Downloaded 09/27/21 03:50 AM UTC 3902 MONTHLY WEATHER REVIEW VOLUME 147

NOVEMBER 2019 P Ú CIK ET AL. 3901

Large Hail Incidence and Its Economic and Societal Impacts across Europe

TOMÁS PÚCIK ,CHRISTOPHER CASTELLANO,PIETER GROENEMEIJER, AND THILO KÜHNE European Severe Storms Laboratory, Wessling, Germany

ANJA T. RÄDLER Georisk Department, Munich RE, Munich, and European Severe Storms Laboratory, Wessling, Germany

BOGDAN ANTONESCU National Institute of Research and Development for Optoelectronics INOE2000, Magurele, Romania, and European Severe Storms Laboratory, Wessling, Germany

EBERHARD FAUST Georisk Department, Munich RE, Munich, Germany

(Manuscript received 20 June 2019, in ﬁnal form 15 August 2019)

ABSTRACT

By 31 December 2018, 39 537 quality-controlled reports of large hail had been submitted to the European Severe Weather Database (ESWD) by volunteers and ESSL. This dataset and the NatCatSERVICE Data- base of Munich RE jointly allowed us to study the hail hazard and its impacts across Europe over a period spanning multiple decades. We present a spatiotemporal climatology of the ESWD reports, diurnal and annual cycles of large hail, and indicate where and how they may be affected by reporting biases across Europe. We also discuss which hailstorms caused the most injuries and present the only case with hail fatalities in recent times. Additionally, we address our ﬁndings on the relation between hail size to the type of impacts that were reported. For instance, the probability of reported hail damage to roofs, windows, and vehicles strongly increases as hail size exceeds 5 cm, while damage to crops, trees, and greenhouses is typically reported with hailstone diameters of 2–3 cm. Injuries to humans are usually reported with hail 4 cm in diameter and larger, and number of injuries increases with increasing hail size. Using the NatCatSERVICE data, we studied economic losses associated with hailstorms occurring in central Europe and looked for long-term changes. The trend in hail losses and the annual number of hail loss days since 1990 to 2018 are compared to that of meteorological conditions favorable for large hail as identiﬁed by ESSL’s Additive Regression Convective Hazards model. Both hail loss days and favorable environments show an upward trend, in particular since 2000.

1. Introduction Brown et al. 2015), and Australia (Yeo et al. 1999; Schuster et al. 2005). While small hail (i.e., less than 2 cm Hail annually causes billions of dollars [U.S. dollars in diameter) can severely impact agriculture (Changnon (USD)] in damage worldwide to crops and property et al. 2009), it rarely causes damage to property. To combined. Damage from individual hailstorms can ex- address the risk hail poses to property, it is necessary to ceed $1 billion USD, as shown by individual cases from look speciﬁcally at the occurrence of larger hail. Un- Europe (Höller and Reinhardt 1986; Kunz et al. 2018), fortunately, there is a lack of temporally and spatially the United States (Changnon and Burroughs 2003; homogeneous records of large hail over Europe. Most climatological studies about hail over Europe (see the Denotes content that is immediately available upon publica- recent review by Punge and Kunz 2016) do not concern tion as open access. large hail, with exceptions of Dessens (1986) for France, Burcea et al. (2016) for Romania, Tuovinen et al. (2009) Corresponding author: TomásPúcik, [email protected] for Finland, and Kahraman et al. (2016) for Turkey.

DOI: 10.1175/MWR-D-19-0204.1 Ó 2019 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses). Unauthenticated | Downloaded 09/27/21 03:50 AM UTC 3902 MONTHLY WEATHER REVIEW VOLUME 147

This is because European weather stations typically do not between the radar-derived hail kinetic energy and the record hail sizes, unlike in China, where the hail size is property losses. This metric is not a reliable indicator of observed at all stations (Li et al. 2018). Hailpad station the observed hail size, as it depends on the concentration networks, often installed in hail-prone areas such as of hail stones, which tends to get smaller for large hail northern Spain, southern France, or northern Italy, have (Cheng and English 1983). Hail size observations are facilitated analyses of the spectrum of observed hail thus still necessary to complement the radar data in sizes (Fraile et al. 1992; Fraile et al. 2003; Sánchez et al. these types of studies. 2009; Palencia et al. 2010), but they often miss the Studies of inflicted damage with respect to observed largest hailstones due to the gaps between the hail pads or radar-derived hail sizes have been done either by (Changnon 1977; Smith and Waldvogel 1989). detailed investigations of individual hail events (Brown Besides weather stations and hailpad networks, ob- et al. 2015) or in a laboratory setting (Marshall et al. servations of large hail can be collected from various 2002). There is a lack of studies on the relationship be- sources, such as voluntary observers, storm spotters, or tween hail size and damage covering multiple cases or media. Collecting such information has been standard multiple years of data, except for crops (Sánchez et al. practice since the 1950s in the United States (Tippett 1996; Changnon et al. 2009). The Institute for Business et al. 2015; Allen and Tippett 2015) and Australia (Allen and Home Safety is currently using hail observations and Allen 2016), though questions have been raised re- from the Community Collaborative Rain, Hail and Snow garding the lack of homogeneity of these datasets (Allen Network (CoCoRaHS) to create such a dataset in the and Tippett 2015). In Europe, the European Severe United States (Giammanco et al. 2014; Reges et al. 2016). Weather Database (Dotzek et al. 2009; Groenemeijer In addition to structural damage, very large hailstones et al. 2017), established in 2006 and managed by the ($5 cm) may cause injury or death to animals or hu- European Severe Storms Laboratory (ESSL), collects mans. Individual cases of injuries or fatalities of people reports of large hail. have been reported in the literature (Calianese et al. Remote sensing provides more spatially homogeneous 2002), including large death tolls for historical hail cases observations of (severe) convective hazards than storm in southern Asia (Cerveny et al. 2017). A comprehensive spotter reports. Hail size metrics derived from the vertical assessment of the injuries or deaths caused by large hail profiles of radar reflectivity have been used to develop has not been published so far for any region. climatologies of large hail in the United States (Cintineo This paper uses archived hail reports, hail loss in- et al. 2012) and the Alpine region (Kunz and Puskeiler surance data, and a probabilistic hail model to fill 2010; Nisi et al. 2016; Nisi et al. 2018). However, several existing gaps in the knowledge of 1) large hail incidence, studies have pointed out limitations of radar in inferring 2) the societal and economic impacts of large hail across hail size (Edwards and Thompson 1998; Cicaetal.2015 ; Europe, and 3) the relationship between hail size and Ortega 2018). Satellites cover larger areas than the radar damage. Section 2 describes the data and methodology observations (Cecil and Blankenship 2012), but the re- used to analyze hail occurrence and its impacts, section 3 sults tend to overestimate hail occurrence in tropics discusses the spatiotemporal characteristics of large hail, (Allen et al. 2015; Ferraro et al. 2015). Observations section 4 deals with the relationship of hail size with the of overshooting tops as indicators of storm severity impacts and section 5 with the injuries and economic were used to develop hail models over Europe (Punge losses caused by hail. Section 6 briefly summarizes the et al. 2017) and Australia (Bedka et al. 2018). In ad- main results. dition, the frequency of large hail on global (Prein and Holland 2018), continental (Rädler et al. 2018), and 2. Data national (Allen et al. 2015) scales has been estimated by identifying favorable large-scale environments in Quality controlled large hail reports were obtained reanalysis datasets. from the ESWD. The quality control system was de- Due to the large economic losses hail can cause, the scribed in detail by Dotzek et al. (2009) and Groenemeijer relationship between inflicted damage and hailstorm et al. (2017). Only the reports that have passed the plau- intensity is another branch of hail-related research, sibility check were included for further analysis, com- which can help in terms of increasing structural resil- prising a total of 39 537 reports by 31 December 2018 ience. A frequently used measure for hailstorm intensity (including the historical reports entered in the database is hail kinetic energy, which can be derived from both retrospectively). Each report of large hail contains infor- radar and hailpad data (Waldvogel et al. 1978; Hohl mation on event location, location accuracy, event time, et al. 2002; Sánchez et al. 2013). However, Schuster et al. event time accuracy and the source of the information. (2006) and Brown et al. (2015) noted weak correlation Optionally, a number of fields can be filled out, such as

