestein, 1993).
To a large degree, the background wind profile and the thermodynamic profile
determine the dominant mode (Newton and Newton, 1959; Patuskkov 1975;
Klemp and Wilhelmson, 1978).
Byers and Braham (1948) first characterized an ordinary (single)
convective cell. These cells frequently initiate when little to no vertical wind
shear is present between the sub-cloud layer and the mid-troposphere (Bluestein,
1993). Initially, the updrafts promote the formation of water droplets, which
grow until they become too heavy to be held aloft by the updrafts (Byers and
Braham, 1948). At this point, the water droplets begin to fall, and a downdraft of
12
cold air resulting from precipitation drag and evaporation is created that undercuts
the updraft. Once the downdraft reaches the ground, it spreads out in all
directions and overwhelms the inflow, leading to storm decay.
Multicells refer to a series of convective cells that initiate in regions of
moderate shear (5-15 m s-1) between the sub-cloud and mid-troposphere layers
from the outflow of adjacent, dissipating cells (Wesiman and Klemp, 1982).
While the outflow still cuts off the inflow of individual cells, stronger forward
wind shear in the background flow prevents this outflow from spreading too
quickly ahead of the cells (Bluestein, 1993). This promotes the formation of a
deeper layer of dense air, which lifts air at the leading edge of the outflow
boundary to the LFC and initiates new cells. This process can repeat many times
as long as the outflow remains sufficiently deep and vigorous.
Supercells form preferentially in high vertical shear (> 25 m s-1 between
the surface and 6 km layer) environments (Marwitz, 1972a). This strong
background shear (and hence horizontal vorticity) is tilted vertically by convective
updrafts, allowing for two counterrotating mesocyclones to develop on the left
and right flanks of the updraft (Lemon and Doswell III, 1979). Non-hydrostatic
vertical pressure perturbations within these mesocyclones lead to dynamic uplift
on both flanks of the initial updraft, which eventually cause storm splitting into
left- and right-movers (Houze et al., 1993). If directional shear is present, one of
these cells will be favoured (Davies-Jones, 1984). Wilhelmson and Klemp (1978)
showed that for a typical veering wind profile, the right-mover is typically
13
preferred. This is commonly observed over the central United States (DaviesJones, 1986). These storms can sustain themselves for long durations ( > 60
minutes) due to their dynamic lifting combining with buoyant lifting resulting
from environmental instability (Houze et al., 1993). Moreover, their rotation
allows for a separation between evaporative downdrafts and the storm inflow,
preventing cell decay that occurs in single- and multi-cell storms (Bluestein,
1993).
1.4: Mountain convection mechanisms
Recalling that mountains serve as hot-spots for convection initiation, we
return our attention to describing the precise mechanisms by which a mountain
can trigger convective storms. Banta (1990) identifies three such mechanisms:
direct (mechanically-forced) lifting, thermally-forced lifting, and obstacle effects.
Here we pay significant attention to describing the first two mechanisms, as they
are particularly relevant for our present study.
1.4.1: Mechanically-forced lifting
This method of lifting, also referred to as “direct orographic lifting”,
involves the forced ascent of a layer of moist air up the windward side of a
mountain to the LFC (Banta, 1990). In order to determine when this situation is
possible, it is useful to examine the ratio of kinetic energy of the airflow to the
potential energy required to surmount the barrier: also known as the Froude
14
number (Chu and Lin, 2000). When this number is large (> 1), air is able to to
surmount the obstacle, leading to strong ascent over the windward slope; when the
Froude number is small (< 1), air detours around the mountain, leading to weaker
ascent upstream of the mountain or in a downwind convergence zone. Thus,
mechanically-forced lifting requires a Froude number > 1 (Hagen et al., 2011).
Previous investigation has shown that convective storms initiated through
mechanically-forced lifting have led to extreme flooding events (e.g., Schroeder et
al., 1977; Maddox et al., 1978; Caracena et al., 1979). Maddox et al. (1978)
compared meteorological aspects at various scales between two extreme flooding
events in the United States: The Big Thompson Flood of 31 July 1976 near
Loveland, CO, and the Rapid City flash flood on 9 June 1972 in Rapid City, SD.
In each of these events, a “very moist” (14 g kg-1) capped layer of air near the
surface was orographically forced up the windward side of the regional mountain
to the LFC as a result of strong winds from the passage of a cold front over the
region (Maddox et. al, 1978). This capping inversion within the layer prevented
the release of moist instability until the mountain mechanically lifted the layer up
to the LFC, and resulted in a convective storm that inundated the surrounding
drainage basin with up to 12 inches of rain over a 4h period (Maddox et al., 1978).
Caracena et al. (1979) developed a conceptual model describing both flooding
events, which was later validated by Nair et al. (1997) in their numerical
simulations of the Rapid City flash flood.
15
1.4.2: Thermally-forced lifting
The initiation of convection via thermal forcing is accomplished when
sufficiently strong updrafts are created over the mountain’s surface and bring
conditionally unstable air to the LFC (Banta, 1990). These updrafts are generated
when the air directly above the mountain’s surface is warmed from solar heating:
this air is significantly warmer than the air at the exact same height away from the
mountain - leading to the development of low pressure directly over the mountain
- which induces surface convergence and corresponding updrafts over the
mountain crest (Banta, 1990).
Substantial attention has been paid to understanding the precise dynamics
behind the initiation of convection through this mechanism (e.g., Raymond and
Wilkening, 1980; Banta, 1990; Damiani et al., 2008; Kottmeier et al., 2008;
Hanley at al., 2011; Wulfmeyer et al., 2011; Barthlott and Kirshbaum, 2012). As
well, studies have addressed the effects of various environmental and atmospheric
factors determining the ability of convective clouds to initiate (e.g., Ookouchi et
al., 1984; Segal et al., 1988; Houze Jr., 1993; Damiani et al., 2008; Hagen et al.,
2011; Hanley et al., 2011). One such factor is the surface energy balance over the
mountain. Banta (1990) argued that because surfaces with higher soil moisture
will lead to higher latent and lower sensible heat fluxes, it is more likely that
thermally-forced circulations will be stronger over drier mountains. This claim
was well justified by previous work done by Segal et al. (1988) and Ookouchi et
al. (1984) who investigated the effects of vegetated versus arid mountain slopes
16
on updraft speed. Observed updraft speeds of 3 m s-1 routinely occurred over
slopes with vegetation, while dry slopes produced up to 6 m s-1 (Segal et al.,
1988). Recent field campaigns such as the Cumulus, Photogrammetric, In Situ,
and Doppler Observations (CuPIDO) experiment of 2006 by Damiani et al.
(2008) empirically validated this effect over the Catalina Mountains of Arizona.
Throughout their campaign, days on which the soil was saturated resulted in
considerably reduced measured updraft speeds (< 6 m s-1), compared to updraft
speeds of up to 9 m s-1 when the soil was dry (Damiani et al., 2008). However,
due to the very dry conditions, this enhanced updraft speed was often still
insufficient to initiate convection over the mountain peaks.
Similarly, recent numerical simulations such as those performed by Hanley
et al. (2011) further highlighted the importance of moisture on convection
initiation. In their study, an ensemble of convection-permitting simulations failed
to initially reproduce a well-observed convective event that occurred during the
Convective and Orographically induced Precipitation Study (COPS; see
Wulfmeyer et al., 2011) over the Black Forest region of Germany in 2007. They
found that their simulations had a moisture deficiency of 2-4 g kg-1, when
compared to the 12-16 g kg-1 of moisture observed during the experiment. This
deficiency arose from low soil moisture, which led to the development of an
overly-deep boundary layer that entrained too much dry air to support convection.
While the surface energy balance plays a key role in allowing convection
to initiate, Hagen et al. (2011) found that it is largely the background wind profile
17
that dictates where convection initiates in mountainous terrain. During the
aforementioned COPS campaign, it was found that in general, convective storms
initiated over the ridge on days when the 925 hPa winds were weak (< 5 m s-1),
and the wind direction in relation to the Vosges mountains changed from
perpendicular to parallel between 925 and 700 hPa, respectively (Hagen et al.,
2011). Conversely, convection initiated on the lee of the mountains when wind
speeds were generally higher at all levels, and wind direction did not change much
with height. Kirshbaum (2011) emphasized that the background wind speed
controls the strength of the mountain convergence, and thus has a direct impact on
convection initiation. Indeed, numerical simulations performed by Banta (1986)
demonstrated that thermally forced circulation becomes weaker and shorter in
duration when the ridge-top wind speeds increased from 4 to 12 m s-1. Banta
(1993) and Kirshbaum (2011) both attribute this to the background winds
ventilating heat away from the mountain.
1.5: Motivation for the current project
While the initiation mechanisms behind convective cells originating over
mountainous terrain have been studied intensely over the past few decades, far
less information is available on how these convective systems evolve after
initiation. Wilson and Roberts (2006) provided valuable insight into the fate of
convective storms after they initiate over flat terrain during the International H20
Project (IHOP), but did not take into consideration the influence of orography.
18
While other campaigns such as CuPIDO and COPS investigated orographic
effects on convective storms, they primarily focussed on the initiation problem
and did not examine how they evolved over time.
In order to properly assess the evolution of a convective storm, it is first
necessary to have a reliable method of tracking one. The first computerized
method capable of tracking convective storms was developed by Crane (1979).
Rosenfeld (1987) was not entirely satisfied by Crane’s method of only tracking
the cores (centroids) of convective cells, so he developed his own cell-tracking
software that would track both isolated and clustered convective cells “in a
physically meaningful manner”. Similarly dissatisfied, Dixon and Weiner (1993)
later developed the thunderstorm identification, tracking, analysis and nowcasting
(TITAN) algorithm to track thunderstorms using volume-scan radar data. To this
day, TITAN remains a popular and reliable method of tracking convective storms
that is still being implemented in current research (e.g., Goudenhoofdt and
Delobbe, 2012). Other algorithms, such as the storm cell identification and
tracking algorithm (SCIT) created by Johnson et al. (1998), focussed on
improving the tracking methods in TITAN to better monitor the evolution of
convective storms over the continental United States through the Weather
Surveillance Radar, 1988, Doppler (WSR-88D) network. Handwerker (2002) also
made an attempt to improve TITAN by developing the TRACE3D algorithm.
While these tracking methods are now available to use, it is only very
recently that they have been implemented to monitor orographically-induced
19
convective storms. Davini et al. (2012) performed a six-year climatology of
storms initiated over the mountainous region of northwestern Italy with the goal
of building a preliminary climatology of storm events to support operational
nowcasting activities. Goudenhoofdt and Delobbe (2012) examined the
statistical characteristics of convective storms in Belgium, and found that regions
with slightly higher convective initiation are related to orography. Surprisingly,
no investigation has yet been carried out over a mountainous region of North
America. It is for this reason that we will investigate the evolution of convective
cells by means of an observational climatology of storms initiated over the Black
Hills of South Dakota during the summer months of 2010-2012.
The Black Hills, being isolated in nature, serve as an ideal natural
laboratory to accurately and objectively measure direct orographic effects.
Indeed, Kuo and Orville (1973) performed the first radar climatology of
summertime convective clouds in this region, though their aim was once again to
identify the primary mechanisms of initiation. Furthermore, the location of these
mountains within the United States Great Plains provides access to warm,
moisture-laden air from the Gulf of Mexico, which can then be lifted both
mechanically and thermally to produce relatively frequent storms. This can been
in Figure 1, which depicts the probability of observing a radar echo exceeding 40
dBZ as a function of solar time of day.
20
The Hills, encased in red, can be seen as having a relatively increased probability
of observing a convective radar echo when compared to the immediate
surroundings specifically during the 12:00 - 13:00 solar hour.
