ArticlePDF Available

Abstract and Figures

This paper presents a thirteen-year hail climatology for Switzerland based on volumetric Radar reflectivity. Two radar-based hail detection products that are used operationally at MeteoSwiss, namely the Probability of Hail (POH) and the Maximum Expected Severe Hail Size (MESHS), have been reprocessed for the extended convective season (April-September) between 2002 and 2014. The result of these two products is a comprehensive hail Distribution map, which highlights regional and local-scale hail characteristics. The map of the annual number of hail days shows a high spatial variability and several maxima over the foothills north and south of the Alps as wells as over the Jura mountains. Directly over the Alps hail frequency exhibits a minimum. Annual hail anomalies show a pronounced variability, which suggests that hail occurrence is strongly controlled by large-scale weather patterns. Furthermore, hail probability exhibits a strong seasonal and diurnal cycle with a maximum in July in the late afternoon. The hail peak over the northern prealpine region occurs approximately two hours earlier compared to the south. A possible explanation is the trigger mechanism between the cold pool initiated by early convective cells over the Jura mountains and the development of cells on the northern slope of the Alps. Since radar-based hail signals are only indirect measurements, statistical verification of the hail detection algorithms is crucial. Damage reports from an automobile insurance company are used as independent dataset. Verification results confirm that radar-based hail algorithms provide valuable information on hail probability. Finally the synoptic-scale hail-driving weather conditions are investigated using a weather type classification based on upper-air flow direction and mean pressure from a NWP model. The results show that six out of nine main synoptic-scale patterns favour the development of hailstorms in Switzerland.
Content may be subject to copyright.
Accepted Article
Spatial and temporal distribution of hailstorm in the Alpine
region: a long-term, high resolution, radar-based analysis
L. Nisi1,2,3,4, O. Martius1,2,3, A. Hering4, M. Kunz5, U. Germann4
1 University of Bern, Oeschger Centre for Climate Change Research, Bern, Switzerland
2University of Bern, Institute of Geography, Bern, Switzerland
3University of Bern, Mobiliar Lab for Natural Risks, Bern, Switzerland
4 Federal Office of Climatology and Meteorology MeteoSwiss, Locarno-Monti, Switzerland
5 Institute of Meteorology and Climate Research (IMK), Karlsruhe Institute of Technology
(KIT), Karlsruhe, Germany
Abstract
This paper presents a thirteen-year hail climatology for Switzerland based on volumetric
radar reflectivity. Two radar-based hail detection products that are used operationally at
MeteoSwiss, namely the Probability of Hail (POH) and the Maximum Expected Severe Hail
Size (MESHS), have been reprocessed for the extended convective season (April-September)
between 2002 and 2014. The result of these two products is a comprehensive hail distribution
map, which highlights regional and local-scale hail characteristics. The map of the annual
number of hail days shows a high spatial variability and several maxima over the foothills
north and south of the Alps as wells as over the Jura mountains. Directly over the Alps hail
frequency exhibits a minimum. Annual hail anomalies show a pronounced variability, which
suggests that hail occurrence is strongly controlled by large-scale weather patterns.
Furthermore, hail probability exhibits a strong seasonal and diurnal cycle with a maximum in
July in the late afternoon. The hail peak over the northern prealpine region occurs
approximately two hours earlier compared to the south. A possible explanation is the trigger
mechanism between the cold pool initiated by early convective cells over the Jura mountains
and the development of cells on the northern slope of the Alps. Since radar-based hail
signals are only indirect measurements, statistical verification of the hail detection
algorithms is crucial. Damage reports from an automobile insurance company are used as
independent dataset. Verification results confirm that radar-based hail algorithms provide
valuable information on hail probability. Finally the synoptic-scale hail-driving weather
conditions are investigated using a weather type classification based on upper-air flow
This article is protected by copyright. All rights reserved.
This article has been accepted for publication and undergone full peer review but has not
been through the copyediting, typesetting, pagination and proofreading process, which
may lead to differences between this version and the Version of Record. Please cite this
article as doi: 10.1002/ qj.2771
Accepted Article
direction and mean pressure from a NWP model. The results show that six out of nine main
synoptic-scale patterns favour the development of hailstorms in Switzerland.
Keywords: Hail, Hailstorms, Convection, Weather radar, Switzerland, Diurnal cycle
1 Introduction
Hail has been a subject of scientific interest for many decades (e.g. Plumandon, 1901;
Changnon, 1978) because of the severe damage it causes to agriculture, buildings and cars.
For a specific location, hail is a low probability high impact weather event (e.g. Delobbe and
Holleman, 2006). Hail is characterized by a strong local-scale variability of the occurrence
and intensity and the small extent of the affected areas referred to as hailstreaks (e.g.
Weisman et al., 1997; Bryan et al., 2003; Sánchez et al. 2013). As a consequence, point
observations of hail are not representative for larger areas. Compared to other atmospheric
parameters such as temperature or precipitation, observational networks need to be at least 10
times denser to capture all hailstreaks (Wieringa and Holleman, 2006). Physical
measurements collected with dense networks are an ideal basis to investigate multi-year hail
occurrence, hail variability and hail trends. Additional information are provided by reports
from trained storm spotters or newspapers, which are archived in different databases, for
example the European Severe Weather Database (ESWD). Several hail climatologies and
analyses have been compiled worldwide during the last decades based on data from human or
automatic observations (e.g. Changnon, 1978; Changnon and Changnon, 2000, Xie et al.,
2008; Zhang et al., 2008; Tuovinen et al., 2009; Xie et al., 2010; Mezher et al., 2012) and
hailpads (e.g. Dessens and Fraile, 1994; Sánchez et al., 1996; Eccel and Ferrari, 1997; Vinet,
2001, Dessens et al., 2001; Fraile et al., 2003; Giaiotti et al., 2003; Počakal et al., 2009;
Berthet et al., 2011; Manzato, 2013; Sánchez et al. 2013).
Unfortunately, only few dense hail observation networks exist worldwide (e.g., in several
parts of France or northern Italy), and Switzerland does not have a ground-based hail
observation network. For this reason, studies on hail occurrence require indirect observations,
primarily radar data or insurance loss data.
Weather radar-based hail detection algorithms can be useful for investigating hail frequency
over larger areas (e.g. Basara et al., 2007; Cintineo et al, 2012) and in regions where long
observation time series, like in Switzerland, do not exist. Hail detection by weather radar has
a long history. In the fifties, first studies were conducted to investigate the presence of hail in
This article is protected by copyright. All rights reserved.
Accepted Article
thunderstorms by means of weather radars (e.g. Donaldson, 1961). A series of studies
analysed the relation between the presence of hail and different radar parameters such as:
vertical reflectivity profiles (Wilk, 1961), height of strong echos (e.g. Douglas 1963),
maximum reflectivity (e.g. Rinehart et al., 1968), height of the 45-dBZ contours in storm
profiles (Mather et al., 1976), height difference between the 45 dBZ reflectivity and the
melting layer (Waldvogel et al., 1979), or vertical integrated liquid water (e.g. Amburn and
Wolf, 1997). Several radar-based methods for estimating the probability of hail (e.g.
Waldvogel et al., 1979), the hail kinetic energy (e.g. Sánchez et al., 2013) and the maximum
expected hailstone size (e.g. Treloar, 1996; Edwards and Thompson, 1998; Witt et al., 1998)
were developed and implemented operationally by weather offices. Summaries of the most
commonly used hail detection techniques using single-polarization C-Band radars are
presented in Holleman (2001), Sánchez et al. (2013) and Kunz and Kugel (2015).
Insurance loss data usually have a high spatial coverage and are available over a long
period, but are affected by several sources of uncertainty (e.g. Willemse, 1995; McMaster,
1999; Changnon et al., 2001; Webb et al., 2001a; Schuster et al., 2005; Kunz and Puskeiler,
2010). They are strongly dependent on non-meteorological characteristics like population
density, object vulnerability and claim handling (Dessens et al., 2009; Mohr and Kunz, 2013).
According to Vinet (2001), damage observations are the results of the combination of “object
vulnerability” (property) and “agent risk” (hail). Nevertheless, these data sets provide one of
the few possibilities for cross-validating radar-based hail observations in areas where
ground-based hail observation networks are not available. For example, Kunz and Kugel
(2015) and Skripniková and Řezáčová (2014) used building loss information for validating
and adjusting radar-based hail detection algorithms.
Switzerland is regularly affected by severe hailstorms causing substantial damage. Despite the
high hail risk exposure, only a few climatological hail investigations based on insurance and
radar data exist (e.g. Houze et al., 1993; Willemse, 1995; Stucki and Egli, 2007). Substantial
improvements of the Swiss radar network offer the unique opportunity to extend and broaden
the existing knowledge. Long-term information on hail occurrence is not only relevant for
agriculture and the insurance industry, but also serves as basis for the advancement of hail
forecasting in Switzerland. Building on climatological knowledge of the temporal and spatial
distribution of hailstorms allows assessing the hail hazard and hail risk for different regions
(Wieringa and Holleman, 2006).
This article is protected by copyright. All rights reserved.
Accepted Article
Forecasting severe storms is a challenging task due the chaotic nature of convective
processes. In particular over complex orography, high spatial and temporal gradients make the
task of observing and nowcasting thunderstorms challenging (e.g. Mecklenburg et al., 2000;
Hering et al., 2004; Rotach et al., 2009; Mandapaka et al., 2012; Nisi et al., 2014). However,
in mountainous areas the orographic forcing is a source of repeatability of precipitation and
convective cells (Foresti et al., 2011), and knowledge about hail occurrence in the past is very
valuable for many applications.
The aim of this study is to investigate the distribution of hail over Switzerland and adjacent
regions, which are characterized by complex terrain and the absence of hail detection
networks (e.g. hailpads). Hail occurrence is assessed by using radar-based observations (full
resolution) over a 13-year investigation period. Volumetric radar data from the Swiss radar
network are combined with information from the regional Numerical Weather Prediction
(NWP) model COnsortium for Small-scale MOdelling (COSMO-CH). Two operational
single-polarization hail detection algorithms are reprocessed for the extended convective
season (April-September) between 2002 and 2014. The results are analysed to address the
following questions: (i) What is the annual, monthly and hourly distribution of hail
occurrence over the prealpine and alpine region? (ii) Is it possible to identify and characterize
the synoptic-scale hail favouring weather conditions? (iii) Does the application of radar-based
hail detection algorithms over a region with complex terrain yield reliable results?