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC NOVEMBER 2019 P Ú CIK ET AL. 3903 maximum hail diameter, hail weight, depth of a hail layer, number of injured/killed people or a description of the event. Unlike Storm Data in the United States, maximum hail diameter is not a mandatory field in the ESWD. In this paper we will refer to large hail as hail with diameter $2cm, very large hail as hail with diameter $5cm and giant hail as hail with diameter $10 cm (Blair et al. 2011). Information about the damage inflicted by hail was acquired from the description field of the report. To match the particular hail sizes with specific types of damage, we searched for object keywords contained in the description field such as ‘‘vegetation,’’ ‘‘vehicle,’’ ‘‘window,’’ ‘‘wind- shield,’’ ‘‘roof,’’ or ‘‘greenhouse’’ and then checked for the damage description associated with that object. The descriptions submitted in languages other than English FIG. 1. Number of large (blue line) and very large (red line) hail reports between 1990 and 2018. were not taken into consideration to prevent the incorrect interpretation. The fraction of entries in the description the probability that a thunderstorm occurs and the prob- field in other languages was 4.5%. Such entries were ability of the severe hazard, given the storm is present. typically entered in German or French and pertained Annual losses and annual number of hail loss days were historical reports from before 2006. compared to the probability of large and very large hail Data on the hail losses were acquired from the based on the ARCHaMo applied to the ERA-Interim NatCatSERVICE database of Munich RE (Kron et al. dataset. The annual frequency was calculated as a sum of 2012), which contains information from more than the 6-hourly probabilities of large and very large hail, 30 000 datasets. Each loss entry includes its type, lo- averaged over a specific area in the 0.75830.758 latitude cation, date, monetary losses, and the description. We and longitude grid in the ERA-Interim dataset. considered only events, where either the type or the description of the event suggested that the loss was at least partly caused by hail. Individual loss events often 3. Spatiotemporal climatology of large hail reports spanned multiple days and comprised multiple severe according to the ESWD convective storms. In such cases, the coordinates of a. Number of (very) large hail days across Europe the loss location point to the place of the greatest loss during the event. When calculating the number of The annual number of large and very large hail re- hail loss days per year, we took into consideration the ports (Fig. 1) shows a steep increase in the 30-yr period whole duration of the loss event. Due to inflation, ur- between 1989 and 2018. Since the ESSL was founded in ban growth and regional wealth differences, we used 2006, the annual number of ESWD large hail reports has the estimated overall losses normalized to 2018 levels tripled, from 750 reports in 2006 to 2400 reports in 2018. of destructible wealth. Normalization is done by con- Since 2012, the number of reports has consistently ex- sidering the Gross Domestic Product discretized on a ceeded 2000, whereas prior to 2000 it never reached 100. 18318 latitude and longitude grid. Kron et al. (2012) A similar pattern is visible for very large hail reports. state that the accuracy of the loss estimates in the The trend in the number of hail reports reflects an in- NatCatSERVICE depends on the insurance penetration crease in the reporting rate due to the increased co- in a given country for a specific hazard. In Europe, the operation between the ESSL and voluntary observer best insurance penetration is in Germany, BENELUX, networks or national (hydro)meteorological institutes Austria, and Switzerland. across Europe. Thus, reports cannot be used to investi- To match the hail-related losses and number of gate the trend in the large hail occurrence, similar to the hail loss days to the frequency of favorable environ- record of large hail in Storm Data for the United States ments for damaging hailstorms, we used the additive (Allen and Tippett 2015). regression hazard model ARCHaMo (Rädler et al. 2018). To perform spatial analysis, we selected reports in the ARCHaMo was developed for central Europe using period from 2006 onward, when a coordinated effort to ERA-Interim (Dee et al. 2011), EUCLID lightning detec- collect reports across the whole of Europe started. Large tion network (Schulz et al. 2016), and the ESWD dataset hail is most frequently reported in the pre-Alpine areas between 2008 and 2016. The model predicts the proba- of central Europe and northern Italy, with up to 4 large bility of a severe hazard as a product of two components: hail days per year (Fig. 2a). A large hail day is defined

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC 3904 MONTHLY WEATHER REVIEW VOLUME 147

FIG. 2. Mean annual number of (a) large and (b) very large days in the 0.58 latitude 3 0.58 longitude grid, smoothed by a Gaussian filter. as a 24-h period, between 0000 and 0000 UTC, with at Spain, southern France and parts of the Balkan peninsula, least one report of large hail. Reports were counted in as they do over southern Germany or southeastern a grid spanning 0.58 longitude 3 0.58 latitude. Central Austria. The discrepancy between the frequency of Europe has the smallest reporting bias and thus also favorable environments and the ESWD reports has shows the highest frequency of large hail days, com- also been noted by Taszarek et al. (2019),andthe pared to southern or eastern Europe (Groenemeijer and observations of overshooting tops (Punge et al. 2017) Kühne 2014). Outside central Europe, local maxima can associated with convective storms corroborate the be found over southern Bulgaria, the Caucasus moun- negative bias of ESWD outside the central Europe. tains, and the Moscow metropolitan area. Very large Hailpad network data (Fraile et al. 2003) show that hail hail occurs less frequently, but the location of maxima larger than 3 cm in diameter occurs every year somewhere shows similar patterns to large hail with the highest in- in the hail pad network of southern France, spanning 811 cidence over southeastern Austria, where very large hail stations over 34 000 km2, again suggesting underreporting is reported almost every second year (Fig. 2b). A local in the current state of ESWD. On the other hand, hailpad maximum between the Black Forest and Swabian Jura data tend to underestimate the largest hail sizes (Fraile in southwestern Germany was also detected by Kunz et al. 2003). For example, the hailpad network over and Puskeiler (2010). Other maxima were found over northeastern Italy collected a maximum hail size of the Po Valley in Italy, over Ore mountains on the border 3.5 cm over an 11-yr period (Palencia et al. 2010), while of the Czech Republic and Germany, and Silesia on the ESWD reports suggest that the area experienced very border of the Czech Republic and Poland. These areas large hail on average at least once every 5 years. Gridding experience on average at least one day with very large the dataset to 0.58 of latitude and longitude and smooth- hail every 5 years. ing with a Gaussian filter has decreased the strong gra- The locations of hail maxima based on studies from dients in the large hail activity between the main Alpine individual countries (i.e., France, Germany, Poland, ridge and pre-Alpine areas. This strong gradient in large Austria, Croatia, or Serbia) shown in Punge and Kunz hail activity between the mountains and premountain (2016) coincide with the maximabasedontheESWD areas was found using the radar data over Switzerland reports. Underreporting in the areas outside of central (Nisi et al. 2016; Nisi et al. 2018). Europe do not allow for a country to country compari- b. Annual cycle son of large hail incidence. Furthermore, because most hail climatologies listed in Punge and Kunz (2016) did Peak month was calculated similar to Groenemeijer not focus exclusively on large hail, a quantitative com- and Kühne (2014), over a grid of 28 of latitude and parison to our study is difficult. Rädler et al. (2018), longitude and a running 3-month mean. Large hail is Taszarek et al. (2018) and Púcik et al. (2017) show reported most frequently in the summer months across that environmental conditions supporting severe thun- much of Europe and the frequency peaks in June over derstorms (or large hail) occur as frequently over eastern three quarters of the continent (Fig. 3). Frequency peaks

