Search

Annual maxima of daily precipitation sums can be typically described well with a stationary generalized extreme value (GEV) distribution. In many regions of the world, such a description does also work well for monthly maxima for a given month of the year. However, the description of seasonal and interannual variations requires the use of non-stationary models. Therefore in this paper we propose a non-stationary modelling strategy applied to long time series from rain gauges in Germany. Seasonal variations in the GEV parameters are modelled with a series of harmonic functions and interannual variations with higher ordered orthogonal polynomials. By including interactions between the terms, we allow for the seasonal cycle to change with time. Frequently, the shape parameter ξ of the GEV is estimated as a constant value also in otherwise instationary models. Here, we allow for seasonal-interannual variations and find that this is benefical. A suitable model for each time series is selected with a step-wise forward regression method using the Bayesian Information Criterion (BIC). A cross-validated verification with the Quantile Skill Score (QSS) and its decomposition reveals a performance gain of seasonal-interannual varying return levels with respect to a model allowing for seasonal variations only. Some evidence can be found that the impact of climate change on extreme precipitation in Germany can be detected, whereas changes are regionally very different. In general an increase of return levels is more prevalent than a decrease. The median of the extreme precipitation distribution (2-year return level) generally increases during spring and autumn and is shifted to later times in the year, heavy precipitation (100-year return level) rises mainly in summer and occurs earlier in the year.

Severe winter wind storms are related with strong impacts. We could show in recent studies that the co-occurrence of extreme wind and precipitation is leading to higher damages of residential buildings in comparison to non-compound events. This was done using ERA5 reanalysis data for the European winter season and daily insurance records of damages for residential buildings over Germany provided by the German Insurance Association (GDV).This study investigates the representation of co-occurrence of extreme wind and precipitation for climate simulations of the Coordinated Regional Climate Downscaling Experiment (CORDEX) for Europe (EURO-CORDEX). We use multi-model ensemble simulations with horizontal resolution of 0.44° and 0.11°. Individual simulations are analysed with respect to the characteristic of compound events for historical projections. Climate change signals for future scenarios are discussed.

Precipitation is commonly analysed from an Eulerian perspective, in which rainfall is considered at a fixed location. Lagrangian analysis of precipitation represents an alternative approach. Here, precipitation objects – for example, convective cells – are identified in a precipitation field and are then tracked through space and time, allowing object properties over the whole life of a convective cell to be collected. This approach offers additional insights into the mechanisms by which convective cells develop and behave across their lifecycle, which would not be evident from standard analysis methods.In this study, we perform Lagrangian analysis of convective cells under different large-scale circulation regimes. Tracking is based on convection-permitting simulations with the COSMO-CLM at 0.025° resolution over central Europe. All identified precipitation objects are tracked through space and time, collecting cell characteristics for each object, e.g. cell area, intensity, distance travelled, etc. Here we associate precipitation objects with categorical synoptic-scale circulation patterns and compare the cell properties between the different categories.

High wind speeds are a major cause for tree fall. However, the impact of wind on trees can depend on other co-occurring or antecedent meteorological conditions. Tree fall can also have an impact on surrounding infrastructure like the railway system. Here, it may damage railway infrastructure and lead to train disruptions. The intensity and frequency of windstorms and other meteorological factors is expected to change in the course of a changing climate. Thus, their impacts on trees and tree damage might change as well. To understand these changes an examination of impact-relevant weather factors and sequences is needed.We obtained a dataset for tree and branch fall events alongside German railway lines from the Deutsche Bahn for the years 2017 to 2021. We use logistic regression to model tree fall probabilities and to identify relevant current and antecedent weather factors, their combinations and their impact on tree fall risk during winter. We use meteorological predictors derived from the ERA5 reanalysis and RADOLAN radar data.High wind speed is identified as the strongest risk increasing factor. However, high daily precipitation and high soil water volume during the tree fall event as well as an antecedent year with high precipitation also increase tree fall risk. A small decreasing effect was found for warm and wet soil conditions in the three preceding months. We found no or only minor effects for daily mean air temperature, storm duration and wind direction.In further steps these factors and their combinations will be assessed in terms of their effect on occurrence probabilities of tree fall events under recent and future climate conditions based on regional climate simulations.