2 Domain and data-sets
2.1. Investigation area
The region under investigation has a complex orography, characterized by deep valleys
with lowest altitude around 100 m.a.s.l. and mountains with peaks above 4000 m.a.s.l. ,
see Figure 1. We considered in the analysis both the whole domain and smaller sub-
regions, see orange areas in Figure 1. The six sub-regions exhibit different terrain as well as
climatological conditions. The Jura (1) is a long south-west to north-east oriented mountain
ridge, the highest peak of which is the Crêt de la Neige at 1720 m.a.s.l.. The northern slopes
of the Alps (2) encompass a mixture of flat and hilly terrain (Swiss Plateau) and mountainous
regions (Prealps) with an average altitude of 580 m.a.s.l. and peaks reaching 1500 m.a.s.l. .
The Alps (3) are characterized by deep valleys and high peaks with altitudes ranging between
400 and more than 4810 m.a.s.l. . The southern prealpine area (4) is dominated by hilly
This article is protected by copyright. All rights reserved.
Accepted Article
terrain.
Baden-
W
Forest).
Figure
1
Prealps
;
the loca
Ital
y
; A
2.2. R
a
In this
used, t
h
from t
study i
s
1998)
h
two ad
d
2015).
T
generat
(2015).
minute
s
To cal
c
(MES
H
Finally the
W
ürttembe
r
1
: Investi
g
a
t
;
3: Alps; 4
tion of the
t
u: Austria.
a
dar data
study two
h
at is Albi
s
h
e third ra
d
s
160 km.
S
h
as been c
o
d
itional ra
d
T
his is the
ion radar
d
The cali
b
s
, whereas
a
c
ulate the
P
H
S), the 4
5
Po valley
(
r
g (Germa
n
t
ion area a
n
: Southern
P
t
wo radars
u
C-band D
o
s
and Mon
t
d
ar on La
D
S
tarting fro
o
mpletely r
d
ar sites to
4
th
genera
t
d
ata proce
s
b
ration is
p
a
n externa
l
P
robability
5
dBz and
5
(
Italy; 5) is
n
y; 6) enc
o
n
d the six s
u
P
realps; 5:
u
sed in this
o
ppler rad
a
t
e Lema, s
e
D
ôle was n
o
m June 20
enewed an
d
improve c
o
t
ion of we
a
s
sing are d
p
erformed
l
calibratio
n
Of Hail (P
O
5
0dBz ech
o
characteri
z
o
mpasses a
u
b-re
g
ions a
Po valley;
6
study. CH:
a
rs from th
e
e Figure 1
o
t include
d
11, the C-
b
d
extende
d
o
verage in
t
a
ther radar
escribed i
n
with an i
n
n
using the
O
H) and t
h
o
top heig
h
z
ed by flat
t
mixture o
f
nal
y
sed in
t
6
: Baden-
W
Switzerlan
d
e MeteoS
w
. For reas
o
d
; rather th
e
b
and Dopp
l
d
adding d
u
t
he inner-a
s in Switz
e
n
Joss et a
l
n
ternal noi
sun is per
f
h
e Maximu
h
t product
t
errain, and
f
flat and
h
t
his paper (
1
W
ürttember
g
d
; D: Ger
m
w
iss radar
n
o
ns of data
e
maximu
m
l
er radar n
e
u
al-polarisa
t
a
lpine regi
o
e
rland. De
t
l
. (1998) a
se source
f
ormed on
a
m Expecte
s
are used
the federa
l
h
illy terrai
n
1
: Jura; 2:
N
g
). Red dots
m
an
y
; Fr: F
r
n
etwork h
a
homogene
i
m
range use
e
twork (Jo
s
t
ion capab
i
o
ns (Germa
n
t
ails on 3
r
d
nd Germa
n
generator
a
daily basi
d Severe
H
(ET45 a
n
l
state of
n
(Black
N
orthern
indicate
r
ance; It:
a
ve been
i
ty, data
d in this
s
s et al.,
i
lity and
n
n et al,
d
and 4
th
n
n et al.
every 5
s.
H
ail Size
n
d ET50
This article is protected by copyright. All rights reserved.
Accepted Article
hereafter). ET45 and ET50 represent the highest altitude at which a radar reflectivity of at
least 45 dBZ and 50 dBZ, respectively, can be detected (Donaldson, 1961). The parameters
ET45, ET50, POH and MESHS are 2-dimensional, gridded Cartesian products. For more
details see Table 1. Since many years echo top height products are used for the diagnosis
of severe convection (e.g. Held, 1978; Waldvogel et al, 1979; Witt et al., 1998) and for
radar-based thunderstorm nowcasting systems (e.g. Dixon and Wiener, 1993; Johnson et al.,
1998; Hering et al., 2008). Quantitative radar estimation of precipitation over complex terrain
has to cope with several major challenges (see Section 3.5).
Table 1: Specifics of the 3rd and 4th Swiss radar generation and related product resolutions.
Before
June 2011
Inbetween
After
June 2012
Radar generation 3rd
Transition
phase
4th
Radar capabilities Doppler Doppler + polarimetric
Radar resolution
(polar data) 1km x 1° 500m x 1°
Spatial resolution of
Cartesian products
2 x 2 km²
1 x 1 km²
Scan Strategy
20 elevations
(-0.3°-+40°) repeated every 5 min
20 elevations
(-0.2°-+40°) repeated every 5 min
2.3. Freezing level height from the regional NWP model COSMO-CH
The radar-based hail detection algorithms POH and MESHS require information on the
freezing level height (hereafter H0). This information is extracted from COSMO-CH analysis
(http://cosmo-model.org/). COSMO-CH is a non-hydrostatic, regional, high-resolution
numerical weather prediction model operated by MeteoSwiss. The horizontal resolutions are
6.6 x 6.6 km² (COSMO-7) and 2.2 x 2.2 km² (COSMO-2), respectively, and the temporal
resolution is 1 hour.
H0 represents the model grid-cell, where the 3D temperature field is equal to 0 °C. During the
investigation periods, the model set-up changed over time. Between March 2002 and March
2008 only one model resolution was available with a 7 x 7 km² mesh-grid, whereas after
March 2008 both COMOS-7 and COSMO-2 were available. Another minor change concerns
This article is protected by copyright. All rights reserved.
Accepted Article
the algorithm for the diagnosis of the freezing level. The bottom-up algorithm was substituted
with a top-down approach in May 2011. This modification had a significant effect during
stable situations with a temperature inversion only, but no effect in case of convective events.
The sensitivity of POH (or MESHS, respectively) to errors in the definition of H0 decreases
with increasing difference between ET45 (or ET50, respectively) and H0. For example, an H0
error of ±1000m produces a variation in POH of ±20-40% if POH<60% and ±10-20% if
POH60%. For MESHS the variation is more linear and correspond to ±1 cm for the all hail
sizes. An H0 error of few hundred of meters produces a small variation only for both POH
and MESHS. A statistical verification of upper air temperature show a bias smaller than 1 °K.
Consequently the bias of H0 is small as well (<100m). However, local precipitation dynamics
inside the model have to be considered. Latent energy from melting and evaporation
processes affects H0 especially in case of intensive convective rain. These physical process
are parametrized in the model. Such small-scale effects on H0 were not considered in the
calibration of hail products when proximity sounding were used (Waldvogel et al., 1979;
Treloar, 1998; Joe et al., 2004). This may have a consequence causing overestimation of hail
(M. Stoll, MeteoSwiss, personal communication, 22.11.2015). However, this process causes a
decrease of H0 of several hundred meters only in rare cases. Therefore, we neglected this
effect in this study, given the low sensibility of POH and MESHS to changes in H0.
2.4. Weather Type classification
Recently a new automatic weather type classification method (hereafter WTC) has been
introduced at MeteoSwiss (Weusthoff, 2011). On a daily basis, 10 classifications based on
two different methods (GrossWetterTypes, GWT, and Cluster Analysis of Principal
Components, CAP) are calculated. WTCs have been calculated using reanalysis data of
ERA40 (Uppala et al., 2005) and ERA-Interim (Dee et al. 2011) back until 1957 and stored in
the MeteoSwiss database. In this study we used a 10 members classification with predefined
weather types based on geopotential, mean wind speed and direction on 500 hPa (see Table
2).
This article is protected by copyright. All rights reserved.
Accepted Article
Table 2: The ten members of the weather types classification used in this study (adapted from
Weusthoff, 2011).
Type Description
WT1 Westerly
WT2 South-westerly
WT3 North-westerly
WT4 Northerly
WT5 North-easterly
WT6 Easterly
WT7 South-easterly
WT8 Southerly
WT9 Low Pressure
WT10 High Pressure
2.5. Insurance loss reports used for verification
The results of the two applied hail detection methodologies were validated using hail loss
data (car) from a major insurance company in Switzerland. In this study car losses
collected over a 10-year period (2003-2012) were used. Daily damage locations and
frequency are provided for each municipality in Switzerland (4-digit postal code zones,
3195 in total). The average area per postal code is 12.9 km2. Note however, that in
mountainous and rural regions postal codes zones may be much larger. On the other hand,
cities encompass several postal codes with smaller areas.
The insurance datasets were corrected manually in order to eliminate obvious errors due to
the recording procedure (e.g. elimination of single claims on days and over regions
without observed rain, see Morel (2014) for more details).
The relation between hail kinetic energy and automobile losses in Switzerland was studied
by Hohl et al. (2002). They show that losses grow exponentially with hail size. However,
it is difficult to determine a size threshold for damaging hail. Indeed, several factors like
car structure, hailstone geometry, hailstone density, presence of rain, impact angle,
horizontal velocity and other meteorological and non- meteorological factors can affect
the extension and the amount of damage. Hohl et al. (2002) found that in general
hailstones with diameters larger than 2 cm produce substantial damage to automobiles.
This article is protected by copyright. All rights reserved.