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC NOVEMBER 2019 P Ú CIK ET AL. 3905

heating, while the coastal maritime areas experience their peak in the autumn to winter months as the Atlantic subtropical ridge shifts southward and deep low pressure systems develop over the Mediterranean Sea. Splitting Europe into two parts, south and north of the 468 latitude, shows some differences in the annual cycle of large hail incidence between the two regions (Fig. 4). The 468 latitude was chosen as it is the ﬁrst latitude to be completely north of the Mediterranean Sea. Both the southern and northern part experience a rapid increase in large hail activity between April and May with the activity peaking around June. The most pronounced difference is the activity in July, which remained on the level of June across northern Europe but decreased markedly over southern Europe by almost 50%. This can be attributed to the migration of the polar frontal zone, separating polar and tropical air mass, northward and the dominance of the subtropical high pressure system inhibit- FIG. 3. Month with most frequent large hail occurrence calculated over 28 latitude 3 28 longitude grid. Grid points with less than ing convection over southern Europe. Another difference is 5 events are not displayed. that the hail activity remained more frequent in the cooler months over southern part of Europe. The annual cycle of very large hail (not shown) is similar to the activity of in July over northeastern Spain, southeastern France, large hail, but the peak month shifts from June to July over northern Italy, United Kingdom, parts of Estonia, or northeastern Spain and parts of central Europe. This shift Russia. Slightly more variation can be found over south- has also been noted by Punge et al. (2017). ern Europe with large hail occurring most frequently in The annual cycle of large hail does not exactly match September over Corsica, February over Crete and the the annual cycle of thunderstorms or other severe southern Ionian Sea, and in May over southern Greece, convective storm phenomena, such as tornadoes. Tor- southern Turkey, and Cyprus. Continental areas experi- nado activity peaks one month later than large hail, in ence the peak thunderstorm season in months with peak July, over continental Europe. Over the Mediterranean,

FIG. 4. The annual cycle of large hail over Europe north of 468N latitude (blue bars) and south of 468N latitude (red bars).

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC 3906 MONTHLY WEATHER REVIEW VOLUME 147

FIG. 5. Diurnal cycle of large hail over Europe for all reports (blue bars) and for hail reports with time accuracy higher than one hour (red bars). tornadoes occur most frequently between August to for subsequent analysis of peak hourly occurrence of October, including the inland areas (Groenemeijer and large hail over Europe. Kühne 2014), where hail typically occurred in May or Large hail typically occurs in the afternoon over most June. This discrepancy has already been addressed by of Europe (Fig. 6). Due to the shift in time of the max- Antonescu et al. (2016) and can be noted when com- imum diurnal heating, the peak hour was later over paring the climatologies of tornadoes and large hail over western Europe compared to eastern Europe, which is Turkey (Kahraman and Markowski 2014; Kahraman corroborated by the findings of Punge and Kunz (2016). et al. 2016). Taszarek et al. (2019) noted two peaks of For example, the peak time of occurrence was between thunderstorm activity over the Mediterranean area: one in late spring, and the other in autumn (particularly near the coast and over the sea). This suggests that the large- scale environment in which storms form over this area must differ between the two peak seasons, as autumn activity produces large hail less frequently than spring. c. Diurnal cycle The diurnal cycle of all large hail reports (Fig. 5) over Europe shows three distinct peaks at 0000, 1200, and 1500 UTC. Because the ESWD reports also contain the information on the time accuracy of the reports, subsequent filtering for reports that have an accuracy within one hour yielded only one peak of 1500 UTC. The peaks at 0000 and 1200 UTC represent the events that occurred at night and day, with exact time unknown (time accuracy between 6 and 24 h), rather than the true peaks in the large hail occurrence. It is important to take time accuracy into consideration when performing time sensitive studies. For instance, only 37.71% of 39 531 FIG. 6. Peak hour (in UTC) of large hail activity on a 28 latitude 3 large hail reports used in this study had time accuracy 28 longitude grid. Grid points with less than 5 events are not higher than one hour. These filtered reports are used displayed.

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC NOVEMBER 2019 P Ú CIK ET AL. 3907

FIG. 8. Very large hail and giant hail reports over Europe. Giant FIG. 7. Cumulative relative frequency of hail sizes sorted into hail reports occurring before (after) 1 Jan 2006 are denoted by bins of 1–1.9 cm, 2–2.9 cm up to 9–9.9 cm category. Hail sizes of orange (red) triangles, while very large hail corresponds to the 101 cm were not considered here. small blue triangles.

1600 and 1800 UTC over much of France and Spain, decrease in the relative frequency of very large hail compared to 1300–1700 UTC over much of central sizes. Between 2005 and 2014, very large hail repre- Europe. However, differences also existed within the sented 2%–3% of all large hail reports, less than half western and eastern half of Europe. The peak time of of the relative frequency found for Europe. occurrence for Great Britain was two hours earlier Giant hail is very rare but has been observed across compared to France, suggesting the differences in the many areas over Europe (Fig. 8). Hail of such size was usual time that convection is initiated over these areas. reported 91 times in the database, 49 times before and The same was true for the Alpine region, as large hail 42 times after 1 January 2006. Overall, giant hail en- occurred most frequently between 1200 and 1500 UTC compassed 0.38% of the reports where a hail size was north of the ridge and between 1600 and 1800 UTC provided. Most of the pre-2006 reports of giant hail south of the ridge. The diurnal cycle of very large hail came from Germany, Czechia and northern Italy. Since (not shown) was similar to large hail, but the peak was 2006, giant hail has been reported from regions north of shifted by 1–2 h toward the evening over central Europe the Caucasus mountains, surrounding the Alps, and and the Balkan Peninsula. from Serbia and Romania. The largest reported hail sizes in recent years are 15 cm on 20 June 2016 at d. Hail size distribution Sanandrei^ in western Romania and 14.1 cm on 6 August Analyses of hail size distributions have been performed 2013 at Undingen in southwestern Germany. using hailpad networks (Sánchez et al. 2009), observa- e. Other measures of hail severity tions from manned stations (Li et al. 2018)orfromthe Storm Data (Allen et al. 2017). Because the ESWD col- Besides the maximum hail size, which is by far the lects severe hail events, most of the hail sizes involve hail most common characteristic submitted to the database, over 2 cm in diameter. 60.79% of large hail reports sub- information on the average hail size, maximum hail mitted to ESWD include information on maximum hail weight or the thickness of the hail cover was included diameter. The remaining reports contain only informa- in some of the reports. Average hail size was present tion about the hail damage or mention that large hail did in 8.55% of the reports and was discontinued from the occur, but size could not be estimated. The hail size of ESWD in 2017. Thickness of the hail cover can still be 2–2.9 cm is the most common hail size category reported, reported to the ESWD, and hail cover reaching over comprising almost 40% of reports, where hail size was 2 cm on the flat surface is considered severe. A total of given (Fig. 7). Only 7% of these reports included maxi- 5.11% of the submitted reports involved presence of mum hail size larger than 5 cm and less than 1% of reports deep hail cover. Significant hail cover can cause major involved hail equal or larger than 8 cm in diameter. Allen problems in traffic and may increase the flash flood and Tippett (2015) analyzed the changes in the hail size threat. For example, during the severe hailstorm in distribution in Storm Data since 1955 and found a Stuttgart on 15 August 1972, hail accumulations on

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC 3908 MONTHLY WEATHER REVIEW VOLUME 147

TABLE 1. Number of reports containing description of damage to different types of objects and with speciﬁed hail size.

