Gregor Ehrensperger
Gregor Ehrensperger has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Tobias Hell
Data Lab Hell GmbH, Zirl, Austria
5 shared publications - Thorsten Simon
Department of Atmospheric and Cryospheric Sciences, University of Innsbruck, Innsbruck, Austria
4 shared publications - Georg Johann Mayr
Department of Atmospheric and Cryospheric Sciences, Universität Innsbruck, Innsbruck, Austria
3 shared publications - Marc-André Falkensteiner3 shared publications
Tobias Hell
3 shared publications- Harald Schellander
Regional Office Tyrol, GeoSphere Austria, Innsbruck, Austria
2 shared publications
Latest Publications
- Open Access
- 25 May 2026
- Published by Elsevier in Weather and Climate Extremes
- Vol. 53
Building a high-resolution hail database for Austria and estimating hail hazard using a metastatistical approach
- 25 May 2026
- Published by Elsevier in Weather and Climate Extremes
- Vol. 53
Open AccessBuilding a high-resolution hail database for Austria and estimating hail hazard using a metastatistical approach
- Open Access
From Radar to Risk: Building a High-Resolution Hail Database for Austria And Estimating Risk Through the Integration of Distributional Neural Networks into the Metastatistical Framework
This study makes significant contributions to the understanding of hail climatology in Austria. First, it introduces a comprehensive database of hailstone sizes, constructed from three-dimensional radar data spanning 2009 to 2022 and calibrated by approximately 5000 verified hail reports. The database serves as foundation for describing the short-term climatology of hail and provides the data necessary for estimating hail risk maps with enhanced spatial resolution and quality. Second, the study enables the spatio-temporal metastitical extreme value distribution (TMEVD) to feature return levels of up to 30 years on a high-resolution grid of 1km x 1km. Key advancements include the adaptation of the TMEVD, which now incorporates atmospheric input variables for robust estimations in data-sparse regions. Additionally, this paper presents a novel methodological approach that utilizes a distributional neural network, tailored with innovative sample weighting to efficiently handle the increased computational demands and complexities associated with modeling the distribution parameters. Together, these contributions provide a valuable resource for future research and risk assessment.
Open AccessFrom Radar to Risk: Building a High-Resolution Hail Database for Austria And Estimating Risk Through the Integration of Distributional Neural Networks into the Metastatistical Framework
This study makes significant contributions to the understanding of hail climatology in Austria. First, it introduces a comprehensive database of hailstone sizes, constructed from three-dimensional radar data spanning 2009 to 2022 and calibrated by approximately 5000 verified hail reports. The database serves as foundation for describing the short-term climatology of hail and provides the data necessary for estimating hail risk maps with enhanced spatial resolution and quality. Second, the study enables the spatio-temporal metastitical extreme value distribution (TMEVD) to feature return levels of up to 30 years on a high-resolution grid of 1km x 1km. Key advancements include the adaptation of the TMEVD, which now incorporates atmospheric input variables for robust estimations in data-sparse regions. Additionally, this paper presents a novel methodological approach that utilizes a distributional neural network, tailored with innovative sample weighting to efficiently handle the increased computational demands and complexities associated with modeling the distribution parameters. Together, these contributions provide a valuable resource for future research and risk assessment.
