Nowcasting severe convection in the Alpine region: the COALITION approach

Abstract

The purpose of the operationally oriented system named the Context and Scale Oriented Thunderstorm Satellite Predictors Development (COALITION) is automatically to detect severe thunderstorms early in their development and consequently help weather forecasters to increase lead times when issuing severe weather warnings. This new object-oriented system integrates data provided by different sources. Data from the Meteosat Second Generation Rapid Scan Service, weather radar and numerical weather prediction, as well as climatology, are utilized by the system. One of its primary purposes is to use all the best operationally available information about convective processes and to integrate it into a heuristic model. Furthermore the orographic forcing, which is often neglected in heuristic nowcasting models, is taken into account and included in the system as an additional convective triggering mechanism. This is particularly important for areas characterized by complex orography like the Alpine region. The COALITION algorithm merges evolving thunderstorm properties with selected predictors. The forecast evolution of the storm is the result of the interaction between convective signatures and surrounding storm environment. Eight different ‘object-environment’ interactions are analyzed in eight modules, providing ensemble nowcasts of thunderstorm attributes (satellite- and radar-based) for the following 60 min. All ensemble nowcasts are then combined through a weighting and thresholding scheme and the results are summarized into a single graphicalmap in order to facilitate user
interpretation. The COALITION nowcast system has an update frequency of 5min. The output highlights the cells having a high probability of severe thunderstorm development within the next 30min. Verification statistics confirm that COALITION is able to nowcast the intensity of developing convective cells with sufficient skill up to a lead time of about 20 min.

Full text

Quarterly Journal of the Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684–1699, July 2014 A DOI:10.1002/qj.2249
Nowcasting severe convection in the Alpine region:
the COALITION approach
L. Nisi,*†P. Ambrosetti and L. Clementi
MeteoSwiss, Locarno-Monti, Switzerland
*Correspondence to: L. Nisi, MeteoSwiss, via ai Monti 146, CH-6605 Locarno-Monti, Switzerland.
E-mail: luca.nisi@meteoswiss.ch
The purpose of the operationally oriented system named the Context and Scale Oriented
Thunderstorm Satellite Predictors Development (COALITION) is automatically to detect
severe thunderstorms early in their development and consequently help weather forecasters
to increase lead times when issuing severe weather warnings. This new object-oriented
system integrates data provided by different sources. Data from the Meteosat Second
Generation Rapid Scan Service, weather radar and numerical weather prediction, as well as
climatology, are utilized by the system. One of its primary purposes is to use all the best
operationally available information about convective processes and to integrate it into a
heuristic model. Furthermore the orographic forcing, which is often neglected in heuristic
nowcasting models, is taken into account and included in the system as an additional
convective triggering mechanism. This is particularly important for areas characterized by
complex orography like the Alpine region. The COALITION algorithm merges evolving
thunderstorm properties with selected predictors. The forecast evolution of the storm is the
result of the interaction between convective signatures and surrounding storm environment.
Eight different ‘object-environment’ interactions are analyzed in eight modules, providing
ensemble nowcasts of thunderstorm attributes (satellite- and radar-based) for the following
60 min. All ensemble nowcasts are then combined through a weighting and thresholding
scheme and the results are summarized into a single graphical map in order to facilitate user
interpretation. The COALITION nowcast system has an update frequency of 5 min. The
output highlights the cells having a high probability of severe thunderstorm development
within the next 30 min. Verification statistics confirm that COALITION is able to nowcast
the intensity of developing convective cells with sufficient skill up to a lead time of about
20 min.
Key Words: convection initiation; orographic triggering; nowcasting; Meteosat Second Generation.
Received 3 April 2013; Revised 12 August 2013; Accepted 6 September 2013; Published online in Wiley Online Library 28
October 2013
1. Introduction
This article describes the Context and Scale Oriented Thunder-
storm Satellite Predictors Development (COALITION) system,
an application providing cell-based 0–60 min nowcasts of thun-
derstorm severity. The application can be classified as belonging
to a group of ‘expert systems’ and has been tested at MeteoSwiss
since May 2012. Since May 2013, COALITION has been used
operationally.
The basic physical mechanisms governing severe thunder-
storms are relatively well understood (e.g. Rosenfeld and Wood-
ley, 2000; Doswell, 2001). Usually, convection forecasts rely
mainly on the analysis of temperature and humidity profiles of the
†Current address: University of Bern, Oeschger Centre for Climate Change
Research, Bern 3012, Switzerland.
troposphere or on derived indices, considering synoptic forcing
as well. Low-level and mid-level moisture, thermodynamic insta-
bility and kinematic parameters like 3 –6 km wind shear are key
convective predictors for the Alpine region (Huntrieser et al.,
1996). Nevertheless, thunderstorms are governed by processes
that range from the synoptic to the microphysical scale and are
considered one of the most challenging and difficult weather
phenomena to predict, especially in the context of operational
weather forecasting. Moreover, they pose serious hazards to soci-
ety and the economy. For example, it is estimated that on average
between 50% and 80% of all weather-related damage in Switzer-
land is caused by strong thunderstorms (Hilker et al., 2010). Hail,
flash floods and severe wind gusts are the main causes.
Forecast verification is necessary for improving thunderstorm
forecasting, yet observational networks resolve convective
phenomena poorly at spatial and temporal scales down to
10–100 m and 1 min. Since the MeteoSwiss operational weather
c
2013 Royal Meteorological Society

Nowcasting Severe Convection in the Alpine region 1685
prediction model (COSMO2-CH: http://cosmo-model.org/)
operates with lower resolution, deep convection is treated
explicitly only on a 2.2 km horizontal scale and shallow convection
is parametrized. Numerical weather predication (NWP) models
face major limitations in resolving the dominant nonlinear
processes and the characteristically strong gradients associated
with thunderstorms adequately (e.g. Weisman, 1997; Bryan et
al., 2003; Kain et al., 2008). Nowcasting and very short-range
forecasting techniques still remain the most skilful forecasting
tools for strong convection at adequate temporal and spatial
resolution. Remote sensing observations and imagery can reveal
specific features at various scales and stages of the thunderstorm
life cycle to help improve nowcasting skill. Typically, the
thunderstorm life cycle can be subdivided into five main steps
(Mecikalski et al., 2012).
(1) The pre-convective stage (‘clear air’), where dry convective
processes take place caused by solar heating of the boundary
layer, low-level advection of warm air and/or low-level
convergence.
(2) The convective initiation stage, where shallow moist
convective processes take place and cloud formation begins
(e.g. cumulus humilis and mediocris). Typically, in this
phase, only warm cloud processes take place.
(3) The deep convection stage, where deep moist convection
takes place, clouds develop to higher altitudes (e.g. towering
cumulus) and cloud glaciation processes initiate.
(4) The mature stage, when the thunderstorm reaches its
maximum intensity. Updraughts feed the storm with warm
moist air and the downdraught begins to form due to the
fallout of precipitation and subsequent evaporative cooling
of low-level air, known as the cold pool.
(5) The dissipating stage, when the storm’s cold pool stabilizes
its environment and updraught formation consequently
ceases as precipitation gradually tapers off.
For over half a century, a number of applications for diagnosing,
monitoring and nowcasting convective storms have been
developed. In the 1950s, the first attempts to track precipitation
echoes and to extrapolate their future position were developed by
analyzing consecutive radar images (Ligda, 1953). Fujita (1968)
later developed a cloud-motion extrapolation technique using
satellite imagery. Since the 1990s, many applications focusing
mainly on the analysis of real-time radar products have been
developed (e.g. Dixon and Wiener, 1993; Steinacker et al., 2000;
Mecklenburg et al., 2000; Lang, 2001; Handwerker, 2002; Hering
et al., 2004; Kober and Tafferner, 2009). The main advantage
of radar-based systems is that the most active part of the
thunderstorm can be identified and analyzed using reflectivity
information and the spatial and temporal resolution is generally
superior to that of satellite imagery. Radar-based nowcasting
techniques that detect, classify and extrapolate the position of
convective cells (based on Lagrangian persistence) for the next
5 –30 min usually perform well and satisfy the needs of operational
forecasters. Nevertheless, radar-based techniques are unable to
identify convection in its early stages (shallow convection and
initial stages of deep convective processes) due to the lack of radar
echo and are often not able to detect thunderstorm initiation and
forecast thunderstorm motion before the storm approaches its
mature stage.
Other supplemental data, such as satellite imagery, can help to
improve thunderstorm detection during its earliest stages. In the
past decade, many satellite-based applications or multisensor
approaches have been developed to improve nowcasting at
various stages of the thunderstorm’s life cycle. Studies of the
pre-convective environment (e.g. Martinez et al., 2007; Koenig
and De Coning, 2009; Goodman et al., 2012), the convective
initiation phase (e.g. Mecikalski and Bedka, 2006; Mecikalski et
al., 2008, 2010, 2013) and the mature stage (e.g. Setv´
ak et al.,
2003; Rosenfeld and Woodley, 2000; Schulz et al., 2009; Bedka et
al., 2010) have been presented. Some applications, in particular
multisensor ones, have the ability to cover more than a single
phase (e.g. Pierce et al., 2000; Mueller et al., 2003; Puca et al., 2005;
Zinner et al., 2008; Auton´
es, 2012). These ‘expert systems’ have
greatly improved our ability to detect the location of convective
cells and to estimate their magnitudes at different stages.
Despite these recent improvements in nowcasting, thunder-
storms are heavily affected by nonlinear and chaotic processes. For
this reason, the convection intensity nowcasts remain notoriously
challenging. Nowadays a satisfying modelling of the convec-
tive processes is still prohibitive. One of the reasons is that the
predictability of these processes is partially limited intrinsically,
because of their chaotic nature. Particularly in regions with com-
plex orography like the Alps, heuristic nowcasting models, which
are implicitly based on conservation assumptions like Lagrangian
persistence, often fail (Mandapaka et al., 2011). Mountain chains
play a crucial role by driving the conditions at the boundary
layer and convective features can be triggered, strengthened or
weakened by orographic forcing (Barthlott et al., 2005; Kottmeier
et al., 2008; Davolio et al., 2009). Furthermore, the foothills of
the Alpine chain represent one of the regions in Europe where
thunderstorm initiation is most likely to happen (Collier and Lil-
ley, 1994; Huntrieser et al., 1996). Thus, nowcasting systems for
regions with complex orography should account for orographic
influence as well as a multiple-sensor technique.
COALITION is focused on the Alpine region in central Europe.
Typically, the thunderstorms in this area are affected by strong
orographic forcing coupled with moderate synoptic-scale forcing.
According to a climatology study performed in the framework of
the Convective and Orographically-induced Precipitation Study
field experiment (COPS: Wulfmeyer et al., 2008; Craig et al., 2012)
and other studies presented by further authors (e.g. Huntrieser et
al., 1996), thunderstorms over the Alpine chain can be classified
into four different types.
(1) Airmass convection: characterized by isolated, stationary
convection, generally in the case of uniform pressure con-
ditions and missing upper-level forcing. Local instability,
combined with low-level convergence or another trigger-
ing mechanism produced by solar radiation and thermal
circulation, results in short-lived deep convective cells.
(2) Forced non-frontal convection: when synoptic-scale,
upper-level disturbances cause widespread fast-moving
convection. Surface fronts are missing, but often low-level
flow convergence and orographic forcing are important for
storm initiation.
(3) Forced prefrontal convection: found ahead of approaching
cold fronts, where the low-level flow often converges
along terrain-locked quasi-stationary prefrontal bands.
Prefrontal convection is often characterized by strong,
regenerating deep convective cells and echo training in
specific areas within the pre-Alpine region. In these cases,
rapidly moving convective cells tend to initiate over
the western pre-Alpine and Jura areas. In some cases,
if the environment is favourable, the convection may
become organized into supercellular thunderstorms and/or
mesoscale convective systems (MCS).
(4) Forced frontal convection: results from deep convective
processes along surface cold fronts, dominated and driven
by synoptic-scale forcing. In these cases the convective
precipitation is often embedded within stratiform and
orographic precipitation.
1.1. Objectives
The goal of COALITION is to provide early identification
of potentially severe thunderstorms in terms of intensity and
location by rapidly combining and modelling the available
predictors. Through a similar ‘ingredient-based’ approach
(Doswell et al., 1996) and a new methodology derived from
the physics of general dynamic systems, COALITION integrates
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