Hail size (cm) Farmland gardens Trees Greenhouses Roofs House windows ,2182 0 0 0 2–2.9 326 90 5 1 0 3–3.9 584 277 40 59 0 4–4.9 392 193 16 103 27 5–5.9 158 47 8 143 47 6–6.9 53 12 9 88 31 7–7.9 37 15 8 56 25 8–8.9 7 4 1 25 15 9–9.9 3 2 2 15 7 .10 6 5 2 22 14 Total 1584 647 91 512 166 Hail size (cm) Vehicle body Vehicle windows Small animals killed Large animals killed People injured ,200 0 0 0 2–2.9 17 0 0 0 3 3–3.9 50 0 3 0 5 4–4.9 64 0 2 1 5 5–5.9 104 7 2 4 14 6–6.9 58 19 2 4 8 7–7.9 54 39 4 3 8 8–8.9 19 16 5 5 7 9–9.9 11 12 0 1 1 .10 19 17 0 1 14 Total 396 110 18 19 65

the order of tens of centimeters clogged the drainage These data can be used to study the relationship system and the subsequent flooding killed 6 people. between the hail sizes and resulting damage. The first While hailstorms producing significant hail accumu- question that is possible to answer is ‘‘At what hail lations receive less attention in the literature than the sizes does a particular type of damage typically occur?’’ hailstorms including very large hail sizes, first efforts Damage to vegetation (crops and trees) occurred most have been undertaken to address this issue (Kalina frequently with a hail size of 3 cm (Fig. 9). Damage to et al. 2016; Kumjian et al. 2019). Hailstone weight was vehicle bodies, house windows and roofs was most included only in 0.6% of all severe hail reports sub- common with a hail size of 5 cm, and to vehicle win- mitted to the ESWD. This metric was more typically dows with a hail size of 7 cm. In contrast to other types used for historical records as the ratio of reports of damage, car window damage was only damage cat- containing hail weight dropped to 0.13% after 2006. egory to occur almost exclusively with hail exceeding 5 cm. Injuries were reported most frequently with hail size of 5 cm, but some cases involved smaller hail, 4. Comparison of hail size to inflicted damage including a few cases where the reported diameter was TheESWDallowsforadescriptionofthetypeof only 2 cm. In these cases, other factors, such as the hail damage. While the type of hail damage is not de- strong wind, may have exacerbated the hail impact. scribed for every single report, it is possible to use this The second question that we can pose is: ‘‘Given the information to analyze their association with a par- hail size, what is the probability that particular type of ticular hail size. The number of instances when damage damage will be reported?’’ Because the frequency of of particular type was reported along with the given hail hail decreases with increasing size (Fig. 7), the hail size size is shown in Table 1. Damage to property was re- which corresponds to the highest probability of impact ported less frequently than damage to vegetation. For maynotcoincidewiththehailsizeatwhichtheimpact example, damage to crops was mentioned 1584 times is most frequently observed. The probability of dam- whereas damage to house windows was mentioned age occurrence climbed steadily toward the largest hail 166 times. Animal and human casualties were reported sizes for most types of damage, except for crops, trees less frequently than other types of damage. Human and greenhouses that can be damaged by smaller hail injuries were reported 65 times and animal deaths were sizes that dominate the dataset (Fig. 10). Giant hail had reported 37 times. the highest likelihood to damage roofs (34%), windows

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC NOVEMBER 2019 P Ú CIK ET AL. 3909

FIG. 9. Relative frequency of impact type as a function of maximum hail diameter.

(22%), car bodies (30%), car windshields (27%), and For example, Marshall et al. (2002) found that most also to injure humans (22%). The values of relative roof types suffered almost complete destruction when frequency and conditional probability of impact given impacted by spherical hail stones with 5 cm diame- the hail size may be biased due to a number of reporting ter in a laboratory setting. Yet, conditional probability issues. First, observers do not consistently report dam- of roof damage by 5 cm hail is only 10.4%. Yeo et al. age even if it occurs. Second, they may feel more moti- (1999) also found higher probability of damage oc- vated to report the damage with increasing hail size currence for the Sydney hailstorm of 1999. 80% of because it is less common and is more impressive. Third, roofs were damaged by hail between 5 and 7 cm and hail reports often do not sample the largest hailstone. all roofs were damaged by hail exceeding 7 cm in di- Blair et al. (2017) compared the hail size reports in the ameter. Broken windshields in vehicles were reported Storm Data to their in situ high-resolution measure- in 30% and 40% of cases for 5–7 and 71 cm hail, comments and found that the largest hail size per hail event pared to 10% and 20% in our study. While such studies in Storm Data was on average negatively biased by 2 cm. likely represent a more accurate account of the rela- Due to the damage reporting biases mentioned above, tionship between hail size and impact compared to our real conditional probabilities of hail damage for a par- study, it is based on a single case. Furthermore, rooﬁng ticular hail size are likely higher than presented here. materials used in the United States and Australia are

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC 3910 MONTHLY WEATHER REVIEW VOLUME 147

5. Injuries and economic losses associated with hailstorms a. Injuries Injuries associated with large hail were reported 96 times. The severity or the circumstances of such injuries is rarely mentioned in the reports, but the text description most typically mentions ‘‘light injuries,’’ ‘‘head injuries,’’ or ‘‘persons taken to hospital.’’ Serious injuries were reported only once, for a hailstorm event in Marbella and Malaga, southern Spain, which occurred in the morning hours of 21 September 2007. The 8–10-cm large hail injured 30 people and three persons were rushed to the hospital with serious head injuries. Re- ports with injuries are concentrated in areas with a high incidence of very large hail, such as southern Germany, northern Italy, the Balkans and southern Russia (Fig. 11a). Surprisingly, almost no injuries were reported from southern Poland, where the frequency of very large hail is also high. On the other hand, injuries were also reported in areas where the ESWD suggests that very large hail is infrequent. This difference may be related either to varying degree of vulnerability to large hail or to how media report on weather-related injuries across Europe. Hailstorms with larger hail sizes cause more injuries. For three hail size categories (2–5, 5–8, and 81 cm), the FIG. 10. Conditional probability of different types of hail damage as a function of maximum hail diameter for living and nonliving mean (median) value of the number of injuries per re- objects. port, given that at least one injury occurred, was 1.77 (1), 4.37 (2), and 38.79 (10), respectively. Number of hail injury reports associated with the three hail size cate- different from Europe, which does not allow a direct gories was 13, 30, and 22. There were 17 reports with 10 comparison with our results. or more injuries and 9 reports with at least 20 injuries. Hailstone weight would likely yield better corre- The 10 events with most injuries are listed in Table 2. spondence to the potential hail impact than hail di- The hailstorm with the highest number of injuries ameter alone, as it can be converted to the kinetic was a hailstorm in Munich, Germany, on 12 July 1984 energy of the hailstone upon its impact. Hail weight with 400 injuries, followed by a hailstorm in Reutlin- can be inferred from the diameter, assuming a perfect gen, Germany, on 28 July 2013 that caused 74 injuries. sphere, but ﬁeld observations show that large hail- Injuries related to hail can also be found in the Storm stones are irregular in shape (Knight and Knight 2005; Data for the United States. As the database was estab- Giammanco et al. 2014; Giammanco et al. 2017). For lished in 1955, the number of hail events with injuries example, Knight and Knight (2005) documented very (291) is higher than in Europe (96). Most hail-related large hailstone measuring 18 cm across, but weighing injuries were concentrated in the Midwest, east of the only around 500 g in contrast to the expected weight Rocky Mountains (Fig. 11b), where very large hail oc- of 2500 g if the hailstone was a perfect sphere. There- curs most frequently (Cintineo et al. 2012; Allen and fore, for a given hail size, large range of kinetic energies Tippett 2015; Allen et al. 2017). The highest number of is possible, depending on the shape and the density of injuries in a single storm event (109) was associated the hailstone (Heymsﬁeld et al. 2014) Large, spongy hail with a hailstorm in Fort Worth, Texas, on 5 May 1995. may produce less damage upon impact than denser, but Similar to Europe, larger hail sizes were associated with smaller hail. Laboratory tests (Marshall et al. 2002), more injuries. The mean (median) number of injuries performed with the perfectly spherical hailstone, then per report, given at least one injury, was 2.72 (1) for 2–5, provide the upper-end damage estimate that hail can 3.59 (1) for 5–8, and 8.34 (2) for 81 cm hail sizes. cause upon impact, as natural hailstones with the same Number of hail injury reports associated with the three diameter will typically weigh less. hail size categories was 132, 115, and 44. The higher