- Open Access
- 26 February 2025
- Published by Copernicus GmbH in Geoscientific Model Development
- Vol. 18 (4) , 1141-1153
Identifying lightning processes in ERA5 soundings with deep learning
Atmospheric environments favorable for lightning and convection are commonly represented by proxies or parameterizations based on expert knowledge such as convective available potential energy (CAPE), wind shear, charge separation, or combinations thereof. Recent developments in the field of high-resolution reanalyses, accurate lightning observations, machine learning (ML), and explainable artificial intelligence (XAI) open possibilities for identifying tailored proxies without prior expert knowledge. This study utilizes a deep neural network trained to match temporally and vertically well-resolved ERA5 soundings of cloud physics, mass-field, and wind-field variables with lightning observations from the Austrian Lightning Information & Detection System (ALDIS). The ML model only receives the raw model atmosphere data as inputs, without incorporating any expert parameters or proxies derived from the model levels. Using and adapting appropriate XAI methods, it is then demonstrated how the inner workings of this well-performing deep learning model can be uncovered to identify physically meaningful patterns within the ERA5 soundings that describe lightning processes. The ERA5 parameters are taken on model levels beyond the tropopause, forming an input layer of approx. 670 features, and the lightning data are transformed to a binary target variable labeling the spatio-temporal ERA5 grid cells as cells with lightning activity and cells without lightning activity. Scaled Shapley additive explanations (SHAP) values are introduced to highlight the atmospheric processes learned by the neural network and show that the model identifies cloud ice and snow content in the upper troposphere and mid-troposphere as very relevant features. As these patterns correspond to the separation of charge in thunderstorm clouds, the deep learning model can serve as a physically meaningful description of lightning. The scaled SHAP values also reveal that, depending on the location, the model additionally learns to correctly classify cells with lightning activity by exploiting mass-field or wind-field variables. This approach also showcases how XAI can be used to accelerate knowledge discovery in areas where expert knowledge is still scarce.
- 26 February 2025
- Published by Copernicus GmbH in Geoscientific Model Development
- Vol. 18 (4) , 1141-1153
Open AccessIdentifying lightning processes in ERA5 soundings with deep learning
Atmospheric environments favorable for lightning and convection are commonly represented by proxies or parameterizations based on expert knowledge such as convective available potential energy (CAPE), wind shear, charge separation, or combinations thereof. Recent developments in the field of high-resolution reanalyses, accurate lightning observations, machine learning (ML), and explainable artificial intelligence (XAI) open possibilities for identifying tailored proxies without prior expert knowledge. This study utilizes a deep neural network trained to match temporally and vertically well-resolved ERA5 soundings of cloud physics, mass-field, and wind-field variables with lightning observations from the Austrian Lightning Information & Detection System (ALDIS). The ML model only receives the raw model atmosphere data as inputs, without incorporating any expert parameters or proxies derived from the model levels. Using and adapting appropriate XAI methods, it is then demonstrated how the inner workings of this well-performing deep learning model can be uncovered to identify physically meaningful patterns within the ERA5 soundings that describe lightning processes. The ERA5 parameters are taken on model levels beyond the tropopause, forming an input layer of approx. 670 features, and the lightning data are transformed to a binary target variable labeling the spatio-temporal ERA5 grid cells as cells with lightning activity and cells without lightning activity. Scaled Shapley additive explanations (SHAP) values are introduced to highlight the atmospheric processes learned by the neural network and show that the model identifies cloud ice and snow content in the upper troposphere and mid-troposphere as very relevant features. As these patterns correspond to the separation of charge in thunderstorm clouds, the deep learning model can serve as a physically meaningful description of lightning. The scaled SHAP values also reveal that, depending on the location, the model additionally learns to correctly classify cells with lightning activity by exploiting mass-field or wind-field variables. This approach also showcases how XAI can be used to accelerate knowledge discovery in areas where expert knowledge is still scarce.
- Open Access
- 14 August 2023
- Published by Elsevier in Weather and Climate Extremes
- Vol. 42
Accounting for seasonality in the metastatistical extreme value distribution
- 14 August 2023
- Published by Elsevier in Weather and Climate Extremes
- Vol. 42
Open AccessAccounting for seasonality in the metastatistical extreme value distribution
- Open Access
- 15 May 2023
- Published by Copernicus GmbH
Identifying Lightning Processes in ERA5 Soundings with Deep Learning
Atmospheric environments favorable for lightning and convection are commonly represented by proxies or parameterizations based on expert knowledge such as CAPE, wind shears, charge separation, or combinations thereof. Recent developments in the field of machine learning, high resolution reanalyses, and accurate lightning observations open possibilities for identifying tailored proxies without prior expert knowledge. To identify vertical profiles favorable for lightning, a deep neural network links ERA5 vertical profiles of cloud physics, mass field variables and wind to lightning location data from the Austrian Lightning Detection & Information System (ALDIS), which has been transformed to a binary target variable labelling the ERA5 cells as lightning and no lightning cells. The ERA5 parameters are taken on model levels beyond the tropopause forming an input layer of approx. 670 features. The data of 2010 - 2018 serve as training/validation. On independent test data, 2019, the deep network outperforms a reference with features based on meteorological expertise. Shapley values highlight the atmospheric processes learned by the network which identifies cloud ice and snow content in the upper and mid-troposphere as relevant features. As these patterns correspond to the separation of charge in thunderstorm cloud, the deep learning model can serve as physically meaningful description of lightning.