The influence of meteorological (pre-) condition on landslide probability in the German low mountain regions is assessed and effects arising from climate change are investigated. The landslide events analysed for this study are taken from the landslide database for Germany (Damm and Klose, 2015) and from an event inventory from the German railway company (Deutsche Bahn). We follow two different approaches in order to determine the influence of atmospheric conditions on hillslope failure.The first approach is based on weather types. Each day is assigned one of 28 Lamb-style weather types. The meteorological variables used to classify the weather types are sea level pressure and anomalies of the atmospheric water content. We were able to identify 4 patterns associated with a statistically significant increase for landslide frequency. The climate change signal of the frequency for the occurrence for these weather types is investigated in a multi-model ensemble of regional climate simulations (EURO CORDEX). The majority of the models shows a decrease in the frequency of those relevant patterns under RCP8.5 scenario conditions. In most models this decrease is, however, not statistically significant. The second approach is based on logistic regression. The logistic regression model was fitted using meteorological observations close to the landslide sites. Conditions at the day of the event as well as the pre-conditions from the days leading up to the event were considered. In order to select the best statistical model we tested a large number of physically plausible combinations of meteorological predictors. Each model was checked using cross-validation. The decision on the final model was based on the value of the logarithmic skill score and on expert judgement. As relevant predictors we identified daily precipitation, frost, and a soil moisture proxy determined from multi-day accumulated precipitation and potential evapotranspiration. The climate change signal is determined by applying the statistical model to the output of a multi-model ensemble of climate scenario simulations. Damm, B. and Klose, M. (2015): The landslide database for Germany: Closing the gap at national level, Geomorphology, 249, 82-93, https://doi.org/https://doi.org/10.1016/j.geomorph.2015.03.021.

Additive Logistic Regression Models for different convective hazards were developed across Europe and a portion of North America using lightning observations, severe weather reports and convective parameters from the ERA5 reanalysis. To model convective hazards, convective initiation was taken explicitly into account by computing, for instance, Phail (probability of hail) as the product of P­storm (probability of convective initiation) and Phailstorm (conditional probability of hail given a storm). We will report on the development of models that are skilful across Europe and North America, and on the regionally dependent skill of convective parameters used as model predictors. To reconstruct the probability of lightning, large hail, and very large hail from 1950 to 2021, the models were applied to the ERA5 reanalysis, at one hourly intervals across both regions. The modelled hazard climatologies are in strong agreement with observed patterns and can accurately resolve local-scale features thanks to the (0.25 x 0.25° degree) spatial resolution of the ERA5 reanalysis. We analysed long-term trends over the 71 year period and detected important differences between the regions. Across North America, 1950–2021 hail trends were found to be weak and mostly non-significant, but a period of enhanced lightning activity (+30% to 1950–2021 average) was detected between 1980 and 1990 across the Central Plains. In Europe, trends are mostly positive and significant, with the highest trend modelled across Northern Italy. Here, the convective activity has seen an abrupt increase with very large hail 3 times as likely in recent years (2012–2021) than in the 1950s. Apart from a sharp increase in frequency, the year-to-year variability has also increased with yearly differences in occurrence exceeding 100% for large and 200% for very large hail compared to the long-term average. In addition to (very) large hail, preliminary results on the development of models for severe wind gusts and tornadoes models along with the corresponding long-term trends will be presented.

The effect of climate change on rockfall in the German low mountain regions is investigated following two different approaches. The first approach uses a logistic regression model that describes the combined effect of precipitation, freeze-thaw cycles and fissure water on rockfall probability. The climate change signal for past decades is analysed by applying the model to meteorological observations. The possible effect of climate change until the end of the century is explored by applying the statistical model to the output of a multi-model ensemble of 23 regional climate scenario simulations. It is found that the number of days per year exhibiting an above-average probability for rockfall has been mostly decreasing during the last decades. Statistical significance is, however, present only at few sites. A robust and statistically significant decrease can be seen in the RCP8.5 climate scenario simulations for Germany and neighbouring regions, locally falling below -10 % when comparing the last 30 years of the 20th century to the last 30 years of the 21st century. The most important factor determining the projected decrease in rockfall probability is a reduction in the number of freeze-thaw cycles expected under future climate conditions. For the second approach four large-scale meteorological patterns that are associated with enhanced rockfall probability are identified from reanalysis data. The frequency of all four patterns exhibits a seasonal cycle that maximizes in the cold half of the year (winter/spring). Trends in the number of days that can be assigned to these patterns are determined both in meteorological reanalysis data and in climate simulations. In the reanalysis no statistically significant trend is found. For the future scenario simulations all climate models show a statistically significant decrease in the number of rockfall promoting weather situations.

Lagrangian analysis of convective precipitation involves identifying convective cells (“objects”) and tracking them through space and time. The Lagrangian approach helps to gain insight into the physical properties and impacts of convective cells and, in particular, how these may respond to climate change. Lagrangian analysis requires both a fixed definition of what constitutes a convective object and a reliable tracking algorithm. Whether the climate-change signals of various object properties are sensitive to the choice of tracking algorithm or to how a convective object is defined has received little attention. Here we perform ensemble pseudo global warming experiments at convection-permitting resolution to test this question. Using two conceptually different tracking algorithms, Lagrangian analysis is systematically repeated with different thresholds for defining a convective object, namely minimum values for object area, intensity and lifetime. We find that the tracking method has no impact on the detected climate-change signal. The criteria for identifying a convective object, however, can have a strong and statistically significant impact on the magnitude of the climate-change signal, for all analysed object properties. For the case considered in our study, this insight reveals that projected changes in the characteristics of convective rainfall vary considerably between cells of differing intensity, area and lifetime; for example, an increase in the area of moderate-intensity cells alongside a decrease for the most intense cells. Our results suggest that for Lagrangian analysis of precipitation in climate models, sensitivity analysis of the climate-change signal in relation to how an object is defined is a useful enhancement.