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC NOVEMBER 2019 P Ú CIK ET AL. 3911

FIG. 11. Hail-related injuries over (a) Europe, based on ESWD, and (b) the United States, based on Storm Data. Color and size of the triangle correspond to the number of injuries. average number of injuries per hail injury event in 669 losses involving hail with a normalized value ex- Europe may result from the higher population den- ceeding $1 million (USD) between 1980 and 2018 across sity over hail-prone areas in Europe compared to Europe. The majority of those loss events were located the United States. within Germany, Austria and Switzerland (Fig. 12). While many injuries associated with hail were re- In particular, urban centers of southern and western ported to the ESWD, deaths directly caused by hail were Germany had a high incidence of normalized hail extremely rare with only one such event since 1990. losses, including several events with loss exceeding Allegedly, 4 people were killed by hail on 20 June 1997 $1 billion (USD). Six out of the top 10 normalized in Apele Vii, Dolj, Romania (Marinica 2003). The same losses (Table 3)occurredinGermany,includingthe event also ranks as the hailstorm with third most in- Munich hailstorm of 12 July 1984 and the hailstorms juries. Witnesses of the event mention giant hail size of 27–28 July 2013. Both events also rank as the hail- (ostrich egg), but no photographs have been found to storms with most injuries (Table 2). The 2013 event loss conﬁrm the actual hail size. A number of indirect deaths comprises losses from two separate damaging hail- due to hail have been reported with other hail events, storms occurring on consecutive days ﬁrst over northern such as road accidents due to slippery road conditions or and then over southwestern Germany. All events deaths during the repair of damaged roofs. listed in Table 3 involved hail exceeding 5 cm and 5 of the top 10 loss events included hail reaching 10 cm b. Financial losses in diameter. Besides harm to persons, hail has a large monetary Comparing the distribution of very large hail re- impact across Europe. NatCatSERVICE database lists ports (Fig. 8) to the distribution of loss events (Fig. 12)

TABLE 2. Ten hailstorms with most injuries across Europe.

Date Country Location Number of injuries Maximum hail size (cm) 12 Jul 1984 Germany Munich 400 10 28 Jul 2013 Germany Reutlingen 74 8 20 Jun 1997 Romania Apele Vii 60 13a 8 Jul 2014 Bulgaria Soﬁa 40 10 23 Jul 1988 France Torcy 33b Not given 22 Jun 2013 Serbia Indija 30 7 21 Sep 2007 Spain Marbella 30 10 4 Aug 2002 Italy Bardolino 20 Not givenc 28 Jun 2005 Germany Fridolﬁng 17 8 20 Jun 2016 Serbia Pancevo 15 9 a Ostrich egg size mentioned, but not corroborated by other reports. b Injuries caused by roof collapsing under the weight of the hail. c Hail weight up to 700 g mentioned.

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC 3912 MONTHLY WEATHER REVIEW VOLUME 147

FIG. 12. Hail-related losses exceeding $1 million (USD) from 1980 to 2018. suggests that the loss database likely underestimates occurred on 8 July 2014 and the Istanbul hailstorm number of events in countries where the incidence of on 28 July 2017. very large hail events is similar to southern Germany, c. Trend in hail losses and its relation to the changes in but the number of loss events is much smaller. Be- environments capable of large hail sides the underreporting of losses, the density and the value of vulnerable assets differ across Europe. According to the NatCatSERVICE, the majority of Thus, it is likely that database currently underesti- losses associated with hail were reported in Germany. mates the number of loss events over regions such Thus, we focus on Germany for which we assume the as southern France, northern Italy, southern Poland, data are more homogenous compared to other Euro- central Serbia, Bulgaria, Romania, and southern pean countries. Because Germany was divided be- Russia. Over eastern Europe the only loss to exceed tween the German Federal Republic and the German $500 million (USD) is the Soﬁa hailstorm that Democratic Republic, and no reports are available

TABLE 3. Ten highest hail-related loss events across Europe. Location denotes the place that experienced the highest loss during the event.

Date Country Location Loss (in millions USD) Maximum hail size (cm) 27–28 Jul 2013 Germany Reultingen 4689 8 12 Jul 1984 Germany Munich 4330 10 7–10 Jun 2014 France/Belgium ^Ile de France 2301 11 28 May–2 Jun 2008 Germany Krefeld 1987 8 23–24 Jul 2009 Switzerland/Austria Romont 1856 10 22–24 Jun 2016 Netherlands Someren 1783 10 19–21 Jun 2013 Germany Lohmar 1295 7 4 Jul 1994 Germany Köln 1032 5 19–24 Jun 2002 Germany Hochstetten 834 10 11 Jul 1984 France Epinal 782 6

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC NOVEMBER 2019 P Ú CIK ET AL. 3913

FIG. 13. Relation of mean annual probability of hail, calculated for a 0.758 ERA-Interim grid box and averaged over the area of Germany with each grid point having the same weight, to the hail-related annual losses (in millions of USD) for (a) large and (b) very large hail and to the number of hail loss days for (c) large and (d) very large hail across Germany between 1990 and 2018. from the latter, the analysis is confined to the time Besides increasing vulnerability, the frequency of se- period after 1990. Both the number of hail loss events vere storm environments capable of producing large andthenormalizedlossesshowanupwardtrend.The hail has increased since 1990, as demonstrated by the annual number of the hail loss days over Germany has ARCHaMo models developed by Rädler et al. (2018) increased substantially, from around 7.5 days in the early using the ERA-Interim data (Fig. 13). The mean an- 1990 to around 15 days in recent years (Figs. 13c,d). nual probability of large hail per ERA-Interim grid Using Mann–Kendall test (Hussain et al. 2019), the point, averaged over Germany, has almost doubled trend was found to be statistically significant (p , 0.001). since 1990. The mean annual probability of very large The trend in the annual hail losses (Figs. 13a,b)also hail has increased as well. Increasing trend in both seems positive, with the average annual hail loss be- large and very large hail probabilities is statistically tween 1990 and 1999 reaching $527.7 million (USD) significant (p , 0.001). The correlation coefficient and between 2006 and 2018 reaching $1398.24 million between the number of hail loss days and the mean (USD), but was found to be statistically insignificant annual probability of large hail and very large hail rea- (p . 0.05). The time series of hail losses over Germany ches 0.65 and 0.69, respectively. Strong, positive corre- is dominated by two outliers, corresponding to the lation is also found using Spearman’s correlation major hail events of 12 July 1984 and of 27 and 28 July coefficient (0.647 and 0.65, p , 0.001) The correla- 2013, respectively. tion between annual hail losses and mean annual prob- The upward trend of hail loss days and number of ability of large/very large hail was lower than with hail losses can be attributed to changes in exposure, vul- loss days, 0.24 and 0.22, respectively. Spearman’s cor- nerability and hazard. Increasing population in the relation coefficient yields values of 0.405 (p 5 0.029) and major urban centers as well as the increase in the 0.327 (p 5 0.084). Low correlation is also due the strong vulnerability of assets has increased potential hail outlier in 2013, which was not reflected in the corre- losses. This effect is similar to the impact of in- spondingly high frequency of favorable large hail envi- creasing population size on the potential impact of ronments in given year. For the United States, Sander tornado outbreaks that was discussed by Strader et al. et al. (2013) shows that high loss events are correlated (2017), Ashley and Strader (2016) and Antonescu with environments characterized by climatologically et al. (2018). very rare combinations of high CAPE and vertical

