SEVERE THUNDERSTORM NOWCASTING IN THE ALPINE REGION
AND
THE COALITION APPROACH
Luca Nisi, Igor Giunta, Paolo Ambrosetti
MeteoSwiss, via ai Monti 146, 6605 Locarno-5-Monti, Switzerland
Abstract
Within the COALITION (Context and Scale Oriented Thunderstorm Satellite Predictors Development)
project an object-oriented model is developed, where data from different sources (e.g. Meteosat
Second Generation raw data and derived products, Weather Radar, Numerical Weather Prediction)
are used to increase the lead time in Nowcasting severe convection. Furthermore, the orographic
forcing (often neglected in heuristic Nowcasting models) is considered as a possible convection
triggering mechanism, in particular over complex terrain like the alpine region. The COALITION
algorithm merges severe convection predictors with evolving thunderstorm properties: the storm
evolution is the result of couplings between convective signatures (objects) and environments.
Benefits of the method are expected due to better linkage of the different phases of a convective
storm.
Introduction
Thunderstorms are atmospheric convective phenomena, where large accumulated thermal potential
energies are rapidly released into dynamic form, through nonlinear processes at different scales.
Observations at adequate resolution (10-100m, 1min) are still poor and explicit modeling is prohibitive
in operational environments. Numerical Weather Prediction models are still inadequate for implicitly
assimilating and resolving the important gradient fields. Nowcasting or very short-range weather
forecast are the dedicated fields of the meteorology deputed to cover this forecasting gap. The basic
physical mechanisms governing thunderstorms are fairly well understood (Doswell 2001, Rosenfeld et
al. 2000 and 2008, Setvák et al. 2003, Mecikalski et al. 2006, 2008, 2010a and 2010b, Bedka et al.
2009 and 2010) and these rely on the analysis of temperature and humidity profiles at upper and lower
layers. The orography, especially in major mountain chains like the Alps, plays an important role,
driving the conditions at boundary layer. The convection elements can be initiated, focused, oriented,
reactivated or inhibited (Barthlott et al. 2005, Davolio et al. 2009, Huntrieser et al. 1996, Kottmeier et
al. 2008, Pucillo et al. 2009) by complex orography. The accurate observation, e.g. through remote
sensing techniques, can reveal specific features at different phases of the thunderstorm lifecycle
(preconvective, convective, deep, mature stage), which can be used for significantly improve forecast
skills.
Many of current heuristic Nowcasting models are based on inertial rules (Eulerian or Lagrangian
persistence) and their algorithms analyze data from only a single observation source (e.g. satellite,
radar).
This kind of models is typically employed for forecasting position and intensity in the next 15/60
minutes of mature thunderstorms, through the identification, tracking and classification of a particular
feature, like rain rate. Such persistence rules however can not sufficiently reproduce important
dynamic features, which are driven by other factors, in particular during the initial phases. This results
in low performances during the convection initiation and mature phase (low probability of detection
and high false alarm rates).
The challenge is how to build up a methodology for integrating physical and heuristic information into
one appropriate model. The user community has big expectations to improve the predictability skills
through a better consideration of the thunderstorm dynamics, i.e. first through a better linkage of the
convection phases and secondly through the combination of different observation sources, like
satellite, radar, numerical models and topography information.
This paper reports on the achievements made over the first two years in the context of the
EUMETSAT Fellowship project at MeteoSwiss. After this short introduction, the document presents
the objectives of the project, the achievements made, preliminary results and the conclusions together
with a short outlook.
Objectives
The final goal of COALITION is to produce early assessments of potential thunderstorms in terms of
severity and location, through the rapid modeling of the available convection signals. Thunderstorms
are governed by processes which are nested from the synoptic to the micro-physical scale.
COALTION extracts and merges information derived from satellite Meteosat Second Generation
(MPEF, SAFNWC), Weather Radars (Max-Reflectivity, Vertical Integrated Liquid content) and static
conditions, like topography (fig 1 and fig 2). The orographic effect is very important in particular in the
alpine region with mountain ridges, valleys driving thunderstorm processes through the whole lifecycle (triggering, reactivation and decaying). The challenge is to build up a methodology for integrating
physical and heuristic information into one appropriate model which combines different observation
sources and links the phases of the convection cycle life.