Accepted Article
3 Methods
3.1. Probability of Hail (POH)
Waldvogel et al. (1979) investigated and verified radar-based hail detection criteria over
Switzerland. They presented a methodology, which considers the vertical distance between
ET45 and H0 as an indication for the presence of hailstones on the ground:
z = ET45 H0 1.4 km . (1)
In this formulation, the height difference z in Equation (1) is a proxy for the zone, where
hail may grow by riming in deep convective storms. The success of this algorithm was further
confirmed by other authors, which were mainly involved in hail suppression experiments (e.g.
Foote and Knight, 1979). Later Witt et al. (1998) showed that this criterion is useful for
estimating the hail probability (POH). For this purpose, they considered the distance z to be
proportional to the probability of hail on the ground. POH provides an estimate of the
presence of hail of any size at the ground, with a scale ranging from 0 (no hail; z < 1.65 km)
to 100% (hail; z > 5.5 km). Several versions of the technique proposed by Waldvogel et al.
(1979) have been tested extensively in different countries showing that it provides reasonable
results for single polarisation radar data (e.g. Kessinger et al.,1995; Holleman, 2001; Delobbe
and Holleman, 2006; Skripniková and Řezáčová, 2014). Furthermore, the algorithm has been
implemented operationally by several weather services (e.g. Witt et al., 1998; Holleman,
2001; Šálek et al., 2004; Betschart and Hering, 2012) and adjusted for different purposes (e.g.
Puskeiler, 2013; Puskeiler et al., 2015; Kunz and Kugel, 2015). The version proposed by
Foote et al. (2005), tested for a C-band radar, is implemented operationally at MeteoSwiss
since 2008 and is used in this study.
3.2 Maximum Expected Severe Hail Size (MESHS)
Radar-based approaches have also been employed for estimating the hailstone size. Treloar
(1998), for example, empirically investigated the relation between some radar and upper air
parameters. Few years later, an empirical algorithm based on the work of Treloar (1998) has
been implemented and run operationally during the Sydney 2000 Forecast Demonstration
project (Joe et al., 2004). The algorithm provides a reasonable estimation of the maximal
hailstone size on the ground using the relation between the height of ET50 and H0. It was
This article is protected by copyright. All rights reserved.
Accepted Article
successfully tested for monitoring severe hailstorms during the Olympic games in 2000
(Webb et al., 2001b). The algorithm implemented operationally at MeteoSwiss was
empirically derived from Figure 3 in Joe et al. (2004). It is used since 2009 for both warning
purposes and climatological analyses. The algorithm estimates the maximal size of hailstones
at the ground for hailstone-diameters equal or larger than 2 cm, which is the critical size for
damage to buildings or automobiles. MeteoSwiss verified the information provided by the
MESHS product using ground-truth hail reports collected from the media over a limited
period (Betschart and Hering, 2012). Because the lack of large and reliable datasets of
hailstone sizes, performing a solid verification of the algorithm is not possible at the moment.
For this reason, hail analysis presented in this manuscript are mainly based on POH.
3.3 Creating a multi-year POH and MESHS data set
A 13-year period was selected from 2002 till 2014. The period has been selected such that
the quality of input data from radar and COSMO-CH is fairly high and homogeneous, and
the number of hail cells in the domain is sufficiently large to calculate frequency statistics
at the desired space-time resolution. POH and MESHS have been reprocessed for the 13
extended convective seasons from April to September using data from COSMO-CH and
over 27 million scans from the two radars Albis and Lema, see Figure 1. Polar 3rd
generation data have been interpolated to the 4th generation polar data. The same value of
a 1000m 3rd generation bin has been assigned to two 500m bins simulating 4th generation
polar data. For the whole investigation period radars’outages were below 4%. Outages time
were mainly due to ordinary maintenance and mostly during fair weather conditions in the
absence of precipitation. Reprocessing was necessary to account for changes in the spatial
resolution of COSMO-CH and the scan program and data format of the new 4th generation
radar network. The goal of the reprocessing was to obtain a homogeneous data base
suitable for the investigation of spatial and temporal patterns in the hail distribution over
Switzerland. For more details see Figure 2.
This article is protected by copyright. All rights reserved.
Accepted Article
Figure
2
sets.
3.4 Ha
i
Daily
P
values
a
5) and
t
day is
d
3 cm o
r
verific
a
of rada
r
using t
h
hour” i
s
discuss
where
p
standar
d
3.5 Di
s
As op
p
simulta
n
domain
discuss
e
2
: Reproce
s
i
l days an
P
OH and
M
a
t every gri
d
t
he number
d
efined eith
e
r
MESHS >
a
tion with i
n
r
-
b
ased hai
h
e normali
z
s
considere
d
annual hai
l
p
i
is the ha
i
d
deviation
s
cussion
o
p
osed to a
l
n
eously ob
. However
e
d. Among
s
sin
g
strate
g
d anomal
i
M
ESHS pro
d
d
point. Fr
o
hail days
p
e
r as POH
>
4 cm (Fig
u
n
surance lo
l algorith
m
z
ed average
d
when PO
l
variability
i
l frequenc
y
(hereafter
S
o
f radar-b
a
l
l other h
a
serve hail
, some ch
a
them we
fi
gy
for crea
i
es
d
ucts were
q
o
m these fi
e
p
er months
a
>
80% (Fig
u
u
re 5). The
8
ss data (Se
c
m
s over the
hourly hai
H > 80%.
S
over Switz

y
of the i
th
S
TD).
a
sed app
r
a
il observi
n
cells dow
n
a
llenges
w
fi
nd ground
tin
g
lon
g
t
e
q
uantified
f
e
lds the nu
m
a
re determi
n
u
re 4, 7, 10
8
0% POH t
h
c
tion 4.5)
a
Alps (Sec
t
l frequenc
y
S
tandardiz
e
erland (Fig
u

,
year, μ is
t
r
oaches o
n
g systems
,
n
to the siz
w
hich can
a
clutter an
d
e
rm, homo
g
f
ro
m
the re
m
ber of hai
l
n
ed (Figur
e
and 11), o
r
h
reshold is
a
nd (ii) the
t
ion 3.5).
T
y
(Figure 9
)
e
d anomali
e
u
re 6):
t
he mean o
v
ver comp
,
a radar
o
e of few
k
a
ffect the
d
bright ba
n
g
enous PO
H
spective m
a
l
days per s
e
e
7). For ea
c
r
MESHS
>
selected
ba
evaluation
T
he diurnal
)
. For each
e
s N (Wilk
s
v
er several
lex terrai
n
o
ffers a u
n
km
2
and m
i
quality of
n
d contami
n
H
and MES
a
xima of t
h
e
ason (Fig
u
c
h grid poi
n
>
2 cm or
M
a
sed on (i) s
of the app
l
cycle is c
o
grid point,
s
, 2006) ar
e
years, and
n
n
ique capa
b
i
nutes ove
r
the data
m
n
ation, part
i
HS data
h
e 5 min
u
re 4 and
n
t, a hail
M
ESHS >
tatistical
l
icability
o
mputed
an “hail
e
used to
(2)
σ is the
b
ility to
r
a large
m
ust be
i
al beam
This article is protected by copyright. All rights reserved.
Accepted Article
blocking, beam shielding, overshooting and widening as well as different attenuation effects
(e.g. Joss et al., 1998; Germann et al., 2006; Villarini and Krajewski, 2010, Mandapaka et al.,
2013). Over complex terrain like the Alps some of these effects (e.g. beam shielding) may
severely modify the observations, see for instance Germann et al. (2006). Bright-band
contamination and beam overshooting only marginally affect POH and MESHS. The former,
a local increase in reflectivity due to the melting of frozen hydrometeors, is an effect that is
mainly relevant for stratiform precipitation, but not for convective cells. Even if occasionally
the bright band may produce reflectivity values larger than 45 dBZ at the altitude of the
freezing level, the difference in height between ET45 and H0 is too small to affect the
identification of hail by radar, see Equation (1). Furthermore, beam overshooting, increasing
with the distance off the radar, also affects both hail algorithms only marginally. The criteria
for a positive detection by POH and MESHS require the ET45 and the ET50 to be located at
mid and high levels of the troposphere. For example, assuming a freezing level at 3000 m,
ET45 has to be 7200 m for a POH value of 80%. In such cases, beam overshooting does not
play a role. Only during the early convective season (April-May), when usually airmass
conditions limit the vertical development of convective cells to around 6000 – 8000 m and
when the freezing level is still low, beam overshooting can be an issue. The spatial
distribution of the hail signals, however, is mainly governed by the events during June and
July (cf. Figure 7), where the effect of beam overshooting is negligible.
The other three issues, namely ground clutter contamination, partial and total beam shielding
and different types of attenuation, may affect the hail products. MeteoSwiss uses a
sophisticated clutter suppression algorithm, based on Doppler and statistical filtering (Joss et
al, 1998), which is continuously improved (Germann et al., 2006; Germann et al., 2015) and
eliminates clutter efficiently. In the 4th generation radar data residual clutter is a marginal
issue. In the 3rd generation radar data a few clutter pixels remain near to mountain peaks.
However, the effect on the hail products is negligible because these radar echos are located at
lower altitudes and do not result in high echotops used as criteria for hail.
A particular type of clutter occurs in inversions with anomalous propagation. The impact on
the hail studies is negligible because anomalous propagation is unlikely in convective
situations and limited to low altitudes.
Wet-radome attenuation affects the data used in this study only marginally. The concomitant
occurrence of convective storms exactly over the radar site and the presence of hailstorms in
the radar domain is rare. Based on our dataset less than 0.2% of all thunderstorms are
This article is protected by copyright. All rights reserved.