1686 L. Nisi et al.
and models the best operationally available information about
convective cells and their surrounding environment. The aim
is to provide convection intensity nowcasts for 60 min. The
output updates frequently, every 5 min, and the cells having a
high probability of becoming severe in the next 15–30 min are
highlighted.
This nowcasting product is intended to support weather
forecasters in the decision-making process for issuing severe
thunderstorm warnings. The user community is very interested
in new methods aimed at improving the accuracy and increasing
the lead time of severe storm warnings. One of the basic goals
is to understand the end users’ needs and to provide them with
reliable, easy-to-use, high-quality information in real time.
1.2. Outline
This article is organized into the following sections. The acqui-
sition of remote sensing data and environmental information,
which are processed and ingested into the algorithm, and their
spatial and temporal resolution characteristics are described
in section 2. The configuration of the COALITION model is
described in detail in section 3, illustrating the eight different
modules modelling thunderstorm intensity parameters with con-
vective environmental information. In section 4, the selection of
the case studies and a show case are presented. Then, in section 5,
after the description of the forecast verification methodology and
the presentation of the skill scores used to evaluate the system,
the verification statistics of the COALITION model are described
and evaluated. Finally, we conclude and summarize the results of
our study in section 6.
2. Datasets
The algorithm ingests data obtained from five different sources:
geostationary meteorological satellites, weather radars, numerical
weather prediction models, climatology and digital terrain
information (Figure 1).
The ingested products can be subdivided into three groups (see
Table 1):
(1) Primary products. Represent the algorithm’s basic data. If
one or more of these products is not available, the algorithm
cannot operate.
(2) Secondary products. Increase the quality and reliability of
the results. If one of more of these products are not
available, the algorithm can still operate but in general the
output is of a lower quality.
(3) Auxiliary products. These data improve the quality of the
graphical output by visualizing the information on the
standard grid of the Swiss radar composite. If one or
more of these products is not available, the algorithm can
still be run and the output information is visualized on a
standard polar-stereographic infrared (IR 10.8 μm) image
over central Europe.
The spatial domain of COALITION is depicted in Figure 2. It
includes the central and western Alpine area between 43.6 –49.3◦N
and 2.9–12.2◦E.
2.1. Meteosat Second Generation (MSG)
The main instrument of the payload of the European Organisation
for the Exploitation of Meteorological Satellites (EUMETSAT)
geostationary satellite Meteosat (Schmetz et al., 2002) is a passive
radiometer referred to as the Spinning Enhanced Visible and
Infrared Imager (SEVIRI). This instrument has 12 spectral bands
at wavelengths between 0.4 and 13.4 μm with a horizontal
resolution of 3 km at the satellite subpoint (SSP). The high-
resolution visible band (HRV) has a horizontal resolution of 1 km
at SSP. The temporal resolution is 5 min for all 12 channels,
thanks to a scanning strategy called Rapid Scan Mode (RSS).
Several visible and infrared channels are used to compute
advanced products, which are then ingested into the COALITION
algorithm. The first product presented in this article is
SATellite Convection AnalySis and Tracking (SATCAST) or
Convective Initiation (CI). CI was originally developed for the
US Geostationary Operational Environmental Satellite (GOES:
Mecikalski and Bedka, 2006) and it was then adapted to Meteosat
Second Generation (MSG: Mecikalski et al., 2010). In order to
Figure 1. COALITION input data for the current version of the algorithm. For acronym descriptions and product classifications, refer to Table 1.
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

Nowcasting Severe Convection in the Alpine region 1687
Table 1.COALITION real-time ingested data subdivided into three groups.
Input data Name Acronym Source Resolution Frequency
Primary
Cloud-Top Temperature CTT Satellite (MSG– RSS) 3 km 5 min
Cloud-Top Temperature and Height CTTH Satellite (MSG– Nowcasting SAF) 3 km 5 min
Rapid Developing Thunderstorms RDT Satellite (MSG–Nowcasting SAF) 3 km 5 min
Vertically integrated Liquid content VIL Radar (Swiss Radar Network) 1 km 2.5 min
Secondary
Convective Initiation CI Satellite (MSG– RSS) 3 km 5 min
Convective Available Potential Energy CAPE NWP (COSMO-2 Switzerland) 2.2 km 180 min
Lightning Climatology LC Lightning (Meteorage) 1 km 1 year
Directional Slope Gradients DGRAD Digital Elevation Model 0.5 km Static
Auxiliary Thunderstorm Radar Tracking TRT Radar (Swiss Radar Network) 1 km 5 min
Max Echo CZC Radar (Swiss Radar Network) 1 km 2.5 min
For each product an abbreviation, the operational dissemination frequency and nominal resolution are presented.
Figure 2. COALITION domain over central Europe. The algorithm is run over
an area of 640 ×710 km2.
identify the location of CI, three physical cloud characteristics
are investigated. Cloud-top cooling, cloud glaciation and cloud
depth are estimated using brightness temperatures from five
infrared channels (8.7, 9.7, 10.8, 12.0 and 13.4 μm) and a cloud
mask product (e.g. from the EUMETSAT Meteorological Product
Extraction Facility (MPEF)). With CI, COALITION assimilates
two further advanced satellite products:
•the Cloud-Top Temperature and Height (CTTH) and
•the Rapid Developing Thunderstorm (RDT) product
(Auton´
es, 2012),
which were developed in the framework of the EUMETSAT Now-
casting Satellite Application Facilities (NWC–SAF) Consortium
and generated locally with the Nowcasting SAF software installed
at MeteoSwiss. These are two automatically generated meteoro-
logical products, where MSG data are combined with data from
numerical models (in our case the freezing level extracted from
the MeteoSwiss regional model COSMO-2). These products pro-
vide information about temperatures and heights of cloud tops
as well as an object-oriented identification of rapid developing
convective cells, respectively. In order to help RDT detect convec-
tive clouds at a very early stage, some default parameters (e.g. the
threshold for minimum thunderstorm cloud area and maximum
cloud-top temperature) in the Nowcasting SAF software have
been modified and regional tuning has been carried out. The
temperature of the cloud top is one of the basic parameters in
COALITION, since, as demonstrated in many studies (e.g. Zinner
et al., 2008; Kober and Tafferner, 2009; Mecikalski et al., 2010,
2012; Roberts and Rutledge, 2003), its rate of change in time
provides an estimation of the storm updraught intensity.
2.2. Swiss weather radar network
The algorithm also ingests the grid-based Vertically Integrated
Liquid (VIL) product provided by three ground-based C-band
polarimetric Doppler radars located in the Swiss Alps. This
product is obtained from radars performing volume-scanning and
represents the three-dimensional characteristics of precipitation
systems, with particular emphasis on the convective ones, in
a two-dimensional display (Greene et al., 1972; Johnson et al.,
1998). VIL is represented by radar columnar reflectivity, which
is then converted into liquid water equivalent. It is a useful
indicator of short-term rainfall and intensity of convection used
in nowcasting methods (Boudevillain and Herv´
e, 2003). The
horizontal spatial resolution of the VIL map is 1 km ×1kmand
the temporal resolution is 2.5 min.
In order to improve the quality of the graphical output of
COALITION, the Max Echo radar reflectivity product (Joss et
al., 1998) and the output of the Thunderstorm Radar Tracking
(TRT) algorithm (Hering et al., 2004; Rotach et al., 2009) are
also taken into account. Max Echo is a grid-based product, which
is retrieved using the maximum detected radar reflectivity in a
vertical column. TRT detects the position of convective cells: it
classifies them according to selected radar parameters (Hering
et al., 2004) and extrapolates their position for the next 60 min
based on Lagrangian persistence rules. TRT is very important
for COALITION, since it provides the storm’s future location
and it is used as an independent data source for the verification
of convection intensity nowcasts (see section 5). At present, the
decision process at MeteoSwiss for issuing severe thunderstorm
warnings relies mainly on TRT, which determines the intensity of
the cells according to thresholds of radar-based parameters e.g.
VIL, Echo Top 45 (dBz) and Max Echo. Once a thunderstorm
is identified as ‘severe’ or ‘moderate’ for at least two consecutive
5 min steps, a warning is issued. Hereafter, the warning lead time
will be defined as the difference between the moment when a
severe thunderstorm warning is issued and the moment of the
first detection of the thunderstorm provided by COALITION.
2.3. Regional NWP model COSMO2-CH
The algorithm incorporates the most unstable Convective
Available Potential Energy (CAPE) indicated by COSMO2-
CH forecasts. COSMO2-CH is a non-hydrostatic, regional,
high-resolution model (2.2 km) operated by MeteoSwiss. CAPE
represents an integral value of the available convective potential
in the troposphere and therefore it is a robust indicator, used
for determining the potential of convective intensity (Emanuel,
1994).
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

