What is the hail climatology along the northeastern Adriatic? Is it possible to predict the synoptics, westerly wind flows, and climatological blocking systems?

General awareness and overall interest regarding hailstorms and hail properties in Europe have increased significantly in the last several decades and have resulted in numerous local, national, and even Europe-wide studies on hail and hail properties. To contribute to this field, we determined the hail climatology in the northeastern (NE) Adriatic region and analyzed its spatial and temporal patterns and performed an objectively derived weather type analysis of ERA5 daily mean data and instability indices. We studied the NE Adriatic region due to its focus on agricultural activities and on quality wine production. Our results are based on approximately 60 years of high spatial resolution measurements collected from 27 stations across complex terrain. The results show (i) high levels of spatial variability, (ii) significant annual variations, and (iii) hail throughout the whole year that (iv) intensifies in summer months. Furthermore, redistribution of hail among seasons (in particular, from summer to spring) was detected. Most significant changes were visible in the June–October period, with a negative trend of −0.06 hail days/year, and the period from November to March exhibited a positive trend of 0.13 hail cases/year. We found that deep cyclonic systems in front of and above our domain were most responsible for hail generation, often supported by southwesterly winds. Additionally, the vast majority of observed hail events occurred in unstable and sheared environments.Over the past several decades in Europe, the general awareness and overall interest in hailstorms and hail properties have significantly increased. Even minor hail events can cause considerable damage to agricultural systems, and as hail increases in size, trees, greenhouses, vehicles, animals, and humans are more strongly affected (Púcik et al., 2019). Thus, any effort toward a better understanding of hail properties can ˇ help reduce damage and risks involved with hail. Most hail properties, such as its annual and diurnal cycles, trends, and interannual and seasonal variations, can be identified through hail climatology. To date, hail climatology at local or national scales has been analyzed for the majority of European countries (Punge & Kunz, 2016). Data used in derivations of hail climatology include hailpad measurements (Manzato, 2011; Pocakal et al., 2009; Sánchez et al., 2009), ˇ station measurements (e.g., Burcea et al., 2016; Curi ´ c & Janc, 2016; Kotinis-Zambakas, 1989; Vinet, 2001; ´ Zhang et al., 2008), hail reports (e.g., Dessens, 1986; Tuovinen et al., 2009; Webb et al., 2001, 2009), radar estimates (Lukach et al., 2017; Nisi et al., 2016; Strzinar & Skok, 2018; Visser & van Heerden, 2000), satellite assessments (Punge et al., 2017), insurance damage data (Vinet, 2001), and global model outputs (Brooks et al., 2003; Hand & Cappelluti, 2011). Overall, the results show the presence of hail over much of Europe from the northern regions of Scandinavia (up to 67.5◦N), where the hail season is shortest (during summer) (e.g., Tuovinen et al., 2009), to the Mediterranean region, where hail can occur throughout the year (e.g., Baldi et al., 2014; Berthet et al., 2011; Punge & Kunz, 2016). In addition, while continental regions are mainly affected by hail during the warmer parts of the year (e.g., April–September), coastal (maritime) regions along the Atlantic Ocean or Mediterranean show different hail frequency distributions. For coastal hail climatology, most cases occur during winter and spring but with relatively few hailstones (Punge & Kunz, 2016; Santos & Belo-Pereira, 2019). According to Punge et al. (2014), the areas with the most frequent hail events are positioned between 39◦N and 50◦N. Hail hot spots are found in perialpine regions of the north and south, followed by the greater Pyrenees Region, Massif Central in France, Apennines in Italy, Dinarides in the Balkan Region, and Carpathians in the Pannonian Basin. Some reported hail frequencies have presented values of 0–2.4 hail days per year according to station measurements (Punge & Kunz, 2016),Figure 1. (a) The studied area of western Croatia. (b) A zoomed view of the NE Adriatic region, including the locations of the study stations. The colored contours represent topography in meters, the blue circle denotes the main weather station in Istria, the blue squares denote continental