Hail signals were identified from radar reflectivity composites using the operational C-band radar network of the DWD. This network currently consists of 17 weather radars (16 prior to 2013 and between 2018 and 2021; see Supplementary Figure S1 for locations and Supplementary Table S1 for further details in the supplementary material). It is now equipped with the latest dual-polarization Doppler technology. To cover the longest possible period, our study used only single-polarization radar data. Note that the upgrade from single- to dual-polarization radar began in 2011 and concluded in 2021 (Wilke et al., 2025). Although 3D radar data with higher spatial and temporal resolutions, as well as finer intensity classes are now available, we exclusively used the so-called PZ product (single-radar CAPPI reflectivity product) with low resolution (DWD, 2024; DWD, 2025), which has been available since 2005, in order to maintain the comparability and homogeneity of the data basis and to enable consistent statements about temporal trends.

This paragraph describes the main radar specifications that remain the same throughout the study period. A volume scan is performed every 15 min at 18 elevation angles ranging from 0.5 ° to 37 ° (Bartels et al., 2005). The scan consists of an intensity mode with a range of 230 km, covering the lowest 4.5 ° , and a Doppler mode with a range of 120 km, covering all elevations above. The original PZ product derived from this scan has an original spatial resolution of 2 km. For our study, it was resampled to a horizontal resolution of 1 × 1 km² and covers an area of 400 ×  400 km² for each radar. It has been available every 15 min since 2005 (since 2013 every 5 min). Radar reflectivity is converted into six discrete reflectivity classes: (7,19], (19,28], (28,37], (37,46], (46,55] and (55, ∞ ) dBZ, at 12 equidistant altitude layers between 1 and 12 km. The two highest classes, (46, 55] and (55, ∞ ) dBZ, are particularly important to our tracking algorithm. Individual PZ volumes were merged into a radar composite to create the tracks by retaining the highest reflectivity values at overlapping volume points.

Overall, data availability is high, ranging between 87% and 99% depending on the radar location and year. The exception is 2005 with 82.3%, as the PZ product was only available starting in May of that year. Outages due to maintenance and, more importantly, the modernization of the radar network, especially in 2013 and 2014, were compensated for by neighboring radars (backup radars). More detailed information including specific values are listed in the Supplementary Section A1. In the case of radar outages, the composite was constructed using all available radars, partially compensating for spatial gaps through the substantial overlap of the individual PZ volumes (see Supplementary Figure S1). A Doppler filter eliminates ground clutter in the data from fixed objects, mainly at low elevations.

The temporal and spatial evolution of PHTs was tracked by applying an adjusted version of the cell tracking algorithm TRACE3D (Handwerker, 2002) to the 3D radar product. For this purpose, Puskeiler et al. (2016) adapted the algorithm for the binned PZ product in Cartesian coordinates, whereas Handwerker (2002) used single-radar 3D continuous reflectivity in spherical coordinates.

Several radar-based hail studies have used a threshold of 55 dBZ, known as the Mason criterion (Mason, 1971), to detect hail in 2D radar data (Schiesser, 1990; Hohl et al., 2002; Kunz and Kugel, 2015; Junghänel et al., 2016; Fluck et al., 2021). Therefore, identifying potential hail-bearing cells in a PZ composite volume works as follows (see also Schmidberger, 2018): 1. High reflectivity areas were identified by the existence of at least four radar bins of the highest class (55, ∞ ) dBZ. 2. The surrounding (46, 55] dBZ bins were then added to form the so-called reflectivity core (RC). 3. To exclude erroneous radar signals, an RC must have a minimum area of 5 ⋅  10⁶ km², and a minimum volume of 3 ⋅  10⁹ km³, and at least 50% of an RC must extend over two elevation levels.

After detection of potential hail-bearing RCs, their weighted centers were tracked over a constant time interval d t to create PHTs. In the PZ product, d t is 15 min. In this step, TRACE3D attempts to establish a temporal connection between the detected RCs in two time steps. To do so, a 2D shift velocity vector v T was calculated for each RC at time t 1 . If an RC has already been identified in a previous time step, the shift vector s T was computed using the positions at t − 1 = t 1 − 2 d t and t 0 = t 1 − d t . If an RC has already been detected in several previous time steps, v T was computed as the weighted sum of the velocity vectors from the previous time steps. If an RC is detected for the first time, the shift velocity vector v T was estimated from the mean velocity of neighboring RCs. Convective cells are assumed to propagate mainly with the mean wind, which does not exhibit significant spatial variation. The new position of an RC was calculated by s T = v T ⋅ d t . Here, the actual RC was identified by searching all possible RCs within a certain radius. The search radius is not fixed; rather, it depends on the distance s T and the distance to the nearest neighboring RC. The above-described method was repeated several times for subsequent scans until a complete track of the barycenter of an RC is determined.