- 15 May 2023
- Published by Copernicus GmbH
Open AccessIdentifying Lightning Processes in ERA5 Soundings with Deep Learning
Atmospheric environments favorable for lightning and convection are commonly represented by proxies or parameterizations based on expert knowledge such as CAPE, wind shears, charge separation, or combinations thereof. Recent developments in the field of machine learning, high resolution reanalyses, and accurate lightning observations open possibilities for identifying tailored proxies without prior expert knowledge. To identify vertical profiles favorable for lightning, a deep neural network links ERA5 vertical profiles of cloud physics, mass field variables and wind to lightning location data from the Austrian Lightning Detection & Information System (ALDIS), which has been transformed to a binary target variable labelling the ERA5 cells as lightning and no lightning cells. The ERA5 parameters are taken on model levels beyond the tropopause forming an input layer of approx. 670 features. The data of 2010 - 2018 serve as training/validation. On independent test data, 2019, the deep network outperforms a reference with features based on meteorological expertise. Shapley values highlight the atmospheric processes learned by the network which identifies cloud ice and snow content in the upper and mid-troposphere as relevant features. As these patterns correspond to the separation of charge in thunderstorm cloud, the deep learning model can serve as physically meaningful description of lightning.
- Open Access
- 15 May 2023
- Published by Copernicus GmbH
Evaluating the generalization ability of a deep learning model trained to detect cloud-to-ground lightning on raw ERA5 data
Atmospheric conditions that are typical for lightning are commonly represented by proxies such as cloud top height, cloud ice flux, CAPE times precipitation, or the lightning potential index. While these proxies generally deliver reasonable results, they often need to be adapted for local conditions in order to perform well. This suggests that there is a need for more complex and holistic proxies. Recent research confirms that the use of machine learning (ML) approaches for describing lightning is promising.In a previous study a deep learning model was trained on single spatiotemporal (30km x 30km x 1h) cells in the summer period of the years 2010--2018 and showed good results for the unseen test year 2019 within Austria. We now improve this model by using multiple neighboring vertical atmospheric columns to also address for horizontal moisture advection. Furthermore data of successive hours is used as input data to enable the model to capture the temporal development of atmospheric conditions such as the build-up and breakdown of convections.In this work we focus on the summer months June to August and use data from parts of Central Europe. This spatial domain is thought to be representative for Continental Europe since it covers mountainous aswell as coastal regions. We take raw ERA5 parameters beyond the tropopause enriched with a small amount of meta data such as the day of the year and the hour of the day for training. The quality of the resulting paramaterized model is then evaluated on Continental Europe to examine the generalization ability.Using parts of Central Europe to train the model, we evaluate its ability to generalize on unseen parts of Continental Europe using EUCLID data. Having a model that generalizes well is a building block for a retrospective analysis back into years where the structured recording of accurate lightning observations in a unified way was not established yet.