stations, the triangles represent climatological stations, and the x symbols denote rain gauge stations. The red dots indicate the stations represented in data set III. The domain shown in (b) displays the area for which prevailing wind directions have been calculated using the weather-type algorithm, and the large black-lined rectangle shows the area used for instability indices.while radar-based estimates have reported values of 0–2 hail days per year (Nisi et al., 2016; Strzinar & Skok, 2018). A recent publication by Punge et al. (2017) provides an estimation of hail frequencies from satellite-based overshooting tops (OTs) and ERA-Interim reanalyses of Europe. The authors reported 10 km × 10 km gridded information for yearly hail frequency estimates that spanned 0–2 hail days (Punge et al., 2017; their Figure 6) and the spatial distributions corresponded well with previous studies. While Croatia is situated in an area where strong impacts of hail are expected (e.g., Punge et al., 2014), the Adriatic Coast has never been analyzed in detail, and its national hail climatology has not yet been developed. Nevertheless, several papers have addressed the hail characteristics of the continental region of Croatia (Figure 1a, northwestern [NW] part of figure), which is an agricultural region that is well equipped with weather radar, meteorological stations, and hailpad networks. The hail properties and their characteristics in the continental part of Croatia are reported in Pocakal and Štalec (2003) and Po ˇ cakal et al. (2009, 2018). ˇ In these papers, the analyses focused on the warm season (i.e., May–September), which is connected with the hail suppression network and hailpad data. The main results show that the spatial and temporal characteristics of hail in the period from 1981 to 2006 show higher hail activity on the windward slopes of mountains. The hail frequencies in Croatia range from 0.1 to 2.4 hail days/season. The first 3 months of the hail season represent 84% of total hail cases, and hail is most active between 14 and 18 h local time. A positive significant trend for hail duration was also reported. In addition, the authors calculated that the average hypothetical length of hail streaks was approximately 1,890 m and that the average damaged area was approximately 0.7 km2. According to analyses of lightning activity by Mikuš Jurkovic et al. (2015), OT analy- ´ ses by Punge et al. (2014) and lightning climatology results (Anderson & Klugmann, 2014; Mikuš et al., 2012; Poelman et al., 2016), the northeastern (NE) Adriatic (Figure 1b) appears to be a very active convective region and is similar to the southern part of the perialpine region (i.e., the Friuli Venezia Giulia region, as reported by Feudale et al., 2013). Therefore, we focus here on the NE Adriatic region, which consists of a section of the Croatian coast that covers the county of Istria and a few neighboring districts of the town of Rijeka. Geographically, the NE Adriatic region is surrounded by the Alps to the north, the Adriatic Sea to the west and south, and Dinarides to the east. Topographically, the Istrian Peninsula is characterized by a very complex terrain with flatlands to the south and west coasts, variable hills (up to 600 m high) with valleys in the central region, and mountainous regions (up to 1,400 m high) to the east and north. Therefore, the NE Adriatic region serves as a unique site for local analysis of hail properties across highly complex terrain.By recognizing the current lack of knowledge of hail characteristics over the NE Adriatic, this study has two objectives. (i) The aim of this study is to provide an analysis of existing hail data and an estimation of the spatial hail distribution in the area of interest, which have not been previously conducted. (ii) A classification of the most frequent and relevant weather patterns in connection with hail reports is provided and reveals weather conditions during hail days. The approach adopted is based on the objective classification method, which allows for analysis of weather patterns not only of past observations but also of future data.We used daily mean ERA5 reanalysis data (1979–2017) to extract weather patterns related to hail events and identified 673 days with hail. We created an objective algorithm for WT classification based on the objective classification given by Bissolli and Dittmann (2001) and incorporated the subjective approach proposed by Poje (1965) to examine weather phenomena in Croatia. Our algorithm generally follows their