1688 L. Nisi et al.
Figure 3. Average number of CG strokes per square kilometre in the Alpine area
calculated with a data period of 10 years (2000 –2010). [Correction added on
26 June 2014 after original online publication: the scale for Figure 3 has been
amended to ‘number of CG strokes per year per km2’.]
2.4. Lightning climatology
Lightning data provided by the Meteorage–European Corpora-
tion for Lightning Detection (EUCLID) network are included
in COALITION. Positive and negative Cloud to Ground strokes
(CG) over the Alpine area over a period of 10 years (2000–2010)
are used to build a climatological database. This product, referred
to in this article as Lightning Climatology (LC), provides a yearly
average number of CG strokes per square kilometre. As illus-
trated in Figure 3, this climatological stroke frequency correlates
strongly with the orography. This correlation has been already
documented by several studies, where lightning data were used
for verification (e.g. De Coning et al., 2011).
2.5. Directional slope gradients
The ability of mountain chains to trigger, invigorate and decay
convective processes is well acknowledged (Huntrieser et al.,
1996; Kottmeier et al., 2008). Nowcasting systems, designed for
regions with complex orography, should take into account terrain
features. The contribution of the topography to precipitation
patterns can be described by an extensive set of topographical
descriptors computed from a digital elevation model (e.g.
Foresti and Pozdnoukhov, 2011). The relationship between the
mesoscale flow, which largely determines the direction and speed
of convective cell propagation, and one of these topographic
descriptors, namely the slope gradient, is very important. In the
Alps, upward velocities caused by upslope flows often act as a
trigger for CI and intensification (Pielke and Segal, 1986; Barthlott
et al., 2005). Area, direction and velocity of convective cells have
been used together with data provided by a digital elevation model
(horizontal resolution of 1 km) in order to estimate directional
slope gradients (hereafter DGRAD). According to a Lagrangian
extrapolation of the position of the cells for the following 30 min,
an average terrain slope of the area intersected by cells is
considered. The relation between topographic descriptors and
precipitation depends on the spatial scale. A sensitivity study
(not carried out in this version of the algorithm) should help
to increase the skill of such topographic predictors. Figure 4
provides two examples of DGRAD: the left panel presents a case
where a uniform westerly flow is assumed, whereas the right panel
assumes a uniform southerly flow.
3. Methodology
3.1. Theoretical description of the algorithm: Hamiltonian
formulation of the model
COALITION assimilates data provided by different sources and
integrates them according to a conceptual model. Two physical
attributes, namely the CTT and the VIL, are selected for assessing
the thunderstorm severity. These two parameters are independent
and have been used in many studies as well as in nowcasting
models for assessing and forecasting the storm intensity and
evolution (e.g. Dixon and Wiener, 1993; Hering et al., 2004;
Puca et al., 2005; Kober and Tafferner, 2009; Zinner et al., 2008;
Mecikalski et al., 2010; Auton´
es, 2012). Rapid cooling of cloud
tops and increasing vertical columnar liquid content are clear
indicators of storm intensification processes. In the algorithm,
cell-based CTT and VIL are forecast independently. The first is
used to assess possible CI and intensification processes of cells
at a very preliminary stage when no rain data, and consequently
no radar observations, are available. VIL content is later used to
forecast further intensification and re-intensification of already
developed storms. The other data products, representing the
convective predictors, describe the surrounding environment of
the thunderstorm cells. The algorithm models the evolution
of CTT and VIL with convective predictors as interacting
elements. Nowcasts of thunderstorm attributes are derived from
the analysis and the modelling of present and past observed
conditions. The core of the algorithm works as an engine, which
links thunderstorm attributes with some selected surrounding
environmental conditions. The selection of these conditions is
based on the real-time availability of the respective products and
on their relation to thunderstorm attributes. The relationship
Figure 4. Directional terrain slope gradients in the Alpine area. Red colours indicate upslope, blue colours downslope. Two examples are shown: on the left a westerly
flow is assumed and corresponding terrain slopes in the west– east direction are depicted. On the right, a southerly flow is assumed and therefore terrainslopesinthe
south– north direction are highlighted.
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

Nowcasting Severe Convection in the Alpine region 1689
Figure 5. Simplified illustration of the COALITION conceptual model. The
object-based approach uses the energy conservation principle from physical
mechanics. The total energy is given by the sum of the kinetic (Ekin)andthe
potential (Epot) energy. The forecast of object attributes is the result of the solution
of Hamilton’s equation (Eq. (2)).
that is applied usually relies on conceptual rules used by weather
forecasters. The COALITION methodology borrows its approach
from the physics of general dynamical systems, where the orbital
evolution of a particle moving within a potential field is commonly
described by Hamilton’s equations. The interaction between
thunderstorm attributes and the surrounding environment is
modelled as a particle–field interacting system, as depicted
schematically in Figure 5.
A Hamiltonian model formulation has already been used to
model thunderstorm processes. Mak (2001) considered a non-
supercell tornado as a non-hydrostatic columnar disturbance in a
layer of constant density fluid over a flat surface and used the basic
form of Hamilton’s principle of least action (Miles and Salmon,
1985) to derive the equations governing fluid systems (Salmon,
1998). The COALITION model is formulated in a similar
way, using Hamiltonian formalism as analogy. In general, the
Hamiltonian function for a time-independent system is given by
H(q,p)=H(q1...qn,p1...pn)=Etot,(1)
where q1,... ,qnis the set of generalized coordinates representing
the location of each element in space (e.g. q1=x;q2=y;q3=z)
and other element attributes (e.g. q4=CTT,... ,qn=VIL).
p1,... ,pnis the set of the corresponding generalized momenta.
Hamilton’s equations are given as:
˙q=∂H
∂p,˙p=−∂H
∂q.(2)
The COALITION algorithm scheme is based on the premise
that the energy of the system is conserved. For each interacting
system formed by the couplet ‘thunderstorm attribute’ and
‘surrounding environment’, the pseudo-total energy is assumed to
remain approximately constant over short time-scales. Hereafter,
the word ‘pseudo’ will be used to distinguish energies estimated
by the COALITION model from true physical energies (e.g.
related to movements or to thermodynamics as analogy). If the
inertial state (kinetic energy) is assumed to be conserved, common
inertial rules of closed systems can be applied. This happens mostly
in cases of mature convective processes, for which nowcasting
algorithms based on Lagrangian persistence are suitable. For all
other cases, where such conservation is violated (in particular at
the initiation and early development stage), the system may no
longer be considered to be closed. Energy losses and gains are then
explained as import or export of energy from the surrounding
environment through dynamical interactions (e.g. see example in
Figure 6). Assuming a system without dissipation, the total energy
is given by the sum of a kinetic and a potential component:
Ekin +Epot =H(q,p)=constant, (3)
where qrepresents the generalized coordinates (CTT, VIL) and p
represents the corresponding momentum (pCTT,pVIL ).
As described in section 3.2, predictors for convective cell
attributes are selected from among those accessible in real-time
available products and according to conceptual rules. Predictors
are used in the model as external potential fields (e.g. CAPE).
Potential components of the system are built up using a
predictor’s characteristics (e.g. value distribution, magnitude,
gradients), whereas the computation of the pseudo-kinetic
component is based on the rate of change of attributes describing
the convective cells (e.g. VIL).
A one-dimensional, time-dependent generalization of a
harmonic oscillator (Eq. (4)) involving pairs of predictors and
predictands is assumed for each COALITION module (eight in
total). A single attribute of the convective cell determines the
rate of change in the inertial part (pseudo-kinetic energy), as well
as the quadratic term appearing in the interaction part of the
equation (potential field).
H(q,p,t)=p2
2m−Af(t)q2,(4)
where Ais a positive constant, mis the mass of the object’s inertia
and f(t) is the correlation function between the object’s attribute
evolution and the external field. The choice of the potential energy
in Hamilton’s equation is based on an effort to simulate the system
in a zero-order approximation. Potentials of far more intriguing
complexity may describe the system better, but it is non-trivial to
track them down in a systematic way; moreover a simpler model
may be more appropriate for operational purposes. Conceptually
speaking, the actual number of degrees of freedom in action as well
as the way their interaction/coupling takes place determines the
level of nonlinearity and, as a consequence, the unpredictability of
the system. In some sense, the actual nonlinearity characterizing
the system might reflect to a certain extent how fast the nowcast
produced by our model disassociates from the actual observation.
In the COALITION model it is heuristically assumed that the
total energy remains constant and equal to zero over time:
H(q,p,t)=0.(5)
Figure 6. Energy losses or gains in the kinetic component are explained as an exchange of energy with the surrounding environment. In this example, the thunderstorm
attribute VIL has been used to estimate the pseudo-kinetic energy, whereas the pseudo-potential energy is estimated from the NWP– CAPE product, whichisusedas
the surrounding environment. Dots represent observations of VIL and the surrounding CAPE of a particular thunderstorm cell, whereas solid lines are an idealized
(hand-drawn) representation of the evolution of the kinetic and potential energy.
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