Figure 1: Probability of observing a radar echo exceeding 40 dBZ as a function of solar time of
day. The region around the Black Hills (encased in the red box), shows an increased probability
compared to its surroundings (image courtesy of Prof. F. Fabry).
Given this relatively increased probability, one might expect that the
region east of the Black Hills would be subject to many severe weather outbreaks.
However, this has not been observed in the last five years. Figure 2 shows the
frequency of Storm Prediction Center (SPC) storm reports issued between 2008
and 2012 for June through August over a portion of the US midwest. While a
maximum in the direct vicinity of the Hills is not observed, this may be a result of
this region being sparsely populated.
21
Figure 2: Frequency of Storm Prediction Centre (SPC) storm reports between 2008-2012 over a
portion of the United States. The approximate location of the Black Hills, SD is shown by the red
ellipse.
By performing a climatology of storms initiated over this region, we aim
to better understand the dominant controls on how a convective storm evolves
after initiation over mountainous terrain. This insight may help to improve the
accuracy of forecasting severe events, and secure the lives and livelihoods of
many in the process.
22
Chapter 2: Methodology
2.1: Obtaining radar data
To begin our investigation of the evolution of convective storms over the
Black Hills, we first consulted online radar imagery made available by the
Mesoscale and Microscale Meteorology (MMM) division of the National Centre
for Atmospheric Research (NCAR; Ahijevych, 2013). These images displayed a
24h loop of composite reflectivity values - the maximum reflectivity recorded in
the vertical column - over the Northern Great Plains region. This data was taken
from the S-band WSR-88D Doppler radar network, operated by the National
Weather Service (NWS) within the National Oceanic and Atmospheric
Administration (NOAA).
For each day in the months of June-August in the years 2010-2012, we
consulted the available radar imagery to broadly determine if a convective cell
initiated in the general vicinity of the Black Hills between 15:00 UTC and 03:00
UTC. If, on a given day, it appeared as though cells initiated independently of
large-scale systems (e.g., directly over or downwind of the Black Hills
themselves), we noted the date and approximate time of cell initiation/dissipation
in a reference table. Each of these was classified as a case (or “event”) requiring
further investigation. In the rare occurrence that radar imagery was not available
for a given day, we did not include this date in our climatology.
23
Next we downloaded two sets of radar data from the National Climatic
Data Centre (NCDC; ncdc.noaa.gov): 230 km (short range) base and composite
reflectivity (N0R and NCR, respectively). These data sets included reflectivity
values on a latitudinal/longitudinal grid, interpolated to a constant spacing of
0.006855˚ in both directions. These data were obtained from the NWS radar
KUDX at a 0.5˚ elevation angle, located on the lee of the Black Hills in Rapid
City, SD (Figure 3). This radar has a 0.5˚ beam width, and a nominal horizontal
resolution of 1 km. Both the N0R and NCR reflectivity fields were provided
every 3-6 minutes. For each event, these radar data sets were first converted to
ASCII files using the application wct-viewer - a component of NOAA’s Weather
and Climate Toolkit (Ansari, 2013) - and then into netCDF files using a
conversion script written in NCAR’s command language (NCL). This allowed us
to easily read and manipulate the radar data in Matlab.
2.2: Storm tracking
To more quantitatively assess the role of the Black Hills in the initiation of
convective systems, we developed and implemented a cell tracking algorithm.
This algorithm automatically determined the track length and duration of all the
convective cells that developed over the Black Hills in each event. While storm
tracking algorithms have been successfully developed and implemented over the
past few decades (e.g., Dixon and Weiner, 1993), we created our own to track
individual convective cells using composite radar reflectivity data. This allowed
24
for superior control over the algorithm itself and avoided technical issues with
external software.
2.2.1: Region of interest
To focus our interest on storms that developed directly over the Black
Hills, we downloaded the terrain of the region from the United States Geological
Survey’s (USGS) GTOPO 30 global 30 arc second elevation data set
(eros.usgs.gov). At the latitude of the Black Hills, 30 arc seconds corresponds to
horizontal and vertical resolutions of approximately 666 m and 927 m,
respectively. This terrain data was converted into an evenly spaced Cartesian grid
to assist with forthcoming numerical calculations. We then defined a quadrilateral
analysis box (175 km x 100 km) over the Black Hills in such a way that it would
encompass terrain contours greater than 1250 m ASL (Figure 3). We required that
a cell must initiate within this box to be considered in our climatology. For those
that did, we evaluated their progression using the algorithm below.
2.2.2: Our algorithm
Our algorithm begins by reading in the NCR data contained within the
netCDF file closest to the time of convective cell initiation for a given event, as
per our preliminary analysis of online radar imagery. The extracted composite
reflectivity values were converted to lie on the same evenly-spaced Cartesian grid
as our terrain. For each point on this grid, reflectivity values less than 40 dBZ
25
m
350
2000
300
y [km]
2500
250
*
Rapid
City, SD
1500
KUDX
1000
200
500
150
150
200
250
x [km]
300
350
Figure 3: Analysis box (blue) over 250 m terrain contours (greyscale) of the Black Hills, in
Cartesian coordinates. The approximate location of Rapid City, SD, and the KUDX radar is
shown by the red asterisk.
were set to 0; this threshold represents a reasonably accurate threshold separating
convective from stratiform precipitation (Tokay and Short, 1995). This revised
reflectivity data was duplicated and stored in two separate arrays: one to preserve
the original reflectivity values, and one to assist in cell identification and tracking.
Using the latter array, the algorithm performed a left-to-right horizontal
examination of all reflectivity values along a constant y. Upon discovery of the
first non-zero value, the original reflectivity value at this grid point was replaced
by a “flag” (an integer) that was initially set to 1. As long as the loop encountered
a directly-adjacent non-zero value, it would assign the same flag to those grid
points. Once a zero value is encountered again, the cells’ boundaries are known
for the given y-index. When this occurs, the flag value increases by 2 to retain an
26
odd-number (see section 2.2.3 below). This process continues for all grid points
in x, with the flag increasing by 2 each time a new non-zero value is encountered.
Upon reaching the last grid point in x, the flag value was retained, and the process
repeated for all grid points in x along the next constant grid point in y.
Once all grid points had been evaluated for their extent in the x-direction,
the algorithm then performs a similar verification loop to connect adjacent cells in
the y-direction. Beginning with the second y-index, we compare its cell locations
to the y-index below. If any cells connect (e.g., if any points share the same xindex across the two y-levels), the index of that entire cell is replaced by the index
at the y-index below. Repeating this process across all y-indices allows all
coherent regions of 40 dBZ to be identified at that time.
Once all cells have been identified at a given time (t1), their grid point
locations are stored in an array (a1), and the flag value contained within each cell
becomes the cell’s ID number. Using the true reflectivity values, we then
calculate the reflectivity-weighted mean cell location - essentially the centre of
gravity of the cell. This location, alongside the time at which this data was
obtained, gets saved to a separate netCDF file corresponding to the cell number.
Then, the reflectivity values from the next available time (t2) are read in. We
repeat the above cell-identification scheme to these new data values, and identify
the location of all cells at this time (stored in a2). The fundamental assumption we
made for tracking the movement of cells over time is that there must be overlap
between the location of cells in a2 and a1. This assumption is justified by Newton
27
and Fankhauser (1964) who found that small cells (~ 8 km in diameter) travel at
speeds of up to 26 kts (13.38 m s-1), which would require ~ 10 minutes for no
overlap to occur; as data is available every 3-6 minutes, this does not pose a
significant threat to our analysis. Thus, the algorithm overlays a2 atop a1; if even
a single grid point overlaps between the location of a specific cell in both arrays,
the cell ID in a2 is replaced by the cell ID at a1. This allows us to methodically
track a cell’s movement from one time to the next. Once all cells have been
identified and re-labeled between time steps, the process is repeated at the next
time step. Anytime direct overlap does not occur, the algorithm retires that cell
ID number to prevent confusion with future cells, and we conclude the cell has
dissipated.
Finally, we computed each cell’s total displacement and duration:
calculating the horizontal distance between the initial and final location of a cell
yields its displacement, while comparing the time at which the cell initiated and
dissipated yields the duration. This calculation was performed for each cell in a
given event, and then for each event in our climatology. A visual comparison
between radar data and storm tracks produced by the algorithm ensures that all
tracks have been accurately depicted.
2.2.3: Cell splitting/merging, and other issues
Not all cells always remain perfectly coherent from initiation to
dissipation. Notable issues include: a cell temporarily dropping below the
28
reflectivity threshold, a cell splitting into two, or two cells merging into one. In
the case of a cell dropping below the reflectivity threshold, we simply claim that
the cell has died - even if it regenerates at a later time. The cell ID associated
with this particular cell is retired, forcing the regenerated cell to have a new ID.
However, if this cell re-initiates outside of our region of interest, it will be omitted
from our analysis.
In the case of one cell splitting into two, a specific addendum to the
algorithm is followed: if two cells appear at a given location in t2 where only one
was present at approximately that same location in t1, we conclude a split took
place. When this occurs, one cell will have retained the initial cell’s ID (c1), while
the second cell is assigned an index of c1+1 (c2); which cell is assigned which
index does not matter. By increasing the cell index by 1 instead of 2, the
algorithm can now check the list of cell IDs for any odd-then-even numbers (e.g.,
61, 62). Having found one, it verifies that the initiation time and location of c2 is
broadly consistent (< 10 minutes and < 10 km) with an entry in the track of c1. If
these conditions are met, the track information of c1 is appended to the track of c2.
This ensures that all cells arising due to a split are tracked back to their parent
cell. It is also possible that c2 (or c1) undergoes subsequent splits. Should this
occur, 0.1 is temporarily added to c2 or c1 as required to prevent confusion with
other cells. Should either of those split again, 0.01 is added to c2 or c1, followed
by the addition of 0.001 for the next split, and so on. Once all splits have been
identified, the algorithm performs the same verification of location and time
29
between cells with a decimal value in their ID, and appends the correct track
information as before. Finally, all cell IDs containing a decimal value are
changed back to an available integer to facilitate labeling.
In the opposite case, we conclude that two cells merge when one cell
exists at t2 in a given area where there were two at t1. Based on the number of
grid points that each cell contains, the ID of the smaller cell will be reassigned to
be one index less than the larger cell - so that it too is an even-number. Once
again, the algorithm will go through and verify the list of cell IDs, but this time
will look for any even-then-odd indices (e.g., 58,59). Because this is similar to
identifying splits, it is the subsequent verification of location and initiation times
between the two cells in question that ensures no cells are mislabeled. If there is
agreement, the algorithm will append the tracks of both individual cells with the
track of the recently-merged cell.
2.3: Event Classification
For each event, all cells initiated over the Hills were classified according
to their track length and duration (Table 1).
LTLD
STLD
STSD
!d " 100 km
!d # 25 km
Neither LTLD
!t " 120 min
!t " 90 min
nor STLD
Table 1: Cell classification based on track length and duration
30
If, on a given event, one or more cells lasted 120 minutes or longer, and
travelled 100 km or more, the event was classified as Long-Track, Long-Duration
(LTLD); the cell with the longest duration was be picked as the “event-defining
cell”. If one or more cells travelled 25 km or less in any 90 minute period, the
event was classified as Short-Track, Long-Duration (STLD); the first cell to
initiate and meet those conditions was picked as the event-defining cell. Finally,
if neither of these conditions were met by any cell, the event was labeled as ShortTrack, Short-Duration (STSD); the cell with the longest duration was again picked
as the event-defining cell. However, if an event could be classified as both STLD
and LTLD, it would be given both classifications and denoted as a hybrid event
(though this was not observed in the present climatology).