recommendations, but we made changes to our determinations of advection types and vorticities. To determine advection types, we used 700-hPa horizontal wind component data and defined 36 possible wind directions that were shifted by 10◦ from each other (Bissolli & Dittmann, 2001). When two thirds of the grid points fell within one quadrant (i.e., 0–90◦, 90–180◦, 180–270◦, and 270–360◦), the center of the sector was defined as the wind index (in ◦). The program determined which main wind direction corresponded with the wind index (NE, SE, SW, NW, or UC unclassified), i.e., the prevailing wind direction. We conducted these computations over a smaller domain (20 points) that covered only the NE Adriatic region (Figure 1b) because, for the purposes of studying hail events, we are only interested in local wind features. The other algorithm uses large-scale parameters and runs over a larger domain that covers the entire Adriatic Region, part of the Mediterranean and the continental area (35–50◦N, 5–21◦E). The vorticity indexes for the upper levels were obtained from a weighted areal mean of ∇2Φ500 (∇2 Laplace operator, Φ500 500-hPa geopotential), and the vorticity values at the mean sea level were obtained from ∇2p, where p is the mean sea level pressure (mslp). A quasi-nongradient field is present when the mean ∇p is less than 0.06 hPa/km. From the vorticity index, we obtained 17 different weather patterns, of which 16 corresponded to cyclones or anticyclones occurring near the surface, while one pattern represented a quasi-nongradient field (as proposed by Poje, 1965). All the cyclonic and anticyclonic weather patterns were defined as deep (a cyclone or anticyclone on the surface and a cyclone or anticyclone above the surface; in Figure 8, we use CC or AA, respectively) or shallow (a cyclone or anticyclone on the surface and an anticyclone or cyclone above the surface; in Figure 8, we, use CA or AC, respectively). Furthermore, with respect to our inner domain, we distinguished among the four sides of prevailing pressure

systems: front, upper, back, and lower. A more detailed description is presented in Belušic Vozila (2018). ´ The thermodynamic indices were computed using the Sounding and Hodograph Analysis and Research Program in Python (SHARPpy) (Blumberg et al., 2017). SHARPpy is a collection of an open-source, upper-air sounding analysis and visualization routines. Based on the profiles of temperature, specific humidity, pressure, height, and wind components from ERA5 reanalysis, SHARPpy was used to compute lifted index (LI), most unstable CAPE (MUCAPE), K-index (KI), deep-layer shear (DLS), and freezing level height. ERA5 data from the surface and 17 pressure levels at 00:00, 06:00, 12:00, 18:00 UTC were used. LI was calculated as the difference between the temperature at 500 hPa and the temperature of an air parcel lifted moist adiabatically to 500 hPa from the surface. MUCAPE was assessed by lifting the parcel with the maximum equivalent potential temperature values in the lowest 400 hPa, i. e., the most unstable parcel. From the temperature difference between 850 and 500 hPa and the moisture content of the lower atmosphere, the KI was determined. DLS was determined as the difference in wind speed between the surface (10 m) and 6 km. Using linear interpolation, the temperature profile was obtained and the height (meters above ground level) of the 0◦C isotherm was obtained. To obtain a comparison between indices on hail days and days without hail, we used the daily mean fields of indices averaged over the NE Adriatic region, as indicated by the black rectangle in Figure 1b.Conclusions: We analyzed data acquired from two independent data sources to identify the spatial and temporal aspects of hail in the NE Adriatic region and to highlight the WTs responsible for this hail. The main conclusions from this analysis are as follows. (i) Station measurements show strong spatial and temporal variability of hail occurrences across the studied domain. We identified three hot spots in our domain with frequencies of 1.75–2.8 hail days/year and several cold spots with frequencies of less than 0.5 hail days/year. Trend analysis revealed a slight but insignificant increase in yearly hail activity. However, it detected a redistribution of hail among the seasons (e.g., MAM, JJA, SON, and DJF), especially between MAM and JJA. While the JJA season and the June–October period recorded negative significant trends similar to reports of the stations measurements (Punge & Kunz, 2016), the period