The user community have big expectations to improve the predictability skills, through a better
consideration of the thunderstorm dynamics, i.e. through a better linkage of the convection phases
and through the combination of different data sources. This will help to increase the quality of severe
weather warnings.
Input data and case selection
For training and tuning the algorithm, in particular for assessing Probability of Detection and False
Alarm Rate, about 60 stormy days, corresponding to more than 200 single thunderstorm cells with
different intensities were selected. The cases have been selected and classified according to four
criteria, namely intensity, extension, duration of the convective cell, and synoptic conditions. The first
three criteria are based on the information given by Thunderstorm Radar Tracking (Hering et al.,
2004), according to thresholds applied on the maximum reflectivity (dBZ), vertical integrated liquid
content (VIL), the height of the 45 dBZ echo top, the cells dimension (area) and time duration.
Through the fourth criterion thunderstorms are classified according to meteorological, synoptic scale,
conditions, which are qualitatively inspected from NWP model analysis.
Extra-tropical thunderstorms are differentiated in three main groups, namely pre-frontal, frontal and
stationary. For the first two groups, which are characterized at large scale by the presence of
important thermodynamical gradients and strong winds, orographic forcing plays a secondary role.
Reversely, topography can be very important in case of stationary thunderstorms, where synoptic
scale conditions are flatter and solar heating dominates. During algorithm development this grouping
helped us to organize and optimize the validation of environment modeling. An example is the clear
sky information, which is available and thus applicable to stationary thunderstorms and less to
prefrontal and frontal regimes.
The above depicted classification results in five thunderstorm families:
1. severe, long living cells;
2. mixed situation (in case of many thunderstorms at the same time over an
extended area, typically in case of an incoming cold front);
3. Localized mixed situations (stationary thunderstorm cells, weak wind shear);
4. Weak convective activities;
5. Possible ambiguous situations (strong stratiform precipitations);
This classification scheme is the result of the initial work and can be subjected to modification and
extension in the future, as we expect to enrich our database with 10-15 stormy days each year. About
half of the selected cases will be used to train the algorithm, whereas the remaining will be applied to
validate the algorithm as time-independent datasets.
Fig 1. COALITION input data at present status. Ingested products: Convection Initiation (CI), SAF
Nowcasting Rapid Developing Thunderstorms (RDT), SAF Nowcasting Cloud Top Height, Radar
based Vertical Integrated content (VIL).
Approach and first results
The methodology of COALITION
The COALITION algorithm models severe convection predictors and evolving thunderstorm properties
as interacting elements. The core of the algorithm is a coupling engine with convective signatures as
attributes of an object (e.g. cloud top temperature, cell area and vertical integrated liquid) and
convective environment as external field. Objects are individual entities generated either by ad-hoc
confinement rules or by given external algorithms (e.g. SAF Nowcasting/Rapid Developing
Thunderstorm). Environments are gridded fields from other observation sources, in which the objects
are embedded. These are selected among the collection of available fields, for which well known
physical or heuristic correlations with the object attributes are given (e.g. cloud top cooling and radar
echoes), and depend on the involved context and scale. Potential fields are built up on these
environment characteristics.
Couplings are defined whenever possible through physical conservation laws. Otherwise, semi
empirical rules (based on forecaster’s experience and/or conceptual models) are applied as
functionals (momentum or energy). The forecast corresponds to the evolution of the object attributes
which results as solution of Hamilton equations of type Eq. (1). The method is very similar to that of
solving the generalized dynamics of a particle within a potential field (fig 2), with minimization of the
action.
Fig 2. Simple illustration of the COALITION model. The object-based approach makes use of the
energy conservation principle from the physical mechanics. The forecast of object attributes results
from the solution of Hamiltonian’s equation (eq. 1).
 
H ( q , p )  H ( q1  q n , p1  p n )  E tot
(eq. 1)
where q1..qn is the set of observed attributes (cloud top temperature, cell-dimension, vertical integrated

liquid, …), used as generalized coordinates, and p is the corresponding momentum, related through:
·

q  H / p

·
p  H / q
(eq. 2)
Energy conservation is forced by adding an external potential field (Eext) to the kinetic component
(Ekin):
This Hamiltonian is integrated over past time steps to obtain forward propagators. For this aim a set of
attempted solutions (ensemble) is built according to uncertainties (object location and attributes) and
to residuals. An ensemble forecast is finally performed. Further improvements are expected by
minimizing the difference between observed and forecasted position and momentum all along the
whole development.