Due to the limited number of reflectivity classes available on the equidistant grid, the cells sometimes exhibit only slight differences in their structure. Consequently, TRACE3D may misinterpret the distances and volume structures of potential subsequent cells when linking cells. This incorrect assignment results in an abrupt change in the track direction from one time step to the next. To address this issue, an additional filter criterion has been implemented: a change in track direction of more than 45 ° with a distance greater than 30 km within 15 min is considered a mismatch and is removed from the dataset. If the mismatch occurs at the beginning or end of a track, the first or last track point is deleted. If the track shift occurs between two RC detections, it was checked to see if two different tracks may be present. This check requires at least three time steps before and after the track shift. Real cell splits and merges, which often involve a significant change in track direction, are unaffected by this correction. Further details on TRACE3D can be found in Handwerker (2002); further details on the adjustment for the PZ product can be found in Schmidberger (2018).

As described above, the adapted version of TRACE3D was applied to the PZ product (Sect. 2.1) from 2005 to 2024 (SHY) to calculate individual PHTs, which form the basis for the subsequent investigations (PHT catalog). Because a cell lifetime of at least 30 min can be assumed for hailstorms and hail formation (Changnon, 1977; Kumjian et al., 2021), a minimum duration of 45 min, i.e., three time intervals, is required to reconstruct a complete PHT. Additionally, some PHTs are completely outside of mainland Germany because radar data were also available in the surrounding area. Furthermore, some other PHTs extend beyond the border if the cells have formed or dissipated in an adjacent country. However, the dataset only includes PHTs that crossed the German border at some point.

Each PHT contains the following information: start time and lifetime of the cell, cell center at each individual detection (latitude and longitude), total track length (distance between start and end points), mean width (of all RCs relative to the direction of movement), mean direction (angle between start and end points), and mean velocity. Based on the detections of the RCs, polygons with a constant track width perpendicular to the direction of travel were determined (for details, see Section A2 in the supplementary material). For the study, these polygons were gridded at a resolution of 1 ×  1 km², forming the basis for further statistical analyses. Some analyses use a coarser grid of 25 ×  25 km².

As part of a comparison between the RCs derived from the volume scan of the PZ product and those obtained from a higher-resolved radar product with a 500-m range bin available for a subdomain, it was observed that the widths of the RCs in the previous version of Schmidberger (2018), which were based on the lower-resolved PZ product, were systematically overestimated. This overestimation was largely independent of the size of the associated RCs. Therefore, the width of the RCs derived from the PZ product was adjusted by applying a correction factor of 0.6. A detailed explanation can be found in the supplement material Section A2.

This study uses insurance data to validate the accuracy of the radar-based PHT catalog. The data source is the German Insurance Association (in German: Gesamtverband der Deutschen Versicherungswirtschaft e. V., GDV), an umbrella organization representing all primary insurers in Germany. The data pertain to standard residential building insurance, covering classic risks such as fire, lightning, and water damage, as well as the natural hazards storm and hail (GDV, 2023). The GDV aggregates claims and contract data from all insurers, providing a comprehensive, standardized database. Statistics are available for each of Germany’s 400 districts from 2005 to 2023, including the number of daily reported claims and the total annual number of contracts. From these, the claim frequency, i.e., the ratio of reported claims to insured risks (i.e., contracts), is calculated. Setting a threshold allows this measure to serve as an objective indicator of the number of hail days in each district.

Compared to manually collected datasets such as ESWD hail reports, GDV insurance data are more homogeneous, objective, and comprehensive. In 2022, the average insurance density for standard residential building insurance in Germany was about 95% (GDV, 2025). Regional differences are minor: Bavaria has the lowest coverage at 88%, followed by Rhineland-Palatinate (91%) and Hesse (92%). In all other states, coverage is at least 97% (Küpfer, 2025).

