TY - JOUR A1 - Seckler, Henrik A1 - Metzler, Ralf T1 - Bayesian deep learning for error estimation in the analysis of anomalous diffusion JF - Nature Communnications N2 - Modern single-particle-tracking techniques produce extensive time-series of diffusive motion in a wide variety of systems, from single-molecule motion in living-cells to movement ecology. The quest is to decipher the physical mechanisms encoded in the data and thus to better understand the probed systems. We here augment recently proposed machine-learning techniques for decoding anomalous-diffusion data to include an uncertainty estimate in addition to the predicted output. To avoid the Black-Box-Problem a Bayesian-Deep-Learning technique named Stochastic-Weight-Averaging-Gaussian is used to train models for both the classification of the diffusionmodel and the regression of the anomalous diffusion exponent of single-particle-trajectories. Evaluating their performance, we find that these models can achieve a wellcalibrated error estimate while maintaining high prediction accuracies. In the analysis of the output uncertainty predictions we relate these to properties of the underlying diffusion models, thus providing insights into the learning process of the machine and the relevance of the output. KW - random-walk KW - models Y1 - 2022 U6 - https://doi.org/10.1038/s41467-022-34305-6 SN - 2041-1723 VL - 13 PB - Nature Publishing Group UK CY - London ER - TY - GEN A1 - Seckler, Henrik A1 - Metzler, Ralf T1 - Bayesian deep learning for error estimation in the analysis of anomalous diffusion T2 - Zweitveröffentlichungen der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe N2 - Sprache Englisch Modern single-particle-tracking techniques produce extensive time-series of diffusive motion in a wide variety of systems, from single-molecule motion in living-cells to movement ecology. The quest is to decipher the physical mechanisms encoded in the data and thus to better understand the probed systems. We here augment recently proposed machine-learning techniques for decoding anomalous-diffusion data to include an uncertainty estimate in addition to the predicted output. To avoid the Black-Box-Problem a Bayesian-Deep-Learning technique named Stochastic-Weight-Averaging-Gaussian is used to train models for both the classification of the diffusionmodel and the regression of the anomalous diffusion exponent of single-particle-trajectories. Evaluating their performance, we find that these models can achieve a wellcalibrated error estimate while maintaining high prediction accuracies. In the analysis of the output uncertainty predictions we relate these to properties of the underlying diffusion models, thus providing insights into the learning process of the machine and the relevance of the output. T3 - Zweitveröffentlichungen der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe - 1314 KW - random-walk KW - models Y1 - 2022 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-586025 SN - 1866-8372 IS - 1314 ER - TY - JOUR A1 - Reinecke, Robert A1 - Trautmann, Tim A1 - Wagener, Thorsten A1 - Schüler, Katja T1 - The critical need to foster computational reproducibility JF - Environmental research letters KW - reproducibility KW - models KW - software KW - Open Science Y1 - 2022 U6 - https://doi.org/10.1088/1748-9326/ac5cf8 SN - 1748-9326 VL - 17 IS - 4 PB - IOP Publ. Ltd. CY - Bristol ER - TY - JOUR A1 - Naliboff, John B. A1 - Glerum, Anne A1 - Brune, Sascha A1 - Péron-Pinvidic, G. A1 - Wrona, Thilo T1 - Development of 3-D rift heterogeneity through fault network evolution JF - Geophysical Research Letters N2 - Observations of rift and rifted margin architecture suggest that significant spatial and temporal structural heterogeneity develops during the multiphase evolution of continental rifting. Inheritance is often invoked to explain this heterogeneity, such as preexisting anisotropies in rock composition, rheology, and deformation. Here, we use high-resolution 3-D thermal-mechanical numerical models of continental extension to demonstrate that rift-parallel heterogeneity may develop solely through fault network evolution during the transition from distributed to localized deformation. In our models, the initial phase of distributed normal faulting is seeded through randomized initial strength perturbations in an otherwise laterally homogeneous lithosphere extending at a constant rate. Continued extension localizes deformation onto lithosphere-scale faults, which are laterally offset by tens of km and discontinuous along-strike. These results demonstrate that rift- and margin-parallel heterogeneity of large-scale fault patterns may in-part be a natural byproduct of fault network coalescence. KW - magma-poor KW - continental lithosphere KW - extension KW - insights KW - margins KW - architecture KW - systems KW - models KW - sea KW - reactivation Y1 - 2019 VL - 47 IS - 13 PB - John Wiley & Sons, Inc. CY - New Jersey ER - TY - JOUR A1 - Metzler, Ralf T1 - Superstatistics and non-Gaussian diffusion JF - The European physical journal special topics N2 - Brownian motion and viscoelastic anomalous diffusion in homogeneous environments are intrinsically Gaussian processes. In a growing number of systems, however, non-Gaussian displacement distributions of these processes are being reported. The physical cause of the non-Gaussianity is typically seen in different forms of disorder. These include, for instance, imperfect "ensembles" of tracer particles, the presence of local variations of the tracer mobility in heteroegenous environments, or cases in which the speed or persistence of moving nematodes or cells are distributed. From a theoretical point of view stochastic descriptions based on distributed ("superstatistical") transport coefficients as well as time-dependent generalisations based on stochastic transport parameters with built-in finite correlation time are invoked. After a brief review of the history of Brownian motion and the famed Gaussian displacement distribution, we here provide a brief introduction to the phenomenon of non-Gaussianity and the stochastic modelling in terms of superstatistical and diffusing-diffusivity approaches. KW - Brownian diffusion KW - anomalous diffusion KW - dynamics KW - kinetic-theory KW - models KW - motion KW - nanoparticles KW - nonergodicity KW - statistics KW - subdiffusion Y1 - 2020 U6 - https://doi.org/10.1140/epjst/e2020-900210-x SN - 1951-6355 SN - 1951-6401 VL - 229 IS - 5 SP - 711 EP - 728 PB - Springer CY - Heidelberg ER - TY - GEN A1 - Ayzel, Georgy A1 - Scheffer, Tobias A1 - Heistermann, Maik T1 - RainNet v1.0 BT - a convolutional neural network for radar-based precipitation nowcasting T2 - Postprints der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe N2 - In this study, we present RainNet, a deep convolutional neural network for radar-based precipitation nowcasting. Its design was inspired by the U-Net and SegNet families of deep learning models, which were originally designed for binary segmentation tasks. RainNet was trained to predict continuous precipitation intensities at a lead time of 5min, using several years of quality-controlled weather radar composites provided by the German Weather Service (DWD). That data set covers Germany with a spatial domain of 900km × 900km and has a resolution of 1km in space and 5min in time. Independent verification experiments were carried out on 11 summer precipitation events from 2016 to 2017. In order to achieve a lead time of 1h, a recursive approach was implemented by using RainNet predictions at 5min lead times as model inputs for longer lead times. In the verification experiments, trivial Eulerian persistence and a conventional model based on optical flow served as benchmarks. The latter is available in the rainymotion library and had previously been shown to outperform DWD's operational nowcasting model for the same set of verification events. RainNet significantly outperforms the benchmark models at all lead times up to 60min for the routine verification metrics mean absolute error (MAE) and the critical success index (CSI) at intensity thresholds of 0.125, 1, and 5mm h⁻¹. However, rainymotion turned out to be superior in predicting the exceedance of higher intensity thresholds (here 10 and 15mm h⁻¹). The limited ability of RainNet to predict heavy rainfall intensities is an undesirable property which we attribute to a high level of spatial smoothing introduced by the model. At a lead time of 5min, an analysis of power spectral density confirmed a significant loss of spectral power at length scales of 16km and below. Obviously, RainNet had learned an optimal level of smoothing to produce a nowcast at 5min lead time. In that sense, the loss of spectral power at small scales is informative, too, as it reflects the limits of predictability as a function of spatial scale. Beyond the lead time of 5min, however, the increasing level of smoothing is a mere artifact – an analogue to numerical diffusion – that is not a property of RainNet itself but of its recursive application. In the context of early warning, the smoothing is particularly unfavorable since pronounced features of intense precipitation tend to get lost over longer lead times. Hence, we propose several options to address this issue in prospective research, including an adjustment of the loss function for model training, model training for longer lead times, and the prediction of threshold exceedance in terms of a binary segmentation task. Furthermore, we suggest additional input data that could help to better identify situations with imminent precipitation dynamics. The model code, pretrained weights, and training data are provided in open