Accepted Article
affecte
d
attenua
t
Path at
t
precipi
t
current
data of
t
Beam
w
distanc
e
order t
o
has bee
n
Partial
b
comput
e
visibili
t
Beam s
the lo
w
over ar
e
worst
c
Lema r
a
radar b
e
Figure
3
mounta
i
and hi
gh
also sho
d
. Non-con
v
t
ion is li
m
it
t
enuation c
a
t
ation. Stat
i
status Me
t
t
he 4
th
gen
e
w
idening is
e
between t
h
o
minimize
n
reduced t
o
b
eam bloc
k
e
s the visib
t
y partial be
s
hielding b
y
w
est radar
b
e
as behind
c
ase examp
a
dar. Beca
u
e
am towar
d
3
: In the d
i
i
n. At a dis
t
h
er. This is
wn.
v
ective rai
n
e
d to few d
e
a
n have an
i
stically sp
e
t
eoSwiss is
e
ration.
another po
t
h
e radar sit
this effect
o
160 km.
k
ing is ano
ility of eac
h
am blocki
n
y
obstacles
b
eam. The
higher mo
u
le, Figure
3
u
se of the
s
d
northeast
i
rection of
N
t
ance of 15
0
the directi
o
n
on the
r
e
cibels (Ge
effect on
h
e
aking this
testing an
t
ential erro
r
e and the l
o
a two-rad
a
ther challe
n
h
pixel of t
h
n
g is correct
and moun
t
se altitude
s
u
ntains tha
t
3
shows th
s
hielding o
f
exceed 8
k
N
NE the b
e
0
km the ra
d
o
n with m
o
r
adar site i
rmann et al
h
ail data w
h
applies o
n
attenuatio
n
r
in radar a
p
o
cation of t
h
a
r composit
e
n
ge especi
a
h
e domain
f
ed at the le
v
t
ain ranges
s
are parti
c
t
are cover
e
e altitude
o
f
a nearby
m
k
m at a dist
a
e
am of Le
m
d
ar can obs
o
st severe s
h
s more fr
e
., 1999).
h
en a hailst
o
n
ly to few
n
correctio
n
p
plications
.
h
e scattere
r
e
has been
a
lly in co
m
f
rom a digi
t
v
el of the p
o
has a dire
c
c
ularly hi
g
e
d only by
o
f the low
e
m
ountain
p
a
nce of 15
0
m
a radar is
erve precip
h
ieldin
g
. Th
e
quent, but
o
rms is loc
hailstorms
n
algorithm
.
This effec
r
(e.g. Cinti
n
used and t
h
m
plex topo
g
t
al terrain
m
o
lar data.
c
t influenc
e
g
h over m
o
a single ra
e
st visible
b
p
eak the al
t
0
km.
severel
y
s
h
itation onl
y
e position
o
here the
r
ated behin
d
in our da
t
using pol
a
t
increases
n
eo et al.,
2
h
e maxim
u
g
raphy. Me
t
m
odel. Bas
e
e
on the al
t
o
untain ra
n
d
ar. By m
e
b
eam of th
e
t
itude of th
e
h
ielded b
y
a
y
at a hei
g
h
t
o
f the Albis
r
esulting
d
intense
t
aset. At
a
rimetric
with the
2
012). In
u
m range
t
eoSwiss
e
d on the
t
itude of
n
ges and
e
ans of a
e
Monte
e
lowest
a
nearb
y
t
of 8 km
radar is
This article is protected by copyright. All rights reserved.
Accepted Article
Taking profit from a two radar composite and limiting the radius to 160 km, the altitudes
of the lowest radar beam over the Alps are generally below 5000 m. Over a limited area in
the central Alps they range between 5000 and 6700 m. POH values of 40, 60 and 80%
correspond to z differences (see Equation 1) of 2400, 3070 and 4200 m, respectively
(Foote et al., 2005). If a typical summertime freezing level located around 3000 m is
considered, issues arise in locations where the altitude of the lowest radar beam exceed 5400,
6070 and 7200 m (for POH of 40, 60 and 80%, respectively). Since for all the presented
investigations a POH threshold of 80% has been used, the shielding issue can be neglected.
Evidently, by decreasing the freezing level height and the POH threshold, this issue may
become more relevant.
Generally speaking, it can be concluded that POH and MESHS are calculated using radar
reflectivities at high altitudes. For this reason, they are less affected by radar issues which
typically affect quantitative precipitation estimation near the ground.
3.6 Verification
The algorithms employed in this study are validated against insurance claim reports by the
method of categorical verification, which is widely applied in meteorology (e.g. Delobbe
et al., 2003; Saltikoff et al. 2010; Skripniková and Řezáčová (2014); Kunz and Kugel, 2015).
A 2 x 2 contingency table (Wilks, 2006) for radar detections and claims reports was used
to compute probability of detection (POD), false alarm rate (FAR) and critical success
index (CSI). Whereas POD describes the ability of the algorithms to detect hail correctly,
FAR indicates the proportion of wrong detections, i.e. when the algorithm identifies hail
over an area but no claims were recorded by the insurance. CSI describes the ability of the
detection algorithms of having simultaneously a high POD and a low FAR. A perfect
score is represented by a POD and CSI equal to unity and a FAR equal to zero.
Recently Kunz and Kugel (2015) and Skripniková and Řezáčová (2014), for example,
validated different hail detection algorithms against loss data provided by building insurance
companies using different skill scores and quality measures from categorical verification
(Wilks, 2006). They showed that in general radar-based hail information (Waldvogel’s
method among them) provided a comparatively high probability of detection (POD), but also
a high false alarm rate (FAR). Similar results were obtained by Delobbe et al. (2005), who
verified the POH algorithm with ground reports. However, it is acknowledged that the
This article is protected by copyright. All rights reserved.
Accepted Article
assessment of the FAR is more complicated compared to the assessment of the POD (e.g.
Delobbe et al., 2003, Saltikoff et al. 2010) or not possible at all (e.g. Delobbe and Holleman,
2006). This is due to the fact that if no damage was recorded for a specific area, this does
not imply that there was no hail. The reason for any discrepancy might also be that
insured losses are controlled by several other factors such as insurance coverage (i.e., the
number of insured objects), land use, vulnerability of insured assets (crops, cars or
buildings) and insurance regulatory practice. Since the radar usually detects hail at higher
altitudes, also melting processes may affect the results. However, since the hailstone’s
surface is proportional to the square of the radius but the volume is proportional to the
cubic, melting may influence only small ice particles such as graupel and small hailstones
(Mahoney et al., 2012).
4 Results and discussion
Based on the two hail detection algorithms presented in the previous section, different
statistics of the estimated hail signals were calculated. The gridded climatological frequency
of hail will be discussed first, followed by the monthly distribution, the diurnal cycle and the
hail frequency associated with different weather types. Finally, a preliminary validation of the
hail products with insurance data is presented. In order to investigate regional differences in
the alpine microclimates, hail occurrences are presented with the full radar resolution. We are
aware that considering the limited length of the investigation period and the relative rarity of
hailstorms, a resolution of 1 km2 is somewhat high and will reveal details that are partly
related to individual events. However, the goal of the study being to present occurrence
statistics without making any assumptions about an underlying statistical-physical model of
hail, the results are presented in the full resolution of the radar grid.
This article is protected by copyright. All rights reserved.
Accepted Article
4.1. Cl
i
Figure
4
km
2
du
r
season.
The se
a
with va
the Alp
s
norther
n
Bavari
a
foothill
s
Alps, h
a
and 2.0
The rel
a
with t
h
variabil
despite
hailstor
m
In addi
t
the fre
q
D = 2,
related
t
i
matologi
c
4
. Left: ave
r
in
g
the pe
a
sonal aver
a
lues betwe
e
s
(north an
d
n
parts, en
h
a
n Alps in
G
s
and in th
e
a
il is rare, l
e
days.
a
ted STD
s
h
e mean fr
e
ity. Except
i
the high fr
e
m
s in gene
r
t
ion to the
P
q
uency of h
a
3 and 4 c
m
t
o the distri
b
c
al frequ
e
ra
g
e numb
e
riod 2002-
2
a
ge number
e
n 2 and 4
d
south) are
h
anced hai
l
G
ermany.
T
e
Po valle
y
e
ss than on
c
s
hown in F
i
e
quency;
a
i
ons are fo
u
e
quency of
r
al, except
f
P
OH, the
M
a
ilstor
m
s i
n
m
were con
s
b
ution of t
h
e
ncy
e
r of days
w
2
014; right:
of hail day
s
hail days p
located m
a
l
frequency
T
o the sout
h
y
(sub-regi
o
c
e per year
,
i
gure 4 ran
g
a
reas with
u
nd only f
o
hail. The S
f
or the Jura
M
ESHS pro
d
n
cluding th
e
s
idered (Fi
g
h
e POH sig
n
w
ith POH
>
STD of t
h
s
estimated
er km². Th
e
a
inly over t
h
is also fo
u
h
of the Al
p
o
ns 4 and
5
,
whereas i
n
g
es betwee
n
higher fre
q
o
r the Jura
a
TD reflect
and the Po
d
uct was c
o
e
sizes on t
h
g
ure 5). Fr
e
n
als presen
t
>
80% per
s
h
e number
o
from POH
e
most app
a
h
e foothills
u
nd over t
h
p
s, distinct
5
in Fig. 1)
n
other are
a
n
0 and 2.
8
q
uencies s
h
a
nd the Po
the well-k
n
valley.
o
mputed o
v
h
e hailstone
s
e
quency m
a
t
ed in Figu
r
s
eason (Ap
r
o
f radar-d
e
> 80% sh
o
a
rent maxi
m
in the prea
l
h
e Jura, th
e
maxima a
r
. Over the
a
s the value
s
8
and corr
e
h
ow also
a
valley, wh
n
own high
a
v
er the sam
e
s
. Three di
a
a
xima of t
h
e 4. For ha
i
r
il - Septem
b
e
rived hail
d
o
ws several
m
a on both
l
pine regio
n
e
Swabian
J
r
e located
a
main chai
n
s
range bet
w
e
lates in m
o
a
high yea
r
ere the ST
D
a
nnual vari
a
e
period to
a
meter thre
s
h
e MESHS
i
lstones gre
a
b
er) and
d
a
y
s per
maxima
sides of
n
. On the
J
ura and
a
long the
n
s of the
w
een 0.5
o
st areas
r
-to-year
D
is low
a
bility of
estimate
s
holds of
are well
a
ter than
This article is protected by copyright. All rights reserved.