- 15 May 2023
- Published by Copernicus GmbH
Open AccessEvaluating the generalization ability of a deep learning model trained to detect cloud-to-ground lightning on raw ERA5 data
Atmospheric conditions that are typical for lightning are commonly represented by proxies such as cloud top height, cloud ice flux, CAPE times precipitation, or the lightning potential index. While these proxies generally deliver reasonable results, they often need to be adapted for local conditions in order to perform well. This suggests that there is a need for more complex and holistic proxies. Recent research confirms that the use of machine learning (ML) approaches for describing lightning is promising.In a previous study a deep learning model was trained on single spatiotemporal (30km x 30km x 1h) cells in the summer period of the years 2010--2018 and showed good results for the unseen test year 2019 within Austria. We now improve this model by using multiple neighboring vertical atmospheric columns to also address for horizontal moisture advection. Furthermore data of successive hours is used as input data to enable the model to capture the temporal development of atmospheric conditions such as the build-up and breakdown of convections.In this work we focus on the summer months June to August and use data from parts of Central Europe. This spatial domain is thought to be representative for Continental Europe since it covers mountainous aswell as coastal regions. We take raw ERA5 parameters beyond the tropopause enriched with a small amount of meta data such as the day of the year and the hour of the day for training. The quality of the resulting paramaterized model is then evaluated on Continental Europe to examine the generalization ability.Using parts of Central Europe to train the model, we evaluate its ability to generalize on unseen parts of Continental Europe using EUCLID data. Having a model that generalizes well is a building block for a retrospective analysis back into years where the structured recording of accurate lightning observations in a unified way was not established yet.
- Open Access
- 15 May 2023
- Published by Copernicus GmbH
Modelling severe hail events over Austria using the metastatistical extreme value distribution
Knowledge about extreme values of severe hail plays an important role in engineering and insurance. The estimation of return levels of severe hail events is challenging, as hail is locally rare and documentation about hail events is not available in a unified way. For instance for the state of Austria GeoSphere provides radar based probabilities of hail (POH) and maxima of expected hail size (MEHS) that only span a period from 2010 onward.Based on this sparse data the application of classical extreme value theory, such as Block-Maxima or Peak over Threshold might be invalid. Instead we use a version of the metastatistical extreme value distribution (MEVD), which was shown to work reasonably well in the context of extreme precipitation events, even with a rather small number of available years used for the estimation in comparison to the recurrence time. More precisely we make an assumption about the underlying probability distribution of the daily maximum POH values. The parameters of the distribution are then modeled as smooth functions of the day of the year and the year of observation, thus employing the framework of generalized additive models for location, scale and shape (GAMLSS). Furthermore we add topographic information (longitude, latitude, altitude) to our model, resulting in a full spatiotemporal model across the whole domain of Austria, from which the return values of the POH, respectively MEHS are calculated.This framework allows for the incorporation of an arbitrary number of additional covariables, as long as they are available on the same grid as the desired output. To illustrate this we use the information of daily precipitation extremes to enrich the model with additional atmospheric information.
- 15 May 2023
- Published by Copernicus GmbH
Open AccessModelling severe hail events over Austria using the metastatistical extreme value distribution
Knowledge about extreme values of severe hail plays an important role in engineering and insurance. The estimation of return levels of severe hail events is challenging, as hail is locally rare and documentation about hail events is not available in a unified way. For instance for the state of Austria GeoSphere provides radar based probabilities of hail (POH) and maxima of expected hail size (MEHS) that only span a period from 2010 onward.Based on this sparse data the application of classical extreme value theory, such as Block-Maxima or Peak over Threshold might be invalid. Instead we use a version of the metastatistical extreme value distribution (MEVD), which was shown to work reasonably well in the context of extreme precipitation events, even with a rather small number of available years used for the estimation in comparison to the recurrence time. More precisely we make an assumption about the underlying probability distribution of the daily maximum POH values. The parameters of the distribution are then modeled as smooth functions of the day of the year and the year of observation, thus employing the framework of generalized additive models for location, scale and shape (GAMLSS). Furthermore we add topographic information (longitude, latitude, altitude) to our model, resulting in a full spatiotemporal model across the whole domain of Austria, from which the return values of the POH, respectively MEHS are calculated.This framework allows for the incorporation of an arbitrary number of additional covariables, as long as they are available on the same grid as the desired output. To illustrate this we use the information of daily precipitation extremes to enrich the model with additional atmospheric information.