1690 L. Nisi et al.
t1t2
time
VIL [kg/m2]
Number of
realizations
i-realization
Figure 7. Assuming statistical stationarity (no change in PDFs), the kinetic
energy is calculated according to the (discrete) rate of change of the convective cell
attribute (in this example the VIL). An energy value is estimated for each possible
realization iwithin the same cell for two consecutive time steps (i.e. taking into
account all combinations of VIL values between the cell at time t1and t2). In the
right panel, the corresponding distribution is presented.
This assumption is based on the intuitive expectation that losses
and gains of the pseudo-kinetic energy relate to and are balanced
by an exchange of energy with the surrounding environment.
One of the tasks in developing the model was to convert
heuristic relations between cell evolution and its surrounding
environment into forecast rules. A model for energy conservation
has been assumed and verified by means of a number of different
functions representing the pseudo-potential energy. The function
providing the best estimation scores was finally selected as the
most appropriate to be implemented in the model. Under the
aforementioned energy-conservation assumption, Eq. (4) can be
solved analytically providing the following two solutions:
q(t)−q(t0)=e±t
t0√2(A/m)f(t)dtwith q(t), q(t0)>0.(6)
In the model, convective cells are not represented by single
values of CTT and VIL, but rather they are described by a set of
kattributes corresponding to a set of kpixels within the confined
cell. The distribution of the attribute over all pixels provides
information about the variability of the attribute within the cell.
Figure 7 shows a developing cell at times t1and t2(with t=
5 min). VIL is selected as the attribute of this cell. For each object
element (pixel), a pseudo-kinetic value for all possible realizations
i(with i=k) is calculated, according to the rate of change of
the considered attribute.
Equation (6) can therefore be generalized as
qk(t)−qk(t0)=e±t
t0√2(A/m)f(t)dtwith qk(t), qk(t0)>0.(7)
The COALITION model nowcasts the evolution of two
attributes of convective cells, namely CTT and VIL. These
nowcasts are particularly important for cells that are detected
in their early stages and likely to develop further. In the
current version, to satisfy the primary user needs, only the
cell development is taken into account; decaying processes are
not considered yet. As a consequence, only the positive solution
(Eq. (6)), which represents the forward propagation of our model
and allows us to forecast increases in thunderstorm intensity, is
taken into account:
qk(t)=qk(t0)+et
t0√2(A/m)f(t)dt,(8)
where
f(t)=f0k+B(t−t0)3/2−f03/2,(9)
f0k=1
2˙
qk(t0)
qk(t0)2
, (10)
B=df
dt+dferr
dt=df
dt+∂ferr
∂σ ∂σ
∂t, (11)
and where σrepresents the predictor. Potential fields steer the
evolution of the object attributes, following the energy conser-
vation assumption. Characteristics of the external environment
(∂σ) are included in the model (Eq. (11)) and are used to explain
the differences between extrapolated (Eq. (9)) and observed
(Eq. (10)) pseudo-kinetic energy. We use the correlation between
the averaged value of these differences (ferr) and the averaged
value of the selected surrounding environment (σ). This cor-
relation takes into account regressive information and is used
to estimate a set of pseudo-kinetic energies for the following
time step, consequently providing an ensemble forecast of the
evolution of the considered cell attribute.
In order to validate the underlying assumptions and the
COALITION methodology, in the first version of the algorithm
a simple function, i.e. a one-dimensional, time-dependent
generalization of a harmonic oscillator (Eq. (4)), was used.
Later, by applying more complex functions and by using more
refined relations between cell attributes and the surrounding
environment, the skill of the method and consequently the results
will improve.
3.2. Algorithm structure and modules
The structure of the algorithm can be summarized in the following
steps.
(1) As a primary input, COALITION ingests the Nowcasting
SAF/RDT product tuned to detect small convective clouds
as well as larger ones (see section 2). For this purpose, the
threshold for maximum cloud-top temperature is increased
from 5 to 10 ◦C and the threshold for the minimum cloud
area decreased from 60 to 25 km−2. Based on this external
information, COALITION selects and confines convective
cells on MSG 10.8 μm infrared images.
(2) Once a convective cell is detected, the parallax correction
is computed using cloud-top height information provided
by Nowcasting SAF. This step is very important in order
to reduce the differences in the geolocation of the cells,
considering the large number and variety of products
ingested by the algorithm (e.g. radar, NWP).
(3) In a third step, the cell is additionally identified and ana-
lyzed using the radar-derived VIL. Cell-based CTT and VIL
attributes are then modelled using several environmental
parameters (predictor fields). In the current version of the
algorithm, eight different ‘attribute– environment’ cou-
ples, derived by semi-empirical rules based on forecasters’
experience and conceptual models, are implemented (Fig-
ure 8 and Table 2). Three of them provide an ensemble
nowcast of the CTT attribute, while the remaining five
provide an ensemble nowcast of the VIL. Nowcasts are
generated for the next 60 min with a time resolution of
5 min. Finally, all 90% quantiles from the ensembles are
then used to produce and provide the end user with a single
output.
As a fully automatic ‘expert system’, COALITION verifies the
input product availability and quality. If the needed data for
one or more modules are missing, COALITION automatically
uses available information from the immediate past or it ignores
the module concerned, depending on the type of the data and
its typical variability in time. For example, if the environmental
information CAPE is not available, the previous value is taken (it
couldbethevalueof5,10or15minearlier),sincethisproduct
does not change considerably over a short time and a small region.
However, if VIL is missing, the modules requiring this input data
are automatically excluded from the model, given that VIL is a
highly variable parameter over time and space.
3.3. Combination of module forecasts
VIL and CTT nowcasts provided by eight modules with a
5 min update cycle are integrated through a weighting and
thresholding scheme similar to a fuzzy logic approach. This
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

Nowcasting Severe Convection in the Alpine region 1691
Figure 8. The eight modules implemented in the current version of COALITION. Products used as external environment are in green, cell attributes to be forecast
are in blue.
Table 2.Description of the eight modules and the related semi-empirical rules that motivated the selection of predictand– predictor pairs.
Module Data combination Semi-empirical rule
1 Evolution of the CTT based on the environment defined in
terms of the Convective Initiation product
Stronger convective initiation signal of a cell →stronger updraught is available for
its cloud-top cooling (towering of the cloud)
2 Evolution of the VIL based on the environment defined in
termsofCTT
Faster cooling of the cloud top of a convective cell →more potential energy is available
for increasing its VIL
3 Evolution of the VIL based on the environment defined in
terms of a lightning climatology
Higher climatological density of cloud to ground lightning over a specific area →more
potential energy is available for increasing the VIL of a convective cell developing in this
area
4 Evolution of the VIL based on the environment defined in
terms of the Convective Initiation product
Stronger convective initiation signal of a cell →more potential energy is available for
increasing its VIL
5 Evolution of the VIL based on the environment defined in
terms of the Convective Available Potential Energy (CAPE)
Higher instability values →more potential energy is available for increasing the VIL of a
convective cell developing in this area
6 Evolution of the CTT based on the environment defined in
terms of orographic information (slope gradients)
A convective cell is moving toward a mountain (orographic forcing) →stronger updraught
is available for its cloud-top cooling (towering of the cloud)
7 Evolution of the VIL based on the environment defined in
terms of orographic information (slope gradients)
A convective cell is moving toward a mountain (orographic forcing) →more potential
energy is available for increasing its VIL
8 Evolution of the CTT based on the environment defined in
terms of the Convective Available Potential Energy (CAPE)
Higher instability values →stronger updraught is available for its cloud-top cooling
(towering of the cloud)
is an efficient methodology for nowcasting the likelihood of
severe storms through the application of a conceptual model
(Roberts et al., 2006). According to subjective experience of local
weather forecasters, the conceptual model is built on the fact that
predictors have a different importance according to the current
stage of the convective cell. Different weights are assigned to the
modules for each possible stage (first guess based on forecaster
experience). By analyzing cell-based characteristics like CTT and
VIL as well as their first derivatives, the system automatically
computes an estimation of the current cell’s stage. A simple
sensitivity study based on 40 randomly selected cases has been
performed for optimizing the weighting scheme. Figure 9 shows
the module weights currently used in the COALITION algorithm.
Large datasets, required for example by neural network systems,
were not necessary. If needed, the weights can be modified in an
easy way for regional tuning purposes. Furthermore, the system
can be extended as soon as additional predictands or improved
modules become available.
4. Results
4.1. Selection of cases
The COALITION application has been run in real time at
MeteoSwiss since May 2012. During the convective season
(May–September 2012), the algorithm has been subjectively
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