The time and distance thresholds in the LTLD and STLD events were
chosen to help distinguish between two types of severe convection: supercells/
multicells (LTLD), and quasi-stationary cells (STLD). Supercells pose a serious
threat to local inhabitants by producing heavy precipitation, hail, strong winds,
and tornadoes (see Lemon and Doswell, 1979) and frequently remain organized
for an hour or more - traveling large distances in the process (Weisman and
Klemp, 1978). To help filter out most non-supercell storms, we opted to use twice
this cited duration in conjunction with a calculated length scale specific to the
Black Hills: this length scale, L, represents the horizontal distance from the
highest point of elevation in the Black Hills to their base (~ 25 km). This length
scale is particularly useful because it gives a good indication of the minimum
31
distance a storm could travel and still remain directly under the mountain’s
influence. We therefore claim that a LTLD cell must travel a minimum distance
of 4L (100 km) in order for it to be sufficiently removed from orographic
influence, and remain coherent for 120 minutes or more.
Equally as hazardous are storms that remain relatively fixed over an area
(quasi-stationary) for an extended period of time, as these can result in disastrous
floods (see Maddox et al., 1978). Schroeder (1977) estimated that orographic
convective precipitation rates leading to intense flooding can vary anywhere
between 50 and 100 mm h-1, with an increased probability of a flood occurring
when the same area receives " 75 mm of rain. Using a conservative estimate of
50 mm h-1, an area is at heightened flood risk after just 90 minutes. For this
reason, we use a lower time threshold in STLD than LTLD events. We use L as
the maximum track length permitted for a cell to be considered quasi-stationary
and to merit the short-track, long-duration (STLD) designation.
2.4: Estimating maximum precipitation
Using base reflectivity (N0R) radar data, we computed the maximum
cumulative precipitation over all grid points within our analysis box (see Figure 3)
during each event. This was done by first reading in the N0R data contained
within the netCDF file closest to the time of convection initiation for a given
event (t1). We converted the base reflectivity value, Zb (dBZ) recorded at each
32
grid point in our analysis box into its associated radar reflectivity factor, z (mm6
m-3) according to the following relationship:
Zb = 10log10(z)
We used z to compute the rainfall rate, R (mm h-1) at each point according to the
following “WSR-88D Convective” z-R relationship best suited for deep summer
convection (Harrison, 2005):
z=300R1.4
We read in the next available set of N0R data at t2, and computed the time elapsed
between t2 and t1 (!t). We assumed the computed rainfall rate held constant over
!t, so the amount of rain that fell, r (mm), over a single grid point in that time was
simply:
r=R x !t
This computation was repeated for all subsequent times in the event cumulatively adding values of r at each grid point - and thus provided an estimate
of the cumulative rainfall during the event. Finally, we determined the maximum
cumulative precipitation recorded for a single point over the entire analysis box
for all events, which can be seen in Appendix A.
While this calculation provides us with a reasonable estimation of
cumulative precipitation, it does not account for different types of hydrometeors
(e.g., hail) that have very high reflectivity values. This likely leads to an
overestimation of the cumulative precipitation recorded, and thus requires caution
when making any direct conclusions.
33
2.5: Obtaining radiosonde data
To differentiate the background conditions between the different event
classes, we investigated atmospheric radiosonde profiles at the time and location
closest to cell initiation. Archived operational radiosonde data are freely available
online from the Department of Atmospheric Science at the University of
Wyoming (Oolman, 2013). Soundings from Rapid City, SD (see location in
Figure 3) are routinely taken every 12h at 00Z and 12Z. Occasional 6Z and 18Z
soundings are also taken. These data supply the following information:
atmospheric pressure (hPa), geopotential height (m), temperature (K), dew point
temperature (K), relative humidity (%), mixing ratio (g/kg), wind direction
(meteo ˚), wind speed (kts), potential temperature (K), and equivalent/virtual
potential temperature (K), taken at ~10 hPa intervals from the surface (usually
around 900 hPa) up to 10 hPa.
The vast majority of soundings that coincided most closely with the
convection initiation were taken at 00Z. In a few cases where 18 Z soundings
were available and better coincided with convection initiation, we chose these
soundings instead. Also, in a few cases where the 00Z sounding showed no
conditional instability but the 12Z did, we chose the 12Z sounding over the 00Z
sounding. Finally, in three events either no sounding was available or no
sounding in the 12-00Z time range showed any conditional instability. Because
no suitable profile that supported convection could be found, we left those events
out of the analysis.
34
For the remaining events, we composited (averaged) all soundings within
each classification category to investigate their differences. This was done by
interpolating each individual sounding to fixed pressure levels, and then taking
the mean of all the interpolated soundings of each class We compared the mean
thermodynamic profiles along with wind vectors at different levels. The latter
was to avoid filtering out important information on wind speed and direction by
averaging.
2.6: Obtaining analysis data
Two final sets of data were obtained from NOAA’s National Operational
Model Archive and Distribution System (NOMADS): North American Mesoscale
(NAM) model analysis data, and the North American Regional Reanalysis
(NARR) model data (nomads.ncdc.noaa.gov). The NAM model is a numerical
weather prediction model run by National Centers for Environmental Prediction
(NCEP) that assimilates data using three-dimensional variational data assimilation
(3DVAR). This model is run four times a day at 00, 06, 12, and 18Z, and is
capable of forecasting up to 84 hours. Currently, it is run with a 12 km horizontal
resolution, and with 1h temporal resolution. The NARR model is a long-term,
dynamically consistent, high-resolution, high-frequency, atmospheric and land
surface hydrology data-set for the North American domain, also run through
NCEP (Mesinger et al., 2006). Using 3DVAR to assimilate data, the model
35
produces 3-hourly outputs between 00 and 21Z with a 32 km horizontal
resolution.
2.7: Model set-up and initialization
We used the Bryan cloud model version 16 (Bryan and Fritsch, 2002) to
investigate the controls on storm evolution. This is a fully non-linear,
compressible, and non-hydrostatic 3D model designed for high-resolution explicit
convection simulations. This model uses a 3rd order forward integration KlempWilhelmson scheme with time-splitting for the stability of acoustic modes.
Centred, sixth-order horizontal advection is used in conjunction with sixth-order
horizontal diffusion to keep grid-scale noise to a minimum. Fifth-order vertical
advection with implicit diffusion is also used. The Coriolis effect was applied to
perturbations from the base state using an f-plane approximation with a value of
10-4 s-1. We applied positive-definite advection to moisture variables to ensure
that water is conserved. We parameterized subgrid-scale turbulence using a
prognostic 1.5-order turbulent kinetic energy (TKE) scheme, while we
parameterized cloud microphysics using the Morrison two-moment scheme (see
Morrison et al., 2005). We prescribe horizontally uniform, sinusoidally timevarying sensible and latent heat fluxes of amplitude 250 and 100 W m-2,
respectively, over a 24h cycle. While arbitrary, these values are based on model
analyses of various events from the database (not shown). We also incorporated
surface drag with a horizontally uniform drag coefficient of Cd = 0.01.
36
The entire domain of this model is 480 km x 480 km x 18 km, with a
constant grid spacing of 1 km in both the x and y directions. Open (radiative)
lateral boundary conditions are used. In the vertical, the nominal resolution is
100 m up to 4 km. Between 4 km and 8 km, a linear stretch in resolution from
100 m to 400 m is implemented. Between 8 km and 18 km, the resolution
remains fixed at 400 m. A wave absorbing layer is used over the uppermost 6 km.
A terrain-following vertical coordinate is used to handle complex terrain.
To set up the simulated terrain, we used the GTOPO 30 data from the
USGS (eros.usgs.gov). We focussed on the terrain with a latitude of 42 to 46˚N
and a longitude of -106 to -101˚E. This data was converted into Cartesian
coordinates, and then re-gridded to a uniform resolution of 1 km with the Black
Hills in the centre (Figure 4a).
Figure 4: Implementing terrain into the Bryan Cloud Model. a) Re-gridded terrain with defined
ellipse b) Undergoing the exponential decay function c) Leveling terrain < 1200 m
d) Final terrain with wavelengths < 6 DX filtered out
37
We opted to not use the exact terrain profile of the Hills, but rather to
implement a method that would smooth the terrain surrounding the hills
themselves and effectively isolate them. The justification for this is twofold:
primarily, we wished to avoid generating large amplitude perturbations at the
lateral boundaries of the grid, and secondly we wished to isolate the response to
the Black Hills in the absence of external complications. To do this, we then
defined an ellipse with a minor axis of ax=50 km, and a major axis of ay=100 km.
This ellipse was rotated counterclockwise by 35˚ so that its major axis was
roughly aligned with the ridge axis of the Black Hills. We defined a Gaussian
function that is equal to unity in the interior of the ellipse and smoothly decreases
to 0 outside of it. This has an e-folding scale equivalent to ax along the minor
axis, and equivalent to ay along the major axis (Figure 4b). The product of this
function and the terrain profile retains the main ridge but allows the surrounding
terrain to be gradually diminished. At this point, we set all terrain less than 1200
m in elevation equal to 1200 m - effectively filling the surrounding terrain around
the hills themselves (Figure 4c). This removes the gentle slope of the terrain in
the wast-west direction characteristic of the Great Plains. We then lowered the
entire terrain by 1200 m so that the height was zero away from the mountains.
Finally, we filtered out any wavelengths less than 6 DX (Figure 4d). This was
done to avoid forcing waves at poorly resolved scales that often lead to gross
errors within the model.
38
Chapter 3: Observational Analysis
3.1: Description of observed events
During the months of June, July, and August in the years 2010-2012, a
total of 53 events occurred in which convective cells were qualitatively identified
as initiating directly over the Black Hills of South Dakota. The storm tracking
algorithm described in Section 2.2.2 was applied to each of these events,
generating the results summarized in Appendix A. Of the 53 events, 28 were
STSD, 10 were STLD, 8 were LTLD, and 7 were removed from the investigation
due to interactions with large-scale systems or from a lack of available data.
Figure 5 displays the distribution of initiation and dissipation times for the events.
The hour of initiation is when the event-defining cell first reached 40 dBZ, while
the time of dissipation is when that same cell dropped below 40 dBZ.
Frequency [%]
Hour of initiation
0.4
0.2
0
18 19 20 21 22 23 00 01 02
Frequency [%]
Hour of dissipation
0.4
0.2
0
18 19 20 21 22 23 00 01 02
UTC hour
Figure 5: Distribution of UTC hours in which the event-defining cell initiated (orange) and
dissipated (blue) for all events
39
From the above figure, we observe a fairly wide distribution event initiation
times, with the maximum frequency occurring between 21:00 and 22:00 UTC.
Given the seven hour offset between the local time in South Dakota (MST) and
UTC, this indicates a slight preference for storms to initiate between 14:00 and
15:00 local time. A similar distribution is seen for when an event-defining cell
dissipates, with the maximum frequency occurring between 23:00 and 23:59 UTC
(16:00 and 16:59 local time). These results suggest diurnally-forced circulations
resulting from solar heating over the mountains play a key role in cell initiation.
This is consistent with many previous studies that found elevated thermal forcing
to be an important convection initiation mechanism during midlatitude summers
(e.g., Raymond and Wilkening, 1980; Damiani et al., 2008; Kottmeier et al., 2008;
Hanley et al., 2011). To exemplify the differences between storm evolutions in
each category, individual case studies are presented below.
3.1.1: The STSD event of July 2nd, 2012
STSD storms are the most commonly observed type of storm in this
investigation, with 61% of all events falling into this category (see Section 2.3 for
event classifications). Owing to their short-lived nature, these storms are unlikely
to pose a serious threat to local inhabitants. The event of July 2nd, 2012 is a prime
example of an STSD storm (Figure 6) and is described below.