- Open Access
- 15 May 2023
- Published by Copernicus GmbH
Accounting for seasonality in trends of extreme precipitation
For the estimation of daily precipitation extremes, the metastatistical extreme value distribution (MEVD) is known to perform superior to classical approaches like the generalized extreme value distribution, especially for small sample lengths. This is due to the fact that the MEVD incorporates all ordinary rainfall events within a block rather than only the extremes, which then emerge from repeated sampling of ordinary events. For daily rainfall extremes, the MEVD combines the Weibull distribution of ordinary daily rainfall events and the number of wet days per year as additional random variable. The MEVD provides yearly distributional parameters, which makes it already capable of analyzing temporal trends in daily precipitation extremes.But still, the MEVD in its current formulation does not take into account the seasonal, i.e. sub-yearly character of ordinary precipitation events. This problem becomes apparent when events originating from fundamentally different precipitation regimes show very similar MEVD parameters.In this contribution we therefore propose to explicitly model both the temporal trend and interannual seasonality of daily rainfall extremes and present an explicitly non-stationary MEVD formulation which is called temporal MEVD or TMEV. The TMEV is then used to derive historical trends of rainfall extremes in Austria. It is shown that the 50-year return level of daily rainfall in Austria has significantly increased over the last 30 years at the majority of Austrian observation sites. Furthermore the temporal change of the extreme value distribution is analyzed with respect to seasonality.
- 15 May 2023
- Published by Copernicus GmbH
Open AccessAccounting for seasonality in trends of extreme precipitation
For the estimation of daily precipitation extremes, the metastatistical extreme value distribution (MEVD) is known to perform superior to classical approaches like the generalized extreme value distribution, especially for small sample lengths. This is due to the fact that the MEVD incorporates all ordinary rainfall events within a block rather than only the extremes, which then emerge from repeated sampling of ordinary events. For daily rainfall extremes, the MEVD combines the Weibull distribution of ordinary daily rainfall events and the number of wet days per year as additional random variable. The MEVD provides yearly distributional parameters, which makes it already capable of analyzing temporal trends in daily precipitation extremes.But still, the MEVD in its current formulation does not take into account the seasonal, i.e. sub-yearly character of ordinary precipitation events. This problem becomes apparent when events originating from fundamentally different precipitation regimes show very similar MEVD parameters.In this contribution we therefore propose to explicitly model both the temporal trend and interannual seasonality of daily rainfall extremes and present an explicitly non-stationary MEVD formulation which is called temporal MEVD or TMEV. The TMEV is then used to derive historical trends of rainfall extremes in Austria. It is shown that the 50-year return level of daily rainfall in Austria has significantly increased over the last 30 years at the majority of Austrian observation sites. Furthermore the temporal change of the extreme value distribution is analyzed with respect to seasonality.
- 9 September 2019
- Published by Springer Nature
- p. 296-301
Evaluating CNNs on the Gestalt Principle of Closure
- 9 September 2019
- Published by Springer Nature
- p. 296-301
Evaluating CNNs on the Gestalt Principle of Closure
- Open Access
Fast algorithms for morphological operations using run-length encoded binary images
This paper presents innovative algorithms to efficiently compute erosions and dilations of run-length encoded (RLE) binary images with arbitrary shaped structuring elements. An RLE image is given by a set of runs, where a run is a horizontal concatenation of foreground pixels. The proposed algorithms extract the skeleton of the structuring element and build distance tables of the input image, which are storing the distance to the next background pixel on the left and right hand sides. This information is then used to speed up the calculations of the erosion and dilation operator by enabling the use of techniques which allow to skip the analysis of certain pixels whenever a hit or miss occurs. Additionally the input image gets trimmed during the preprocessing steps on the base of two primitive criteria. Experimental results show the advantages over other algorithms. The source code of our algorithms is available in C++.
Open AccessFast algorithms for morphological operations using run-length encoded binary images
This paper presents innovative algorithms to efficiently compute erosions and dilations of run-length encoded (RLE) binary images with arbitrary shaped structuring elements. An RLE image is given by a set of runs, where a run is a horizontal concatenation of foreground pixels. The proposed algorithms extract the skeleton of the structuring element and build distance tables of the input image, which are storing the distance to the next background pixel on the left and right hand sides. This information is then used to speed up the calculations of the erosion and dilation operator by enabling the use of techniques which allow to skip the analysis of certain pixels whenever a hit or miss occurs. Additionally the input image gets trimmed during the preprocessing steps on the base of two primitive criteria. Experimental results show the advantages over other algorithms. The source code of our algorithms is available in C++.