1692 L. Nisi et al.
0
0.5
1
1.5
2
2.5
3
3.5
CTT >= 220°
VIL < 5 kg/m2
CTT >= 220°
VIL >= 5 kg/m2
dCTT >= 3 °
dVIL <= 5 kg/m2
CTT >= 220°
VIL >= 5 kg/m2
dCTT >= 3 °
dVIL >= 5 kg/m2
CTT >= 220°
VIL >= 5 kg/m2
dCTT < 3 °
dVIL < 5 kg/m2
CTT >= 220°
VIL >= 5 kg/m2
dCTT < 3 °
dVIL >= 5 kg/m2
CTT < 220°
VIL >= 5 kg/m2
CTT < 220°
VIL < 5 kg/m2
weights
MOD1
MOD2
MOD3
MOD4
MOD5
MOD6
MOD7
MOD8
Figure 9. Module weights for different stages of convective processes. Absolute values of CTT and VIL are used to assess warm/cold cloud tops and light/heavy
precipitation; first derivatives approximated by finite differences are used to identify important rates of change in time, for example strong cooling of the cloud tops
or strong rain/hail intensification.
Figure 10. The distribution of geopotential height (black bold lines), temperature (red lines) and relative humidity at 500 hPa (ECMWF) on 1 August 2012 at
1200 UTC shows the trough located over the Atlantic and a short-wave, thermal signature (red circle) determining increased instability over central and western
Europe.
verified by weather forecasters and independently and objectively
using the output of the TRT system. Since the main purpose of
the algorithm consists of automatically analyzing and modelling
convective cell parameters using surrounding environmental
characteristics, the verification was performed independently of
the weather situation and life duration of the cells. 80 convective
cases were considered. Based on the severity classification
provided by TRT (Hering et al., 2008), 40 weak and 40 moderate
to severe thunderstorms were randomly selected from the
archive.
4.2. Example of a real-time application: 1 August 2012
This section provides an example of a COALITION nowcast
for 1 August 2012. The synoptic situation was dominated by a
long-wave quasi-stationary trough over the Atlantic Ocean and
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

Nowcasting Severe Convection in the Alpine region 1693
~300 KM
15:05 UTC 15:05 UTC 15:05 UTC
15:05 UTC 15:05 UTC 15:05 UTC
15:35 UTC 15:35 UTC 15:35 UTC
16:05 UTC 16:05 UTC 16:05 UTC
15 25 35 45 dBz
Cloud top cooling
t+15 min
Convection intensification
t+30 min
Severe Convection
development t+30 min
WEAK
MODERATE
SEVERE
A
B
Figure 11. COALITION severe thunderstorm nowcasts and radar estimation references for the western Alpine region on 1 August 2012. Only some selected frames
with a 15 min time step (between 1505 and 1605UTC) are shown. The left column shows the COALITION nowcast, the central column shows the maximum radar
reflectivity and the right column shows the thunderstorm classification provided by TRT.
western Europe (Figure 10). A short-wave thermal trough moving
around the depression from west to east crossed central Europe
and affected the Alpine region.
Over central Europe, moderate south-westerly flow brought
very warm, moist unstable air. Initially in the Alpine region there
were dry conditions because of weak Foehn winds from the south.
During the afternoon, moisture and instability increased and at
the same time an upper-level disturbance crossed central Europe
from west to east. This moderate synoptic-scale forcing triggered
the development of strong thunderstorms over the French pre-
Alpine region. Then, later in the evening and overnight, the entire
northern Alpine region was affected by convective activity.
The left column in Figure 11 shows COALITION nowcasts. The
yellow colour indicates locations where only cloud-top cooling
processes were forecast for the next 15 min. Orange and red
colours are indicators of VIL intensification for the next 30 min:
convective cells are highlighted with these colours when the
nowcasts indicate that (i) CTT will decrease by at least 5 K within
the next 15 min (yellow), (ii) VIL will increase by 15 – 25 kg m−2
(orange) and (iii) VIL is expected to increase by more than
25 kg m−2in the next 30 min (red).
Radar products, which are used as reference for verification,
show that at 1505 UTC two convective cells (highlighted with
a red square in Figure 11) have already initiated. At that time,
COALITION predicted an important decrease of the CTT for
the northernmost cell Aand an increase of the VIL values for
the second one B. The first cell was not yet classified by TRT,
whereas the second was classified as ‘weak’ (right column). At
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

1694 L. Nisi et al.
Figure 12. Nowcasted VIL over lead time for four different thunderstorm cells (two severe and two weak ones). In the plots, nowcasts provided by five modules for
the selected reference (represented by the vertical black line at lead time =0) are presented. VIL observations at different lead times are also shown (bold dashed line).
1535 UTC, the radar products showed that cell Bindeed increased
its intensity (max values greater than 50 dBz) and it was already
classified as ‘moderate’ by TRT. At that time, the COALITION
nowcast indicated that a probable further intensification up to
‘severe’ convection was expected within the following 30 min.
The rapid cooling forecast at 1505 UTC for cell Awas confirmed.
In fact, the cell slightly increased its intensity and was classified as
‘weak’ by TRT. If compared with cell B, however, the analysis of
the surrounding environment (especially the CI product, CAPE
and orographic triggering) showed that during this period there
were less favourable conditions supporting severe convective
development. By 1535 UTC, COALITION indicated a probable
intensification of this cell within the next 30 min as conditions for
development became more favourable. For every 5 min thereafter,
between 1535 and 1605 UTC, the nowcast system continuously
indicated that the two cells would increase in intensity (not shown
in Figure 11). At 1605 UTC, cell Bwas classified as ‘severe’ by TRT
and cell A after a splitting process, was classified as ‘moderate’.
For this particular case study, COALITION forecasts verified
correctly. A more general verification of the system, where a large
number of cases were analyzed, is presented in section 5.
5. Performance statistics
5.1. Verification of quantitative module forecasts
In this section, quantitative nowcasts provided by the eight
COALITION modules are validated against observations. The
panels in Figures 12 and 13 present four randomly selected
convective cells. According to the classification provided by TRT,
two of them (cases 1 and 2, depicted in the upper panels of
Figures 12 and 13) developed into severe thunderstorms. The
remaining two (cases 3 and 4, depicted in the bottom panels of
Figures 12 and 13) developed into weak thunderstorms.
The plots show a comparison of COALITION’s VIL and CTT
nowcasts provided by different modules for lead times between
5 and 60 min. Figure 12 shows the VIL forecasts provided by
modules 2, 3, 4, 5 and 7, whereas Figure 13 shows the CTT
forecastsprovidedbymodules1,6and8(seesection3fora
detailed description of the modules).
VIL and CTT observation references are highlighted on the
plots with a vertical line. For the severe ones, the forecast reference
is represented by the VIL value observed when the thunderstorm
cell was first recognized as severe by TRT. For the weak ones,
the forecast reference is given by the time when the maximum
observed VIL value is observed, considering the whole life cycle
of the cell.
As expected, for both weak and severe cases the differences
within the nowcast ensemble become larger as the lead time
increases. Deviations among the modules are a consequence of the
different influence of the external environments on the forecast
thunderstorm attribute: some environments may support the
development of severe convection, whereas at the same time
other environments inhibit it.
As suggested by Wilks (1995), Mean Absolute Error (MAE) and
nowcast bias errors are calculated for lead times of up to 30 min
(Table 3). Considering the results of the analysis of 40 severe cells,
nowcasts generally show good skill up to 20 min before reaching
the mature stage [MAE <8.1 (kg m−2), bias >−3.4 (kg m−2)].
Explosive cells constitute an exception, since for this kind of
storm the warning lead time is reduced to 5–10 min. The main
difficulty is the handling of extremely large and rapid increases
of the VIL: in some explosive thunderstorms it jumps by about
30 –50 kg m−2within less than 10 min. Although the environment
shows favourable conditions for severe convection, it is very
difficult to model and nowcast this extremely rapid change of cell
attributes. Considering VIL nowcasts for weak thunderstorms,
the prediction skill remains good for longer lead times, in general
up to 30 min [MAE <7.1 (kg m−2), bias >−3.8 ( m−2)]. In these
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