Fig 3. Data sources and some possible couplings currently identified for COALITION.
Fig 4. Illustration of the data assimilation principle into COALITION. Residuals and measurements
uncertainty are included to produce an ensemble forecast. New observations are used to correct the
forecast.
In order to present first results a case study is considered. In this paper a thunderstorm cell developing
over a mountainous region in Northern Italy on 12. July 2010 between 12:00 and 14:00 is used to
illustrate the COALITION methodology. The kinetic component of the total energy of this cell, which
has to be correlated with the potential field, is calculated and analyzed. In the implementation of the
first module we assume a simplified one dimensional harmonic oscillator, where the cloud top
temperature is used as generalized coordinate, i.e. appearing as quadratic form in the interaction term
(potential field) and as quadratic form of its rate of change in the inertial part (kinetic energy):
H (q, p, t ) 
p2
 f (t )q 2
2m
(eq. 3)
where q represents an observed thunderstorm attribute (e.g. CTT), p the correspondent momentum,
m a mass of the object inertia and f(t) a function of the correlation between the object attribute
evolution and the external field (e.g. CI). The quadratic term in q simulates a potential field for a one
dimensional harmonic oscillator. Consequently in this module only the thunderstorm towering, without
the horizontal displacement, is forecasted.
In a first phase of the project the object was selected by means of confinement rules (e.g. convexity
analysis and thresholding) applied on 10.8μm brightness temperature from MSG in Rapid Scan mode.
At the moment we have substituted this confinement algorithm with the more reliable information
coming from the SAF-Nowcasting Rapid Developing Thunderstorm (RDT). This new information
allows a better handling of difficult cases, like merging and splitting of thunderstorm cells.
Fig 5. Thunderstorm objects (red) are selected by means of confinement rules. The algorithm based
on convexity analysis and thresholding on 10.8μm brightness temperature from MSG in Rapid Scan
mode.
The kinetic energy is calculated for each object portion (pixel), namely accounting all possible
realizations (see fig 6), which brings the cell at time t1 to the cell at time t2 (t1, t2 .are two successive
time steps).
Number of
realizations
i


 CTT 
 t
i
Fig 6 and 7. The kinetic energy is calculated according to the rate of change of the object attribute (in
this example cloud top temperature). It is estimated for each possible realization i of the same
thunderstorm cell (object) for two successive time steps. On the right the related histogram is shown.
In case that the inertial state (kinetic energy) can be assumed as conserved, usual inertial rules of
closed systems can be applied. This mostly happens in case of mature convective processes, for
which Nowcasting algorithms based on Lagrangian persistence are suited. For all other cases, where
conservation is violated (in particular at initiation and early development stage), the system may no
longer be considered as closed. Energy losses/gains are then explained as import/export of energy
from the surrounding environment, through dynamical exchanges. Fig 8 shows the evolution of kinetic
energy expressed in percentiles for a 2 hours time interval (12:00 - 14:00 UTC). Three time-windows
are selected at different convection phases. Time intervals identified with A show an increase of the
kinetic energy. If the Cloud Top Temperature is used as object attribute, this indicates that during
these periods the thunderstorm is evolving in the vertical direction (indicator of cooling). Changes in
the slope of these percentiles give second order indications about accelerating and decelerating
processes. Time intervals indentified with B indicates that the thunderstorm is almost conservative (no
evolution). The last highlighted time interval C shows an energy decrease, that can be interpreted as
indication of warming processes and/or an homogenizing distribution of cloud top temperatures
(decreasing convexity of the thunderstorm cloud).
Fig 8. Evolution of the vertical component of the kinetic energy over 2 hours period (12:00 and 14:00
UTC on 12 July 2010); a percentile representation is used and units (K2) are arbitrary but consistent
with the simplified model; highlighted are time intervals A, B, C (cf. text).