Weather conditions can affect the probability of road accidents. However, the effect can be different, depending on the vehicle type, the type of collision or road characteristics at the location of the accident. The aim of this study is to quantify the combined effects of traffic volume and different meteorological parameters on hourly probabilities of different accident types using generalized additive models. These models are based on police reports, which are available at the level of administrative districts, as well as reanalysis data and radar-based precipitation. Using tensor product bases, we model non-linear relationships and combined effects of different meteorological parameters. We evaluate the increase in relative risk of different accident types in case of precipitation, sun glare and high wind speeds.The largest effect of snow is found in case of single-truck accidents, while rain has a larger effect on single-car accidents. Precipitation particularly increases the relative risk of run-off-road accidents, as well as accidents at curves and descending road sections. A increasing effect of sun glare on accident probability was found in case of multi-car accidents, in particular in case of rear-end crashes. High wind speeds increase the risk of single-truck accidents and, for all vehicle types, the risk of collisions with objects blown on the road. A comparison of the predictive skill of models with and without meteorological variables shows an improvement of scores of up to 24%. This makes the models suitable for applications in real-time traffic management or impact-based warning systems and have the potential to improve risk perception and behavior of warning recipients.

Severe storm events are one of the most damaging natural hazards in Central Europe and thereby under special focus in disaster risk management. One key element of risk reduction is the vulnerability. While in risk assessments vulnerability is often assumed as constant, in reality different drivers lead to dynamic developments. To take the step from constant to dynamic vulnerability a detailed understanding of the processes and impacts of these influences is necessary. One crucial driver on vulnerability is the hazard itself. In theory vulnerability is likely to increase immediately after an event (De Reiter et al. 2020), whereas on long term the mean vulnerability decreases due to improved reconstruction (UNISDR, 2018).In this study we quantify the effects of storm events on the dynamics of vulnerability, with a focus on the impact of the intensity of previous events as well as the duration between events. In a first step vulnerability curves are derived from the relation of storm events and damage using generalized linear models. For the damage the loss ratio (loss/insured sum), extracted from a 23 yearlong data set of daily storm and hail damages for insured residential buildings in Germany on county level provided by the German Insurance Association, is used. The storm event is described by the three parameter daily maximum wind gust, duration of the storm event and gust factor, all calculated from ERA5 reanalysis. The threshold for the occurrence of an event is the exceedance of the 98th percentile of the wind gust. In the next step the generalized linear model is extended with the storm intensity of the previous event and the duration between two events as additional covariates. First results confirm the thesis of vulnerability decrease due to previous events. In further analyses, the results will be quantified in more detail and key findings of the vulnerability dynamics will be elaborated.

Historical temperature measurements are the basis of global climate datasets like HadCRUT4. This dataset contains many missing values, particularly for periods before the mid-twentieth century, although recent years are also incomplete. Here we demonstrate that artificial intelligence can skilfully fill these observational gaps when combined with numerical climate model data. We show that recently developed image inpainting techniques perform accurate monthly reconstructions via transfer learning using either 20CR (Twentieth-Century Reanalysis) or the CMIP5 (Coupled Model Intercomparison Project Phase 5) experiments. The resulting global annual mean temperature time series exhibit high Pearson correlation coefficients (≥0.9941) and low root mean squared errors (≤0.0547 °C) as compared with the original data. These techniques also provide advantages relative to state-of-the-art kriging interpolation and principal component analysis-based infilling. When applied to HadCRUT4, our method restores a missing spatial pattern of the documented El Niño from July 1877. With respect to the global mean temperature time series, a HadCRUT4 reconstruction by our method points to a cooler nineteenth century, a less apparent hiatus in the twenty-first century, an even warmer 2016 being the warmest year on record and a stronger global trend between 1850 and 2018 relative to previous estimates. We propose image inpainting as an approach to reconstruct missing climate information and thereby reduce uncertainties and biases in climate records.From:Kadow, C., Hall, D.M. & Ulbrich, U. Artificial intelligence reconstructs missing climate information. Nature Geoscience 13, 408–413 (2020). https://doi.org/10.1038/s41561-020-0582-5The presentation will tell from the journey of changing an image AI to a climate research application.