Nowcasting Severe Convection in the Alpine region 1695
Figure 13. Nowcasted CTT over lead time for four different thunderstorm cells (same as in Figure 12). In the plots, nowcasts provided by three modules for the
selected reference (highlighted with the vertical black line) are presented. CTT observations at different lead times are also shown (bold dashed line).
Table 3.Mean absolute error and bias for different lead times and for both VIL and CTT nowcasts.
Lead time (min)
VIL forecast (kg m−2) CTT forecast (K)
Weak Severe Weak Severe
MAE BIAS MAE BIAS MAE BIAS MAE BIAS
51.1−0.4 2.1−1.4 2.9 −0.5 3.1 −0.5
10 2.3 −0.6 4.3−2.2 3.0 −0.7 3.8 −0.9
15 2.9 −2.4 5.6−2.6 3.3 −0.9 4.3 −1.7
20 4.1 −2.9 8.1−3.4 3.6 −1.3 4.6 −2.1
25 5.9 −3.3 9.5−3.8 4.2 −2.3 5.1 −2.5
30 7.1 −3.8 11.0−4.4 4.8 −2.4 8.0 −3.4
40 weak and 40 severe thunderstorms have been considered.
cases, most of the surrounding environments agree in showing
less favourable conditions for the development of severe cells;
consequently, the variability among the modules is smaller.
Figure 13 shows the comparison of CTT nowcasts for the same
thunderstorms as shown in Figure 12. The forecast reference
has been defined using similar criteria to those described above.
For severe thunderstorms, the reference corresponds to the CTT
value observed at the moment when the thunderstorm cell is
first recognized as severe by TRT. For weak ones, the forecast
reference is given by the CTT value observed at the time when the
maximum VIL value (considering the whole life cycle of the cell)
was observed. Generally, for both weak and severe convective
cases, the CTT nowcasts provided by the three modules show
good skill for lead times up to 30 min [MAE <8 (K) and bias
>−3.4 (K) for severe thunderstorms and MAE <4.8 (K) and
bias >−2.4 (K) for weak ones]. For a better understanding of
the skill, the variability of the thunderstorm CTT and VIL will
be analyzed. As demonstrated by the analysis of the life cycle
of a large number of convective cells, the distribution of CTT
values remains more constant compared with the distribution of
VIL values. This property, together with the fact that the CTT
attribute is less variable in time compared with the VIL, results in
an improved forecast performance of the corresponding modules.
However, the good skill of the CTT nowcasts for longer lead times
is mitigated by higher false alarm ratios (hereafter FARs); in fact,
the VIL attribute is more robust and reliable for discriminating
between severe thunderstorm cells and weak ones.
5.2. Verification of COALITION nowcast
In this section, the likelihood for a severe thunderstorm provided
by COALITION, which is generated using the combination of all
module forecasts (as described in section 3.3) and provided to the
weather forecasters, is verified against TRT. The goal is to assess
the probability of detections and the FARs. 80 cases from the 2012
convective season have been considered. At MeteoSwiss, forecast-
ers use TRT and COALITION synergistically for issuing severe
thunderstorm warnings. Based on objective radar parameters,
the operational and validated TRT system provides information
about the current cell intensity and an extrapolation of its future
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

1696 L. Nisi et al.
position. COALITION identifies the cells having a high proba-
bility of becoming severe during the next 30 min. For this reason,
TRT is considered an independent ground truth in the evaluation
of the COALITION algorithm. In particular, the evaluation
process should provide answers to the following questions:
(1) what is the skill of COALITION in nowcasting severe
thunderstorms; and
(2) do we improve lead time for issuing severe thunderstorm
warnings using COALITION forecasts?
Thunderstorm forecasts were evaluated using typical skill
scores applied in meteorology to verify weather predictions
(e.g. Huntrieser et al., 1996). Strong thunderstorms have been
used to assess the probability of detection (POD), whereas weak
thunderstorms were used for assessing FARs. Furthermore, the
critical success index (CSI) was calculated for different lead times.
POD describes the ability of the forecasting system to predict
severe thunderstorms: if the occurrence is always forecast (perfect
score), the POD is equal to 1. FAR indicates the proportion of
thunderstorm forecasts that did not develop into severe cases. A
perfect score is represented by a FAR equal to 0, indicating that
forecasts were always followed by the occurrence. CSI describes
the ability of the forecasting system to have at the same time a
high POD and a low FAR. A perfect CSI score equals 1, while 0
indicates that there are no correct forecasts. The three skill scores
were assessed according to a 2 ×2 contingency table for forecasts
and observations (Wilks, 1995).
Algorithm performances are described in Table 4. COALITION
severe thunderstorm nowcasts were compared with the
thunderstorm intensity classification provided by TRT. Again,
as described in section 5.1, the lead time represents the time
lag between the first identification of the upcoming severe
development provided by COALITION and the first detection
of a severe thunderstorm by TRT. As presented in Table 4
skill scores decrease with increasing lead time. In the first
20 min, the POD decreases from 100% to 60%, whereas the
FAR increases from 0% to 44.2% and the CSI decreases from
100% to 40.7%. For lead times longer than 20 min, however, skill
scores decrease very rapidly: the algorithm tends to overestimate
the intensity evolution of convective cells, but in some cases it
missed the severe ones. In general, considering that COALITION
provides fully automated identification and nowcasting of strong
thunderstorms, the verification results are promising. Algorithms
providing nowcasts with a POD greater than 60% and a FAR
below 40% with an averaged lead time of about 15 –20 min
are considered reliable and potentially useful for operational
forecasting purposes (e.g. Kober and Tafferner, 2009). For this
reason, it can be concluded that COALITION nowcasts add value
to operational thunderstorm nowcasts: in favourable conditions,
Table 4.Performance of COALITION for predicting development of severe
thunderstorms.
Lead time POD FAR CSI
50.925 0.26 0.698
10 0.8 0.385 0.533
15 0.725 0.42 0.475
20 0.6 0.442 0.407
25 0.375 0.487 0.277
30 0.225 0.64 0.161
35 0.125 0.689 0.098
40 0.05 0.8 0.042
45 0.025 0.8 0.023
50 0.025 0.834 0.022
55 0 0.843 0
60 0 0.85 0
Comparison of three skill scores (POD, FAR and CSI) for lead times between 5
and 60 min. Bold values indicate lead times for which the COALITION forecast
can be considered useful. For this summary statistics, 80 case studies have been
considered.
Figure 14. Frequency distribution of the lead times in identifying severe
thunderstorms by COALITION. This histogram considers 40 cells that effectively
developed into a severe thunderstorm.
severe weather alerts could be issued up to 20 min earlier than
without the use of COALITION.
The distribution of 40 severe thunderstorms versus the lead
time is presented in Figure 14. The figure shows that about half
of all thunderstorms were identified 15–25 min before they were
classified as severe by TRT. Only two cases were detected with
lead time greater than 35 min, but on the other hand only three
cases were not detected at all. Considering all cases, a typical lead
time of about 20 min is obtained. Reviewing the different cases, it
can be noted that missed forecasts or a detection with very small
lead times (5–10 min) are often related to extremely explosive
cells or convection embedded in cold fronts.
5.3. Discussion
The first real-time test phase of COALITION during the 2012
convective season and the detailed analysis of a large number
of thunderstorm cells demonstrated that the nowcasts are
generally useful for operational forecasting, providing support
for convective intensity nowcasts. The continuity in time
(consistency) of COALITION nowcasts is also an important
factor to consider. Cells for which COALITION provides a
consistent nowcast for two or more consecutive time steps have
in fact a higher probability of increasing their intensity and
reaching the severe stage. This probability is much lower when
the nowcasts are less coherent in time. Usually, in the first case,
all the surrounding environmental conditions (more than 80%)
are consistent and agree with each other, showing favourable
conditions for intensification. In the second case, the different
environmental predictors provide less consistent information,
resulting in a lower consistency between the different modules.
Some of them indicate favourable conditions for the development
and intensification of convective cells, whereas others indicate less
favourable conditions, yielding a lower prediction skill.
The system showed some difficulties in handling convective
development in some specific weather conditions. Typically, the
case of strong synoptic-scale forcing in proximity to cold fronts
and embedded convective cells is not well forecast, or their
prediction has very short lead times (5 min). The reason has to
do with the quality of some of the primary input products, in
particular RDT. The identification of convective cells through
a thresholding scheme and tracking based on the overlap-
mechanism applied in the RDT algorithm are less suitable,
especially in this case of dense cloudiness in a frontal band and
in the absence of significant cloud-top temperature gradients.
COALITION can deal fairly well with merging and splitting
cells; however, some misdetections originated from embedded
convection with a large number of convective cells. For this reason,
COALITION performed best in cases with airmass convection,
forced non-frontal and prefrontal convection, since the cells rarely
become organized into larger mesoscale structures.
As mentioned in section 2, using RDT for early identification
of convective processes requires modification and tuning. It was
observed that, without these modifications, when the cells were
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