40
Figure 6: a-c) Composite radar reflectivity (coloured) atop 250 m terrain contours (greyscale) for
select times during the STSD event of July 2nd, 2012 d) Storm tracks for all convective cells
At 19:51 UTC, the first cell surpassing the 40 dBZ reflectivity threshold
initiated on the southwestern side of the Hills. This cell dissipated within the
hour, while new cells initiated around 21:51 (Fig. 6a). At this time, three main
cells are observed: cell 1 had just initiated, while cells 2 and 3 have already
existed for 25 and 60 minutes, respectively. By 22:23, cell 3 entirely dissipated,
while cells 1 and 2 have traveled approximately 30 km to the east (Fig. 6b). Note
that cell 1 grows in both size and intensity, increasing from 40 to 50 dBZ, while
the reflectivity of cell 2 begins to decrease. By 22:51 (Fig. 6c), the reflectivity of
both remaining cells has decreased further as they continue to move eastward with maximum values around 45 dBZ. By 23:18, the reflectivity of these cells
has fallen below 40 dBZ, and no further cells exceeded this threshold for the
41
remainder of the day. The tracks of all cells observed on this day, as detected by
the storm-tracking algorithm, are shown in Fig. 6d. Cell 1 travelled the largest
distance out of any cell observed that day (49.8 km) and lasted the longest (87
minutes), thus serving as the event-defining cell. A total of 8 cells initiated over
the Hills, all of which were STSD.
3.1.2: The STLD event of June 30th, 2012
We observed only 10 STLD events in our investigation, corresponding to
22% of all events (see Section 2.3 for event classifications). These events are
unique because, by remaining quasi-stationary, they are capable of generating
heavy precipitation and localized flooding (e.g., Schroeder 1977; Maddox et al.,
1978; Caracena et al., 1979). Indeed, events classified as STLD had the highest
maximum cumulative precipitation values over a single point within our analysis
box out of all the storm categories, with the mean being 7.05 cm (see Appendix
A). This is noticeably higher than both the mean STSD and LTLD maximum
cumulative precipitation amounts of 2.91 and 4.77 cm, respectively. A prime
example of an isolated STLD event took place on June 30th, 2012 (Figure 7).
At 19:30 UTC, the first convective cell (labelled “1” in the figure)
exceeding 40 dBZ formed on the southwestern side of the Black Hills. By 19:43,
this cell had doubled in size, and attained a maximum reflectivity of 60 dBZ
(Figure 7a). Fifty minutes later (20:34 UTC) this cell had only travelled 8 km
southeast of its initiation point and sustained its high reflectivity value over this
42
Figure 7: a-d) As in Figure 6, but for the STLD event of June 30th, 2012
time (Figure 7b). By 21:16, this cell finally detached from the Hills and travelled
toward the southeast (Figure 7c). At this time, a second cell (2) initiated just
north of the initiation point of cell 1. This cell was not sustained for very long (46
minutes in total), while cell 1 was sustained for a total of 133 minutes. Cell 1, the
event-defining cell for this event, traveled only 22.3 km in that time period, and
thereby met the conditions for an STLD classification. The tracks of all cells
forming within the analysis box are shown in Figure 7d, and can be seen to cover
a remarkably small region. During this event, the maximum cumulative
precipitation recorded over a single point was 7.91 cm.
43
3.1.3: The LTLD event of August 27th, 2011
In our investigation, only 17% of the events were classified as LTLD (see
Section 2.3 for event classifications), making these events the least commonly
observed. While not a necessary condition for a LTLD event, the process of storm
splitting is most frequently observed in this category: 71% of LTLD events were
found to have at least one cell initiate and undergo a split, in contrast to 50% of
STLD and 64% of STSD events. These results are not surprising, as cell splitting
is a common characteristic of supercell storms (see Klemp and Wilhelmson, 1978)
which are likely the dominant mode of convection in this category. Note that not
all splitting events are associated with supercells - a large cell can break up into
two cells for a variety of reasons. This likely explains the large number of
splitting events even in the STSD and STLD categories. The event from August
27th, 2011 best exemplifies the characteristics of this type of storm (Figure 8).
At around 20:50 UTC, a convergence line clearly formed along the ridge
of Black Hills as evident by weak reflectivity echoes and doppler velocity (not
shown). Shortly thereafter, the cell 1 initiated to the east of this convergence line
at 21:06 UTC. Approximately 20 minutes later, around 21:34 UTC, cell 2
initiated roughly 75 km to the north (Figure 8a). Over the next half hour, both
cells continuously exceeded 55 dBZ in reflectivity, and traveled in a southsoutheastward direction. At 22:04 UTC (Figure 8b), cell 1 underwent a split into
two cells: a smaller, left-mover (3) that was only sustained for 25 minutes, and a
larger, right-mover (1), which was sustained and traveled southeastward.
44
Figure 8: a-d) As in Figure 6, but for the LTLD event of August 27th, 2011
At 22:24, cell 2 underwent a similar splitting process into two cells: a smaller leftmover (4), and a larger, right-mover (2). At 22:34 (Figure 8c), all four existed
simultaneously. By 22:54, the left-moving supercells (3 and 4) had already
dissipated, while the right-moving supercells (1 and 2) remained coherent and
exceeded the 40 dBZ threshold until 00:04 UTC on August 28th, 2011. As cells 1
and 2 were sustained for nearly 180 minutes each, and travelled a total of 126 and
134 km, respectively, this was a “golden” LTLD event. The resultant tracks of all
cells are shown in Figure 8d.
45
3.2: Data source(s)
To quantitatively assess the differences in storm evolution between the
STSD, STLD, and LTLD events, we have investigated the differences in the
background wind and thermodynamic profiles of these events using observational
radiosonde data taken from Rapid City, SD (hereafter referred to as RAP). While
RAP sounding data provides the most accurate representation of the wind profile
at a single point, these soundings are only taken every 12h between 00Z and 12Z,
with the occasional sounding produced at 06Z and 18Z. This is a significant
period of time between soundings, and as a result may not reflect the exact
conditions during an event. As well, the low-level winds in these soundings do
not necessarily provide an accurate picture of the undisturbed background flow, as
they can be affected by local mountain-induced circulations. Two other available
data sources - NAM model analysis and NARR reanalysis data - offered 3D
representations of the atmospheric flow, and had higher time resolutions than the
soundings (6h and 3h, respectively). To examine whether these gridded analysis/
reanalysis products could be used to complement RAP data, we assessed their
validity by comparing them directly to RAP data at the same times.
3.2.1: Wind comparison
We first compared the wind speed and direction at three levels using NAM
and NARR data against RAP sounding data for all events within each storm
category. Figure 9 shows a series of wind roses at 250, 500, and 800 hPa using
46
RAP, NAM, and NARR data for all events in the STSD category. These pressure
levels represent the upper-, mid-, and lower-levels of the atmosphere, respectively.
Table 2 summarizes the average deviation in wind speed and direction from RAP
data to the NAM and NARR data to better quantify these variations. Figures 10
and 11 show wind roses as in Figure 9, but for all events in the STLD and LTLD
categories, respectively. Tables 3 and 4 summarize the average deviations as in
Table 2, but for the STLD and LTLD categories, respectively.
Examining first the wind roses for those events in the STSD category
(Figure 9), there is a generally good consistency between the RAP and NAM
results, while NARR is noticeably different. Indeed, there is a 50˚ or greater
average deviation in wind direction between RAP and NARR data across all
selected levels for these STSD events (Table 2). This is not unique to the STSD
category; average deviations in wind direction between RAP and NARR in the
STLD and LTLD categories, especially at the 800 hPa level, are equally as high
(see Table 3 and Table 4, respectively). Conversely, we can see that the NAM
analysis is much more consistent with RAP observations, especially in the midand upper-levels, across all categories of storms. The largest overall error in wind
direction is found at the 800 hPa level, where localized forcing due to the
mountain may be responsible for the disagreement. Despite this, the relatively
small deviation between RAP and NAM data at mid- to upper-levels suggests that
the latter may usefully complement the former.
47
RAP
NAM
NARR
250 hPa
500 hPa
800 hPa
Figure 9: Wind speed and direction at 250, 500, and 800 hPa from radiosondes (RAP),
NAM, and NARR data for all STSD events
Level
NAM
NARR
250 hPa
4.17˚
2.68 m s-1
65.71˚
13.67 m s-1
500 hPa
6.74˚
1.47 m s-1
54.12˚
6.99 m s-1
800 hPa
18.28˚
1.72 m s-1
78.30˚
3.92 m s-1
Table 2: Average deviation from RAP in wind direction and speed per event
in the STSD category at various pressure levels
48
RAP
NAM
NARR
250 hPa
500 hPa
800 hPa
Figure 10: Wind speed and direction at 250, 500, and 800 hPa from radiosondes (RAP),
NAM, and NARR data for all STLD events
Level
NAM
NARR
250 hPa
7.77˚
2.68 m s-1
52.40˚
14.25 m s-1
500 hPa
6.80˚
1.84 m s-1
31.62˚
9.15 m s-1
800 hPa
21.35˚
1.80 m s-1
99.01˚
3.23 m s-1
Table 3: Average deviation from RAP in wind direction and speed per event
in the STLD category for various pressure levels
49
RAP
NAM
NARR
250 hPa
500 hPa
800 hPa
Figure 11: Wind speed and direction at 250, 500, and 800 hPa from radiosondes (RAP),
NAM, and NARR data for all LTLD events
NAM
NARR
250 hPa
13.95˚
3.39 m s-1
39.27˚
10.54 m s-1
500 hPa
3.40˚
2.33 m s-1
38.04˚
7.03 m s-1
800 hPa
26.61˚
1.52 m s-1
67.95˚
4.84 m s-1
Table 4: Average deviation from RAP in wind direction and speed per event
in the LTLD category for various pressure levels
50
3.2.2: CAPE/|CIN| comparison
To compare thermodynamic aspects of the three data sets, we also
compared values of CAPE and |CIN| from the NAM model analysis and NARR
reanalysis data sets directly to RAP data. In Figure 12a-c, histograms of CAPE
are shown for RAP, NAM, and NARR data sets, respectively. In Figure 13a-c,
histograms of |CIN| are shown for RAP, NAM, and NARR data sets, respectively.
Within all data sets, events occurred in which CAPE was 0 J kg-1: six events in the
RAP sounding data, two in the NAM, and eight in the NARR. These events were
omitted from the histograms below to prevent skewing the results. That 0 J kg-1
of CAPE was recorded for a given event is a testament to the lack of
representativity of some soundings for the convective event and/or the influence
of mountain processes (e.g., organized ascent, moistening through evaporation) in
modifying the local flow and changing its stability properties.
For any category of storm (e.g., STSD), the distribution of calculated
CAPE values is not consistent amongst the RAP, NAM, and NARR data sets
(Figure 12a-c). In the RAP data, STSD storms had CAPE < 250 J kg-1 60% of the
time. This is in sharp contrast with 0% of events in the NAM, and 15% in the
NARR having CAPE < 250 J kg-1 for the same STSD events. Similar extreme
variations exist in the remaining STLD and LTLD categories, with the NARR
consistently showing lower values of CAPE than RAP and NAM.
51
Figure 12: a) Distribution of CAPE values during each event per category using RAP data
b) As in a), but for NAM data c) As in a) but for NARR data
52
Figure 13: a) Distribution of |CIN| values during each event per category using RAP data
b) As in a), but for NAM data c) As in a) but for NARR data
53
The same comparison of |CIN| values for each source of data shows an even
greater inconsistency amongst the data sets (Figure 13a-c). In the RAP data,
STLD storms had > 200 J kg-1 of |CIN| 50% of the time. This is again in sharp
contrast with 10% of events in the NAM, and 0% in the NARR having > 200 J
kg-1 of |CIN| for the same STLD events.