The energy losses/gains are then explained with the action of an external potential field which is
pumping “energy” into the system from outside. In the next task we will force our model total energy
(Hamilton) to be conserved, in order to be able to interpret kinetic energy variations through potential
energy variations. The COALITION algorithm includes several object-environment modules. In order
to consider as many as possible sources of forecast uncertainties, e.g. at observation, conceptual and
processing level, special attention is being paid for quantifying the uncertainties intrinsic to data and
those produced by the algorithm. An ensemble approach is applied on the governing equations.
First prototype module (SAT – SAT)
We started implementing a prototype module combining different information derived from the same
satellite (Meteosat in Rapid Scan mode), which guarantied us to focus on the basic algorithm, avoiding
synchronization and geolocation problems. This initial module couples the Convection Initiation
product (Mecikalski et al. 2006, 2008, 2010a and 2010b, Siewert et al. 2010) to a thunderstorm cloud
top parameter (object attributes), namely the cloud top temperature (CTT):
Fig 9. From two satellite products (Convection Initiation and Cloud Top Temperature) we estimate a
pseudo kinetic energy and a pseudo potential energy.
The potential field V(CI(t)) is then built up as distribution function of the environment characteristics (CI
product), and steers the evolution of the object parameter, via energy conservation (Hamilton).
Several inspections of the CI product revealed that a direct usage in COALITION, i.e. frame by frame
and at pixel scale, would strongly bias the result, due to its signal variability. On the other hand higher
order features like signal persistence and signal focalizing came out to play an important role as
precursors. We thus made use of the flexible framework offered by COALITION for accordingly
reshaping the CI potential field. The heuristic rule applied in this module can be summarized in: the
“clearer” the Convection Initiation signal is, the more energy is available for the cloud cooling (towering
of the cloud). Several analyses led us to define a CI signal coherence, a combined measure of time
persistence (accumulation over frames) and space coherence (distribution over the object in form of
percentiles), which has then been tested in four forms:
i)
   CI100%  CI 50% 1
ii)
   1  CI100%  CI 50% 1
iii)
   e  CI
iv)
  e
100% CI 50%

CI 50% 
 CI
 1 100%

CIscale


Version iv) is illustrated in fig 10 and shows the most promising skills.
16
# CI interest fields
14
12
10
CI 100%
CI 50%
8
e^(-1+(CI100%-CI50%)/CIscale) * dt
6
4
2
0
11:31
12:00
12:28
12:57
13:26
13:55
14:24
Time (UTC)
Fig 10. Convection Initiation (CI), number of positive interest fields of one thunderstorm cell: ( )
maximal and ( ) median value. In blue is represented the related cumulated signal coherence ( ) (cf.
text, version iv). The horizontal axe represents the time; the right axe represents values (without unit)
for the signal coherence.
The correlation between kinetic energy and the cumulated signal coherence is introduced in the
Hamilton equation. By solving equation (3) analytically we obtain forward propagators which allow to
forecast the evolution of the object attributes, in this case the Cloud Top Temperature. In this paper
only results of the second prototype module are shown and discussed.
Second prototype module (SAT – RAD)
In the second module a satellite product is combined with a radar product. The Cloud Top
Temperature (environment) is coupled with a thunderstorm radar parameter, namely the Vertical
Integrated Liquid (VIL). The potential field V(CTT(t)) is then built up as distribution function of the
environment characteristics, and steers the evolution of the object parameter, via energy conservation
(Hamilton).
Fig 11. From a satellite product (Cloud Top Temperature) and a radar product (Vertical Integrated
Liquid content) we estimate a pseudo kinetic energy and a pseudo potential energy.
In the implementation of the second module we assume (like in the previous module, see eq. 1) a
simplified one dimensional harmonic oscillator, where the vertical integrated liquid is used as
generalized coordinate, i.e. appearing as quadratic form in the interaction term (potential field) and as
quadratic form of its rate of change in the inertial part. Kinetic energy is calculated for each object
portion (VIL pixel), namely accounting all possible realizations (see fig 6), which brings the cell at time
t1 to the cell at time t2.
The correlation between kinetic energy and the Cloud Top Temperature function (g(t)= CTT50% CTT20%, where the CTT50% value represents the median over the thunderstorm cell and CTT20% 20-
percentile) is introduced in the Hamilton equation. The heuristic rule applied in this module can be
summarized in: the smaller the difference between the median and 20-percentile temperature (cloud
expansion), the more energy is available for increasing the vertical integrated liquid. By solving
analytically the equation (eq. 3) we obtain forward propagator which allows forecasting the evolution of
the object attributes, in this second module the Vertical Integrated Liquid.