Nowcasting Severe Convection in the Alpine region 1697
first detected by RDT the first rain signals had already been
detected by C-band ground radars. On the other hand, this
tuning resulted in a strong increase in the number of false alarms.
The ability of COALITION to integrate this information with
other data sources allows it to reduce the number of false alarms
considerably. Extremely rapidly developing cells, however, are
usually detected by COALITION with a shorter lead time than
in the average. This is simply due to the very high rate of growth
characterizing these explosive storms.
6. Summary and conclusions
In this article, a newly developed object-oriented, heuristic
approach to nowcasting severe convective storms over complex
terrain has been presented. Severe convective cells are effectively
identified in their early stages using a new, operational, convective
intensity nowcasting system. The best available information in
real time collected from satellites, ground-based weather radars,
NWP, climatological data and digital terrain data is assimilated
into the model. Cloud characteristics like cloud-top cooling
rate, cloud depth and cloud glaciation and instability indices,
precipitation intensities, climatological lightning information
and terrain slopes are used as predictors. All this information
is integrated using a conceptual model and a Hamilton’s equation
based formulation. The goal is to assess environmental conditions
favouring growing storms and to model the evolution of selected
storm intensity parameters, namely the vertically integrated liquid
content (VIL) and the cloud-top temperature (CTT). The system
was specially designed and tuned for the Alpine region and
several techniques are applied in order to tackle the challenge
of forecasting convective processes in complex terrain areas.
In its current version, the algorithm relies on eight modules
that integrate thunderstorm attributes with parameters of the
surrounding convective environment. The system is able to
handle the lack of one or more input datasets automatically
and it provides the user, in real time, with information about the
quality of the nowcast.
Using TRT as an independent reference, a preliminary cross-
verification has been carried out on randomly selected cases for
lead times ranging from 5–60 min. The quantitative analysis of
thunderstorm attribute nowcasts and the related skill scores show
promising results. COALITION nowcasts can be considered a
reliable tool for operational forecasting. Skill scores demonstrate
that the algorithm can help with early detection of convective
cells, which are likely to increase in intensity and become severe,
rather than simply highlight large areas prone to convection.
For this reason, the concept and methodology are suitable for
short-term nowcasts; this adds value for operational forecasts of
severe thunderstorms and aids in the severe weather warning
decision-making process.
The underlying idea of COALITION is to explain energy losses
and gains within cells in the context of energy exchanges with the
surrounding environment, according to an assumed total energy
conservation rule. This interaction is described as a dynamical
process. The fact that module results become much less reliable
after 30–40 min (values diverge considerably) indicates that
this methodology cannot be used for nowcasting purposes
with lead times greater than 30 min. The rapid decrease in
forecast skill has to be related to the nonlinearity and the chaotic
nature of convective processes. According to Lorenz (1966), the
predictability of an atmospheric phenomenon depends strongly
on its lifetime. In particular, the lifetime of thunderstorm cells
that are not organized in large-scale structures is very short and
therefore the predictability decreases rapidly during the first
30 min (Wilson et al., 1998; Wieringa and Holleman, 2006).
The system has been developed and tuned in particular for
developing convective cells in their early stages. COALITION is
an easily configurable system and it allows multiple extensions. Its
modularity allows users to implement and test new input products
in an intuitive manner. One possible way to improve the system’s
useful warning lead time would be through the implementation
of other newly developed algorithms. In the same manner as
done with RDT, other products providing information about the
locations of convective cells in their early stages (e.g. Puca et
al., 2005; Zinner et al., 2008; Mecikalski et al., 2010) or clear-air
pre-convective features (e.g. Petersen et al., 2010) should be tested.
COALITION has been validated on 80 cases during the
summer of 2012. Independent of the weather pattern, 40 weak
and 40 severe thunderstorms were selected. The skill scores
presented in this article demonstrate that COALITION nowcasts
can help the forecaster to anticipate the issuing of severe
thunderstorm warnings by about 20 min. The results of this article
demonstrate that multisensor approaches can provide added value
to convective intensity forecasting. The implementation of the
system over new regions requires the availability of primary data
sources. Furthermore, a regional tuning of the system is needed.
Thanks to the modularity of the algorithm, the user can decide
which modules of the COALITION algorithm to use. The skill
scores presented in this article refer to a system tuned for the
Alpine region and with all available data sources.
Efforts are ongoing to incorporate additional data types
(particularly other environmental data) in order to improve
the algorithm’s skill to forecast cell regeneration and decaying
processes better. Real-time lightning information (number of
cloud-to-cloud and cloud-to-ground strokes) and low-level
moisture convergence, a crucial condition for long-lasting
thunderstorms as well as for re-intensification processes (Doswell
et al., 1996), are good candidates for future extensions of the
algorithm. Furthermore, the employment of monthly and hourly
lightning climatology instead of using a single global climatology
should be considered. This predictor will help to introduce
seasonal and diurnal tuning in the algorithm and therefore
improve the skill of the forecast. In general, the addition of
meaningful predictors should help to increase the overall skill of
the system and increase its forecast lead time.
Concerning decaying processes, there are two possibilities
to expand the algorithm. The first includes the formulation
of the basic Hamiltonian equation in a statistical form, where
the uncertainties are included in the algorithm. The second
includes the possibility of modifying the basic equation by
including a dissipation term (presented in section 3). During
the development stage of a cell, this term is considered to be very
small compared with the other terms and is therefore neglected.
This assumption is confirmed by the fact that the total energy
of the object–environment system remains constant over time.
However, during the mature stage of a cell, the dissipation term
becomes more and more important since the total energy no
longer seems to be conserved.
Acknowledgements
The authors thank I. Giunta (ThurGIS, Office for Geoinforma-
tion, Switzerland) for contributing to the development of the
concepts, algorithms and numerics of the COALITION project,
as well as providing support in the review of the involved scien-
tific documentation. COALITION was developed by MeteoSwiss
with major funding from the EUMETSAT Research Fellow-
ship Program. We kindly thank EUMETSAT for supporting the
project and providing satellite data, the entire staff of MeteoSwiss
Locarno-Monti for their input and help, J. Mecikalski (Univer-
sity of Alabama) for the SATCAST algorithm and M. Koenig
(EUMETSAT) for scientific support.
References
Auton´
es F. 2012. ‘Algorithm theoretical basis document for ‘rapid devel-
opment thunder storms’’, Nowcasting Satellite Application Facility
(NWC-SAF) Report Issue 2 Rev. 3, Meteo France: Toulouse, France.
http://www.nwcsaf.org/indexScientificDocumentation.html (accessed 9
October 2013).
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