The complexity of the climate system calls for a combined approach of different knowledge areas. For that, increasingly larger projects need a coordinate effort that fosters an active collaboration between members. On the other hand, although the continuous improvement of numerical models and larger observational data availability provides researchers with a growing amount of data to analyze, the need for greater resources to host, access, and evaluate them efficiently through High Performance Computing (HPC) infrastructures is growing more than ever. Finally, the thriving emphasis on FAIR data principles [1] and the easy reproducibility of evaluation workflows also requires a framework that facilitates these tasks. Freva (Free Evaluation System Framework [2, 3]) is an efficient solution to handle customizable evaluation systems of large research projects, institutes or universities in the Earth system community [4-6] over the HPC environment and in a centralized manner. Freva is a scientific software infrastructure for standardized data and analysis tools (plugins) that provides all its available features both in a shell and web environment. Written in python, is equipped with a standardized model database, an application-programming interface (API) and a history of evaluations, among others:An implemented metadata system in SOLR with its own search tool allows scientists and their plugins to retrieve the required information from a centralized database. The databrowser interface satisfies the international standards provided by the Earth System Grid Federation (ESGF, e.g. [7]). An API allows scientific developers to connect their plugins with the evaluation system independently of the programming language. The connected plugins are able to access from and integrate their results back to the database, allowing for a concatenation of plugins as well. This ecosystem increases the number of scientists involved in the studies, boosting the interchange of results and ideas. It also fosters an active collaboration between plugin developers. The history and configuration sub-system stores every analysis performed with Freva in a MySQL database. Analysis configurations and results can be searched and shared among the scientists, offering transparency and reproducibility, and saving CPU hours, I/O, disk space and time. Freva efficiently frames the interaction between different technologies thus improving the Earth system modeling science. This framework has undergone major refactoring and restructuring of the core that will also be discussed. Among others:Major core Python update (2.7 to 3.9). Easier deployment and containerization of the framework via Docker. More secure system configuration via Vault integration. Direct Freva function calls via python client (e.g. for jupyter notebooks). Improvements in the dataset incorporation. References:[1] https://www.go-fair.org/fair-principles/[2] Kadow, C. et al. , 2021. Introduction to Freva – A Free Evaluation System Framework for Earth System Modeling. JORS. http://doi.org/10.5334/jors.253[3] gitlab.dkrz.de/freva[4] freva.met.fu-berlin.de[5] https://www.xces.dkrz.de/[6] www-regiklim.dkrz.de[7] https://esgf-data.dkrz.de/projects/esgf-dkrz/

Extratropical windstorms are amongst the highest rated perils for the European continent. Extreme wind speeds of these synoptic scale systems occur primarily in the winter season and often cause damage to buildings, forests and infrastructure, and thus can have large socio-economic impacts.In our studies of extratropical windstorms in the CMIP6 model ensemble, we found remarkable trends of opposite sign in the wind speed during the historical period. More specifically, we found a continuous increase in the surface wind speed in the early historical period between 1850 and 1920, and an even stronger decrease thereafter until the present.In a case study with one of the models (MPI-ESM) we found that the trends in the wind speed relate to a trend of opposite sign in the roughness length, thus the wind speed increases in eras with a decrease in the surface roughness (and tree fraction) and vice versa. While this relationship is expected and physically reasonable, it appears that the interaction of surface parameters with the atmosphere was different in CMIP5 climate models, as there is no comparable reaction of surface wind speeds to the trends in surface parameters (e.g. tree fraction).Since the historical era serves as the reference for any derived climate change signal, these trends might affect the amplitude of the changes in a future climate and the derived conclusions. Also, state of the art climate change signals regarding storminess might need to be reconsidered with this newly represented land-atmosphere interaction in the models.We further explore this phenomenon by eliminating the influence of the roughness on the wind speed and investigate the effect that this correction has on the appearance of climate change signals of extratropical windstorms.

Over the last decades the change in the Arctic climate resulted in related sea-ice retreat and a much faster warming of the Arctic compared to the global average (Arctic amplification). These changes in sea ice can affect the large-scale atmospheric circulation over the mid-latitudes, in particular atmospheric blocking, and – mediated by the changes in blocking – the frequency and severity of related extreme events. As a step towards a better understanding of changes in weather and climate extremes over Central Europe (C-EU) associated with Arctic climate change, we study the linkage between periods of low and high Arctic sea ice area and the frequency of winter cold days in C-EU. Since frequency of winter cold days in C-EU is associated with atmospheric blocking, especially over the Ural and Scandinavian region, we investigate frequency changes of cold days with respect to the occurrence of blocking in different Euro-Atlantic regions by composite analysis based on ERA5 reanalysis data. To separate the resulting changes from the global warming trend and associated Arctic sea ice loss, monthly sea ice area data is first detrended and then divided by the median into two parts representing either low or high sea ice periods. The frequency of occurrence of cold days with respect to both sea ice periods is then calculated and compared. The same procedure is applied to cold days occurring during a blocked day in certain regions to analyze the change of linkage between atmospheric blocking and cold days induced by different sea ice area periods. Preliminary results indicate an increased occurrence of cold days in Central Europe during low sea ice periods, which is enhanced during the occurrence of Ural Blocking.