Fig 12 and 13 show the COALITION VIL forecast for lead times between 5 and 60 minutes. Different
thunderstorm cases are analysed. In both diagrams the horizontal red line represents a threshold
which divides the severe thunderstorm (over the line) from the weak ones (under the line). The vertical
green line represents the VIL observation of the thunderstorm cells at the selected reference time. For
the severe ones, the reference time represent moment when the thunderstorm cell was recognised as
severe by the Thunderstorm Radar Tracking system (Hering et al., 2004), a Nowcasting system for
detecting and tracking thunderstorms with radar data used operationally at MeteoSwiss. The diagram
shows that for the weak ones the COALITION forecast has a good skill for lead times up to 30 minutes
and the false alarm rate very low. For the severe ones the forecast skill decreases with the increase of
the lead time: up 10 to 15 minutes some thunderstorm are miss-forecasted, consequently the
probability of detection decreases.
Fig 12. Forecasted Vertical Integrated Liquid (maximal expected value) for different lead times. 9
different case studies are presented.
In fig 13 two “difficult” cases are presented. Both cases represent severe thunderstorm cells
developed in a mesoscale convective system (MCS). The evolution of thunderstorms in MCS is very
dynamic and the environment in which they are embedded is complex. Thunderstorms cells influence
each other by merging and splitting phenomena which are very frequent. COALITION has difficulties
in handling such complex cases. The skill of the resulting forecast is quite low, the class changes
between weak and severe thunderstorms is frequent and sometimes the difference between two
forecast with 5 minutes difference is very large. Fig 14 shows a first assessment of the COALITION
model. The false alarm rate is calculated analyzing 13 different thunderstorm cases. For lead-time till
30 minutes the false alarm rate is acceptable. Good skill scores for 5 and 10 minutes lead time,
between 15 and 30 minutes the FAR show values between 30% - 40%.
Fig 13. Same as fig 12, but only two “difficult” cases are shown.
FAR
1
0.8
0.6
0.4
0.2
0
0
10
20
30
40
50
60
Fig 14. First assessment of the COALITION model based on 13 different case studies.
lead
time
70
Conclusions
This paper reports on COALITION, a newly developed approach to forecast severe convective storms
by collecting and assimilating information from different data sources into a simplified model. Two
prototype modules are available. First results on the quantification and evolution of object attributes
and related skill scores obtained from different case studies are promising. Validation has been done
on 13 cases for different lead-times (5 to 60 min). Currently we are including all selected cases of the
database for a statistical evaluation of the algorithm skill. Ongoing work includes the correlation of
other potential fields to object attributes and the summarizing of the results of different modules in a
probability map. In particular we will include the topographic information by means of a lightning
climatology.
Acknowledgements
Warm thanks to:
• R. Stuhlmann (EUMETSAT Fellowship program)
•
EUMETSAT Central Application Facility (realtime and archived Meteosat data)
•
SAF/Nowcasting Consortium (software and supports)
•
GEPARD J.Scheiber KG (software development Kit)
•
J. Mecikalski (Convection Initiation algorithm)
•
M. Koenig (EUMETSAT MET Division) for the very precious scientific and technical
support
•
MeteoSwiss staff:
•
L. Clementi (informatic solutions)
•
P. Ambrosetti and M. Gaia (resources and administration)
•
the whole staff of Locarno-Monti for their help
Bibliography
[1] Barthlott, C., Corsmeier, U., Meißner, C., Braun, F. and Kottmeier, C., 2005: The influence of
mesoscale circulation systems on triggering convective cells over complex terrain. Atm. Research, 81,
150-175
[2] Bedka, K., Brunner, J., Dworak, R., Feltz, W., Otkin, J. and Greenwald, T., 2009: Objective
Satellite-Based Detection of Overshooting Tops Using Infrared Window Channel Brightness
Temperature Gradients. J. Appl. Met. Clim., 49, 181-202
[3] Bedka, K., Cronce, L.M., Feltz, W.F., Heidinger, A.K., Pavolonis, M.J. and Sieglaff J.M., 2010:
Convective Storm Initiation Using Satellite-Based Box-Averaged Cloud-Top Cooling and Cloud-Type
Trends. DOI 10.1175/2010JAM2496.1
[4] Davolio, S.,Buzzi, A.. and Malguzzi, P., 2009: Orographic triggering of long lived convection in
three dimensions. Meteorol Atmos Phys 103, 35–44.