1698 L. Nisi et al.
Barthlott C, Corsmeier C, Meißner C, Braun F, Kottmeier C. 2005. The
influence of mesoscale circulation systems on triggering convective cells
over complex terrain. Atmos. Res. 81: 150– 175.
Bedka K, Brunner J, Dworak R, Feltz W, Otkin J, Greenwald T. 2010.
Objective satellite-based detection of overshooting tops using infrared
window channel brightness temperature gradients. J. Appl. Meteorol. 49:
181– 202.
Boudevillain B, Herv´
e A. 2003. Assessment of vertically integrated liquid
(VIL) water content radar measurement. J. Atmos. Oceanic Technol. 20:
807– 819.
Bryan GH, Wyngaard JC, Fritsch JM. 2003. Resolution requirements for
the simulation of deep moist convection. Mon. Weather Rev. 131:
2394– 2416.
Collier CG, Lilley RBE. 1994. Forecasting thunderstorm initiation in north-west
Europe using thermodynamic indices, satellite and radar data. Meteorol.
Appl. 1: 74– 84.
Craig GC, Keil C, Leuenberger D. 2012. Constraints on the impact of radar
rainfall data assimilation on forecasts of cumulus convection. Q. J. R.
Meteorol. Soc. 138: 340– 352.
Davolio S, Buzzi A, Malguzzi P. 2009. Orographic triggering of long lived
convection in three dimensions. Meteorol. Atmos. Phys. 103: 35– 44.
De Coning E, Koenig M, Olivier J. 2011. The combined instability index: A
new very-short range convection forecasting technique for southern Africa.
Meteorol. Appl. 18: 421– 439.
Dixon M, Wiener G. 1993. TITAN: Thunderstorm identification, tracking,
analysis, and nowcasting – a radar-based methodology. J. Atmos. Oceanic
Technol. 10: 785– 797.
Doswell CA III. 2001. Severe convective storms – An overview. Severe Convective
Storms, Meteorological Monographs 28: 1– 26. American Meteorological
Society: Boston, MA.
Doswell CA III, Brooks HE, Maddox RA. 1996. Flash flood forecasting: An
ingredients-based methodology. Weather and Forecasting 11: 560–581.
Emanuel KA. 1994. Atmospheric Convection. Oxford University Press: New
York, NY.
Foresti L, Pozdnoukhov A. 2011. Exploration of Alpine orographic
precipitation patterns with radar image processing and clustering
techniques. Meteorol. Appl. 19: 407– 419.
Fujita T. 1968. Present status of cloud velocity computations from ATS-1 and
ATS-3. COSPAR Space Res. 9: 557– 570.
Goodman SJ, Gurka J, DeMaria M, Schmit TJ, Mostek A, Jedlovec G, Siewert
C, Feltz W, Gerth J, Brummer R, Miller S, Reed B, Reynolds RR. 2012.
The GOES-R proving ground. Accelerating user readiness for the next-
generation geostationary environmental satellite system. Am. Meteorol. Soc.
7: 1029– 1040.
Greene D, Robert R, Clark A. 1972. Vertically integrated liquid water –a new
analysis tool. Mon. Weather Rev. 100: 548– 552.
Handwerker J. 2002. Cell tracking with TRACE3D –a new algorithm. Atmos.
Res. 61: 15– 34.
Hering AM, Morel C, Galli G, S´
en´
esi S, Ambrosetti P, Boscacci M. 2004.
Nowcasting thunderstorms in the Alpine region using a radar based adaptive
thresholding scheme. In Proceedings of 3rd European Conference Radar in
Meteorology and Hydrology (ERAD), 6–10 September 2004, Visby, Sweden.
1– 6. Copernicus: Goettingen, Germany.
Hering AM, Germann U, Boscacci M, S´
en´
esi S. 2008. Operational nowcasting
of thunderstorms in the Alps during MAP D-PHASE. In Proceedings of
5th European Conference on Radar in Meteorology and Hydrology (ERAD),
30 June– 4 July 2008, Helsinki, Finland. 1 – 5. Copernicus: Goettingen,
Germany.
Hilker N, Badoux A, Hegg C. 2010. Unwetterschaeden in der Schweiz im Jahre
2009. Wasser Energ. Luft 102: 1– 6 (in German).
Huntrieser H, Schiesser HH, Schmidt W, Waldvogel A. 1996. Comparison
of traditional and newly developed thunderstorm indices for Switzerland.
Weather and Forecasting 12: 108– 125.
Johnson JT, Mackeen PL, Witt A, Mitchell ED, Stumpf GJ, Eilts MD, Thomas
KW. 1998. The storm cell identification and tracking algorithm: An
enhanced WSR-88D algorithm. Weather and Forecasting 13: 263–275.
Joss J, Schaedler B, Galli G, Cavalli R, Boscacci M, Held E, Della Bruna G,
Kappenberger G, Nespor V, Spiess R. 1998. Operational Use of Radar for
Precipitation Measurements in Switzerland. vdf Hochschulverlag AG ETH
Zuerich: Zuerich, Switzerland.
Kain JS, Weiss SJ, Bright DR, Baldwin ME, Levit JJ, Carbin GW, Schwartz
CS, Weisman ML, Droegemeier KK, Weber DB, Thomas KW. 2008.
Some practical considerations regarding horizontal resolution in the
first generation of operational convection-allowing NWP. Weather and
Forecasting 23: 931– 952.
Kober K, Tafferner A. 2009. Tracking and nowcasting of convective cells using
remote sensing data from radar and satellite. Meteorol. Z. 1: 75–84.
Koenig M, De Coning E. 2009. The MSG global instability indices product and
its use as a nowcasting tool. Weather and Forecasting 24: 272–285.
Kottmeier C, Kalthoff N, Bathlott C, Corsmeier U, Van Baelen J, Behrendt
A, Behrendt R, Blyth R, Coulter R, Crewell S, Di Girolamo P, Dorninger
M, Flamant C, Foken T, Hagen M, Hauck C, Hoeller H, Konow H, Kunz
M, Mahlke H, Mobbs S, Richard E, Steinacker R, Weckwerth T, Wieser A,
Wulfmeyer V. 2008. Mechanisms initiating deep convection over complex
terrain during COPS. Meteorol. Z. 6: 931– 948.
Lang P. 2001. Cell tracking and warning indicators derived from operational
radar products. In Proceedings of 30th International Conference on Radar
Meteorology, 19– 24 July 2001, Munich, Germany: 245 – 247. American
Meteorological Society: Boston, MA.
Ligda MG. 1953. ‘The horizontal motion of small precipitation areas as observed
by radar’, Technical Report 21. Massachusetts Institute of Technology,
Department of Meteorology: Cambridge, MA.
Lorenz EN. 1966. Atmospheric Predictability, Advances in Numerical Weather
Prediction. Travellers Research Center, Inc.: Hartford, CT.
Mak M. 2001. Nonhydrostatic barotropic instability: Applicability to
nonsupercell tornadogenesis. J. Atmos. Sci. 58: 1965– 1977.
Mandapaka P, Germann U, Panziera L, Hering A. 2011. Can Lagrangian
extrapolation of radar fields be used for precipitation nowcasting over
complex Alpine orography? Weather and Forecasting 27: 28– 49.
Martinez MA, Velazquez M, Manso M, Mas I. 2007. Application of LPW and
SAFNWC/MSG satellite products in pre-convective environments. Atmos.
Res. 83: 366– 379.
Mecikalski JR, Bedka KM. 2006. Forecasting convective initiation by
monitoring the evolution of moving cumulus in daytime GOES imagery.
Mon. Weather Rev. 134: 49– 78.
Mecikalski JR, Bedka KM, Paech SJ, Litten LA. 2008. A statistical evaluation
of GOES cloud-top properties for nowcasting convective initiation. Mon.
Weather Rev. 136: 4899– 4914.
Mecikalski JR, Mackenzie WM, Koenig M, Muller S. 2010. Use of Meteosat
Second Generation infrared data in 0– 1 hour convective initiation
nowcasting. Part 1. Infrared fields. J. Appl. Meteorol. 49: 521–534.
Mecikalski JR, Bedka KM, Koenig M. 2012. ‘Best practice document’.
EUMETSAT, ESSL Convection Working Group. http://essl.org/cwg/?p=178
(accessed 9 October 2013).
Mecikalski JR, Li X, Carey LD, McCaul EW, Coleman TA. 2013. Regional
comparison of GOES cloud-top properties and radar characteristics in
advance of first-flash lightning initiation. Mon. Weather Rev. 141:55–74.
Mecklenburg S, Joss J, Schmid W. 2000. Improving the nowcasting of
precipitation in an Alpine region with an enhanced radar echo tracking
algorithm. J. Hydrol. 239: 46– 68.
Miles J, Salmon R. 1985. Weakly dispersive nonlinear gravity waves. J. Fluid
Mech. 157: 519– 531.
Mueller C, Saxen T, Roberts R, Wilson J, Betancourt T, Dettling S, Oien N, Yee
J. 2003. NCAR auto-nowcast system. Weather and Forecasting 18: 545–561.
Petersen RA, Aune R, Rink T. 2010. Objective short-range forecasts of
the preconvective environment using SEVIRI data. Proceedings the 2010
EUMETSAT Meteorological Satellite Conference, 20– 24 September 2010,
Cordoba, Spain. 1– 7. EUMETSAT: Darmstadt, Germany.
Pielke RA, Segal M. 1986. Mesoscale Meteorology and Forecasting, American
Meteorological Society: Boston, MA.
Pierce CE, Hardaker PJ, Collier CG, Haggett CM. 2000. GANDOLF: A system
for generating automated nowcasts of convective precipitation. Meteorol.
Appl. 7: 341– 360.
Puca S, Biron D, De Leonimbus L, Melfi D, Rosci P, Zauli F. 2005. A neural
network algorithm for the nowcasting of severe convective systems. In
Proceeding of CIMSA 2005-IEEE International Conference on Computing
Intelligence for Measurement System Applications, 20– 22 July 2005, Giardini
Naxos, Greece. 81– 84. IEEE: Piscataway, NJ.
Roberts RD, Burgess D, Meister M. 2006. Developing tools for now-
casting storm severity. Weather and Forecasting 21: 540– 558, doi:
http://dx.doi.org/10.1175/WAF930.1.
Roberts RD, Rutledge S. 2003. Nowcasting storm initiation and growth using
GOES-8 and WSR-88D data. Weather and Forecasting 18: 562–584.
Rosenfeld D, Woodley WL. 2000. Deep convective clouds with sustained
supercooled liquid water down to −37.5 ◦C. Nature 405: 440– 442.
Rosenfeld D, Woodley WL, Lerner A, Kelman G, Lindsey DT. 2008. Satellite
detection of severe convective storms by their retrieved vertical profiles of
cloud particle effective radius and thermodynamic phase. J. Geophys. Res.
113: D04208, doi: 10.1029/2007JD008600.
Rotach MW, Ambrosetti P, Ament F, Appenzeller C, Arpagaus M, Bauer
HS, Behrendt A, Bouttier F, Buzzi A, Corazza M, Davolio S, Denhard M,
Dorninger M, Fontannaz L, Frick J, Fundel F, Germann U, Gorgas T,
Hegg C, Hering A, Keil C, Liniger MA, Marsigli C, McTaggart-Cowan R,
Montaini A, Mylne K, Ranzi R, Richard E, Rossa A, Santos-Munoz D, Schaer
C, Seity Y, Staudinger M, Stoll M, Volkert H, Walser A, Wang Y, Werhahn
J, Wulfmeier W, Zappa M. 2009. MAP D-PHASE: Real-time demonstration
of weather forecast quality in the Alpine region. Bull. Am. Meteorol. Soc. 90:
1321– 1336.
Salmon R. 1998. Lecture on Geophysical Fluid Dynamics. Oxford University
Press: Oxford, UK.
Schmetz J, Pili P, Tjemkes S, Just D, Kerkmann J, Rota S, Ratier A. 2002. An
introduction to Meteosat Second Generation (MSG). Bull. Am. Meteorol.
Soc. 83: 977– 992.
Schultz CJ, Carey LD, Petersen WA. 2009. Preliminary development and
evaluation of lightning jump algorithms for the real-time detection of
severe weather. J. Appl. Meteorol. 48: 2543–2563.
Setv´
ak M, Rabin RM, Doswell CA, Levizzani V. 2003. Satellite observations
of convective storm top features in the 1.6 and 3.7/3.9 m spectral bands.
Atmos. Res. 67– 68C(:): 589 –605.
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)

Nowcasting Severe Convection in the Alpine region 1699
Steinacker R, Dorninger M, Wolfelmaier F, Krennert T. 2000. Automatic
tracking of convective cells and cell complexes from lightning and radar
data. Meteorol. Atmos. Phys. 72: 101–109.
Wieringa J, Holleman I. 2006. If cannons cannot fight hail, what else? Meteorol.
Z. 15: 659– 669.
Weisman ML, Skamarock WC, Klemp JB. 1997. The resolution dependence of
explicitly modeled convective systems. Mon. Weather Rev. 125: 527– 548.
Wilks DS. 1995. Statistical Methods in the Atmospheric Sciences: An Introduction.
Academic Press: Amsterdam.
Wilson JW, Crook NA, Mueller CK, Sun J, Dixon M. 1998. Nowcasting
thunderstorms: A status report. Bull. Am. Meteorol. Soc. 79: 2079– 2099.
Wulfmeyer V, Behrendt A, Bauer H-S, Kottmeier C, Corsmeier U, Blyth A,
Craig GC, Schumann U, Hagen M, Crewell S, Di Girolamo P, Flamant
C, Miller M, Montani A, Mobbs SD, Richard E, Rotach MW, Arpagaus
M, Russchenberg H, Schlussel P, Koenig M, Gaertner V, Steinacker R,
Dorninger M, Turner DD, Weckwerth T, Hense A, Simmer C. 2008. The
Convective and Orographically induced Precipitation Study: A research
and development project of the World Weather Research Program for
improving quantitative precipitation forecasting in low-mountain regions.
Bull. Am. Meteorol. Soc. 89: 1477– 1486.
Zinner T, Mannstein H, Tafferner A. 2008. Cb-TRAM: Tracking and
monitoring severe convection from onset over rapid development to mature
phase using multi-channel Meteosat-8 SEVIRI data. Meteorol. Atmos. Phys.
101: 191– 210.
c
2013 Royal Meteorological Society Q. J. R. Meteorol. Soc. 140: 1684– 1699 (2014)