Today's climate science is being driven by IT more than ever. Earth system models on high-performance computers (HPC) are common tools for researching the past and projecting it into the future. In addition to that, statistical modelling is reborn thanks to modern computer architectures equipped with artificial intelligence (from ensemble to deep learning). Future advances in machine learning will also shape climate research through analysis tools, prediction techniques, signal and event classification, post-processing, Model Output Statistics (MOS), evaluation and verification, etc. This presentation will look at nowadays research about the future (part one) and the past (part two) of our climate system using AI/ML ideas and technologies in combination with numerical climate models - from two publications accordingly. A special focus will be on the importance of climate science, where the needs are, and how to choose the AI/ML hammer wisely:(1) FUTURE: Derived from machine (ensemble) learning and bagging, a new hybrid climate prediction technique called 'Ensemble Dispersion Filter' is developed. It exploits two important climate prediction paradigms: the ocean's heat capacity and the advantage of the ensemble mean. The Ensemble Dispersion Filter averages the ocean temperatures of the ensemble members every three months, uses this ensemble mean as a restart condition for each member, and further executes the prediction. The evaluation shows that the Ensemble Dispersion Filter results in a significant improvement in the predictive skill compared to the unfiltered reference system. Even in comparison with prediction systems of a larger ensemble size and higher resolution, the Ensemble Dispersion Filter system performs better. In particular, the prediction of the global average temperature of the forecast years 2 to 5 shows a significant skill improvement. Kadow, C., Illing, S., Kröner, I., Ulbrich, U., and Cubasch, U. (2017), Decadal climate predictions improved by ocean ensemble dispersion filtering, J. Adv. Model. Earth Syst., 9, 1138– 1149, doi:10.1002/2016MS000787. (2) PAST: Nowadays climate change research relies on climate information of the past. Historic climate records of temperature observations form global gridded datasets like HadCRUT4, which is investigated e.g. in the IPCC reports. However, record combining data-sets are sparse in the past. Even today they contain missing values. Here we show that artificial intelligence (AI) technology can be applied to reconstruct these missing climate values. We found that recently successful image inpainting technologies, using partial convolutions in a CUDA accelerated deep neural network, can be trained by 20CR reanalysis and CMIP5 experiments. The derived AI networks are capable to independently reconstruct artificially trimmed versions of 20CR and CMIP5 in grid space for every given month using the HadCRUT4 missing value mask. The evaluation reaches high temporal correlations and low errors for the global mean temperature. Kadow, C., Hall, D.M. & Ulbrich, U. Artificial intelligence reconstructs missing climate information. Nat. Geosci. 13, 408–413 (2020). https://doi.org/10.1038/s41561-020-0582-5

While it is known that severe winter wind storms are related with strong impacts, this study investigates the enhanced impact of compound precipitation and wind extremes. Therefore, we analyse the co-occurrence of extreme wind and precipitation using ERA5 reanalysis data for the European winter season. Co-occurring events are defined by simultaneous threshold exceedance of daily wind speed and precipitation in same or neighbouring areas.For the quantification of impacts, we are using daily insurance records of damages for residential buildings over Germany provided by the German Insurance Association (GDV). Using the definition of co-occurring extremes, those damage records can be grouped into compound and non-compound events. Analysing insurance loss data between 1997-2016 allows comparisons of the distribution of both groups. There are much more events in the non-compound group. On the other hand, the distribution of the compound group is shifted towards higher damages with an increased median of a factor of ten.

In this study we assess the influence of changes in the relevant meteorological conditions on the probability for rockfall in German low mountain regions. The study is based on data from a rockfall data base for Germany (Rupp and Damm, 2020) and a data set supplied by the Deutsche Bahn (German railway company) covering the periods 1838-2018 and 2015-2020, respectively. In a first approach, a logistic regression model for the probability of rockfall at a given location developed by Nissen et al. 2021 was applied to gridded meteorological station observations (RADOLAN and EOBS) ranging from the year 1950 to 2019. The logistic regression model quantifies the influence of daily precipitation, a proxy for pore water and freeze-thaw cycles on rockfall probability. A probability forecast was made for each day and location. The day-to-day variability in rockfall probability at the individual sites is high. Thus, the sign of the trends is site specific, but the majority of sites is showing a negative trend over the 70-year period investigated. The significance of the trends at most sites is below the 95% level. Sites at which the trend is statistically significant almost all show a negative trend, down to -4% per decade in terms of the annual number of days with a higher than climatological hazard. The mean probability decreased by as much as -2.3% per decade. The second approach is based on large-scale weather patterns. An analysis identified 3 weather pattern that occur on average at 9% of all days but include 19% of the days on which a rockfall event occurred. The trend in the number of these patterns was determined for the last 40 years. It suggests a decrease by -2.2% per decade and is not statistically significant.