[5] Doswell, A. Charles III, 2001: Severe Convective Storms. American Meteorological Society, 28, 50
[6] Hering, A. M., C. Morel, G. Galli, S. Sénesi, P. Ambrosetti, and M. Boscacci, 2004: Nowcasting
thunderstorms in the Alpine region using a radar based adaptive thresholding scheme. In:
Proceedings of Third European Conference Radar on Hydrology (ERAD), Visby (Sweden),
Copernicus, 206-211.
[7] Huntrieser, H., Schiesser, H.H.,Schmidt, W.and Waldvogel, A., 1996: Comparison of Traditional
and Newly Developed Thunderstorm Indices for Switzerland. Weather and Forecasting, 12, 108-125
[8] Koenig, M., E. De Coning 2009: The MSG Global Instability Indices Product and its use as a
Nowcasting Tool, Convection Working Group Homepage, http://convection.satreponline.org/gii.php
[9] Kottmeier, C., Kalthoff, N., Bathlott, C., Corsmeier, U., Van Baelen, J., Behrendt, A., Behrendt, R.,
Blyth, A., Coulter, R., Crewell, S., Di Girolamo, P., Dorninger, M., Flamant, C., Foken, T., Hagen, M.,
Hauck, C., Höller, H., Konow, H., Kunz, M., Mahlke, H., Mobbs, S., Richard, E., Steinacker, R.,
Weckwerth, T., Wieser, A. and Wulfmeyer, V., 2008: Mechanisms initiating deep convection over
complex terrain during COPS. Meteorologische Zeitschrift, Vol. 17, No. 6, 931-948
[10] Mecikalski, J. R. and Bedka, K.M., 2006: Forecasting convective initiation by monitoring the
evolution of moving cumulus in daytime GOES imagery. Mon. Wea. Rev. (IHOP Special Issue,
January 2006), 134, 49-68.
[11] MecikalskyiJ. R., Bedka, K.M., Paech, S.J. and Litten, L.A., 2008: A Statistical Evaluation of
GOES Cloud-Top Properties for Nowcasting Convective Initiation. American Met. Soc. (IHOP Special
Issue, December 2008), 4899-4914.
[12] Mecikalski, J. R., MacKenzie Jr, W. M.,Koenig, M., Mueller, S., 2010a: Cloud-Top Properties of
Growing Cumulus prior to Convective Initiation as Measured by Meteosat Second Generation. Part 1.
Infrared Fields. J. Appl. Meteor. Climate, 49, 521-534
[13] Mecikalski, J. R., MacKenzie Jr, W. M.,Koenig, M., Mueller, S., 2010b: Cloud-Top Properties of
Growing Cumulus prior to Convective Initiation as Measured by Meteosat Second Generation. Part 2.
Visible Fields. J. Appl. Meteor. Climate, 49, 2544-2558
[14] Pucillo A., Giaiotti, D.B., and Stel, F., 2009: Ground wind convergence as source of deep
convection initiation. Atmospheric research, 93, 437-445.
[15] Rosenfeld, D. and Woodley, W.L., 2000: Deep Convective Clouds with Sustained Supercooled
Liquid Water Down to –37.5°C. Nature, 405, 440-442.
[16] Rosenfeld D., Woodley, W.L, Lerner, A., Kelman, G. and Lindsey, D.T., 2008: Satellite detection
of severe convective storms by their retrieved vertical profiles of cloud particle effective radius and
thermodynamic phase. Journal of Geophysical Res., 113, D04208.
[17] Setvák, M., Rabin, R.M., Doswell, C.A. and Levizzani, V., 2003: Satellite observations of
convective storm top features in the 1.6 and 3.7/3.9 μm spectral bands. Atmos. Research, 67- 68C,
589-605
[18] Siewert C., Koenig M., Mecikalski J., 2010: Application of Meteosat Second Generation Data
Towards Improving the Nowcasting of Convective Initiation. DOI 10.1002/met.176