Despite its strengths, GDV data are not free of uncertainty. The number of damage reports is particularly influenced by the vulnerability of buildings and the density of buildings in a region. Modern buildings with solar panels, conservatories, or skylights are more susceptible to damage than older buildings (Stucki and Egli, 2007). Additionally, claim dates often differ by ±  1 day from the actual event, because reports are not always assigned precisely. The most restrictive criterion is that the data only includes hail events involving larger, damage-relevant hailstones ( ≥  2 cm; e.g., Hohl et al., 2002; Brown et al., 2015; Xiao et al., 2025). A further limitation is that storm and hail claims are recorded jointly, so these damage causes are indistinguishable. Nevertheless, a distinction can be made based on the seasonal distribution of these extremes: storm damage (defined as wind speeds ≥  18 m/s) occurs mainly in winter, while hail damage is almost exclusively reported in summer (Küpfer, 2025). The few summer storm claims usually stem from SCSs with convective gusts or derechos, which often produce hail simultaneously (Pacey et al., 2021). Pure synoptic storms are rare in this season (Mohr et al., 2017).

Because SCSs are linked to large-scale atmospheric circulation patterns, we analyzed their relationship with the Atlantic-European weather regimes developed by Grams et al. (2017). These regimes were calculated using the ERA5 reanalysis (Hersbach et al., 2020) from the European Centre for Medium-Range Weather Forecasts (ECMWF) at 12 UTC (Grams et al., 2025; Grams, 2026). The regimes are derived by applying empirical orthogonal functions (EOFs) to mean geopotential height fields (Z500) followed by k-means clustering, yielding seven objectively defined large-scale atmospheric patterns (Grams et al., 2017). Four of these are blocking regimes characterized by positive geopotential height anomalies: Atlantic Ridge (AR), European Blocking (EuBL), Scandinavian Blocking (ScBL), and Greenland Blocking (GL). The regime GL is similar to the negative phase of the North Atlantic Oscillation (NAO; Beerli and Grams, 2019), while EuBL and ScBL represent two variants of the classical blocking pattern (Büeler et al., 2021). The remaining three regimes–the Atlantic Trough (AT), Zonal Regime (ZO), and Scandinavian Trough (ScTr) – are cyclonic regimes with negative geopotential height anomalies and enhanced cyclonic activity. They correlate to varying degrees with the positive NAO phase, with ZO showing the strongest resemblance. An additional “no regime” category is included for days without a dominant large-scale pattern, representing the climatological mean state and capturing transitional or mixed-flow situations. More details are given in Grams et al. (2017) and Büeler et al. (2021).

The spatial distribution of days with PHTs in Germany reveals distinct regional patterns, influenced by both climatic and orographic features, as evidenced by the high spatial variability in the mean annual number of PHTs per grid point (Figure 2). The median value for Germany as a whole is 0.3 PHT days per year, with the 25th and 75th percentiles at 0.2 and 0.4, respectively. On the larger scale, a pronounced north-to-south gradient is apparent. In northern Germany, the average is 0.1–0.3 per grid point per year, whereas in southern regions, most values range from 0.3 to 0.6 per grid point per year. A peak of 1.45 occurs near Reutlingen (south of Stuttgart), a well-known hail hotspot in Germany. This large-scale pattern is mainly influenced by the distance to the sea (North Sea and Baltic Sea) and the degree of continentality, which affects the climatological distribution of atmospheric stability in Europe in summer (Mohr and Kunz, 2013; Taszarek et al., 2018; Feldmann et al., 2025b).

In addition, the distribution is shaped by several regional substructures related to orographic features, particularly in mountainous areas and, most notably, downstream of the mountains. For example, pronounced maxima are observed downstream of the Black Forest, over the Swabian Jura and the Bavarian Prealps (see Figure 1, which shows the various low mountain ranges). Supercells with a high potential for large hail are particularly common in the Lech Valley of the Bavarian Prealps (DWD, 2021). Based on high-resolution simulations, Feldmann et al. (2025a) also found an increased frequency of supercells in both the Bavarian Prealps and the lee side of the Black Forest. An increased PHT likelihood is also observed near mountain ranges such as the Ore Mountains, the Hessian Highlands, and the Rhenish Massif. In contrast, some regions in Germany statistically experience fewer than one PHT event every 2 years (84.5% of grid points). The spatial variability of PHT days is primarily driven by flow deviations around orographic obstacles, thermally induced wind systems, and localized convergence of moisture transport (de la Torre et al., 2015; Punge and Kunz, 2016; Allen et al., 2020; Fischer et al., 2025a). Kunz and Puskeiler (2010) and Fluck et al. (2021), for example, hypothesized that the increased hail frequency downstream of the Black Forest (Germany) and the Massif Central (France) is due to relatively high atmospheric instability and low wind speeds in pre-convective conditions. This results in Froude numbers below 1, allowing the air to flow partly around the mountains at low levels. A zone of horizontal flow convergence downstream is created that may initiate convection. The Swabian MOSES field campaign in 2023 (Handwerker et al., 2025) partly confirmed this hypothesis, finding that low-level convergence downstream of the southern Black Forest coincided with convective cell development during several intensive observation periods.