Precipitation is typically analysed from an Eulerian perspective, in which rainfall is considered at a fixed location. Lagrangean analysis of precipitation represents an alternative approach. Here, precipitation objects – for example, convective cells – are identified in a precipitation field and are then tracked through space and time, allowing object properties over the whole life of a convective cell to be collected. For the study of precipitation under climate change, this approach may offer additional insights into the mechanisms by which precipitation increases or decreases; for example, changes in cell lifetime or areal extent.In this study, we analyse the climate-change response of convective cells’ properties by combining Lagrangean tracking with the pseudo global warming (PGW) modelling approach. A 14-day period of unusually high convective activity over central Europe is first simulated under observed conditions in an 18-member ensemble with the COSMO-CLM at convection-permitting resolution (2.8 km). The ensemble is then re-simulated under PGW conditions, created by modifying the initial and boundary conditions based on an RCP8.5 end-of-century scenario. All convective objects are then systematically tracked through space and time. Here we present the response to warming of the convective cell characteristics for the study period, and explore the variability of these changes across the full distribution of objects. Cell characteristics considered include cell area, intensity, volume, lifetime and distance travelled.

The number of damaging events caused by natural disasters are increasing because of climate change. Projects of public interest such as ClimXtreme (Climate Change and Extreme Events [1, 2]), aim to improve our knowledge of extreme events, the influence of environmental changes and their societal impacts. ClimXtreme takes a holistic approach this problem through different knowledge areas. For that, projects like this need a coordinate effort from many interdisciplinary groups. On the other hand, the continuous improvement of numerical models and increase on observational data availability provides researchers with a growing amount of data to analyze, and the need for greater resources to host, access, and evaluate them efficiently through High Performance Computing (HPC) infrastructures is growing more than ever. Finally, the emphasis these last years on FAIR data principles [3] and the easy reproducibility of evaluation workflows also requires a framework that facilitates these tasks. Freva (Free Evaluation System Framework [4, 5]) is an efficient solution to handle customizable evaluation systems of large research projects, institutes or universities in the Earth system community [6-8] over the HPC environment and in a centralized manner. Freva is a scientific software infrastructure for standardized data and analysis tools (plugins) that provides all its available features both in a shell and web environment. Written in python, is equipped with a standardized model database, an application-programming interface (API) and a history of evaluations, among others: An implemented metadata system in SOLR with its own search tool allows scientists and their plugins to retrieve the required information from a centralized database. The databrowser interface satisfies the international standards provided by the Earth System Grid Federation (ESGF, e.g. [9]). An API allows scientific developers to connect their plugins with the evaluation system independently of the programming language. The connected plugins are able to access from and integrate their results back to the database, allowing for a concatenation of plugins as well. This ecosystem increases the number of scientists involved in the studies, boosting the interchange of results and ideas. It also fosters an active collaboration between plugin developers. The history and configuration sub-system stores every analysis performed with Freva in a MySQL database. Analysis configurations and results can be searched and shared among the scientists, offering transparency and reproducibility, and saving CPU hours, I/O, disk space and time. The usage of Freva in the context of ongoing large projects like ClimXtreme will be discussed. Additionally, major updates of the software, system deployment, and core functionalities will be presented. References: [1] https://www.fona.de/de/massnahmen/foerdermassnahmen/climxtreme.php [2] https://www.climxtreme.net/index.php/en/ [3] https://www.go-fair.org/fair-principles/ [4] Kadow, C. et al. , 2021. Introduction to Freva – A Free Evaluation System Framework for Earth System Modeling. JORS. http://doi.org/10.5334/jors.253 [5] gitlab.dkrz.de/freva [6] freva.met.fu-berlin.de [7] https://www.xces.dkrz.de/ [8] www-regiklim.dkrz.de [9] https://esgf-data.dkrz.de/projects/esgf-dkrz/

Historical temperature measurements are the basis of global climate datasets like HadCRUT4. This dataset contains many missing values, particularly for periods before the mid-twentieth century, although recent years are also incomplete. Here we demonstrate that artificial intelligence can skilfully fill these observational gaps when combined with numerical climate model data. We show that recently developed image inpainting techniques perform accurate monthly reconstructions via transfer learning using either 20CR (Twentieth-Century Reanalysis) or the CMIP5 (Coupled Model Intercomparison Project Phase 5) experiments. The resulting global annual mean temperature time series exhibit high Pearson correlation coefficients (≥0.9941) and low root mean squared errors (≤0.0547 °C) as compared with the original data. These techniques also provide advantages relative to state-of-the-art kriging interpolation and principal component analysis-based infilling. When applied to HadCRUT4, our method restores a missing spatial pattern of the documented El Niño from July 1877. With respect to the global mean temperature time series, a HadCRUT4 reconstruction by our method points to a cooler nineteenth century, a less apparent hiatus in the twenty-first century, an even warmer 2016 being the warmest year on record and a stronger global trend between 1850 and 2018 relative to previous estimates. We propose image inpainting as an approach to reconstruct missing climate information and thereby reduce uncertainties and biases in climate records. As published in: Kadow, C., Hall, D.M. & Ulbrich, U. Artificial intelligence reconstructs missing climate information. Nat. Geosci. 13, 408–413 (2020). https://doi.org/10.1038/s41561-020-0582-5 Newest developments around the technology will be presented.