Accepted Article
2 cm, t
h
maxim
a
the hig
h
only ra
d
Figure
5
(botto
m
Annual
three d
souther
n
frequen
indicat
e
h
e frequen
c
a
decrease
t
h
agreemen
t
d
ar-derived
5
: Avera
g
e
n
m
right) per
s
mean freq
u
ifferent su
b
n
prealpin
e
n
cy was ac
c
e
years whe
r
c
y maxima
t
o 1.4 days
t
of POH a
n
hail days o
r
n
umber of
d
s
eason (Apr
u
ency ano
m
b
-regions:
e
and alpin
c
umulated
r
e hail was
reach 1.8
d
(for D > 3
n
d MESH
S
r
events es
t
d
a
y
s with
M
il - Septem
b
m
alies were
for the en
t
e regions
(
over all p
i
more (less
)
d
ays per se
a
cm) and 0.
6
S
(D > 2 c
m
t
imated wh
e
M
ESHS > 2
c
b
er) and k
m
calculated
f
t
ire domai
n
(
Figure 6,
m
i
xels of th
e
)
frequent
c
a
son. For l
a
6
days per
m
), we cons
i
e
n POH > 8
c
m (top), >
m
2
durin
g
20
0
f
or the wh
o
n
(Figure
6
m
iddle an
d
e
regions.
c
ompared t
o
a
rger hailst
o
season (for
i
der in the
s
0%.
3 cm (bott
o
0
2-2014.
o
le investig
a
6
, top), th
e
d
botto
m
).
Positive (
n
o
the multi-
y
o
nes, the f
r
D > 4 cm
)
s
ubsequent
o
m left), an
d
a
tion perio
d
e
northern
The displa
y
n
egative) a
n
y
ear avera
g
r
equency
)
. Due to
sections
d
> 4 cm
d
and for
and the
y
ed hail
n
omalies
g
e. In the
This article is protected by copyright. All rights reserved.
Accepted Article
entire
d
in the
s
anomal
i
STD.
Figure
6
re
g
ions
re
g
ions
)
d
omain, hai
l
s
outh (ST
D
i
es larger t
h
6
Annual s
t
1+2 (midd
l
)
.
l
frequency
D
North
= 1.4
4
h
an 1 STD,
t
andardize
d
l
e) and sub
is highly v
4
*STD
South
)
whereas 2
0
d
anomalies
-regions 4
+
ariable. ST
D
. The year
s
0
13 and 20
(POH > 8
0
+
5 (bottom;
D
in north
e
s
of 2003,
14 show n
e
0
%) for th
e
see Fi
g
ure
e
rn regions
i
2008 and
2
e
gative ano
m
e
entire do
m
1 for the
l
i
s greater t
h
2
009 show
m
alies exc
e
m
ain (top)
a
l
ocation of
h
an STD
positive
e
eding -1
a
nd sub-
the sub-
This article is protected by copyright. All rights reserved.
Accepted Article
The decrease in hail frequency since 2009 in the entire domain is mainly due to a decrease of
hail in the northern alpine area (Figure 6, middle). Four years (2005, 2007, 2010 and 2012)
show a strong north-to-south anomaly. These deviations suggest that some of the general
weather situations that favour the development of hailstorms in the North are not the same as
for the South.
In general, the alpine orography has a direct influence on the distribution of hailstorms;
despite of the random nature of hail (Mezher et al., 2012), the identified hot spots suggest that
orographic forcing mechanisms and related flow convergence are decisive for hailstorm
formation. The repeatability of precipitation events over areas with complex orography has
been used to develop a nowcasting systems based on past analogues (e.g. Panziera and
Germann, 2010). The high spatial variability of hail days together with a maximum over and
downstream of hilly terrain (or in its close proximity) as well as the minimum over the highest
mountains were also found in other studies performed over regions with complex orography
(Changnon and Changnon, 2000; Garcia-Ortega et al., 2007; Počakal et al., 2009; Kunz and
Puskeiler, 2010; Berthet et al., 2011; Cintineo et al., 2012; Eccel et al., 2012; Mezher et al.,
2012; Berthet et al., 2013; Merino et al., 2013). The reduced convective activity over the
central chain of the Alps was also found by van Delden (2001) using observations from
synoptic weather stations and by Nisi et al. (2014) using cloud to ground lightnings. The
triggering and updraft enhancement process is particularly important over the foothills in the
vicinity of the Alps or generally over hilly terrain (e.g. Barthlott et al., 2005; Kottmeier et al.,
2008; Davolio et al., 2009). In those regions low level warm and moist air that origins from
the plains or from the Mediterranean south of the Alps are forced to lift due to low level
convergences. These convergence zones are caused by the flow deviations at the hills, by flow
around regime, or by the outflow of previously developed thunderstorms cells in the alpine
valleys. For example, it is well known that the Po valley and the southern Prealpine region is
dominated by anabatic – katabatic wind systems (Morgan 1973; Gladich, 2011). For these
reasons, the foothills of the alpine chain represent one of the regions in Europe with the most
frequent convection initiation (Collier and Lilley, 1994; Huntrieser et al., 1996; van Delden,
2001).
This article is protected by copyright. All rights reserved.
Accepted Article
4.2 M
o
The ra
d
June/Ju
l
distinct
were o
unfavo
u
frequen
wherea
s
conditi
o
shear c
o
Highes
t
Figure
7
Septem
b
The m
o
Alps w
i
o
nthly dis
t
d
a
r
-derived
l
y (Figure
spatial pat
t
bserved d
u
u
rable for s
e
n
cy increas
e
s
in June i
t
o
ns in the
m
o
nditions,
b
t
hail frequ
e
7
: Avera
g
e
b
er) and k
m
o
nth of July
i
th values
o
t
ribution
hail frequ
e
7). In Apr
i
t
erns. Fro
m
u
ring the
l
e
vere conv
e
e
s slightly
b
t
increases
m
ean, not
o
b
ut also sta
t
e
ncy can be
number of
m
2
durin
g
2
0
is the mos
t
o
f about 1.
8
e
ncy show
s
i
l and Sept
m
October t
o
l
ast 13 ye
a
e
ction due
t
b
oth in the
more or l
e
o
nly pre-fr
o
t
ionary air
m
found ove
r
radar-deri
v
0
02-2014.
t
active mo
n
8
hail days.
s
a pronou
n
ember, hai
l
o
March, h
a
a
rs). In t
h
t
o the low
northern p
a
e
ss over t
h
o
ntal and
f
m
ass conve
c
r
the Jura a
n
v
ed hail da
y
n
th with th
e
Most pro
n
n
ced seaso
n
l
storms are
a
ilstorms a
r
h
ese month
magnitude
a
rt of the
A
h
e whole d
o
f
rontal em
b
c
tion are c
a
n
d the alpin
e
y
s (POH >
e
maximum
n
ounced are
n
al cycle
w
quite rare
r
e extremel
y
s, the air
m
of the laps
e
A
lps and in
o
main. Du
e
b
edded stor
m
a
pable to p
r
e
foothills i
n
80%) for
e
frequency
the maxi
m
w
ith a max
i
and occur
y
rare (onl
y
m
asses are
e
rate. In
M
southern
G
e
t
o more
m
s with hi
g
r
oduce hail
n
the Po va
l
e
ach month
o
n both sid
e
m
a in the vi
c
i
mum in
without
y
5 cases
usually
M
ay, hail
G
ermany,
unstable
g
h wind
in June.
l
ley.
(April -
e
s of the
c
inity of
This article is protected by copyright. All rights reserved.
Accepted Article
the Al
p
howev
e
an imp
o
hail fre
q
Alps,
b
Medite
r
Similar
also co
n
Figure
8
Other
s
b
etwee
n
regions
et al.,
2
during
J
(2003)
a
May a
n
showin
g
airmass
and wi
n
develo
p
and the
This is
o
of col
d
hypoth
e
p
s, i.e. on t
h
e
r, hail freq
u
o
rtant role
f
q
uency is l
o
b
ut less in
t
r
ranean Se
a
results we
r
n
firmed by
t
8
: Total nu
m
s
tudies for
n
mid-June
like north-
e
2
009) or
C
J
une or Jul
y
a
nd Berthe
t
n
d a secon
g
the mini
m
conditions
n
d shear c
o
p
ment and
w
reduced ef
f
o
ften obser
v
d
air mass
e
e
sized
b
y
M
h
e norther
n
u
ency is re
d
f
or the initi
o
wer in the
t
he South,
a
still provi
d
r
e found in
t
he monthl
y
m
ber of clai
m
Switzerlan
d
and mid-
J
e
astern Ital
y
C
zechia (S
k
y
(Southw
e
t
et al. (20
1
dary one i
m
um in hail
required f
o
o
nditions, s
t
w
ithout we
f
ect of mel
t
v
ed in the
n
e
s and th
e
M
ezher et al.
n
and south
d
uced com
p
ation of se
v
entire do
m
where wa
r
d
e favoura
b
Griffith (1
9
y
distributi
o
m
s per mon
t
d
(Admira
t
J
uly, which
y
(Giaiotti
e
k
ripniková
a
e
st German
y
1
1) found a
n July. Al
occurrence
o
r sustaini
n
t
orms are
o
ll-structure
d
t
ing, graup
e
n
orthern pa
r
e
increase
d
(2012).
ern side.