Despite the fact that CAPE and |CIN| are very sensitive to the choice of
the level(s) used for the parcel, the inconsistency of their distributions in the
NAM and NARR data sets when compared to RAP soundings is far too large for
us to confidently uses these model analyses/reanalyses as a supplement to RAP
data. We have therefore chosen to use RAP data exclusively in forthcoming
analyses of the wind profiles and thermodynamics during each event.
3.3: Background-wind differences between categories
To broadly investigate the different wind profiles between the three storm
categories, we compare mean velocities over three layers of the atmosphere: 0-3
km (low-level), 3-6 km (mid-level), and 6-12 km (upper-level). Because the 40
dBZ threshold broadly distinguishes deep convection from other precipitation
types, the winds at all three levels likely influence the storm evolutions of all
cases. Figure 16 depicts this by means of a wind barb plot that uses RAP
sounding data for all 53 events taken at the time closest to convection initiation.
On each plot, the computed average wind speed is shown per category, and per
level, to more easily quantify differences in each category. It should be noted that
54
for results presented within the 0-3 km layer, we are referring to wind speed and
direction from the surface up to 3 km in the atmosphere. Given that the terrain
itself at Rapid City is already roughly at 1 km, this corresponds to roughly 2 km
above the surface, which extends about 800 m above the maximum elevation of
the Black Hills.
We begin by examining the average wind speeds over each layer in Figure
14. The LTLD events have the strongest wind speeds of all three categories over
each layer. These differences are most pronounced at middle levels.
Figure 14: Wind barb plots of average wind speed (m s-1) and meteorological direction (˚) for
low-, mid-, and upper-levels in STSD, STLD, and LTLD events using RAP data.
55
Note also the comparatively weak mid-level winds for storms under the STLD
classification. The average wind speed in this layer for STLD events is
approximately 2 m s-1 less than STSD events, and nearly 6 m s-1 less than LTLD
events. These results are consistent with previous work done by Maddox et al.
(1978) and Carcena et al. (1979) who found that slow moving, quasi-stationary
flooding events over the Rapid City and Big Thompson basins had between 7.7
and 9.3 m s-1 mid-level (500 hPa) wind speeds. These values vary significantly
from the computed 500 hPa wind speed by Maddox et al. (1978) for a typical
Great Plains thunderstorm (25 m s-1). This suggests that weaker wind speeds at
mid-levels play a key role in the development of a quasi-stationary storm.
Another key feature of Figure 14 is the large variation in low-level wind
direction amongst the STSD, STLD and LTLD events. Recalling the influence of
local topography on the low-level winds, we cannot reliably claim that this
presented data is a reliable representation of the background undisturbed flow.
Nonetheless, some interesting differences are apparent. While STSD events show
no definitive preference for wind direction in the lowest 2 km, both STLD and
LTLD events show a tendency for the winds to be southerly/southeasterly, or
northerly/northwesterly. These wind directions align roughly with the major axis
of the Black Hills (approximately 30˚ counterclockwise from due North).
However, with such a low sample size of events within each of these categories,
and the local wind perturbations induced by the terrain, it is impossible to draw a
convincing conclusion from these results. Nonetheless, the premise that low-level
56
flow oriented along the axis of the Hills could impact the evolution of convective
cells is a novel hypothesis that will be tested by means of numerical simulations
in Chapter 4.
3.3.1: Differences in vertical wind shear
Variations in the vertical shear between the categories are presented in
Table 5 and Table 6, respectively. While the differences in our computed vertical
wind shear between the categories are modest, note that LTLD events experience
over 7 m s-1 more vertical wind shear between the surface and 6 km when
compared to the STLD events. A similar trend is found over the 3-6 km layer,
above which local flow modifications associated with the mountain would not
have much effect. Table 6 also shows the speed shear (e.g., the difference in wind
speed between two levels regardless of direction) is highest for the LTLD events,
Level
STSD
STLD
LTLD
0-6 km
21.52
18.76
26.32
3-6 km
17.93
17.16
25.73
Table 5: Values of vertical wind shear (m s-1) per storm category using RAP data
for the 0-6 and 3-6 km levels
Level
STSD
STLD
LTLD
0-6 km
11.40
10.13
16.01
3-6 km
8.85
8.17
11.09
Table 6: Values of vertical speed shear (m s-1) per storm category using RAP data
for the 0-6 and 3-6 km levels
57
and lowest for STLD between both the 0-6 and 3-6 km levels. Results from these
two tables suggest that vertical shear has a large impact on the evolution of a
convective cell: stronger shears favour longer-lived cells with longer tracks. This
is consistent with the well-known tendency for stronger wind shear to favour
more organized and long-lived convective storms over flat terrain (e.g.,
Patushkov, 1975; Weisman and Klemp, 1984). That our results are consistent
with previous studies performed over flat terrain would suggest that the same
principles governing the longevity and track length of storms appear to apply over
complex terrain. To confirm the role of vertical wind shear on the evolution of
convective cells, variations in vertical wind shear on storm evolution will also be
tested by means of numerical simulations presented in Chapter 4.
3.4: Thermodynamic analysis
We also investigated the thermodynamic profiles of the different
categories to determine if any significant differences could be detected. This
involved first examining the average thermodynamic profile for all events in each
category (Figure 15). We will also revisit the distribution of CAPE and CIN in
each category using RAP data (Figure 12a and Figure 13a) to discuss trends.
3.4.1: Average thermodynamic profiles
Figure 15 shows very subtle differences in the average temperature profile
between each storm category. For example, the STLD events have slightly
58
warmer average temperatures aloft between the 850 and 550 hPa levels relative to
the other storm categories. They also have a higher mean dew-point temperature
between the surface and 800 hPa levels. This again is broadly consistent with
Maddox et al. (1978) who found a shallow moist layer near the surface with much
drier conditions aloft in the Rapid City sounding taken just before the initiation of
the quasi-stationary storm on 9 June 1972. These slightly elevated low-to-mid
level temperatures suggest that STLD events may have stronger boundary layer
inversions, which may serve to restrict convection to the region of strong local
forcing over the mountain ridge.
Figure 15: Average thermodynamic profiles (temperature and dew point) per storm category.
Blue represents STSD events, green for STLD, and red for LTLD.
59
Other subtle differences in Figure 15 include the tendency for STSD and LTLD
events to be drier closer to the surface, while LTLD events are driest between 550
and 350 hPa. The latter result is broadly consistent with Gilmore and Wicker
(1998) who investigated the influence of mid-tropospheric dryness on simulated
supercell morphology and evolution. They found that supercells with high
vertical wind shear and/or a high altitude dry air layer resulted in larger downdraft
dilution - leading to the development of weaker outflow - which favoured a
sustained updraft, and enhanced the longevity of the storm. However, their
simulations did not take into account effects of topography, which may also
influence the evolution of storms over the Black Hills.
3.4.2: Differences in CAPE/|CIN|
We now revisit the distributions of CAPE and |CIN| presented in Figure
12a and Figure 13a, respectively, to discuss differences between each category. In
Figure 12a, there is a tendency for STSD events to have smaller values of CAPE
(< 250 J kg-1) compared to the STLD and LTLD events, while LTLD events tend
to have higher values (1000-2000 J kg-1). These results broadly agree with the
trend for less severe storms to have lower CAPE than more severe storms
(McCaul and Weisman, 2001). However, the small sample sizes of events within
each category - and the removal of six events in which CAPE was 0 J kg-1 - may
be obscuring the real trends.
60
In Figure 13a, we see a very weak trend for |CIN| to be the largest in the
STSD and STLD categories, frequently exceeding 200 J kg -1. While |CIN| values
of this magnitude generally prohibit convection from initiating, a sufficiently
strong lifting mechanism - specifically an orographic lifting mechanism - would
still allow storms to develop. Given that locally-forced initiation over the Black
Hills was a necessary condition for an event to be added to the database, it is not
surprising that |CIN| values are relatively high.
Calculations of the Bulk Bulk Richardson Number (BRN) using CAPE
and wind shear values between the 0-6 km layer were also performed as a way to
potentially distinguish between supercell and multicell storms. While LTLD
events had the lowest BRN on average out of all the categories (indicative of
supercell storms), the small sample size of events within this category prevents
this from being a robust result.
Although the thermodynamics profiles surely play a key role in the
evolution of convection, we did not observe clear enough differences between the
average thermodynamic soundings or the distributions of CAPE and |CIN| to
propose any physical hypotheses for how these differences might influence storm
evolution. Thus the forthcoming numerical modeling chapter (Chapter 4)
focusses on the role of the background wind on storm evolution.
61
Chapter 4: Numerical Analysis
4.1: Motivation
The results from the observational climatology presented in Chapter 3
revealed modest differences in the background wind profiles between the STSD,
STLD, and LTLD events. As a result of a limited sampling size of the
climatology, it is fair to question whether or not these results are robust. This
issue can be addressed in two main ways: either by increasing the sample size of
events by extending the observational climatology over a larger period of time
(e.g., from three to five years), or by testing the hypotheses drawn from the
existing climatology by performing numerical simulations. While the former
approach will be pursued in the near future, the latter approach will be followed in
this in this chapter.
As a result of an infinite parameter space of atmospheric flow, the
experiments presented in this chapter will focus on the sensitivity of convective
evolution to the background horizontal wind and vertical shear. As demonstrated
in Chapter 3, the comparison of thermodynamic properties did not reveal any
clear differences between the STSD, STLD, or LTLD regimes. However, there
were more obvious differences in the low-level wind direction and low- to midlevel wind shear between the different categories (see Figure 14). Hence we have
opted to investigate the role of the horizontal wind in greater detail. Furthermore,
due to the location of Rapid City near the near vicinity of the Black Hills, it yet
62
remains unclear whether these differences reflect the background or ridgemodified flow. By performing numerical simulations, it becomes possible to
explore, in a systematic way, the sensitivity of the storm evolution to changes in
the wind profile.
4.2: Experimental setup
The numerical simulations were performed with the Bryan cloud model
version 16 (see Section 2.7 for a detailed description of the model set-up and
initialization). The initial conditions used in these simulations are based on the
full set of 50 observed events between 2010 and 2012 for which a RAP sounding
was available. The thermodynamic and wind profiles in the control case are
simply the averages of those 50 soundings taken at Rapid City, SD at 12Z. This
sounding time was chosen to give a realistic depiction of the early-morning
conditions before solar heat fluxes are applied. A series of six additional
sensitivity simulations were conducted to examine the impact of changes to the
wind profile on storm evolution. These cases, which all use the same
thermodynamic profile (Figure 16) are described in detail here and are
summarized in Table 7.
In Case 2, we halved the vertical wind shear in the 3-6 km layer (to not
disrupt the low-level flow giving rise to the storm initiation) in order to evaluate
the hypothesis that weak mid-level winds favour more stationary, STLD-type
events (see Table 5; Figure 17). Similarly, in Case 3 we doubled the
63
Figure 16: Average thermodynamic profile used in each simulation,
along with the wind profile for the control case.