Previous hail statistics for Germany, based on both 2D or 3D radar data, reveal patterns of hail frequency that are similar to those found in our study. This is especially true with regard to the north-to-south distribution of hail frequency and the hotspot regions (Kunz and Puskeiler, 2010; Kunz and Kugel, 2015; Puskeiler, 2013; Puskeiler et al., 2016; Junghänel et al., 2016; Schmidberger, 2018; Fluck et al., 2021; Wilke et al., 2025; Feldmann et al., 2025a). Slight differences in the results are due to several factors: (a) different hail detection methods or criteria, (b) different radar data (2D or 3D, single or dual-polarimetric) and (c) other data sources included (e.g., weather station data, severe weather reports, lightning data). The underlying study periods are usually shorter (maximum 10 years for all of Germany). For example, the most noticeable discrepancy to Wilke et al. (2025) is that the maximum south of Stuttgart is more pronounced in our study than that over the Bavarian Prealps. This difference can primarily be attributed to the different hail detection methods applied by Wilke et al. (2025), who used vertically integrated ice (VII) and maximum estimated size of hail (MESH) derived from radar reflectivity, rather than to the shorter 6-year analysis period (2018–2023; not shown).

We use categorical verification with a 2 ×  two contingency table and the GDV insurance data to assess how closely the radar-based hail statistic in Section 3.1, i.e., a proxy derived from remote sensing, aligns with events that occurred. Please note that the insurance data also not fully reflect reality (see Section 2.4). From this table, we compute common skill metrics (Wilks, 2006): the Probability of Detection (POD), measuring the share of correctly forecasted events; the False Alarm Ratio (FAR), indicating incorrect forecasts; and the Heidke Skill Score (HSS; Heidke, 1926), a normalized measure accounting for chance agreement. Ideally, POD and HSS would be close to one, reflecting perfect detection and maximum skill, while FAR would approach zero, indicating an absence of false alarms.

The verification period spans from May 2005 to August 2023, when both datasets are available. To avoid potential storm damage reports in April or September (see Section 2.4), the validation only considers the four summer months most relevant to hail: May, June, July, and August (see Section 3.3). The maximum spatial allocation distance is 20 km between a track and an affected district. An insurance-based hail day is defined by two thresholds, derived from sensitivity analyses that improve the HSS results: at least five claims per district per day (to filter out days with too few reports, especially in small counties with few contracts) and a claim frequency of 0.0001 (to exclude an excessive number of false reports in densely populated regions such as Berlin or Hamburg). Both thresholds address different distortions and help prevent misclassifications caused, for example, by incorrect time allocation.

The evaluation shows that the modified TRACE3D algorithm (Section 2.2) can essentially reproduce hail statistics in Germany (Figure 3). For the entire study area, the results yield a median HSS value of 0.47, with an interquartile range of 0.41–0.52 (25th/75th percentiles). The corresponding POD values are 0.43, ranging from 0.35 to 0.51 (25th/75th percentiles), while the FAR values are 0.36, ranging from 0.27 to 0.49 (25th/75th percentiles). The latter demonstrates that many tracks occur without any associated claim reports. However, this does not necessarily mean that the method is flawed. It may instead indicate that the hail was small and therefore did not cause any damage. Note that the TRACE3D-based algorithm was originally designed to detect hail in general (cf. Puskeiler et al., 2016; Schmidberger, 2018), including smaller events. For this purpose, the original calibration used insurance data from the agricultural sector, in which hailstones larger than 5 mm are considered relevant in terms of causing damage.