O
p
ared to Ju
n
v
ere conve
c
m
ain. The d
e
r
m and mo
b
le conditi
o
9
72). The s
t
o
n of insura
n
t
h for the p
e
et al., 19
8
is close t
o
e
t al., 2003
)
a
nd Řezáč
o
y
; Mohr, 2
0
bimodal d
i
l climatol
o
in autumn,
n
g severe c
o
o
ften short
l
d
hail core
e
l showers
a
r
t of the Al
p
d
frequenc
y
O
ver the Ju
r
n
e. This su
g
c
tive storm
s
e
crease is p
a
ist conditi
o
o
ns for the
d
t
rong decre
a
n
ce claim r
e
e
riod 2003-
2
8
5) found
a
o
our resul
)
, the conti
n
o
v
á
, 2014)
0
13). Furth
i
stribution,
o
gies over
in winter
a
o
nvection a
r
l
iving sing
l
. Because
o
a
re commo
n
p
s, but also
y
of grau
p
r
a and in s
o
g
gests agai
n
s
over the
f
a
rticularly
e
o
n due to
t
d
evelopme
n
a
se in hail
a
e
ports (Fig
u
2
012.
a
maximu
m
ts. Investi
g
n
ental part
identified
ermore, in
with an ab
central Eu
r
a
nd in early
r
e not me
t
,
l
e cells wit
h
o
f relativel
y
n
especiall
y
in other re
g
p
el shower
o
uthwest
G
n
that the
A
f
oothills. In
e
vident nor
t
t
he presenc
n
t of sever
e
a
ctivity in
A
u
re 8).
m
in hail oc
c
g
ations for
of Croatia
(
the hail
m
France Fra
solute max
i
r
ope are c
o
in the spri
n
i.e. high i
n
h
a limite
d
y
low tem
p
y
early in th
e
g
ions. The
p
has alrea
d
G
ermany,
A
lps play
August,
t
h of the
e of the
e
storms.
A
ugust is
c
urrence
adjacent
(
Počakal
m
aximum
ile et al.
i
mum in
o
nsistent
n
g. If the
n
stability
d
vertical
p
eratures
e
spring.
p
resence
d
y been
This article is protected by copyright. All rights reserved.
Accepted Article
4.3 Diurnal cycle
Hourly radar-derived hail frequency for the entire domain and for the six sub-regions features
a distinct diurnal cycle during all months and regions with a maximum in the late afternoon
and a minimum in the morning hours (Figure 9). Most hail events occur in the afternoon
hours between 13 and 18 UTC (i.e., approx. 11:00 and 16:00 local time), whereas a minimum
is evident in the early morning. From May to July, the three months when hail is more
frequent, the diurnal cycle peaks slightly later south of the Alps compared to the northern sub-
regions. For example, over the northern Prealps the peak is between 15 and 17 UTC, whereas
it is two hours later over the southern Prealps.
This article is protected by copyright. All rights reserved.
Accepted Article
This article is protected by copyright. All rights reserved.
Accepted Article
Figure
9
regions.
April-S
e
80% is
d
9
: Hourly r
a
The dia
g
r
a
e
ptember a
n
d
etected in
t
a
dar-deriv
e
a
ms show th
n
d 2002-20
1
t
he re
g
ion.
e
d hail freq
u
e diurnal c
y
1
4. A hail
h
u
enc
y
(nor
m
y
cle of hail
d
h
our is defi
n
m
alized) for
d
a
y
s betwee
n
ed as the
h
the entire
d
n 00 and 24
h
our when
d
omain an
d
UTC for t
h
a POH val
u
d
six sub-
h
e period
u
e above
This article is protected by copyright. All rights reserved.
Accepted Article
In the
S
Mesosc
convec
t
evenin
g
or invi
g
of wat
e
occurre
n
night w
Some s
conseq
u
not stat
i
on ave
r
homog
e
of deep
Figure
1
Septem
b
conside
r
S
outh, late
a
ale desce
n
t
ive cells i
n
g
changes i
n
g
orate conv
e
e
r vapour
fr
n
ce over F
r
ere also fo
u
s
ub-regions
u
ence of th
e
i
stically re
p
r
age the hi
g
e
neous area
s
moist con
v
0: Time (U
b
er 2002-2
0
r
ed.
a
fternoon -
n
t of cold
n
the Preal
p
n
the diurn
a
e
ctive cells
.
fr
om July t
o
r
iuli Venez
i
u
nd over so
u
show larg
e
e
small nu
m
p
resentativ
e
g
hest hail
f
s
(Figure 1
0
v
ection.
TC) of the
h
0
14. Onl
y
p
i
evening s
t
air from
t
p
ine area (
M
a
l wind regi
m
.
Furtherm
o
o
Septemb
e
i
a Giulia in
u
th-wester
n
e
r variabili
t
m
ber of hai
l
e
. The spati
f
requency
h
0
), but also
h
i
g
hest rad
a
i
xels affecte
t
orms are r
e
t
he Alps
a
M
organ, 19
m
es can pr
o
o
re, Giaiott
i
e
r is the re
the northe
a
n
France by
t
y in some
l
storms inv
o
al distribut
i
h
as been r
e
sharp grad
i
a
r-derived
h
d b
y
at lea
s
e
latively fr
e
a
nd thund
e
73). Gladi
c
o
duce flow
i
et al. (200
ason for a
n
a
stern part
Dessens (1
9
months (e
o
lved. In t
h
i
on of the
t
e
corded, s
h
i
ents reflec
t
h
ail frequen
c
s
t three hai
l
e
quent duri
n
e
rstorm ou
t
c
h et al. (2
0
convergen
c
3) found t
h
n
increase
d
of Italy. S
e
9
86).
.g. Jur
a
in
h
ese cases,
r
t
wo-hour p
e
h
ow severa
l
t
ing the par
t
cy
per km
2
f
l
storms wit
h
n
g summer
t
flows ma
y
0
11) found
c
es that ma
y
h
at a greate
r
d
evening
h
e
condary m
a
Augus
t
).
T
r
elative ma
x
e
riods (UT
C
l
larger an
d
t
ly stochast
i
f
or the peri
o
h
POH > 8
0
months.
y
trigger
that the
y
trigger
r
amount
h
ailstorm
a
xi
m
a at
T
his is a
x
ima are
C
), were
d
almost
i
c nature
o
d April-
0
% were
This article is protected by copyright. All rights reserved.
Accepted Article
The Swabian Jura and the Jura mountains in western part of Switzerland show the earliest hail
peaks (10-14 UTC). Differential heating, orographic triggering and local low-level
convergence may explain the early convection initiation over these regions (e.g. Kottmeier et
al., 2008). Another possible explanation is the triggering of hailstorms by pre-frontal uplift
(Schemm et al., submitted). Since most of the fronts enter the investigation area from the
west, related convection must develop earlier in those areas. Outflow boundaries of the early
convective development over the Jura mountains may act as additional triggering mechanism
for the subsequent formation of convective cells over the western regions, where the peaks are
between 12-16 UTC. These cells usually move towards the east during the afternoon,
affecting the central part of the northern prealpine region (Luzern - Zurich) in the late
afternoon with peaks between 16-20 UTC. These mechanisms may explain the pronounced
west-to-east gradient visible in Figure 10 in the northern part of the Alps.
In the southern prealpine area, a larger spatial variability of the hail peak hour is found
compared to the northern parts. Over the same region, an interesting line of hail signals peaks
late in the night (pink-violet colours corresponding to 02-06 UTC) is visible along the
southern border of the prealpine foothills (right part of the investigation sub-region 4). This
night-time convective development may be explained by low-level convergence produced by
katabatic wind systems (Morgan 1973; Gladich, 2011).
In the Po valley a large, almost homogeneous area with a peak frequency between 18 and 22
UTC is evident. Over flat areas, the absence of important triggering mechanisms such as
differential heating or orographic uplift may delay convection initiation and, consequently,
the formation of severe storms.
4.4 Hail occurrence and weather types
Typical circulation patterns are investigated using large-scale weather types (WT) provided
by MeteoSwiss (Weusthoff, 2011). A 10-class WT classification based on geopotential and
wind direction at 500 hPa was used here. Since the frequency of the different WTs is not
homogeneous, the results were normalized by the number of days for each weather type.
WT1, WT2 and WT8, which correspond to westerly, south-westerly and southerly flow, are
the most frequent WTs. Highest hail frequency is found during the south-western flow regime
(WT2), with maximum frequency values of about 5% per grid point (Figure 11). Similar hail-
favouring WTs were identified in several other studies for other areas in central and southern
This article is protected by copyright. All rights reserved.
Accepted Article
Europe
and Pu
s
flow (
W
the Nor
t
is foun
d
For the
over ce
Howev
e
Figure
1
differe
n
each cl
a
(wester
l
(WT5);
4.5
V
e
r
(e.g. Bider
s
keiler, 201
0
W
T1) are le
s
t
h and Sou
t
d
at a large
d
types WT
5
e
ntral Euro
p
e
r, some lo
c
1
1: Norma
l
n
t weather t
y
a
ss. Onl
y
th
ly
flow (W
T
northerl
y
f
l
r
ification
o
1954; Wil
l
0
; Merino
e
s
s favoura
b
t
h, but also
d
istance of
5
, WT4 and
p
e, radar-d
e
c
alized ma
x
l
ized frequ
e
y
pes for th
e
e six weath
e
T
1); south-
w
l
ow (WT4);
o
f rada
r
-
b
l
emse, 199
5
e
t al., 2013;
b
le for hail.
ove
r
the B
a
the Albis r
a
WT3, na
m
e
rived hail
x
ima are als
o
e
nc
y
of rad
a
e
period A
p
e
r t
y
pes w
h
w
esterl
y
flo
w
north-west
e
b
ased hail
5
; Hohl and
Berthet et
a
During so
u
a
varian Alp
s
a
dar locatio
n
m
ely north-e
signals oc
c
present i
n
a
r-derived
h
p
ril-Septem
b
h
ich show a
w
(WT2);
s
e
rl
y
flow (
W
algorith
m
Schiesser,
a
l., 2013).
D
u
therly flo
w
s
. Note, ho
w
n
, thus the
r
asterly, no
r
c
ur less fre
q
n
the southe
r
h
ail da
y
s, i
.
b
er 2002-20
hail frequ
e
s
outherl
y
fl
o
W
T3); cf. Ta
b
m
s with in
s
2001; Kun
z
D
ays chara
c
w
(WT8), h
a
w
ever, that
r
esults may
r
the
r
ly and
n
q
uent over
r
n Prealps
f
.
e. da
y
s wi
t
14. N is th
e
e
nc
y
in exce
o
w (WT8);
b
le 2).
s
urance d
z
et al., 20
0
c
terized by
w
a
il occurs b
the latter
m
be not reli
a
n
orth-west
e
the entire
f
or WT4 an
d
t
h POH >
8
e
number o
f
ss of 1% a
r
north-east
e
a
ta
0
9; Kunz
w
esterly
oth over
m
aximum
a
ble.
e
rly flow
domain.
d WT5.