repositories as an input for such future studies. T3 - Zweitveröffentlichungen der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe - 964 KW - weather KW - models KW - skill Y1 - 2020 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-472942 SN - 1866-8372 IS - 964 ER - TY - JOUR A1 - Ayzel, Georgy A1 - Scheffer, Tobias A1 - Heistermann, Maik T1 - RainNet v1.0 BT - a convolutional neural network for radar-based precipitation nowcasting JF - Geoscientific Model Development N2 - In this study, we present RainNet, a deep convolutional neural network for radar-based precipitation nowcasting. Its design was inspired by the U-Net and SegNet families of deep learning models, which were originally designed for binary segmentation tasks. RainNet was trained to predict continuous precipitation intensities at a lead time of 5min, using several years of quality-controlled weather radar composites provided by the German Weather Service (DWD). That data set covers Germany with a spatial domain of 900km × 900km and has a resolution of 1km in space and 5min in time. Independent verification experiments were carried out on 11 summer precipitation events from 2016 to 2017. In order to achieve a lead time of 1h, a recursive approach was implemented by using RainNet predictions at 5min lead times as model inputs for longer lead times. In the verification experiments, trivial Eulerian persistence and a conventional model based on optical flow served as benchmarks. The latter is available in the rainymotion library and had previously been shown to outperform DWD's operational nowcasting model for the same set of verification events. RainNet significantly outperforms the benchmark models at all lead times up to 60min for the routine verification metrics mean absolute error (MAE) and the critical success index (CSI) at intensity thresholds of 0.125, 1, and 5mm h⁻¹. However, rainymotion turned out to be superior in predicting the exceedance of higher intensity thresholds (here 10 and 15mm h⁻¹). The limited ability of RainNet to predict heavy rainfall intensities is an undesirable property which we attribute to a high level of spatial smoothing introduced by the model. At a lead time of 5min, an analysis of power spectral density confirmed a significant loss of spectral power at length scales of 16km and below. Obviously, RainNet had learned an optimal level of smoothing to produce a nowcast at 5min lead time. In that sense, the loss of spectral power at small scales is informative, too, as it reflects the limits of predictability as a function of spatial scale. Beyond the lead time of 5min, however, the increasing level of smoothing is a mere artifact – an analogue to numerical diffusion – that is not a property of RainNet itself but of its recursive application. In the context of early warning, the smoothing is particularly unfavorable since pronounced features of intense precipitation tend to get lost over longer lead times. Hence, we propose several options to address this issue in prospective research, including an adjustment of the loss function for model training, model training for longer lead times, and the prediction of threshold exceedance in terms of a binary segmentation task. Furthermore, we suggest additional input data that could help to better identify situations with imminent precipitation dynamics. The model code, pretrained weights, and training data are provided in open repositories as an input for such future studies. KW - weather KW - models KW - skill Y1 - 2020 U6 - https://doi.org/10.5194/gmd-13-2631-2020 SN - 1991-959X SN - 1991-9603 VL - 13 IS - 6 SP - 2631 EP - 2644 PB - Copernicus Publ. CY - Göttingen ER - TY - JOUR A1 - Schoppa, Lukas A1 - Sieg, Tobias A1 - Vogel, Kristin A1 - Zöller, Gert A1 - Kreibich, Heidi T1 - Probabilistic flood loss models for companies JF - Water resources research N2 - Flood loss modeling is a central component of flood risk analysis. Conventionally, this involves univariable and deterministic stage-damage functions. Recent advancements in the field promote the use of multivariable and probabilistic loss models, which consider variables beyond inundation depth and account for prediction uncertainty. Although companies contribute significantly to total loss figures, novel modeling approaches for companies are lacking. Scarce data and the heterogeneity among companies impede the development of company flood loss models. We present three multivariable flood loss models for companies from the manufacturing, commercial, financial, and service sector that intrinsically quantify prediction uncertainty. Based on object-level loss data (n = 1,306), we comparatively evaluate the predictive capacity of Bayesian networks, Bayesian regression, and random forest in relation to deterministic and probabilistic stage-damage functions, serving as benchmarks. The