Das Wetter kann einen erheblichen Einfluss auf das Reiseverhalten der Menschen haben. Um diesen Zusammenhang zu quantifizieren und nutzbar zu machen, wird untersucht, ob meteorologische Variablen die Vorhersage des Verkehrsaufkommens in Deutschland verbessern können. Dazu werden stündlichen Verkehrszählungen an 1.400 Stationen an Bundesstraßen und Autobahnen verwendet, wobei zwischen Motorrädern, Pkw, Transportern und Lkw unterschieden wird. Es wird untersucht, inwieweit sich der mittlere quadratische Fehler von Poisson-Regressionsmodellen für stündliche Verkehrszählungen durch die Berücksichtigung von Niederschlag, Temperatur, Bewölkung und Windgeschwindigkeit verringert. In einem iterativen Verfahren werden den Modellen Prädiktorvariablen hinzugefügt, wenn diese den mittleren quadratischen Fehler reduzieren. Wir zeigen, dass insbesondere Motorradzahlen stark wetterabhängig sind. Auf Bundesstraßen reduziert sich der mittlere quadratische Fehler von Modellen für stündliche Motorradzahlen um bis zu 60%, wenn meteorologische Prädiktorvariablen zum Modell hinzugefügt werden. Eine detaillierte Analyse des Modells für Motorradzahlen zeigt nicht-lineare Beziehungen zwischen den meteorologischen Variablen und den Motorradzählungen. Für Pkw zeigt sich eine Wetterabhängigkeit insbesondere in touristischen Regionen wie Badeorten und Naturparks. Für Lieferwagen und Lkw zeigt sich nur eine geringe Wetterabhängigkeit. Die Ergebnisse eröffnen Möglichkeiten für unterschiedliche potenzielle Anwendungen, wie z. B. die Verbesserung der Routenplanung in Navigationssystemen, die Implementierung in Verkehrsmanagementsystemen, die Vorausplanung von Besucherzahlen in touristischen Gebieten oder zur Abschätzung der Exposition in Verkehrsunfallstudien.

In Großstädten sowie städtische Regionen sind die Auswirkungen von Extremereignissen (z. B. Starkregen) besonders stark spürbar. Da sich die durch gefährliche atmosphärische Ereignisse verursachten Probleme aufgrund des anthropogenen Klimawandels voraussichtlich verschärfen werden, ist eine angemessene Anpassungsplanung der städtischen Infrastruktur erforderlich. Dies erfordert nicht nur die weitere Erforschung möglicher Auswirkungen unter veränderten Klimabedingungen, sondern auch eine Überprüfung der praktischen Umsetzbarkeit für die betroffenen Interessengruppen und Akteure. Im Rahmen des BMBF-Forschungsprogramms "Stadtklima im Wandel" ([UC]²) setzen wir Starkniederschlagsereignisse mit den zugehörigen Feuerwehreinsätzen über Berlin statistisch in Beziehung. Die Niederschlagsdaten basieren dabei auf zeitlich hoch aufgelösten Radardaten. Die Einsatzdaten der Berliner Feuerwehr sind zwar auch in hoher zeitlicher und örtlicher Auflösung verfügbar, aber die Anzahl der registrierten Ereignisse ist gering (

The object-based method SAL (Structure, Amplitude and Location) was adapted for investigating the errors of forecasts of extreme 10‑m wind gusts associated with winter storms in Germany. It has been applied to a statistically downscaled version of the 51 member ECMWF (European Centre for Medium Range Weather Forecasts) operational ensemble forecast. The horizontal resolution of both downscaled data and of the German weather service's operational analysis data used for verification is 7 km. Forecast errors are subdivided in terms of storm intensity, location and extent. After identifying a set of storm events, objects of moderate and intense 10‑m wind gusts were identified with a local percentile-based threshold (90th percentile for moderate and 98th percentile for intense gust objects). Depending on the intensity of the storm, the gust objects differ in terms of size, shape and intensity. The characteristics of the ensemble forecasts of 10‑m wind gusts can basically be assessed in two different ways. Individual forecast members can be evaluated with respect to the location, intensity and extent of the gust field, and then address the ensemble characteristics by the score distributions. Alternatively, the gust fields' location, intensity and extent can be evaluated by directly using the ensemble mean forecast instead of the individual members. The results of the identified set of storms clearly indicate a high case-to-case variability in the predictability of 10‑m wind gusts objects, particularly when focusing on the structure of intense wind gust objects. It is found, that the gust fields' location and overall intensity can be better estimated from the ensemble mean forecast, compared to the individual forecast members. From a forecaster's perspective this means, that a storms' location and intensity can be well estimated by considering the ensemble mean wind forecasts. Considering the structure of the gust objects, results are different. While for longer lead times, there also seems to be a benefit from applying ensemble averaging, at short lead times the ensemble mean forecast performs equally or worse than most of the individual forecast members. The amplitude error is often the smallest component of the three error types. The findings are particularly relevant when deriving warning information, by giving guidance to forecasters when interpreting ensemble forecasts for severe storms.