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC 3914 MONTHLY WEATHER REVIEW VOLUME 147 wind shear. Over Germany, the years with the highest and 8 losses exceeded $1 billion (USD). Both annual hail losses did not feature correspondingly elevated values losses and annual number of days with hail losses have in- of mean annual probability of large/very large hail. This creased between 1990 and 2018 over Germany. Part of this suggests the highest losses in this region represent un- increase may be attributed to the increase in the exposure, fortunate events of severe storms impacting major urban but it is shown that the large hail risk has also increased areas rather than prolonged presence of high CAPE and based on the ERA-Interim dataset. As the hail losses in- strong vertical wind shear resulting in multiple outbreaks crease over Europe, the future work on hailstorms should of severe weather. The study domain of Sander et al. (2013) further explore the impacts of these events to increase the was larger (United States east of the Rocky Mountains resilience of communities. The authors foresee an update to compared to Germany), so in their case largest losses were this study as the reporting homogeneity improves in future driven by widespread outbreaks of convective storms, and more reports include the speciﬁcation of an impact spanning multiple states, rather than single local events as caused by large hail. in our study. Furthermore, they considered losses from all severe convective storm hazards and did not limit their Acknowledgments. The contributions of TomásPúcik, study only to the losses driven mainly by hail. Christopher Castellano, Pieter Groenemeijer, Anja T. Rädler, and Eberhard Faust were carried out within the project Analysis of Changes in the Risk of Severe con- 6. Conclusions vective Storms in Europe (ARCS), by Munich Re and By 31 December 2018, 39 537 quality-controlled the German Ministry of Education and Research reports of large hail had been submitted to the ESWD (BMBF) under Grant 01LP1525A. The contribution of and the annual number of reports had tripled since Thilo Kühne was carried out as part of the project the founding of the ESSL in 2006. The highest number Severe Thunderstorm Evaluation and Predictability in of annual large hail days were found over the pre-Alpine Climate Models (STEPCLIM), part of the MiKLIP areas including southeastern Austria and southern research program, supported by the German Federal Germany. Large hail activity peaked in June and July Ministry of Education and Research (BMBF) under in the afternoon hours across much of Europe, except Grant 01LP1117A. The contribution of Bogdan Antonescu in several Mediterranean areas, where the peak was was carried under funding from the Romanian STAR shifted to the cooler season. However, due to the tem- Program Ctr. 162/20.07.2017b and from the National poral and spatial inhomogeneities in reporting, it is Core Programe Contract PN 33/2018. The authors would currently not possible to use the database to construct a like to thank three anonymous reviewers for their reliable large hail climatology across Europe. ESSL is comments that helped to improve the manuscript. improving the reporting homogeneity by establishing formal cooperation with national meteorological in- REFERENCES stitutes and voluntary observer networks across Europe. Some of the reports contained description of dam- Allen, J. T., and M. K. Tippett, 2015: The characteristics of United States hail reports: 1955–2014. Electron. J. Severe age caused by large hail. Damage to crops and trees Storms Meteor., 10 (3), http://www.ejssm.org/ojs/index.php/ was reported most frequently with the hail sizes of 2 ejssm/article/viewArticle/149. and 3 cm, while damage to roofs, cars and windows ——, and E. R. Allen, 2016: A review of severe thunderstorms in was typically reported with larger hail size of 4 to 6 cm. Australia. Atmos. Res., 178–179, 347–366, https://doi.org/ The probability of damage to roofs, vehicles and 10.1016/j.atmosres.2016.03.011. ——, M. K. Tippett, and A. H. Sobel, 2015: An empirical model windows steeply increased with increasing hail size, relating US monthly hail occurrence to large-scale meteoro- reaching highest values for giant hail, but remained rel- logical environment. J. Adv. Model. Earth Syst., 7, 226–243, atively constant for damage to vegetation and green- https://doi.org/10.1002/2014MS000397. houses. To streamline the study of impacts reported to ——, ——, Y. Kaheil, A. H. Sobel, C. Lepore, S. Nong, and the ESWD, the online submission form has been sub- A. Muehlbauer, 2017: An extreme value model for U.S. hail size. Mon. Wea. Rev., 145, 4501–4519, https://doi.org/10.1175/ stantially reworked in 2018. Injuries caused by hail were MWR-D-17-0119.1. reported 96 times in the database, compared to the 291 Antonescu, B., D. M. Schultz, F. Lomas, and T. Kühne, 2016: cases over United States listed in Storm Data. Larger Tornadoes in Europe: Synthesis of the observational data- hail sizes were associated with greater number of in- sets. Mon. Wea. Rev., 144, 2445–2480, https://doi.org/10.1175/ juries over both regions. Since 1990, only one case with MWR-D-15-0298.1. ——, J. G. Fairman Jr., and D. M. Schultz, 2018: What is the worst directly caused hail fatalities was reported in Europe. that could happen? Reexamining the 24–25 June 1967 tornado Large economic losses due to the hailstorms have been outbreak over western Europe. Wea. Climate Soc., 10, 323– reported to the NatCatSERVICE database of Munich RE 340, https://doi.org/10.1175/WCAS-D-17-0076.1.

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC NOVEMBER 2019 P Ú CIK ET AL. 3915