Figure 17: The wind profiles used for each simulation as described in Table 7
64
vertical wind shear to determine whether stronger shear favoured LTLD events
(Figure 17). While these changes in vertical shear are stronger than that observed,
this enhancement allows for a clearer appreciation of the potential sensitivities. In
Case 4, we rotated the 0-3 km winds counterclockwise by 60˚ from the control to
produce north-northwesterly winds roughly aligned with the ridge of the Hills
(Figure 17). The amount of rotation from the control was linearly decreased from
60˚ to 0˚ over the 3-6 km layer such that the wind direction at 6 km and above
remained identical to the control. This rotation was intended to represent the
ridge-parallel, low-level wind direction observed in STLD and LTLD events. In
Case 5, we combined the reduced mid-level shear from Case 2 with the low-level
wind rotation in Case 4, which best matched the wind profiles from the STLD
events. Similarly, in Case 6 we combined the stronger mid-level shear from Case
3 with the low-level wind rotation from Case 4, which best matched the wind
profiles from the LTLD events. Finally, in Case 7 we rotated the 0-3 km winds
clockwise by 120˚ from the control to produce south-southeasterly winds aligned
with ridge of the Hills - again linearly decreasing this degree of rotation back to
0˚ over the 3-6 km layer - and combined this with the reduced mid-level shear
from Case 2. As in Case 5, this was done to better represent the wind profile of
observed STLD events, while simultaneously testing the sensitivity of the
orientation of ridge-parallel winds (e.g., north-northwesterly versus southsoutheasterly) on the evolution of convective cells.
65
Case
Description
1
Control: average of 50 event soundings at 12Z
2
Vertical wind shear halved between 3-6 km
3
Vertical wind shear doubled between 3-6 km
4
0-3 km winds rotated counterclockwise by
60˚(rotation decreases linearly to 0˚ over 3-6 km)
5
Same as Case 4, except vertical wind shear
between 3-6 km is halved
6
Same as Case 4, except vertical wind shear
between 3-6 km is doubled
7
Same as Case 5, except the 0-3 km winds are
rotated clockwise by 120˚ (rotation decreases
linearly to 0˚ over 3-6 km)
Table 7: A summary of the initial conditions used in each simulation
4.3: Results
The model simulations were analysed by applying the storm-tracking
algorithm of Section 2.2.2 to the simulated composite reflectivity field.
4.3.1: Control (Case 1)
For the control simulation using the unaltered wind profile, the first
convective cell with reflectivity > 40 dBZ initiated around 19:30 UTC (Figure
18a). This cell (1) rapidly grew in size over the next 50 minutes, and began
traveling eastward while maintaining a reflectivity of 50 dBZ. At 20:20 UTC,
(not shown) a secondary cell (2) initiated in almost the exact same area as the
first. By 21:10, cell 1 had travelled 50 km to the east from its point of initiation,
while cell 2 elongated horizontally (Figure 18b). At 21:45, cell 1 had dissipated
66
Figure 18: a-c) Composite radar reflectivity (coloured) atop 100 m terrain contours (greyscale) for
select times in Case 1 d) Storm tracks for all convective cells in Case 1
entirely, while cell 2 began to move eastward. By 22:50, cell 2 remained coherent
and above our 40 dBZ reflectivity threshold (Figure 18c), though it began to show
signs of breaking apart and weakening. A third and final cell (not shown) initiated
and dissipated within 60 minutes after this point, and all convection stopped by
23:50 UTC. The tracks of this simulation are shown in Figure 18d. No cell
travelled greater than 51.7 km from start to finish. This is a similar average
distance to those cells falling under the STSD category in our observed events,
and is broadly similar to the observed STSD event described in Section 3.1.1.
67
4.3.2: Changes to mid-level wind shear (Cases 2 and 3)
In Case 2, convection initiation occurred 40 minutes later than in the
control (20:20 UTC), and produced two distinct cells instead of one (Figure 19a).
Within 50 minutes, cell 2 travelled approximately 30 km eastward and began to
rapidly dissipate (Figure 19b). Conversely, cell 1 underwent a reforming process
during this time, and remained quite close to its point of initiation. Between
21:10 and 22:00 UTC, a third cell (3) initiated just north of cell 1 and propagated
eastward. At 22:00 UTC (Figure 19c), cells 1 and 3 began to drop in reflectivity,
and both dissipated within 15 minutes. A fourth and final cell (4) initiated outside
of our analysis box at this time, and all convection ceased by 23:15 UTC.
Figure 19: a-d) As in Figure 18, but for Case 2
68
Compared to the control case, this simulation produced shorter tracks (maximum
of 41.5 km; Figure 19d), with slightly shorter durations (maximum of 110
minutes). These values both correspond to the STSD classification. Apparently,
halving the vertical shear in the 3-6 km layer leads to similar cell evolutions as the
control case. This suggests that, although weaker mid-level wind speed and
vertical shear is a characteristic of the STLD events (see Table 5), it is not a
sufficient condition for these events to develop.
In Case 3, the first cells initiated at 20:00 UTC in the northeastern section
of the Hills (not shown), and dissipated entirely 60 minutes later. By 21:00 UTC,
another cell (1) had initiated in roughly the same location and rapidly increased in
size (Figure 20a). From this point on, cell 1 began to travel eastward at a nearly
constant speed, and continued to expand while maintaining a maximum
reflectivity of 60 dBZ. By 21:55 UTC (Figure 20b), cell 1 had travelled outside
our analysis box, and began to show signs of splitting (though it still remained a
single cell at this time). The split took place at 22:25 UTC, and can be most
clearly seen at 22:50 UTC with the right-mover (cell 1) and the left-mover (cell 2)
identified in Figure 20c. While cell 2 dissipated at 23:10 UTC, cell 1 retained its
coherent structure and propagated eastward. Cell 1 was still above the 40 dBZ
reflectivity threshold at the end of the simulation, and produced a storm track that
exceeded the axes limits presented in Figure 20d.
69
Figure 20: a-d) As in Figure 18, but for Case 3
In total, cell 1 lasted 195 minutes and travelled 148.35 km. This exceeded
the thresholds for a LTLD event. The split that took place in this particular
simulation is reminiscent of that observed in the LTLD event of August 27th, 2011
(see Section 3.1.3), though the overall duration of the cell in the simulation was
approximately 45 minutes longer than that observed event. This is a promising
indication that doubling the vertical wind shear in the 3-6 km layer profoundly
impacted the evolution of the convective cells. In addition, due to flow deflection
around the mountain, the low-level winds just east of the mountains (e.g., close to
Rapid City) were actually oriented south-southeasterly, while the background flow
was southwesterly (not shown). This suggests that background winds parallel to
the ridge are not required for development of LTLD cells.
70
4.3.3: Changes to low-level wind direction (Cases 4 through 7)
In Case 4, a convective cell initiated along the southern end of the Hills at
19:05 UTC - 25 minutes earlier than the control. By 19:30 UTC, cell 1 in Figure
21a had only increased in size and reflectivity, but did not travel a noticeable
distance in any direction. By 20:50 UTC (Figure 21b), cell 1 had travelled in a
east-southeastward direction, and began to dissipate as it elongated and decreased
in reflectivity. At the same time, a secondary cell (2) initiated just north of cell 1,
which followed an almost identical track to cell 1 over the next 80 minutes
(Figure 21c). Ten minutes later, cell 2 rapidly dissipated, while cells 3 and 4
initiated outside of our analysis box. These cells would be the precursors to future
convective cells that were sustained until the end of the simulation.
Figure 21: a-d) As in Figure 18, but for Case 4
71
Compared to the control, the cells initiated in Case 4 had very similar track
lengths (Figure 21d), and were all similarly classified as STSD. Aside from a
change in the initiation location of the cells to the downwind edge of the ridge,
changes only to the low-level wind direction did not appear to have a significant
impact on the cell’s evolution.
In Case 5, the first convective cell (1) formed in the southeastern region of
the Hills at 19:10 UTC (Figure 22a). By 20:40 UTC (Figure 22b), this cell had
split into two distinct cells: cell 1 remained quasi-stationary, while cell 2 travelled
approximately 40 km southeast. By 22:10 UTC, cell 1 had dropped below the 40
dBZ threshold, while cell 2 remained coherent at the southeastern boundary of our
analysis box (Figure 22c). At the same time, cells 3 and 4 initiated along the
Figure 22: a-d) As in Figure 18, but for Case 5
72
southern boundary of the Hills . These two cells were sustained for 90 minutes
more until all cells dissipated at 23:35 UTC. Cell 1 was identified as STLD
because it did not travel more than 25 km from its point of origin in a 90 minute
period. Its total displacement from start to finish was 42 km (Figure 22d), and it
remained above the 40 dBZ threshold for a total of 175 minutes. That a STLD
cell developed is a key difference from all previous simulations. This suggests
that the combination of decreasing the wind shear between 3-6 km and altering
the low-level wind direction to be parallel to the major ridge of the mountain is
important for producing a quasi-stationary cell.
In Case 6, the first cell (1) initiated at 19:10 UTC, with a second (2)
initiating just south of the first 20 minutes later (Figure 23a). Over the next 50
minutes, both of these cells moved eastward until they reached the easternmost
edge of our analysis box, where they began to dissipate. By 20:25 UTC, two
additional cells (3 and 4) had initiated in roughly the same areas as 1 and 2
(Figure 23b). Like cells 1 and 2, cells 3 and 4 also followed an eastward
trajectory over the next 50 minutes and dissipated in roughly the same area. A
fifth and final cell (5) initiated at 21:15 UTC in the vicinity of where cell 1
initiated (Figure 23c). This too travelled eastward, and dissipated 55 minutes later
along the southeastern boundary. Cells continued to initiate and then dissipate
outside of the analysis box until the simulation ended.
73
Figure 23: a-d) As in Figure 18, but for Case 6
As in Case 4, this particular simulation produced cells that did not last for longer
than 90 minutes, and only travelled a maximum distance of 73 km. Unlike Case
5, there was no observable period in which a cell remained quasi-stationary, so
this event was classified as STSD. Thus, despite the presence of even stronger
vertical wind shear than in Case 3 (due to the wind turning with height), the cells
lasted for a shorter time and did not reach the LTLD thresholds. The inability of
these cells to persist is a subject that will be investigated in future work.
In Case 7, the first convective cell initiated over the northeastern section of
the Hills at 18:45 UTC (Figure 24a) - 45 minutes before initiation occurred in the
control case. By 19:30 UTC (Figure 24b) the cell grew significantly in size, but
had not moved far from its initial location. Indeed, by 20:15 UTC cell 1 had
74
barely left the direct vicinity of the hills (Figure 24c), though it began to dissipate.
At this time, a second cell (2) initiated just west of where cell 1 initiated. This
cell underwent a similar track as the first, though it dissipated within 75 minutes.
As shown by the tracks in Figure 24d, a few cells were also initiated at the
northwestern edge of the Hills. These cells travelled eastward slowly until they
dissipated 80 minutes later. In this case, multiple cells tended to remain quasistationary. Unlike Case 5, where cells had periods of remaining quasi-stationary
but still ended up traveling a large distance, cells in this case were even more
stationary, as they traveled less than 25 km in their entire lifetime.
Figure 24: a-d) As in Figure 18, but for Case 7
75
These results reinforce the notion that decreased vertical wind shear with ridgeparallel low-level winds can lead to quasi-stationary cells, while also
demonstrating that the direction of ridge-parallel winds (e.g., from the southeast
or northwest) can affect how stationary cells remain. A summary of all results
presented thus far for all simulation cases is shown in Table 8 below, accompanied
by the maximum cumulative precipitation recorded over the analysis box.
Case
Event
Classification
Maximum
Duration (min)
Maximum Track
Length (km)
Maximum
Precipitation (cm)
1
STSD
135
51.7
1.65
2
STSD
110
41.5
2.37
3
LTLD
195
148.4
1.69
4
STSD
95
49.8
1.61
5
STLD
175
42.1
2.99
6
STSD
90
73.3
1.25
7
STLD
100
22.4
3.49
Table 8: Summary of all cases with their respective classifications,
maximum durations, track lengths, and cumulative precipitation
4.4 Discussion
The observations showed subtle differences in wind profiles between the
different categories. These simulations were designed to assess the sensitivity to
those differences, and to evaluate whether the changes in wind profile alone could
potentially explain the different observed storm evolutions. Relatively subtle
wind changes, it was found, could indeed explain these observed storm
evolutions.
