<p>Hail is a pronounced natural hazard in Germany. Nevertheless, major hail events are quite rare and there is a lack of information in hail occurrence and size and its spatiotemporal distribution. Measurement sensors that are able to detect hail (e.g. disdrometers) are in principle available in Germany, but the spatial density of those stations is far lower than the typical spatial extent of hail events. Furthermore, sensors for hail size estimation are still in evaluation stage and currently only located at a few selected places. Hail reports based on professional and particularly amateurish eyewitness become increasingly important. But besides a certain degree of subjectivity in the reported hail size, highly populated areas might be overrepresented compared to rural and sparsely populated areas. Areal information from weather radar networks can overcome this issue with a high spatiotemporal resolution. Because of the high update frequency and fast availability of radar data, an automatic hail detection and hail size estimation might provide valuable hints to forecasters and supports the warning decision process.       <br />The Deutscher Wetterdienst (DWD) utilizes a C-Band dual-polarimetric weather radar network consisting of 17 radar stations that provide ten volume scans and a terrain-following low-elevation scan every five minutes. The operationally used hydrometeor classification algorithm HYMEC processes data of reflectivity, differential reflectivity and co-polar correlation coefficient to distinguish between hail and other hydrometeors. With this classification a hail distribution over Germany can already be derived. For the analysis of hail sizes, the Maximum Expected Size of Hail (MESH) and a method based on Vertical Integrated Ice (VII) are used. The latter method is motivated by a linear relation between maximum hail size and VII proposed by our forecasters based on their practical experience.        <br />This contribution will give an overview on the statistics of hail occurrence and hail size using the aforementioned algorithms in Germany during the convective seasons 2021 and 2022. Also, selected case studies are discussed in more detail. The results are compared against hail observations from manned and automatic weather stations, reports from the European Severe Weather Database and user reports from DWD’s WarnWetter-App.</p>

<p>Heavy rainfall caused by convection over small catchments represents a major challenge in flood forecasting. In Germany, each federal state runs its own flood forecasting center providing operational forecasts obtained from hydrological models. This system works well for large and medium-sized catchments. However, hydrological response happens fast for small catchments and operational models do not perform well. In many cases, these models only use hourly rainfall observations and NWP forecasts but do not take into account rainfall nowcasts.</p> <p>We aim at supporting German flood forecasting agencies in a co-design approach. Specifically, we intend to provide a novel post-processing product containing information about the extremity of catchment-specific areal rainfall. First, a nationwide catchment delineation is performed for each pixel of a 50 m x 50 m grid using a digital elevation model, while catchments smaller than 10 km² and greater than 1000 km² are discarded. Next, we perform an upscaling of catchment information to a 1 km x 1 km grid. Finally, we compute areal rainfall for each pixel using the underlying catchment geometry from rain gauge adjusted radar observations as well as seamless rainfall forecasts resulting from the SINFONY project.</p> <p>We consider various accumulation durations and perform a recalculation of areal rainfall for all catchments using a 20-year dataset of radar-derived rainfall. A partial series of areal rainfall accumulation results for each catchment and we fit extreme value distributions in order to provide information about the expected statistical return period of rainfall events and their potential hydrological impact in real-time. Extreme value statistics with time series of only 20 years is not sufficient to obtain reliable return periods for very rare events. In order overcome this limitation, we include long-term rain gauge statistics and combine them with areal rainfall recalculations in a reasonable manner.</p> <p>The resulting spatial distribution of areal rainfall with corresponding return periods based on an exhaustive collection of catchments appears to be a useful visualization technique to identify small catchments affected by heavy rainfall and can be computed in real-time using SINFONY products. In order to illustrate the benefit of this approach, we will show a comprehensive case study for the catastrophic flooding event on 14<sup>th</sup> of July 2021.</p>

Object-based cell detection and tracking algorithms provide a useful tool for analyzing current and past storm properties and movement. KONRAD3D, an algorithm which recently became operational at DWD, uses radar scans of different elevation angles to derive 3-dimensional cell objects with specific properties.  Nowcasting of storm position by displacing it using the current cell movement seems to be straightforward, however, predicting the life cycle or specific storm properties such as lifetime, maximum future severity, and hail occurrence and size is difficult. We aim at analyzing the potential of machine learning techniques (ML), in particular random forest and gradient boosting, to provide these predictions using KONRAD3D cell properties in combination with NWP and lightning activity data. We perform a recalculation of KONRAD3D for a 6-year time period, where we considered several other data sources to compute specific storm environment and attributes. For instance, NWP data from the ICON-EU model is used to characterize the convective environment, while lightning data is used quantify electrical activity. Next, we filter resulting storm detections regarding our three prediction tasks using ML: (1) Maximum expected future cell severity: all storms are considered. (2) Longevity of quasi-stationary cells: only storms with a minimum speed below a threshold are considered. (3) Hail size: all storms are considered. For tasks 1 and 2, we analyze the prediction performance for different storm ages, i.e. cell attributes at initial detection, after 15 min, and 30 min during the life cycle are used as predictors for ML. Maximum hail size (task 3) is estimated using individual cell detections. In order to achieve a final comparison of methods, we perform a 10-fold cross validation Longevity of quasi-stationary storm events is difficult to predict since there many events with a short duration and only very few with a long duration. The distribution of maximum expected severity is less skew and therefore prediction scores are better. Random forest and gradient boosting are preferred over artificial neural networks, since decision tree methods are easier and faster to train and also provide better cross validation scores. In case of maximum future severity, a substantial improvement compared to a simple persistence forecast is achieved. ML-based prediction of maximum hail size delivers reasonable results and is able to outperform classical methods, such as MESH. However, prediction uncertainty remains high in most cases and needs to be quantified in order to generate meaningful predictions.

Small-scale, convective heavy rainfall events present a major challenge in flood forecasting, primarily affecting smaller catchments. At such scales, flood forecasts are challenging due to short response times, a lack of stream gauges and limitations of operationally used hydrological models. These models are typically designed for larger catchments, therefore we explore alternative strategies to support the German flood forecasters. The AREA product (Areal Rainfall Extremity Assessment) aims at complementing the flood forecasters’ workflow based on hydrological modeling by rapidly identifying catchments affected by strong rainfall. We intend to provide a novel post-processing product containing information about the extremity of catchment-specific areal rainfall. First, a nationwide catchment delineation is performed for each pixel of a 50 m x 50 m digital elevation model, selecting only catchments smaller with an area between 10 and 500 km². The catchments and stream network are subsequently upscaled to the operational radar grid with a resolution of 1 km. Finally, areal rainfall is computed for each pixel using the underlying catchment geometry from rain gauge-adjusted radar observations as well as seamless rainfall forecasts resulting from the SINFONY project with lead times of up to 12h. In order to estimate the extremity of a catchment rainfall event in real-time, we derive return periods based on extreme value statistics. To that end, we consider various accumulation durations and perform a recalculation of areal rainfall for all catchments using a 20-year dataset of radar-derived rainfall. Given the short and limiting observation period of radar data, we attempt to combine the obtained extreme value distributions with existing, regionalized long-term rain gauge statistics (DWD-KOSTRA), in order to estimate longer return periods. The resulting spatial distribution of areal rainfall with corresponding return periods based on an exhaustive collection of catchments appears to be a useful visualization technique to identify small catchments affected by heavy rainfall. The product will be illustrated based on the analysis of specific case studies.