company loss data stem from four postevent surveys in Germany between 2002 and 2013 and include information on flood intensity, company characteristics, emergency response, private precaution, and resulting loss to building, equipment, and goods and stock. We find that the multivariable probabilistic models successfully identify and reproduce essential relationships of flood damage processes in the data. The assessment of model skill focuses on the precision of the probabilistic predictions and reveals that the candidate models outperform the stage-damage functions, while differences among the proposed models are negligible. Although the combination of multivariable and probabilistic loss estimation improves predictive accuracy over the entire data set, wide predictive distributions stress the necessity for the quantification of uncertainty. KW - flood loss estimation KW - probabilistic modeling KW - companies KW - multivariable KW - models Y1 - 2020 U6 - https://doi.org/10.1029/2020WR027649 SN - 0043-1397 SN - 1944-7973 VL - 56 IS - 9 PB - American Geophysical Union CY - Washington ER - TY - GEN A1 - Naliboff, John B. A1 - Glerum, Anne A1 - Brune, Sascha A1 - Péron-Pinvidic, G. A1 - Wrona, Thilo T1 - Development of 3-D rift heterogeneity through fault network evolution T2 - Postprints der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe N2 - Observations of rift and rifted margin architecture suggest that significant spatial and temporal structural heterogeneity develops during the multiphase evolution of continental rifting. Inheritance is often invoked to explain this heterogeneity, such as preexisting anisotropies in rock composition, rheology, and deformation. Here, we use high-resolution 3-D thermal-mechanical numerical models of continental extension to demonstrate that rift-parallel heterogeneity may develop solely through fault network evolution during the transition from distributed to localized deformation. In our models, the initial phase of distributed normal faulting is seeded through randomized initial strength perturbations in an otherwise laterally homogeneous lithosphere extending at a constant rate. Continued extension localizes deformation onto lithosphere-scale faults, which are laterally offset by tens of km and discontinuous along-strike. These results demonstrate that rift- and margin-parallel heterogeneity of large-scale fault patterns may in-part be a natural byproduct of fault network coalescence. T3 - Zweitveröffentlichungen der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe - 1183 KW - magma-poor KW - continental lithosphere KW - extension KW - insights KW - margins KW - architecture KW - systems KW - models KW - sea KW - reactivation Y1 - 2019 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-524661 SN - 1866-8372 IS - 13 ER - TY - JOUR A1 - Nevill, Alan M. A1 - Negra, Yassine A1 - Myers, Tony D. A1 - Sammoud, Senda A1 - Chaabene, Helmi T1 - Key somatic variables associated with, and differences between the 4 swimming strokes JF - Journal of sports sciences N2 - This study identified key somatic and demographic characteristics that benefit all swimmers and, at the same time, identified further characteristics that benefit only specific swimming strokes. Three hundred sixty-three competitive-level swimmers (male [n = 202]; female [n = 161]) participated in the study. We adopted a multiplicative, allometric regression model to identify the key characteristics associated with 100 m swimming speeds (controlling for age). The model was refined using backward elimination. Characteristics that benefited some but not all strokes were identified by introducing stroke-by-predictor variable interactions. The regression analysis revealed 7 "common" characteristics that benefited all swimmers suggesting that all swimmers benefit from having less body fat, broad shoulders and hips, a greater arm span (but shorter lower arms) and greater forearm girths with smaller relaxed arm girths. The 4 stroke-specific characteristics reveal that backstroke swimmers benefit from longer backs, a finding that can be likened to boats with longer hulls also travel faster through the water. Other stroke-by-predictor variable interactions (taken together) identified that butterfly swimmers are characterized by greater muscularity in the lower legs. These results highlight the importance of considering somatic and demographic characteristics of young swimmers for talent identification purposes (i.e., to ensure that swimmers realize their most appropriate strokes). KW - swim speed KW - talent identification KW - limb dimensions KW - ratios KW - allometric KW - models KW - log-linear regression Y1 - 2020 U6 - https://doi.org/10.1080/02640414.2020.1734311 SN - 0264-0414 SN - 1466-447X VL - 38 IS - 7 SP - 787 EP - 794 PB - Routledge, Taylor & Francis Group CY - London ER -