from November to March exhibited a positive trend with a 0.99 significance level. Trend analyses also showed high levels of sensitivity to the selection of the studied period, suggesting that very long time series are needed. (ii) On the annual scale, hail occurred throughout the year, with most events occurring in the warmer months. However, some stronger events were sporadically reported in the cold(er) times of the year. In addition, a local maximum of hail events was detected in November across all data sets and time periods. Such a signal seems to be influenced by the Adriatic Sea, as it coincides with a local maximum in waterspout appearances along the entire Adriatic coast (Renko et al., 2016) and has not been reported for the continental region of Croatia. Two maxima in the diurnal cycle for strong hail (12:00 and 16:00 h) and three maxima of hail over the coastal area (08:00, 12:00, and 16:00 h) were also observed. When we consider the orographic configuration of the region and the dominance of the SW advection type for hail generation, most hail activity that occurs in the interior parts of Istria is presumably orographically induced. Additionally, often influenced by the northern and eastern mountains and by local wind regimes, convective clouds that form over the peninsula may drift over Istria and create new cells during their life cycle. (iii) Convective clouds may often move toward the lowlands of the west coast and especially to the NW regions, where we found higher levels of hail activity relative to the SW part of the coast. The morning (08:00 h) maximum is probably influenced by differences in overnight sea and land cooling. Still, it is necessary to conduct further detailed analyses of hail days using numerical simulations. (iv) Objectively derived WTs provide strong insight into synoptic activities, which are responsible for hail in the NE Adriatic region. Most hail events are related to cyclonic WTs and to events starting at the lower and front parts of cyclones. Approximately 55% of hail is associated with SW wind advection, which is found in WT1, WT4, and WT6 (Figure 10). Anticyclonic WTs associated with hail still offer reasonable results and suggest colder air advection from the NW. Additionally, the instability indices confirmed that the majority of hail days fall under highly unstable (MUCAPE between 600 and 1,000 J/kg) and sheared (DLS between 8 and 20 m/s in the first 6 km) environments that correspond with the WT classification and assure the quality of the observed data.

Weather Types :By analyzing all the WTs for the period available from the ERA5 reanalysis data (1979–2017), we obtained a general picture of both circulation patterns with a limited dominance of cyclonic systems (48%), followed by anticyclonic systems (42%), while the remaining 10% were recognized as quasi-nongradient systems (Figure 8a). The most dominant signal was observed from the deep anticyclonic systems located above our domain (56% within anticyclonic WTs). In contrast, cyclones were more evenly distributed, with WT3s exhibiting the rarest occurrence. From data set I, we extracted all the hail days for the studied period (1979–2017) and computed the WTs for each day (Figure 8b). We used numbers to differentiate between positions and types (e.g., WT1, WT2, WT3, and WT4 for the front, top, back, and lower sides of cyclones affecting the Istrian region, respectively, and the same for WT5–8 for anticyclonic systems). Colors denote the system intensity levels (high intensities are indicated by red bars, and low intensities are indicated by blue bars). In Figure 9, we show composites of high-intensity WTs, including the wind vectors from 700 hPa, and for comparison with Figure 10, we show strong-intensity WTs on hail days.Figure 8b shows that approximately 78.5% of hail can be attributed to cyclonic activity, which is expected since cyclonic weather provides a good source of lift, higher values of CAPE and favors moist air advection (Santos & Belo-Pereira, 2019). Furthermore, 14.5% of hail is attributed to anticyclonic activity and 7.0% is associated with quasi-nongradient activity. WT4, which represents the lower sides of deep and shallow cyclones, is found to be associated with 46% of all cyclonic hail days (Figure 8b). When we compare our hail WTs to the distribution of the 17 WTs for the whole period, we obtain the percentage hail occurrence for a particular WT (Figure 8c). In contrast to what is shown in Figure 8b, hail occurs most frequently along the backs of