76
Halving the vertical wind shear over the 3-6 km layer of the control wind
profile had little impact on the evolution of storms compared to those in the
control. This result suggests that although weaker mid-level wind speed and
vertical shear is a characteristic of the STLD events, it is not a sufficient condition
for these cells to develop. When the mid-level shear was doubled, LTLD cells
similar to those in observed events developed. This is a promising indication that
an increase in vertical wind shear over this layer greatly impacts the longevity of
convective storms over mountains. Furthermore, due to flow deflection around
the mountains, low-level winds at Rapid City were actually oriented southsoutheasterly when these LTLD cells developed, while the background winds
were oriented southwesterly. This suggests that background winds parallel to the
ridge are not required for the development of LTLD cells.
Aligning the low-level winds parallel to the ridge of the Hills also had
little impact on the evolution of storms compared to those in the control.
However, when the low-level winds were similarly aligned and the mid-level
shear was halved, cells remained quasi-stationary. This result suggests that STLD
events require a combination of ridge-parallel low-level winds and decreased midlevel vertical wind shear to develop. When the low-level winds were parallel to
mountain ridge and the mid-level shear was doubled, LTLD cells did not develop
as expected. Although this was inconsistent with the observations from RAP data,
it reinforces the finding from Case 3 that background winds parallel to the ridge
are not required for the development of STLD cells. Further investigation is
77
required to better interpret that result. Finally, when the mid-level shear was
halved and the low-level winds were aligned southeasterly (instead of
northwesterly), STLD storms still developed, albeit with shorter tracks. This
result suggests that the direction of the ridge-aligned winds also affects convective
cell evolution. As with the observed STLD events, the highest maximum
cumulative rainfall amounts were recorded in the STLD Cases 5 and 7. Although
these values are nearly half that calculated for the observational events, this is
likely due to an overestimation of the precipitation amounts for the observed
events due to the presence of hail.
In summary, the simulations reveal that relatively subtle changes in
background winds can lead to large changes in storm evolution and attendant
hazards to life and property. This is not to say that thermodynamics are
unimportant, because conditional instability and sufficient humidity are necessary
conditions for convection. However, these results suggest that if those conditions
are met, the wind profile can determine the outcome of the event. While a
physical interpretation of these results cannot be provided at this time, it will
remain be top priority in forthcoming research.
78
Chapter 5: Conclusions and future work
5.1: Conclusions
We have performed an observational and numerical investigation of
convective storms over the Black Hills of South Dakota to determine the factors
that dictate the evolution of convective storms initiated over the mountainous
terrain. After applying a storm-tracking algorithm to composite reflectivity radar
data for each event, we classified them as either short-track, short-duration
(STSD), short-track, long-duration (STLD), or long-track, long-duration (LTLD)
according to the maximum duration and track length of a cell. Using radiosonde
data from Rapid City, SD closest to the time of convection initiation, we
investigated the differences in the background wind and thermodynamic profiles
for all events in each category. While no physical hypotheses could be proposed
from the very small differences between the average thermodynamic profiles of
each category, analysis of the background winds showed modest differences in the
low-level (0-3 km) wind direction and mid-level (3-6 km) wind speeds and
vertical wind shear. In particular, STLD events in this layer had the weakest
average wind speed and vertical shear (7.9 m s-1 and 17.16 m s-1, respectively)
while LTLD events had the strongest (13.7 m s-1 and 25.73 m s-1, respectively).
Moreover, a tendency was observed for LTLD and STLD events to have low-level
winds roughly aligned with the mountain ridge, while no direction was preferred
in the STSD events. Owing to the relatively small sample size of events within
79
each category (particularly STLD and LTLD), these observational results are open
to question. Hence, to evaluate their robustness, we also performed a series of
numerical simulations in which the wind profiles were systematically changed.
Using the Bryan cloud model version 16, we performed seven simulations
that assessed the impact and sensitivity to changes in the 3-6 km wind shear and
low-level wind direction of the background wind profile. In general, the results
from these simulations were consistent with observations. We found that a
halving of the vertical wind shear over the 3-6 km layer, combined with ridgeparallel low-level winds led to the robust development of STLD storms. When
we doubled the shear over the 3-6 km layer, but did not alter the low-level wind
direction, we observed the development of LTLD storms. However, when the
shear was doubled and the low-level winds were parallel to the mountain ridge,
LTLD storms did not develop. This is inconsistent with the observational results,
and will require further investigation. Given the general agreement between the
observations and numerical simulations, these results suggest that winds may play
a dominant role in the evolution of convective storms, provided conditional
instability and sufficient moisture exists. Finally, we found that the expectation
from previous research that strong low- to mid-level shear favours long-lived cells
over flat terrain also holds true over mountains.
80
5.2: Future work
Substantial work must still be done to physically interpret these results.
Of particular importance is understanding precisely why a LTLD event did not
develop in the simulation where the background wind profile best reflected the
observed average conditions for this storm category. However, it should be noted
that in a case where LTLD storms developed, the low-level winds were in fact
locally oriented parallel to the ridge crest due to orographic forcing. Furthermore,
we wish to specifically determine if the mode of organization of LTLD cells is
indicative of supercells or multicells. It is also highly desirable to determine if the
same rules governing convection over flat terrain also apply to mountain-forced
convection. Additionally, we aim to understand precisely why ridge-parallel lowlevel winds favour the development of STLD storms.
To improve statistical sampling and allow for subtle but important trends
to emerge from the noise, we will lengthen our climatology. We will also seek out
other sources of data that will allow us to get a better representation of the lowlevel background wind, which isn’t always represented by RAP soundings.
Possible sources include Doppler radar wind velocity data, surface wind
observations, and NAM analysis data. Additionally, we wish ascertain the depth
of each convective storm by examining their echo tops. Doing so will allow us to
determine whether or not the 12 km layer used in the current background wind
analysis is appropriate for the convective storms initiated in this region, and give
us the opportunity to refine our analysis as needed. Finally, we will assess the
81
impact of changes to the 40 dBZ threshold used in the current investigation for
defining a convective cell.
82
References
Ahijevych, D., 2013: “Image Archive: meteorological case study selection kit”.
Retrieved from: http://locust.mmm.ucar.edu/.
Ansari, S., 2013: “NOAA’s weather and climate toolkit”. Retrieved from:
http://www.ncdc.noaa.gov/oa/wct/.
Banta, R. M., 1986: Daytime boundary layer evolution over mountainous terrain.
Part II: Numerical studies of upslope flow duration. Monthly Weather
Review, 114 (6), 1112-1130.
Banta, R. M., 1990: Atmospheric processes over complex terrain: The role of
mountain flows in making clouds. Meteorological Monographs, 23,
229-283.
Barthlott, C., and D. J. Kirshbaum, 2012: Sensitivity of deep convection to terrain
forcing over mediterranean islands. Quarterly Journal of the Royal
Meteorological Society, 00 (1), 2-23.
Bluestein, Howard B. Synoptic-dynamic meteorology in the mid-latitudes.
Volume II: Observations and theory of weather systems. New York, NY:
Oxford University Press, Inc., 1993. Print.
Bryan, G. H., and J. M. Fritsch, 2002: A benchmark simulation for moist
nonhydrostatic numerical models. Monthly Weather Review, 130 (12),
2917-2928.
Byers, H. R., and R. R. Braham, 1948: Thunderstorm structure and circulation.
Journal of meteorology, 5 (3), 71-86.
Caracena, F., et al., 1979: Mesoanalysis of the Big Thompson storm. Monthly
Weather Review, 107 (1), 1-17.
Carbone, R. E., et al., 1990: The Generation and propagation of a nocturnal squall
line. Part I: Observations and implications for mesoscale predictability.
Monthly Weather Review, 118 (1), 26-45.
83
Chu, C-M., and Y-L. Lin, 2000: Effects of orography on the generation of
propagation of mesoscale convective systems in a two-dimensional
conditionally unstable flow. Journal of the Atmospheric Sciences, 57 (23),
3817-3837.
Colle, B. A., 2004: Sensitivity of orographic precipitation to changing ambient
conditions and terrain geometries: an idealized modeling perspective.
Journal of the Atmospheric Sciences, 61 (5), 588-606.
Crane, R. K., 1979: Automatic cell detection and tracking. IEEE Transactions on
geoscience electronics, Vol. GE-17, 250-262.
Damiani, R., et al., 2008: The Cumulus, Photogrammetric, In Situ, and Doppler
Observations experiment of 2006. Bulletin of the American
Meteorological Society, 89 (1), 57-73.
Davies-Jones, R. P., 1984: Streamwise vorticity: The origin of updraft rotation in
supercell storms. Journal of the Atmospheric Sciences, 41 (20),
2991-3006.
Davies-Jones, R. P., 1986: Tornado dynamics. Thunderstorm Morphology and
Dynamics, 2nd ed., E. Kessler, Ed., University of Oklahoma Press,
197-236.
Davini, P., et al., 2012: Radar-based analysis of convective storms over
northwestern Italy. Atmosphere, 3 (1), 33-58.
Dixon, M., and G. Weiner, 1993: TITAN: Thunderstorm identification, tracking,
analyzing and nowcasting -- a radar-based methodology. Journal of
Atmospheric and Oceanic Technology, 10 (4), 785-797.
Fovell, R. G., and Y. Ogura, 1989: Effect of vertical wind shear on numerically
simulated multicell storms. Journal of the Atmospheric Sciences, 46 (20),
3144-3176.
Gilmore, M. S., and L.J. Wicker, 1998: The influence of midtropospheric dryness
on supercell morphology and evolution. Monthly Weather Review, 126
(4), 943-958.
Goudenhoofdt, E., and L. Delobbe, 2012: Statistical characteristics of convective
storms in Belgium derived from volumetric weather radar observations.
Journal of Applied Meteorology and Climatology., in press.
84
Hagen, M., van Baelen J., and E. Richard, 2011: Influence of the wind profile on
the initiation of convection in mountainous terrain. Quarterly Journal of
the Royal Meteorological Society, 137 (S1), 224-235.
Handwerker, J., 2002: Cell tracking with TRACE3D---a new algorithm.
Atmospheric Research, 61 (1), 15-34.
Hanley, K. E., et al., 2011: Ensemble predictability of an isolated mountain
thunderstorm in a high-resolution model. Quarterly Journal of the Royal
Meteorological Society, 137 (661), 2124-2137.
Harrison, J. B., 2005: “Doppler radar meteorological observations. Part B:
Doppler radar theory and meteorology”. Retrieved from: http://
www.roc.noaa.gov/WSR88D/PublicDocs/fmh-11B-2005.pdf
Hohenegger, C., and Christoph Schär, 2007: Atmospheric predictability at
synoptic versus cloud-resolving scales. Bulletin of the American
Meteorological Society, 88 (11), 1783-1793.
Houze, R. A. Jr., 1993: Cloud Dynamics. 501-538 pp., Academic, San Diego,
California. Print.
Houze, R. A. Jr., et al., 1993: Hailstorms in Switzerland: left movers, right
movers, and false hooks. Monthly Weather Review, 121 (12), 3345-3370.
Houze, R. A. Jr., 2012: Orographic effects on precipitating clouds. Reviews of
Geophysics, 50 (1), 1-47.
Johnson, J. T., et al., 1998: The storm cell identification and tracking algorithm:
an enhanced WSR-88D algorithm. Weather Forecasting, 13 (2), 263-276.
Kirshbaum, D. J., 2011: Cloud-resolving simulations of deep convection over a
heated mountain. Journal of the Atmospheric Sciences, 68 (2), 361-378.
Kirshbaum, D. J., and D. R. Durran, 2004: Factors governing cellular convection
in orographic precipitation. Journal of the Atmospheric Sciences, 61 (6),
682-698.