A rockfall dataset for Germany is analysed with the objective of identifying the meteorological and hydrological (pre-) conditions that change the probability for such events in Central Europe. The factors investigated in the analysis are precipitation amount and intensity, freeze-thawing cycles as well as sub-surface moisture. As there is no suitable observational dataset for all relevant sub-surface moisture types (e.g. water in rock pores and cleft water) available, simulated soil moisture and parameterised pore water are tested as substitutes. The potential triggering factors were analysed both for the day of the event as well as for the days leading up to the event. It is found that the most important factor influencing rockfall probability in the research area is precipitation amount at the day of the event but the water content of the ground on that day and freeze-thawing cycles in the days prior to the event also influence the hazard probability. Comparing results with simulated soil moisture and parameterised pore water revealed that precipitation minus potential evaporation evaluated for a weekly period performs well as a proxy for the relevant sub-surface moisture types. A logistic regression model was built, which considers individual potential triggering factors as well as their interactions. Using this model the effects of meteorological conditions on rockfall probability in the Central European low mountain ranges can be quantified. The model suggests that precipitation is most efficient, when the moisture content of the ground is high. An increase of daily precipitation from its local 50th percentile to its 90th percentile approximately doubles the probability for a rockfall event under median sub-surface moisture conditions. When the moisture content of the ground is at its 95th percentile the same increase in precipitation leads to a four-fold increase in rockfall probability. The occurrence of a freeze-thaw cycle in the preceding days can further increase the rockfall hazard. The most critical combination can be expected in the winter season after a freeze-thaw transition which is followed by a day with high precipitation amounts and takes place in a region preconditioned by a high level of sub-surface moisture.

Winter windstorms are among the most dangerous and costly natural hazards in Central Europe. Their ability to cause tree and branch fall leads to disruptions and damages along railway systems. Along the German railway network the Deutsche Bahn is preventively removing about 30.000 trees per year. Still, each year a multiplicity of disturbances occur which may lead to delays, economic damages or even train collisions.A data set with vegetation disturbance events between 2017 and 2020 along the German railway system is provided by the Deutsche Bahn. The aim of this study is to use exploratory statistics as well as machine learning methods like regression techniques or decision trees to explore the relationship between vegetation damages and meteorological parameters like wind gusts, precipitation or temperature. Additionally, tree related factors and surrounding conditions like ground frost and soil moisture will be taken into account. Finally, we want to derive critical thresholds and combinations of weather parameters leading to significant damage risk. We find a positive relationship between vegetation disturbance and wind speeds. Especially strong winter storms leave a very clear signal in the disturbance time series. For example, the highest numbers of vegetation disturbances occurred during the winter storms Sabine (10.02.2020, 515 events) and Friderike (18.01.2018, 360 events). During winter storm days the majority of events occurs in those areas affected by high wind speeds. Tree fall disturbances peak during the winter storm season between January and March, while branch fall disturbances peak between June and August. However, a high number of events occurs also during times of low wind speeds. Additionally, high wind speeds do not necessarily lead to vegetation disturbances. It is clear that other meteorological and tree related factors need to be taken into account. Compound events as well as previous weather and soil conditions are expected to affect wind throw risks.

Weather conditions affect both road traffic volume and the probability of road accidents. The aim of this study is improve the understanding of both effects as well as their interactions. In a first step, we develop generalized linear models for hourly road traffic counts at 1400 traffic stations on German federal roads and highways. It is distinguished between different vehicle types, including motorbikes, cars, delivery vans and trucks. Different meteorological variables are derived from reanalysis and radar data. The impacts of these variables on the predictive skill of the models is analyzed. In particular models for motorbike counts show large improvements, if meteorological predictors are added to the model. At weekends in the afternoon the mean squared errors of modeled motorbike counts are reduced by up to 60%. Temperatures around 25°C, no precipitation, low cloud cover and low wind speeds lead to the highest motorbike counts. In a second step, the information derived from the traffic models is used to improve models for hourly probabilities of road accidents. These models are based on police reports, which are available at the level of administrative districts, and can now explicitly take traffic volume into account. It is shown that in particular winter conditions like precipitation and freezing temperatures lead to a significant increase in accident probability. Especially the probabilities of roadway departures show an increase under such conditions. The models presented in this study are suitable for the integration in risk-based warning systems and have the potential to improve risk perception and behavior of warning recipients.

Freva – Free Evaluation System Framework for Earth system modeling is an efficient solution to handle evaluation systems of research projects, institutes or universities in the climate community. It is a scientific software framework for high performance computing that provides all its available features both in a shell and web environment. The main system design is equipped with the programming interface, history of evaluations, and a standardized model database. Plugin – a generic application programming interface allows scientific developers to connect their analysis tools with the evaluation system independently of the programming language. History – the configuration sub-system stores every analysis performed with the evaluation system in a database. Databrowser – an implemented meta data system with its advanced but easy-to-handle search tool supports scientists and their plugins to retrieve the required information of the database. The combination of these three core components, increases the scientific outcome and enables transparency and reproducibility for research groups using Freva as their framework for evaluation of Earth system models.