shallow cyclonic systems, although this particular WT occurs very rarely. Figure 8b shows only 14 hail days (WT3 blue). On the other hand, a deep cyclone in WT4 (see also Figure 10, WT4) has a 15% chance of producing hail over the observed area. The front sections of deep cyclonic systems (Figure 10, WT1) present a greater than 13% chance to generate hail, while anticyclonic WTs present a 5% chance (WT5 and WT6) or less. However, composites of anticyclonic WTs on days with hail (Figure 10) still provide environments for hail generation. WT5 and WT8 both have NW wind components over the NE Adriatic region, which may advect colder air masses and produce unstable environments that are favored by convection, while WT6 shows SW advection. We should note that hail that occurs during a specific WT can also be misinterpreted due to the occurrence of fast-moving smaller low-pressure systems that can produce several WTs within a single day. Unfortunately, we were not able to account for these variation with our algorithm, as we used daily mean data; thus, our algorithm emphasizes WTs with the strongest signals (on a particular day). The advection types at 700 hPa, as described in section 2.2, are shown in Figure 11 for all the days of the ERA5 period (blue bars) and for the hail days (red bars) of the ERA5 period. The y-axis shows the relative frequencies of particular advection types relative to the total number of advection types. In general, SW winds are dominant for both all days and hail days and are associated with 55% of all hail events, which is reasonable and expected since such air masses come directly from the Mediterranean Sea and Adriatic Sea. Such distributions for both all days and hail days correspond well with the wind directions composite shown in Figures 9 and 10 in combination with Figure 9a if we consider the NE Adriatic region. In particular, WT1, WT4, WT6, and NG-WT hail days all have mean wind vectors located in the SW quadrant. Additionally, our results agree strongly with the subjectively applied classifications of the WT and flow conditions for days with deep convection provided by Mikuš et al. (2012) when analyzing lightning climatology. In that paper, in the area that includes Istria (which is somewhat larger than that examined in this study), lightning usually occurs as a result of low-pressure formation (e.g., cyclones and troughs) on approximately 3/4 of days with deep convection. Furthermore, the prevailing large-scale wind regimes are traced from the southwest for half of the days and NW winds on 1/4 of the days during which lightning occurs, echoing the results for this region for days with hail.Additionally, we extracted the values of all the indices for days with hail and compared them with those for all nonhail days using the box and whisker plots shown in Figure 11. The red boxes represent hail days, while the blue boxes represent nonhail days. The median values are the horizontal lines in the boxes, the box edges are the 25th and 75th percentiles, and the whiskers represent the 5th and 95th percentiles. All the indices show the differences between hail days and nonhail days. The most significant differences are for LI, MUCAPE and KI. Low LI values and high MUCAPE and KI values confirm that hail forms in a very unstable environment with moderate to high wind shear. The MUCAPE values roughly correspond to those shown in Púcik et al. (2015), who used radiosonde ˇ data and ESWD data to examine severe environments. The results for 0◦ height (Figure 12c) confirm that all the hail events recorded by observers had acceptable freezing level heights. The absolute minimum of the mean daily freezing levels on days with hail was 645 m (not shown), and 95% of hail days had a mean daily freezing level higher than 1,160 m. On the other hand, there is also an upper limit of freezing level height (e.g., 4,000 m) above which no hail was reported. Such results reinforce our previous findings that hail activity in the NE Adriatic region is, in fact, present throughout the year. When dealing with severe environments, CAPE/MUCAPE and DLS are frequently used indicators of severe weather (e.g., Púcik et al., 2015; Santos & Belo-Pereira, 2019), and their combination often highlights the ˇ thresholds for severe weather occurrences. We compared these two parameters for both hail days (data set I) and nonhail days, and the results are shown in Figure 13. Each colored square shows the probability of a particular parameter combination within the observed set. For low DLS values (