Ashley, W. S., and S. M. Strader, 2016: Recipe for disaster: How the Meteor., 25, 35–47, https://doi.org/10.1175/1520-0450(1986) dynamic ingredients of risk and exposure are changing the 025,0035:HISFIH.2.0.CO;2. tornado disaster landscape. Bull. Amer. Meteor. Soc., 97, 767– Dotzek, N., P. Groenemeijer, B. Feuerstein, and A. M. Holzer, 2009: 786, https://doi.org/10.1175/BAMS-D-15-00150.1. Overview of ESSL’s severe convective storms research using the Bedka, K. M., J. T. Allen, H. J. Punge, M. Kunz, and European Severe Weather Database ESWD. Atmos. Res., 93, D. Simanovic, 2018: A long-term overshooting convective 575–586, https://doi.org/10.1016/j.atmosres.2008.10.020. cloud-top detection database over Australia derived from Edwards, R., and R. L. Thompson, 1998: Nationwide com- MTSAT Japanese advanced meteorological imager obser- parisons of hail size with WSR-88D vertically integrated vations. J. Appl. Meteor. Climatol., 57, 937–951, https:// liquid water and derived thermodynamic sounding data. doi.org/10.1175/JAMC-D-17-0056.1. Wea. Forecasting, 13, 277–285, https://doi.org/10.1175/1520- Blair, S. F., D. R. Deroche, J. M. Boustead, J. W. Leighton, B. L. 0434(1998)013,0277:NCOHSW.2.0.CO;2. Barjenbruch, and W. P. Gargan, 2011: A radar-based assess- Ferraro, R., J. Beauchamp, D. Cecil, and G. Heymsfield, 2015: A ment of the detectability of giant hail. Electron. J. Severe prototype hail detection algorithm and hail climatology Storms Meteor., 6 (7), http://www.ejssm.org/ojs/index.php/ developed with the Advanced Microwave Sounding Unit ejssm/article/viewArticle/87. (AMSU). Atmos. Res., 163, 24–35, https://doi.org/10.1016/ ——, and Coauthors, 2017: High-resolution hail observations: j.atmosres.2014.08.010. Implications for NWS warning operations. Wea. Forecasting, Fraile, R., A. Castro, and J. L. Sanchez, 1992: Analysis of hailstone 32, 1101–1119, https://doi.org/10.1175/WAF-D-16-0203.1. size distributions from a hailpad network. Atmos. Res., 28, Brown, T. M., W. H. Pogorzelski, and I. M. Giammanco, 2015: 311–326, https://doi.org/10.1016/0169-8095(92)90015-3. Evaluating hail damage using property insurance claims ——,C.Berthet,J.Dessens,andJ.L.Sánchez, 2003: Return periods of data. Wea. Climate Soc., 7, 197–210, https://doi.org/10.1175/ severe hailfalls computed from hailpad data. Atmos. Res., 67–68, WCAS-D-15-0011.1. 189–202, https://doi.org/10.1016/S0169-8095(03)00051-6. Burcea, S., R. Cica, and R. Bojariu, 2016: Hail climatology and Giammanco, I. M., T. M. Brown, M. R. Kumjian, and A. J. trends in Romania: 1961–2014. Mon. Wea. Rev., 144, 4289– Heymsfield, 2014: Observations of hailstone sizes and shapes 4299, https://doi.org/10.1175/MWR-D-16-0126.1. from the IBHS Hail Measurement Program: 2012–2014. 27th Calianese, E. J., J. K. Jordan, E. B. Curran, A. R. Moller, and Conf. on Severe Local Storms, Madison, WI, Amer. Meteor. G. Woodall, 2002: The Mayfest high-precipitation supercell of Soc., 16B.2., https://ams.confex.com/ams/27SLS/webprogram/ 5 May 1995—A case study. Preprints, 21st Conf. on Severe Paper255294.html. Local Storms, San Antonio, TX, Amer. Meteor. Soc., P3.4, ——, B. R. Maiden, H. E. Estes, and T. M. Brown-Giammanco, https://ams.confex.com/ams/SLS_WAF_NWP/techprogram/ 2017: Using 3D laser scanning technology to create digital paper_47099.htm. models of hailstones. Bull. Amer. Meteor. Soc., 98, 1341–1347, Cecil, D. J., and C. B. Blankenship, 2012: Toward a global clima- https://doi.org/10.1175/BAMS-D-15-00314.1. tology of severe hailstorms as estimated by satellite passive Groenemeijer, P., and T. Kühne, 2014: A climatology of torna- microwave imagers. J. Climate, 25, 687–703, https://doi.org/ does in Europe: Results from the European Severe Weather 10.1175/JCLI-D-11-00130.1. Database. Mon. Wea. Rev., 142, 4775–4790, https://doi.org/ Cerveny, R. S., and Coauthors, 2017: WMO assessment of weather 10.1175/MWR-D-14-00107.1. and climate mortality extremes: Lightning, tropical cyclones, ——, and Coauthors, 2017: Severe convective storms in Europe: tornadoes, and hail. Wea. Climate Soc., 9, 487–497, https:// Ten years of research at the European Severe Storms Labo- doi.org/10.1175/WCAS-D-16-0120.1. ratory. Bull. Amer. Meteor. Soc., 98, 2641–2651, https:// Changnon, S. A., 1977: The scales of hail. J. Appl. Meteor., 16, doi.org/10.1175/BAMS-D-16-0067.1. 626–648, https://doi.org/10.1175/1520-0450(1977)016,0626: Heymsfield, A. J., I. M. Giammanco, and R. Wright, 2014: Terminal TSOH.2.0.CO;2. velocities and kinetic energies of natural hailstones. Geophys. Res. ——, and J. Burroughs, 2003: The tristate hailstorm: The most Lett., 41, 8666–8672, https://doi.org/10.1002/2014GL062324. costly on record. Mon. Wea. Rev., 131, 1734–1739, https:// Hohl, R., H. H. Schiesser, and D. Aller, 2002: Hailfall: The re- doi.org/10.1175//2549.1. lationship between radar-derived hail kinetic energy and hail ——, D. Chagnon, and S. D. Hilberg, 2009. Hailstorms across the damage to buildings. Atmos. Res., 63, 177–207, https://doi.org/ nation: An atlas about hail and its damages. Contract Rep. 10.1016/S0169-8095(02)00059-5. 2009-12, 95 pp. Höller, H., and M. E. Reinhardt, 1986: The Munich Hailstorm of Cheng, L., and M. English, 1983: A relationship between hailstone July 12, 1984—Convective development and preliminary concentration and size. J. Atmos. Sci., 40, 204–213, https://doi.org/ hailstone analysis. Beitr. Phys. Atmos., 59, 1–12. 10.1175/1520-0469(1983)040,0204:ARBHCA.2.0.CO;2. Hussain, and Coauthors, 2019: pyMannKendall: A python package Cica, R., S. Burcea, and R. Bojariu, 2015: Assessment of severe for nonparametric Mann Kendall family of trend tests. J. Open hailstorms and hail risk using weather radar data. Meteor. Source Software, 4, 1556, https://doi.org/10.21105/joss.01556. Appl., 22, 746–753, https://doi.org/10.1002/met.1512. Kahraman, A., and P. M. Markowski, 2014: Tornado climatology of Cintineo, J. L., T. M. Smith, V. Lakshmanan, H. E. Brooks, and K. L. Turkey. Mon. Wea. Rev., 142, 2345–2352, https://doi.org/ Ortega, 2012: An objective high-resolution hail climatology of 10.1175/MWR-D-13-00364.1. the contiguous United States. Wea. Forecasting, 27, 1235–1248, ——, S¸ . Tilev-Tanriover, M. Kadioglu, D. M. Schultz, and P. M. https://doi.org/10.1175/WAF-D-11-00151.1. Markowski, 2016: Severe hail climatology of Turkey. Mon. Wea. Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Con- Rev., 144, 337–346, https://doi.org/10.1175/MWR-D-15-0337.1. figuration and performance of the data assimilation system. Quart. Kalina,E.A.,K.Friedrich,B.C.Motta,W.Deierling,G.T.Stano, J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828. and N. N. Rydell, 2016: Colorado plowable hailstorms: Synop- Dessens, J., 1986: Hail in southwestern France. Part I: Hailfall tic weather, radar, and lightning characteristics. Wea. Forecasting, characteristics and hailstrom environment. J. Climate Appl. 31, 663–693, https://doi.org/10.1175/WAF-D-15-0037.1.

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC 3916 MONTHLY WEATHER REVIEW VOLUME 147