Moreover, regional differences exist. Good HSS values between 0.5 and 0.7 are found in southern, central, and eastern Germany (Figure 3A). The best performance is found in Saxony-Anhalt (median: 0.55, 25th/75th percentiles: 0.52–0.57), Hesse (median: 0.54, 25th/75th percentiles: 0.50–5.55), and Baden-Württemberg (median: 0.52, 25th/75th percentiles: 0.49–0.58). Lower accuracy appears near some state borders, for example, in the eastern corner of Bavaria (near Passau; see Figure 1) and in Saarland, mainly due to limited radar coverage and outages (see supplementary material Section A1). Skills are also reduced in western North Germany, especially Schleswig-Holstein, which has the lowest median HSS value of 0.31 (25th/75th percentiles: 0.25–0.39). In addition to the issue of radar availability, this could also be due to the fact that the GDV data in this region considers storm damage in some cases. Due to their proximity to the sea, these areas are more exposed to such events. Despite the inherent uncertainties in both datasets, the calculated metrics yield results comparable to those of other studies (e.g., López and Sánchez, 2009; Kunz and Kugel, 2015; Puskeiler et al., 2016; Voormansik et al., 2017; Schmidberger, 2018; Kopp et al., 2024). Overall, these findings underscore the reliability of the modified TRACE3D algorithm in reproducing radar-based hail statistics in Germany, establishing a solid basis for subsequent evaluations.

An analysis of the temporal distribution of the PHT catalog reveals clear seasonal patterns and a preference for specific months of the year (Figure 4). Most of the PHTs occurred in the warm summer season, with a peak in July (31.6%), followed by June (26.4%), August (22.2%), and May (12.8%). Only a small proportion occurred in April (1.2%) and September (5.8%). Almost half of the PHTs (43.1%) reached the detection threshold of 55 dBZ between 13 and 17 UTC (not shown). In general, these temporal patterns align well with the findings of previous studies in Germany (e.g., Schmidberger, 2018; Fluck et al., 2021) and fit into the European context: The hail season generally begins in April/May and ends in August/September (Punge and Kunz, 2016). However, there are regional differences in the main peaks: in central Europe, France, the Alps, and the Po Valley, hail activity is mainly concentrated from June to August. In contrast, in the Mediterranean region (from Spain to Turkey), the highest frequency generally occurs in autumn (Punge and Kunz, 2016; Laviola et al., 2022).

Most of the PHTs are shorter than 50 km (Figure 5A), with a median length of 24.0 km, while the 25th and 75th percentiles are 15.5 and 39.4 km, respectively. In terms of width, the tracks typically measure 3.2 km on average (median), ranging from 2.2 to 4.7 km (25th/75th percentiles; Figure 5B). The PHTs persist on average about 60 min, with durations ranging between 45 and 75 min (25th/75th percentiles; Figure 5C). The statistical characteristic of the mean track speed per cell is 33.0 km h⁻¹ on average (median), ranging from 21.7 to 46.4 km h⁻¹ (25th/75th percentiles; Figure 5D). Note that in environments with many convective cells, links between consecutive time steps can influence the calculated track speed if the algorithm jumps between neighboring cell cores. Such jumps, which may occur on days with a high density of closely spaced cells (e.g., in mesoscale convective systems), can artificially increase the recorded displacement. Even though the track still represents an area potentially affected by hail, these jumps can lead to an overestimation of track speed. For this reason, we have not specified a maximum value for track speed (as it was observed especially at the upper end of the distribution). On average, however, the values remain realistic, as confirmed by plausibility checks.

Figure 6 clearly illustrates that the majority of tracks propagate from southwest to northeast (53% PHTs are between 202.5° and 270.0°). This aligns with various studies on SCSs that link heavy rain or hail events in central Europe with mid-tropospheric flow characteristics (e.g., Wapler and James, 2015; Piper et al., 2019; Merino et al., 2019; Whitford et al., 2024), which in turn influence the movement direction of the associated cells. Additionally, the advection of warm, moist and thus often unstable air masses from the south and southwest is well-known to create optimal settings for convection-favoring conditions over western and central Europe (van Delden, 2001; Mohr et al., 2019; Barras et al., 2021).

These findings are further supported when the analyses are combined with large-scale weather patterns such as the Atlantic-European weather regimes according to Grams et al. (2017). The results (Figure 7) show that half (50.0%) of days with PHTs are primarily associated with the three blocking regimes: Scandinavian Blocking (23.4%), European Blocking (13.2%), and Greenland Blocking (13.3%). All of these regimes support the convection-favoring large-scale atmospheric condition with a south-to-southwesterly mid-tropospheric flow direction over central Europe. In contrast, the remaining four regimes play a minor role (21.5%): Atlantic Trough (7.2%), Atlantic Ridge (3.1%), Zonal Regime (5.0%), and Scandinavian Trough (6.2%). Days without a clearly defined regime (so-called no regime), representing the climatological mean, account for about 28.5%. Similar relationships between these weather regimes and SCSs in general have already been observed in central Europe (Mohr et al., 2019; Augenstein, 2025). Overall, this indicates that PHTs occurred predominantly under blocking regimes (50%) or during periods without a clearly defined large-scale flow pattern (“no regime”), together accounting for approximately 80% of all PHT days.