8
0%, for
f
da
y
s in
r
e shown
e
rl
y
flow
This article is protected by copyright. All rights reserved.
Accepted Article
Since r
a
observ
a
investi
g
from a
n
of a hi
g
loss da
t
does n
o
does n
o
Overal
l
a good
or less
u
Figure 1
squares)
We the
r
cars is
a
da
r
-
b
ased
h
a
tions is r
e
g
ation area
,
n
automob
i
g
her vulner
t
a may be
a
o
t allow to
o
t allow to
e
l
the super
p
agreement
u
rbanized
r
2: POH dai
l
on 23 July
2
r
efore per
f
highly p
r
h
ail detecti
o
e
quired. S
i
,
a prelimi
n
i
le insuran
c
ability; ho
w
a
ffected b
y
perform v
e
valuate n
o
p
osition of
t
(Figure 1
2
r
egions, w
h
l
y maximum
2
009 (left).
f
ormed a q
u
r
obable, n
a
o
n algorith
m
i
nce no
d
n
ary assess
c
e compan
y
w
ever, the
y
y
this addit
i
erification
s
o
n-severe s
t
t
he insura
n
2
an examp
h
ere few ro
a
(color shad
i
u
antitative
a
mely the
m
s are bas
e
d
irect hail
ment of th
e
y
. Compar
e
y
have the
d
i
onal sourc
e
s
over are
a
t
orms with
n
ce claim r
e
le for 23 J
u
a
ds exist, t
h
i
ng) and nu
m
verificatio
n
25 most
e
d on prox
y
measure
m
e
POH an
d
e
d to build
i
d
isadvanta
g
e
of uncert
a
s with ver
y
hailstone
s
e
ports with
u
ly 2009).
h
e agreem
e
m
ber of car
c
n
only ove
r
populated
y
dat
a
, veri
fi
m
ents are
d
MESHS s
k
i
ngs, cars
h
g
e to be m
o
ainty. Furt
h
y
low pop
u
s
izes < 2c
m
daily max
Only over
e
nt is low.
c
laims per p
o
r
areas wh
e
urban are
fi
cation wit
h
available
k
ills uses l
h
ave the a
d
o
bile and i
n
h
er on, thi
s
u
lation den
m
.
POH valu
e
mountaino
o
stal code z
o
e
re the pre
s
as in Swi
h
ground
for the
oss data
d
vantage
n
surance
s
dataset
sity and
e
s shows
us areas
o
ne(black
s
ence of
tzerland
This article is protected by copyright. All rights reserved.
Accepted Article
(Appe
n
these
d
indicat
e
The pu
r
and M
E
90
th
pe
r
been u
MESH
S
Figure
1
events
(
perfor
m
Figure
high P
O
the oth
e
FAR is
90). A
al. (20
0
signifi
c
Saltiko
f
Netherl
a
equal t
o
it can
b
detecti
o
Switze
r
Mang
e
n
dix A). F
o
d
ays the P
O
e
d that ther
r
pose of t
h
E
SHS thre
s
r
centiles a
m
sed. It w
o
S
, but unfo
r
1
3: Compa
r
(
ri
g
ht axis)
m
ed a
g
ainst
c
13 shows
t
O
D (in this
r hand, th
visible fo
r
similar
b
e
h
0
7) or Kess
i
c
antly with
f
f et al. (2
0
a
nds all ev
o
80%. Acc
b
e expecte
o
ns are val
r
land, first
e
t al, 2011)
o
r the veri
fi
O
H, the M
E
e was hail
o
h
e prelimin
a
s
holds rela
t
m
ong the P
O
o
uld be in
t
r
tunately i
n
r
ison of thr
e
detected
w
c
ar insuran
c
t
hree skills
study PO
D
e FAR is
a
r
increasin
g
h
aviour of
t
i
nger et al.
POH valu
e
0
10) and
D
ents with
h
ording to t
h
d
that the
idated aga
i
results fro
m
showed th
a
fi
cation, m
o
E
SHS and
/
o
ver at lea
s
a
ry verific
a
t
e to the o
c
O
H or ME
S
t
eresting t
o
n
the insur
a
e
e skill sco
r
w
ith differe
n
c
e reports o
scores for
D
0.84) i
s
a
lso high (
g
POH thr
e
t
he FAR
w
(1995). T
h
e
s greater t
h
D
elobbe et
a
h
ail stones
l
h
e finding
s
FAR for
l
i
nst crop
d
m
a pilot
s
a
t soft hail
o
re than 2
0
/
or insura
n
s
t one of th
a
tion is to
c
currence
o
S
HS value
s
o
validate
a
nce datase
t
r
es (POD, F
A
n
t POH an
ver 25 sele
c
different
P
s
found for
0.7 for P
O
e
sholds (0.
5
w
as observe
h
is is an in
d
h
an 70-80
%
a
l. (2005);
l
arger than
s
of Saltiko
l
ower PO
H
d
amage, w
h
s
tudy empl
or even g
r
0
0 days ha
v
n
ce claim r
ese areas.
investigat
e
o
f damage
s
distributi
o
the hailst
o
t
this infor
m
A
R and CS
d MESHS
c
ted urban
a
P
OH and
M
all POH t
h
O
H 70%
5
4 for PO
H
d by other
d
ication th
a
%
. A furth
e
which sho
w
2 cm corr
e
ff et al. (2
0
H
values
w
h
ich occu
r
oying aut
o
r
aupel are
o
v
e been se
l
eports (
1
e
how well
to automo
b
o
n of the c
o
o
ne dimen
s
m
ation is n
o
I, left axis)
thresholds.
a
reas betwe
e
M
ESHS thr
e
h
resholds (
F
). An evid
e
H
= 80% a
n
authors, f
o
a
t damage
d
e
r confirma
t
w
ed that i
n
e
sponded t
o
0
10) and D
e
w
ill decreas
e
r
s for sma
l
o
matic hail
o
ften detec
t
l
ected; for
1
0 per urb
a
the differ
e
b
iles. For
t
o
nsidered a
r
s
ions prov
o
t availabl
e
and numb
e
The verifi
c
e
n 2003 and
e
sholds. O
v
F
igure 13,
l
e
nt decrea
s
n
d 0.49 fo
r
o
r example
d
ue to hail
i
t
ion is pro
v
n
Belgium
o
a POH o
f
e
lobbe et al
.
e
if the r
a
l
l hail alr
e
detectors
(
t
ed for lo
w
each of
a
n area)
e
nt POH
t
his, the
r
ea have
ided by
e
.
e
r of hail
c
ation is
2012.
v
erall, a
l
eft). On
s
e of the
r
POH
Aran et
i
ncrease
v
ided by
and The
f
at least
.
(2005),
a
dar hail
e
ady. In
(
Löffler-
w
er POH
This article is protected by copyright. All rights reserved.
Accepted Article
values (i.e. 20-50%). The values of the CSI are in the range of other studies, e.g., by
Hollemann (2001) or Kunz and Kugel (2015). CSI values in excess of 0.4 for POH > 80%
confirm the plausibility of this threshold considered in our study.
Skills scores for the MESHS (Figure 13, right) show slightly lower PODs ( 0.74), but
much lower FARs ( 0.53). On the one hand, it can be expected that higher MESHS values
result in a higher POD (i.e. increased connection to damage to automobiles). On the other
hand, the population of MESHS classes is not equally distributed (few cases of MESHS-60
mm compared to the number of MESHS-20 mm cases). Therefore the POD does not increase
linearly with increasing MESHS thresholds. Because of small populations, the skill scores for
MESHS values greater than 40 mm should be taken with care.
The validation based on insurance loss data is challenging and the results have to be analysed
carefully. Kunz and Kugel (2015) discuss several reasons for the high FARs. First, the POH
algorithm was originally developed by comparing radar observations with hailpads data,
which are able to detect smaller hailstones. Therefore, high POH values (i.e. > 70%) can be
reached even if no damaging hailstones for cars, i.e. hail stones 2cm, are present.
Furthermore, the spatial resolution and errors of both datasets may influence the skill scores
considerably. For example, in case of strong winds, hailstone drift represents an error source
for the radar data (Schuster et al., 2006). Concerning insurance claim reports, uncertainties in
the spatial and temporal allocation of the damage and uncertainty in the presence of cars
(population density, insurance contract distributions) are main error sources. Since for our
investigation domain no other hail observations exist, insurance claim reports represents the
best available verification data for hail.
5 Summary and conclusions
A 13-year hail assessment has been conducted by reprocessing and homogenizing volumetric
radar data (from April to September) over Switzerland and adjacent countries. Radar-based
hail detection algorithms have been used to investigate the spatial distribution and frequency
of hail signals with a high spatiotemporal resolution. Indirect hail observations based on radar
reflectivity and melting height as proxy data is valuable especially in areas where no ground
observations are available like the area considered in this study.
This article is protected by copyright. All rights reserved.
Accepted Article
The results in terms of radar-derived hail days during the 13-year period are located along the
foothills of the Alps in the northern and southern prealpine region, over north-western
Switzerland, the Jura as well as over southern Germany and the Bavarian Alps. Maximum
values range between 2 and more than 4 hail days per summer season. It has been found that
the variability in hail occurrence in the North is greater compared to the South (144% of the
STD).
Monthly radar-derived hail maps show a distinct seasonal cycle with the maximum for June
and July. During May and August differences in the hail frequency are found for areas north
and south of the Alps. During late spring, hailstorm activity is more pronounced in the North.,
whereas in late summer, hail is more frequent in the South, especially over the Po valley,
where the Mediterranean sea still provides the warm and moist air favourable for the
development of hailstorms.
An evident diurnal cycle is found for all six sub-regions from April to September. Clear
spatial differences in the time of the day, when hail is most frequent, are found. Hail occurs
earlier in the day in the western part of Switzerland and in the Swabian Jura. In the northern
prealpine area a clear west-to-east gradient is visible, whereas in the southern prealpine area
hail tends to occur later on the day. However, in these areas a larger variability is found.
Several weather types are favourable for hail formation. Considering wind direction at 500
hPa, hail occurs most frequently with south-westerly flow regimes. For both sides of the Alps
the second most hail-producing wind regime is represented by the westerly flow. In this case,
hail occurrences and distributions are strongly reduced, especially in the southern areas.