Knight, C. A., and N. C. Knight, 2005: Very large hailstones from Convective Hazard Model (AR-CHaMo). J. Appl. Meteor. Cli- Aurora, Nebraska. Bull. Amer. Meteor. Soc., 86, 1773–1782, matol., 57, 569–587, https://doi.org/10.1175/JAMC-D-17-0132.1. https://doi.org/10.1175/BAMS-86-12-1773. Reges, H. W., N. Doesken, J. Turner, N. Newman, A. Bergantino, Kron, W., M. Steuer, P. Löw, and A. Wirtz, 2012: How to deal and Z. Schwalbe, 2016: CoCoRaHS: The evolution and ac- properly with a natural catastrophe database—Analysis of complishments of a volunteer rain gauge network. Bull. Amer. flood losses. Nat. Hazards Earth Syst. Sci., 12, 535–550, https:// Meteor. Soc., 97, 1831–1846, https://doi.org/10.1175/BAMS-D- doi.org/10.5194/nhess-12-535-2012. 14-00213.1. Kumjian, M. R., Z. J. Lebo, and A. M. Ward, 2019: Storms producing Sánchez, J. L., R. Fraile, J. L. de La Madrid, M. T. de La Fuente, large accumulations of small hail. J. Appl. Meteor. Climatol., 58, P. Rodríguez, and A. Castro, 1996: Crop damage: The hail size 341–364, https://doi.org/10.1175/JAMC-D-18-0073.1. factor. J. Appl. Meteor., 35, 1535–1541, https://doi.org/10.1175/ Kunz, M., and M. Puskeiler, 2010: High-resolution assess- 1520-0450(1996)035,1535:CDTHSF.2.0.CO;2. ment of the hail hazard over complex terrain from radar ——, and Coauthors, 2009: Characterization of hailstone size spectra in and insurance data. Meteor. Z., 19, 427–439, https://doi.org/ hailpad networks in France, Spain, and Argentina. Atmos. Res., 10.1127/0941-2948/2010/0452. 93, 641–654, https://doi.org/10.1016/j.atmosres.2008.09.033. ó í ——, U. Blahak, J. Handwerker, M. Schmidberger, H. J. Punge, ——, L. L pez, E. Garc a-Ortega, and B. Gil, 2013: Nowcasting of S. Mohr, E. Fluck, and K. M. Bedka, 2018: The severe hail- kinetic energy of hail precipitation using radar. Atmos. Res., storm in southwest Germany on 28 July 2013: Characteristics, 123, 48–60, https://doi.org/10.1016/j.atmosres.2012.07.021. impacts and meteorological conditions. Quart. J. Roy. Meteor. Sander, J., J. F. Eichner, E. Faust, and M. Steuer, 2013: Rising vari- Soc., 144, 231–250, https://doi.org/10.1002/qj.3197. ability in thunderstorm-related U.S. losses as a reflection of Li, X., Q. Zhang, T. Zou, J. Lin, H. Kong, and Z. Ren, 2018: Climatology changes in large-scale thunderstorm forcing. Wea. Climate Soc., of hail frequency and size in China, 1980–2015. J. Appl. Meteor. 5, 317–331, https://doi.org/10.1175/WCAS-D-12-00023.1. Climatol., 57, 875–887, https://doi.org/10.1175/JAMC-D-17-0208.1. Schulz, W., G. Diendorfer, S. Pedeboy, and D. R. Poelman, 2016: The Marinica, I., 2003: Fenomene Meteorologice extreme ^ın Oltenia European lightning location system EUCLID–Part 1: Perfor- (Extreme Meteorological Phenomena in Oltenia) (in Roma- mance analysis and validation. Nat. Hazards Earth Syst. Sci., 16, nian). MJM Press, 280 pp. 595–605, https://doi.org/10.5194/nhess-16-595-2016. Marshall, T. P., R. F. Herzog, S. J. Morrison, and S. R. Smith, 2002: Schuster, S. S., R. J. Blong, R. J. Leigh, and K. J. McAneney, 2005: Characteristics of the 14 April 1999 Sydney hailstorm based on Hail damage threshold sizes for common roofing and siding ground observations, weather radar, insurance data and materials. 21st Conf. on Severe Local Storms, San Antonio, emergency calls. Nat. Hazards Earth Syst. Sci., 5, 613–620, TX, Amer. Meteor. Soc., P3.2, https://ams.confex.com/ams/ https://doi.org/10.5194/nhess-5-613-2005. SLS_WAF_NWP/techprogram/paper_45858.htm. ——, ——, and K. J. McAneney, 2006: Relationship between radar- Nisi, L., O. Martius, A. Hering, M. Kunz, and U. Germann, 2016: Spatial derived hail kinetic energy and damage to insured buildings for and temporal distribution of hailstorms in the Alpine region: A severe hailstorms in Eastern Australia. Atmos. Res., 81,215– long-term, high resolution, radar-based analysis. Quart. J. Roy. 235, https://doi.org/10.1016/j.atmosres.2005.12.003. Meteor. Soc., 142, 1590–1604, https://doi.org/10.1002/qj.2771. Smith, P. L., and A. Waldvogel, 1989:On determinations of ——, A. Hering, U. Germann, and O. Martius, 2018: A 15-year hail maximum hailstone sizes from hailpad observation. J. Appl. streak climatology for the Alpine region. Quart. J. Roy. Me- Meteor., 28, 71–76, https://doi.org/10.1175/1520-0450(1989) teor. Soc. , 144, 1429–1449, https://doi.org/10.1002/qj.3286. 028,0071:ODOMHS.2.0.CO;2. Ortega, K. L., 2018: Evaluating multi-radar, multi-sensor products Strader, S. M., W. S. Ashley, T. J. Pingel, and A. J. Krmenec, 2017: for surface hailfall diagnosis. Electron. J. Severe Storms Me- Observed and projected changes in U.S. tornado exposure. teor., 13 (1), http://www.ejssm.org/ojs/index.php/ejssm/article/ Wea. Climate Soc., 9, 109–123, https://doi.org/10.1175/WCAS- viewArticle/163. D-16-0041.1. Palencia, C., D. Giaiotti, F. Stel, A. Castro, and R. Fraile, 2010: Taszarek, M., H. E. Brooks, B. Czernecki, P. Szuster, and Maximum hailstone size: Relationship with meteorological K. Fortuniak, 2018: Climatological aspects of convective pa- variables. Atmos. Res., 96, 256–265, https://doi.org/10.1016/ rameters over Europe: a comparison of ERA-Interim and j.atmosres.2009.08.011. sounding data. J. Climate, 31, 4281–4308, https://doi.org/ Prein, A. F., and G. J. Holland, 2018: Global estimates of damaging 10.1175/JCLI-D-17-0596.1. hail hazard. Wea. Climate Extremes, 22, 10–23, https://doi.org/ ——, and Coauthors, 2019: A climatology of thunderstorms across 10.1016/j.wace.2018.10.004. Europe from a synthesis of multiple data sources. J. Climate, 32, ú P cik, T., and Coauthors, 2017: Future changes in European se- 1813–1837, https://doi.org/10.1175/JCLI-D-18-0372.1. vere convection environments in a regional climate model Tippett, M. K., J. T. Allen, V. A. Gensini, and H. E. Brooks, 2015: ensemble. J. Climate, 30, 6771–6794, https://doi.org/10.1175/ Climate and hazardous convective weather. Curr. Climate JCLI-D-16-0777.1. Change Rep., 1, 60–73, https://doi.org/10.1007/s40641-015-0006-6. Punge, H. J., and M. Kunz, 2016: Hail observations and hailstorm Tuovinen, J. P., A. J. Punkka, J. Rauhala, H. Hohti, and D. M. characteristics in Europe: A review. Atmos. Res., 176–177, Schultz, 2009: Climatology of severe hail in Finland: 1930– 159–184, https://doi.org/10.1016/j.atmosres.2016.02.012. 2006. Mon. Wea. Rev., 137, 2238–2249, https://doi.org/10.1175/ ——, K. M. Bedka, M. Kunz, and A. Reinbold, 2017: Hail fre- 2008MWR2707.1. quency estimation across Europe based on a combination of Waldvogel, A., B. Federer, W. Schmid, and J. F. Mezeix, 1978: The overshooting top detections and the ERA-INTERIM re- kinetic energy of hailfalls. Part II: Radar and hailpads. J. Appl. analysis. Atmos. Res., 198, 34–43, https://doi.org/10.1016/ Meteor., 17, 1680–1693, https://doi.org/10.1175/1520-0450(1978) j.atmosres.2017.07.025. 017,1680:TKEOHP.2.0.CO;2. Rädler, A. T., P. Groenemeijer, E. Faust, and R. Sausen, 2018: De- Yeo, S., R. Leigh, and I. Kuhne, 1999: The April 1999 Sydney tecting severe weather trends using an Additive Regressive Hailstorm. Aust.J.EmergencyManage., 14 (4), 23–25.

Unauthenticated | Downloaded 09/27/21 03:50 AM UTC