To study the spatial variability of events with different characteristics, the PHT catalog was aggregated onto a coarser 25 ×  25 km² grid, ensuring a sufficient number of events in all grid points. Considering only PHTs of at least 50 km (approximately the 83rd percentile of the track length distribution), the spatial distribution of the PHTs (Figure 8A) remains similar to that of all tracks (Figure 2), although only 17.0% of all tracks reach this length. Nevertheless, a conspicuous maximum is visible southeast of Reutlingen with almost four events per grid point. Higher values also occur in southern Bavaria around Munich, as well as in the area between Frankfurt, Kassel, and Cologne. PHTs reaching 100 km in length are rare, accounting for only 5.0% of all PHTs, which results in fewer events per grid point (Figure 8B). In the same areas, maxima of approximately 1.5 PHTs per year are observed, but differences compared to the surrounding regions are minor. In the end, this highlights that long hail tracks can occur across all parts of Germany.

In order to examine whether the results in Figure 6 show a uniform distribution across Germany or if regional differences exist, PHTs with a direction between 0 ° and 180 ° were defined in this study as eastern tracks, and those between 180 ° and 360 ° as western tracks. As shown in Figure 6, western tracks dominate the dataset (80.0%). The spatial distribution of the annual number of hail events with western directions (Figure 8C) is close to that of all tracks, with a clear maximum of more than nine events southeast of Reutlingen. High values also occur in the Bavarian Prealps. The strong maximum over the Swabian Jura supports the hypothesis of flow convergence in this zone for (south)westerly flow directions (see Section 3.1). In contrast, the distribution of eastern tracks (20.0%) shows different patterns (Figure 8D): No maxima are found over southern Germany. The highest numbers, with values below two PHTs per year, occur in eastern Germany, north of Frankfurt, and north of Cologne. Overall, the number of eastern tracks is significantly smaller than that of western tracks. The small regional variations suggest a more uniform spatial distribution across Germany.

Investigating the temporal clustering of extreme events, which can often be linked to persistent large-scale weather patterns, is also relevant, as periods with repeated occurrences (compound events) can lead to cumulative impacts (e.g., Zscheischler et al., 2020; Xoplaki et al., 2025; Küpfer et al., 2025). The following analysis examines whether days with detected PHTs cluster in time, either on the daily scale (spatial clustering) or with respect to their persistence across consecutive days (serial clustering). On a daily scale, 61.3% of all days in the study period had no PHTs across Germany (not shown). Of the remaining days with at least one PHT, most contain only a few tracks (Figure 9A): 28.5% show one or two PHTs, about half (47.2%) have up to five, and roughly 65% have no more than 10 tracks. However, there are also some days with significantly higher amounts. For instance, on 30 May 2017, a record of 94 PHTs occurred over large parts of Germany. Additionally, some other days feature a substantial number of PHTs. Approximately 2.4% of all days with tracks have at least 50 PHTs per day. Most of these days fall in June and July.

The serial clustering can provide information about the persistence of the underlying atmospheric conditions and thus also about an ongoing hazard. As with the PHT amounts per day (above), the following statistics refer to the whole of Germany. Therefore, two consecutive hail days do not mean that a particular location was hit by potential hail twice, only that a potential hailstorm occurred somewhere in Germany on both days. In extreme cases, this could mean that 1 day, a PHT occurred in the far north of Germany, and the next day, it occurred in the far south. Of the days with at least one PHT, 41.6% appear isolated (Figure 9B), meaning that the day before and the day after have no PHT anywhere across Germany. In 20.8% of the cases, two consecutive days have at least one PHT, in 12.0% they have three consecutive days. Longer continuous periods are uncommon, as a single day without PHTs resets the count. The longest continuous period with PHTs was 14 days, and this was observed three times during the 20-year study period. Notably, two of these episodes occurred in quick succession with only a brief interruption, spanning from 18 June to 1 July 2021 and from 4 July to 17 July 2021. Both periods were accompanied by different blocking regimes (cf. Kunz et al., 2022). Initially, there were 4 days with no regime, followed by a prolonged blocking phase. During this phase, the atmospheric circulation alternated between European blocking, Scandinavian blocking, and then European blocking again. The third 14-day period occurred from 26 May to 8 June 2016, when atmospheric blocking caused also favorable convective conditions over 2 weeks (cf. Piper et al., 2016).