Kirshbaum, D. J., and R. B. Smith, 2008: Temperature and moist-stability effects
on midlatitude orographic precipitation. Quarterly Journal of the Royal
Meteorological Society, 134 (634), 1183-1199.
85
Klemp, J. B, and R. B. Wilhelmson, 1978: Simulations of right- and left-moving
storms produced through storm splitting. Journal of the Atmospheric
Sciences, 35 (6), 1097-1110.
Kottmeier, C., et al., 2008: Mechanisms initiating deep convection over complex
terrain during COPS. Meteorologische Zeitschrift, 17 (6), 931-948.
Kuo, J-T., and H. D. Orville, 1973: A radar climatology of summertime
convective clouds in the Black Hills. Journal of Applied Meteorology,
12 (2), 359-368.
Lemon, R. L., and C. A. Doswell III, 1979: Severe thunderstorm evolution and
mesocyclone structure as related to tornadogenesis. Monthly Weather
Review, 107 (9), 1184-1197.
Lin, Y-L., et al., 2001: Some common ingredients for heavy orographic rainfall.
Weather Forecasting, 16 (6), 633-660.
Lilly, D. K., 1986: The structure, energetics and propagation of rotating
convective storms. Part II: Helicity and storm stabilization. Journal of the
Atmospheric Sciences, 43 (2), 126-140.
Maddox, R. A., et al., 1978: Comparison of meteorological aspects of the Big
Thompson and Rapid City flash floods. Monthly Weather Review, 106 (3),
375-389.
Markowski, P. M., and Y. P. Richardson. Mesoscale Meteorology in Midlatitudes.
Chichester: Wiley-Blackwell, 2010. Print.
Marwitz, J. D., 1972a: The structure and motion of severe hailstorms. Part I:
Supercell storms. Journal of Applied Meteorology, 11 (1), 166-179.
Marwitz, J. D., 1972b: The structure and motion of severe hailstorms. Part II:
Multi-cell storms. Journal of Applied Meteorology, 11 (1), 180-188.
McCaul, E. W. Jr., and M. L. Weisman, 2001. The sensitivity of simulated
supercell structure and intensity to variations in the shapes of
environmental buoyancy and shear profiles. Monthly Weather Review,
129 (1), 664-687.
Mesinger, F., et al., 2006: North American Regional Reanalysis. Bulletin of the
American Meteorological Society, 87 (3), 343-360.
86
Miglietta, M. M., and R. Rotunno, 2009: Numerical simulations of conditionally
unstable flows over a mountain ridge. Journal of the Atmospheric
Sciences, 66 (7), 1865-1885.
Moncrieff, M. W., and M. J. Miller, 1976: The dynamics and simulation of
tropical cumulonimbus and squall lines. Quarterly Journal of the Royal
Meteorological Society, 102 (432), 373-394.
Morrison, H., et al., 2005: A new double-moment microphysics parameterization
for application in cloud and climate models. Part I: Description. Journal
of the Atmospheric Sciences, 62 (6), 1665-1677.
Nair, U. S., Hjelmfelt, M. R., and R. A. Pielke, 1997: Numerical simulation of the
9-10 June 1972 Black Hills storm using CSU RAMS. Monthly Weather
Review, 125, 1753-1766.
National Climate Data Centre, 2013: “NCDC NEXRAD Data Inventory Search
(Order Historical NEXRAD Weather Data”. Retrieved from: http://
www.ncdc.noaa.gov/nexradinv/chooseday.jsp?id=kudx.
National Operational Model Archive and Distribution System, 2013: “NOAA
National Operational Model Archive & Distribution System - NOMADS
Home Page”. Retrieved from: http://nomads.ncdc.noaa.gov/data.php.
Newton, C. W., and H. R. Newton, 1959: Dynamical interactions between large
convective clouds and environment with vertical shear. Journal of
Meteorology, 16 (5), 483-496.
Newton, C. W., and J. C. Fankhauser, 1964: On the movements of convective
storms, with emphasis on size discrimination in relation to water-budget
requirements. Journal of Applied Meteorology, 3 (6), 651-668.
Ookouchi, Y., et al., 1984: Evaluation of soil moisture effects on the generation
and modification of mesoscale circulations. Monthly Weather Review,
112 (11), 2281-2292.
Oolman, L., 2013: “Atmospheric Soundings”. Retrieved from:
http://weather.uwyo.edu/upperair/sounding.html
Pastushkov, R. S., 1975: The effects of vertical wind shear on the evolution of
convective clouds. Quarterly Journal of the Royal Meteorological Society,
101 (428), 281-291.
87
Raymond, D., and M. Wilkening, 1980: Mountain-induced convection under fair
weather conditions. Journal of the Atmospheric Sciences, 37 (12),
2693-2706.
Richter, J. H., and P. J. Rasch, 2008: Effects of convective momentum transport
on the atmospheric circulation in the community atmosphere model,
version 3. Journal of Climate, 21 (1), 1487-1499.
Rosenfeld, D., 1987: Objective method for analysis and tracking of convective
cells. Journal of Atmospheric and Oceanic Technology, 4 (3), 422-434.
Schroeder, T. A., 1977: Meteorological analysis of an Oahu flood. Monthly
Weather Review, 105 (4), 458-468.
Segal, M., et al., 1988: Evaluation of vegetation effects on the generation and
modification of mesoscale circulations. Journal of the Atmospheric
Sciences, 45 (16), 2268-2292.
Tokay, A., and D. A. Short, 1996: Evidence from tropical raindrop spectra of the
origin of rain from stratiform versus convective clouds. Journal of
Applied Meteorology, 35 (3), 355-371.
United States Geological Survey, 2013: “GTOPO30”. Retrieved from: http://
eros.usgs.gov/#/Find_Data/Products_and_Data_Available/gtopo30_info.
Weisman, M. L., and J. B. Klemp, 1982: The dependence of numerically
simulated convective storms on vertical wind shear and buoyancy.
Monthly Weather Review, 110 (6), 504 –520.
Weisman, M. L., and J. B. Klemp, 1984: The structure and classification of
numerically simulated convective storms in directionally varying wind
shears. Monthly Weather Review, 112 (12), 2479-2498.
Weisman, M. L., and J. B. Klemp, 1986: Characteristics of convective storms.
Mesoscale Meteorology and Forecasting, American Meteorological
Society, 331-358.
Wilde, N. P, Stull, R. B., and E. W. Eloranta, 1985: The LCL zone and cumulus
onset. Journal of Applied Meteorology and Climatology, 24 (7), 640-657.
Wilhelmson, R. B., and J. B. Klemp, 1978: A numerical study of storm splitting
that leads to long-lived storms. Journal of the Atmospheric Sciences,
35 (10), 1974-1986.
88
Wilson, J. W., and D. L. Megenhardt, 1997: Thunderstorm initiation, organization,
and lifetime associated with Florida boundary layer convergence lines.
Monthly Weather Review, 125 (7), 1507-1525.
Wilson, J. W., and R. D. Roberts, 2006: Summary of convective storm initiation
and evolution during IHOP: Observational and modeling perspective.
Monthly Weather Review, 134 (1), 23-47.
Wilson, J. W., and W. E. Schreiber, 1986: Initiation of convective storms at radar
observed boundary-layer convergence lines. Monthly Weather Review,
114 (12), 2516-2536.
Wulfmeyer, V., et al., 2011: The convective and orographically-induced
precipitation study (COPS): The scientific strategy, the field phase, and
research highlights. Quarterly Journal of the Royal Meteorological
Society, 137 (S1), 3-30.
Zhang et al., 2003: Effects of moist convection on mesoscale predictability.
Journal of the Atmospheric Sciences, 60 (9), 1173-1185.
89
Appendix A
List of all events retained in the observational climatology, with their associated times of
initiation/dissipation, storm category, maximum duration, track length, and precipitation:
Event Date
Time of
Initiation1
(UTC)
Time of
Dissipation1
(UTC)
Storm
Category
Maximum
Duration1
(min)
Maximum
Track
Length1
(km)
Maximum
Precipitation2
(cm)
2010-06-21
18:49
21:08
STLD
139
34.56
7.44
2010-06-26
19:00
20:37
STLD
97
24.97
6.55
2010-07-29
19:39
21:48
STLD
129
25.07
7.90
2010-08-01
21:07
23:43
STLD
156
38.75
7.56
2010-08-02
20:04
20:22
STSD
18
3.95
1.25
2011-06-13
21:13
22:17
STSD
64
20.22
2.32
2011-06-24
21:00
23:29
STSD
149
95.89
1.90
2011-06-25
23:05
23:32
STSD
27
19.32
1.55
2011-06-26
19:35
23:00
LTLD
205
171.21
5.22
2011-07-01
20:44
23:28
STSD
164
57.827
1.70
2011-07-04
21:43
23:57
STLD
134
21.73
6.36
2011-07-07
18:49
22:07
STSD
198
78.67
2.88
2011-07-08
21:03
21:40
STSD
37
20.93
2.90
2011-07-18
20:16
21:34
STSD
78
45.03
1.43
2011-07-27
21:01
23:18
STSD
137
62.57
3.41
2011-07-28
19:19
23:34
LTLD
255
106.74
4.62
2011-07-31
22:45
23:50
STSD
65
108.62
1.02
2011-08-01
21:10
23:14
STSD
124
49.91
2.22
2011-08-04
21:42
22:51
STLD
102
22.22
5.93
2011-08-05
19:46
22:39
LTLD
173
107.51
4.17
2011-08-07
21:48
23:30
LTLD
102
103.72
4.74
2011-08-10
18:03
21:51
STSD
228
79.56
1.43
2011-08-11
18:32
21:29
LTLD
177
154.12
5.69
90
Event Date
Time of
Initiation1
(UTC)
Time of
Dissipation1
(UTC)
Storm
Category
Maximum
Duration1
(min)
Maximum
Track
Length1
(km)
Maximum
Precipitation2
(cm)
2011-08-14
20:16
22:35
STSD
139
93.74
1.21
2011-08-18
22:31
23:36
STSD
65
63.18
1.78
2011-08-27
21:20
23:57
LTLD
157
105.29
5.41
2011-08-28
19:22
20:27
STSD
65
49.14
0.54
2012-06-15
21:51
23:39
STSD
108
75.55
1.44
2012-06-20
18:01
20:05
STSD
124
77.83
2.60
2012-06-25
22:00
23:48
STLD
108
16.36
6.74
2012-06-30
19:43
20:57
STLD
97
14.95
7.91
2012-07-02
22:05
23:28
STSD
83
23.77
1.51
2012-07-11*
00:00
01:01
STSD
61
21.66
2.93
2012-07-12
18:49
20:48
STSD
119
67.07
1.50
2012-07-17
19:27
21:03
STLD
96
24.34
7.09
2012-07-18
20:34
21:01
STSD
27
11.09
1.42
2012-07-21
20:42
22:56
STSD
134
82.55
2.19
2012-07-23
21:00
23:57
STSD
177
79.75
1.91
2012-07-24*
00:00
01:57
STSD
117
43.10
2.38
2012-07-25
18:27
20:39
STSD
132
85.18
3.89
2012-07-31
19:28
21:23
STSD
115
60.29
2.36
2012-08-07
20:23
21:10
STSD
47
31.30
1.81
2012-08-09
22:23
23:03
STSD
40
20.35
2.26
2012-08-10
21:48
00:05
LTLD
137
145.58
4.30
2012-08-11*
00:03
01:56
LTLD
113
134.53
5.63
2012-08-26
19:55
20:56
STSD
61
18.98
2.95
* Denotes an event that occurred the following morning (UTC)
1
Of the event-defining cell (see Section 2.3)
2 As
calculated in Section 2.4
91