However, the results show that hail can also occur on a more local scale during southerly,
north-easterly, northerly and north-westerly flow regimes.
The two considered radar-based hail detection algorithms were validated with motor loss data
from an insurance company. The results show that both POH and MESHS are reliable proxies
for hail detection, yielding a POD higher than 75%.Several uncertainties related to the
insurance claim reports result in a high FAR, especially for low POH values. However, for
POH values greater than 80%, which is used in this study, the FAR decreases rapidly to
almost 50%. Despite of a slight overestimation of radar-derived hail occurrence against
insurance claims reports, it can be concluded that the hail distributions presented in this work
are fairly robust and reliable.
The challenges in using radar-based approaches over complex terrain have been described
previously, showing that several effects can affect the radar measurements. However, most of
This article is protected by copyright. All rights reserved.
Accepted Article
them do not or only marginally affect the radar products used in this study. The fact that two
or more hail events can affect the same area in one day results in a slight underestimation of
the number of hail events since monthly and yearly distributions base on daily hail maxima.
However, over the considered period 0.02% of the area has been affected by two hailstorms
(POH>80%) in one day. By using MESHS or reducing the time aggregation (hourly
composites) this issue is even less important. The two empirical algorithms (POH, MESHS)
used in this study are based on the idea that strong radar reflectivities are mainly caused by
large hydrometeors. The POH has been verified and used operationally since several years,
providing reliable results (e.g. Delobbe et al., 2005; Saltikoff et al., 2010; Skripniková and
Řezáčová, 2014; Kunz and Kugel, 2015) and independent verifications with insurance data
are presented in section 4.5. Because of the scarcity of hail size observations over Switzerland
in the past, further effort for the verification of the MESHS is needed in the near future. Next
steps include its verification by means of data collected with (1) the new automatic hail-
sensors network in Switzerland (Löffler-Mang et al, 2011) and (2) hail crowd-sourcing data
(e.g. Elmore et al, 2014).
Furthermore, an object-based analysis of hailstorms characteristics and their environmental
conditions over Switzerland is going to be performed using a radar based thunderstorm
tracking algorithm (Hering et al., 2008). The overall goal is to investigate the conditions and
the characteristics which leads to the formation and intensification of hailstorms. Expected
findings will be potentially interesting for thunderstorm nowcasting purposes.
This article is protected by copyright. All rights reserved.
Accepted Article
A. Ap
Fi
g
ure
A
of the
r
Genève,
de-Fon
d
les-Bain
Ackn
o
The au
t
Ambro
s
Schem
m
the rev
i
Mobili
a
Bern) f
o
Nisi fo
r
pendix
A
.1: the 25
m
r
adar based
Basel, Lau
s
d
s, Freiburg,
n
s, Zug, Krie
n
o
wledg
m
t
hors than
k
s
etti, S. Za
n
m
(Univers
i
i
ew of the
s
a
r) for pro
v
o
r the prep
a
r
the graphi
c
m
ost popula
t
hail identi
f
s
anne, Bern,
Schaffhaus
e
n
s.
m
ents
k
M. Bosc
a
n
ini (Meteo
S
i
ty of Berg
e
s
cientific d
o
v
iding insur
a
a
ration and
p
c
design.
t
ed urban ar
e
f
ication. Fo
l
Witerthur,
L
e
n, Chur, Ve
r
a
cci and L
S
wiss), A.
M
e
n) for scie
n
o
cumentati
o
a
nce loss
d
p
reliminar
y
e
as in Switz
e
l
lowing ag
g
L
uzern, St.
G
r
nier, Neuc
h
. Clementi
M
artynov
a
n
tific discu
o
n. We kin
d
d
ata and rel
a
y
analysis o
f
e
rland (red
a
g
lomerations
G
allen, Luga
n
h
âtel, Uster,
(MeteoS
w
a
nd A. Zisc
h
ssions as
w
d
ly thank
M
a
ted suppo
r
f
the insura
n
a
reas) select
e
have been
n
o, Biel, Th
u
Sitten, Lanc
w
iss) for te
c
h
g (Univer
s
w
ell as for
p
M
.
K
ünzler
r
t and S.
M
n
ce claim r
ed
for the ve
considered
u
n, Köniz, L
y, Emmen,
Y
c
hnical su
p
s
ity of Ber
n
p
roviding s
u
and L. Th
o
M
orel (Univ
eport datas
e
rification
: Zürich,
a Chaux-
Y
verdon-
p
port, P.
n
) and S.
u
pport in
o
mi (Die
ersity of
e
t and S.
This article is protected by copyright. All rights reserved.
Accepted Article
References
Admirat P, Goyer GG, Wojtiw L, Carte EA, Roos D, Lozowki E. 1985. A comparative study
of hailstorms in Switzerland, Canada and South Africa. J. Climatol., 5 : 35 – 51.
Amburn SA, Wolf PL. 1997. VIL density as a hail indicator. Wea. Forecasing, 12: 473 – 478.
Aran M, Sairouni A, Bech J, Toda J, Rigo T, Cunillera J, Moré J. 2007. Pilot project for
intensive surveillance of hail events in Terres de Ponent (Lleida). Atmos. Res., 83: 315-335.
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.
Basara JB, Cheresnick DR, Mitchell D, Illston BG. 2007. An Analysis of Severe Hail Swaths
in the Southern Plains of the United States. Transactions in GIS. 11(4): 531-554.
Berthet C, Dessens J, Sánchez JL. 2011. Regional and yearly variations of hail frequency and
intensity in France. Atmos. Res., 100: 391 – 400.
Berthet C, Wesolek E, Dessens J, Sánchez JL. 2013. Extreme hail day climatology in
Southwestern France. Atmos. Res., 123: 139 – 150.
Betschart M, Hering A. 2012. Automatic Hail Detection at MeteoSwiss. Arbeitsbericht
MeteoSchweiz Nr. 238, 59 pp.
Available: http://www.meteoschweiz.admin.ch/content/dam/meteoswiss/en/Ungebundene-
Seiten/Publikationen/Fachberichte/doc/ab238.pdf (accessed 01 Sep 2015).
This article is protected by copyright. All rights reserved.
Accepted Article
Bider M. 1954. Statistische Untersuchungen über die Hagelhäufigkeit in der Schweiz und ihre
Beziehungen zur Großwetterlage. Archiv für Meteorologie, Geophysik und Bioklimatologie,
6: 66-90. (only in German)
Bryan GH, Wyngaard JC, Fritsch JM. 2003. Resolution requirements for the simulation of
deep moist convection. Mon. Wea. Rev. 131 : 2394 – 2416.
Changnon SA. 1978. The climatology of hail in the United States. Hail: A Review of Hail
Science and Hail Suppression, Meteor. Monogr: Amer. Meteor. Soc., 38 : 107–128.
Changnon SA, Changnon D. 2000. Long-Term Fluctuations in Hail Incidences in the United
States. J. Climate. 13: 658 – 664.
Changnon D, Changnon SA, Changnon SS. 2001. A method for estimating crop losses from
hail in uninsured periods and regions. J. Appl. Meteor. 40: 84 – 91.
Cintineo JL, Smith TM, Lakshmanan V, Brooks HE, Ortega KL. 2012. An objective high
resolution hail climatology of the contiguous United States. Wea. Forecasting. 27: 1235–
1248.
Collier CG, Lilley RBE. 1994. Forecasting thunderstorm initiation in north-west Europe using
thermodynamic indices, satellite and radar data. Meteorol. Appl. 1: 74–84.
Davolio S, Buzzi A, Malguzzi P. 2009. Orographic triggering of long lived convection in
three dimensions. Meteorol. Atmos. Phys. 103: 35–44.
This article is protected by copyright. All rights reserved.
Accepted Article
Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S, Andrae U, Balmaseda
MA, Balsamo G, Bauer P, Bechtold P, Beljaars ACM, van de Berg L, Bidlot J, Bormann N,
Delsol C, Dragani R, Fuentes M, Geer AJ, Haimberger L, Healy SB, Hersbach H, Hólm EV,
Isaksen L, Kållberg P, Köhler M, Matricardi M, McNally AP, Monge-Sanz BM, Morcrette J-
J, Park B-K, Peubey C, de Rosnay P, Tavolato C, Thépaut J-N, Vitart F. 2011. The ERA-
Interim reanalysis: configuration and performance of the data assimilation system. Q.J.R.
Meteorol. Soc., 137: 553–597. doi: 10.1002/qj.828
van Delden A. 2001. The synoptic setting of thunderstorms in western Europe. Atmos. Res.,
56: 89 – 110.
Delobbe L, Dehenauw D, Hamid K, Neméghaire J. 2003. Hail detection using radar
observations: case studies in the summer 2002. Scientific and Technical Publication Royal
Meteorological Institute of Belgium Nr. 029, 39 pp
Delobbe L, Holleman I, Dehenauw D, Neméghaire J. 2005. Verification of radar-based hail
detection products. In Proceedings of the W.W.R.P. Symposium on Nowcasting and Very
Short Range Forecasting (WSN05), 5-9 September 2005, Toulouse, France, P8.07.
Delobbe L, Holleman I. 2006. Uncertainties in radar echo top heights used for hail detection,
Meteor. Appl.. 13: 361– 374.
Dessens J. 1986. Hail in Southwestern France: I: Hailfall Characteristics and Hailstorm
Environment. J. Climate Appl. Meteor. 25: 35–47.
Dessens J, Fraile R. 1994. Hailstone size distributions in southwestern France. Atmos. Res.,
33: 57–73.
This article is protected by copyright. All rights reserved.
Accepted Article
Dessens J, Fraile R, Pont V, Sánchez JL. 2001. Day-of-the week variability of hail in
southwestern France. Atmos. Res. 59-60: 63 – 76.
Dessens J, Berthet C, Sánchez JL. 2009. Yearly fluctuation of hail precipitation in France. In
Proceeding of 5th European Conference on Severe Storms, 12-16 October, Landshut,
Germany.
Dixon M, Wiener G. 1993. TITAN: Thunderstorm identification, tracking, analysis, and