TY - GEN A1 - Seleem, Omar A1 - Ayzel, Georgy A1 - Bronstert, Axel A1 - Heistermann, Maik T1 - Transferability of data-driven models to predict urban pluvial flood water depth in Berlin, Germany T2 - Zweitveröffentlichungen der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe N2 - Data-driven models have been recently suggested to surrogate computationally expensive hydrodynamic models to map flood hazards. However, most studies focused on developing models for the same area or the same precipitation event. It is thus not obvious how transferable the models are in space. This study evaluates the performance of a convolutional neural network (CNN) based on the U-Net architecture and the random forest (RF) algorithm to predict flood water depth, the models' transferability in space and performance improvement using transfer learning techniques. We used three study areas in Berlin to train, validate and test the models. The results showed that (1) the RF models outperformed the CNN models for predictions within the training domain, presumable at the cost of overfitting; (2) the CNN models had significantly higher potential than the RF models to generalize beyond the training domain; and (3) the CNN models could better benefit from transfer learning technique to boost their performance outside training domains than RF models. T3 - Zweitveröffentlichungen der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe - 1323 Y1 - 2023 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-589168 SN - 1866-8372 IS - 1323 SP - 809 EP - 822 ER - TY - JOUR A1 - Seleem, Omar A1 - Ayzel, Georgy A1 - Bronstert, Axel A1 - Heistermann, Maik T1 - Transferability of data-driven models to predict urban pluvial flood water depth in Berlin, Germany JF - Natural Hazards and Earth System Sciences N2 - Data-driven models have been recently suggested to surrogate computationally expensive hydrodynamic models to map flood hazards. However, most studies focused on developing models for the same area or the same precipitation event. It is thus not obvious how transferable the models are in space. This study evaluates the performance of a convolutional neural network (CNN) based on the U-Net architecture and the random forest (RF) algorithm to predict flood water depth, the models' transferability in space and performance improvement using transfer learning techniques. We used three study areas in Berlin to train, validate and test the models. The results showed that (1) the RF models outperformed the CNN models for predictions within the training domain, presumable at the cost of overfitting; (2) the CNN models had significantly higher potential than the RF models to generalize beyond the training domain; and (3) the CNN models could better benefit from transfer learning technique to boost their performance outside training domains than RF models. Y1 - 2023 U6 - https://doi.org/10.5194/nhess-23-809-2023 SN - 1684-9981 SN - 1561-8633 VL - 23 IS - 2 SP - 809 EP - 822 PB - Copernicus CY - Göttingen ER - TY - GEN A1 - Seleem, Omar A1 - Ayzel, Georgy A1 - Costa Tomaz de Souza, Arthur A1 - Bronstert, Axel A1 - Heistermann, Maik T1 - Towards urban flood susceptibility mapping using data-driven models in Berlin, Germany T2 - Zweitveröffentlichungen der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe N2 - Identifying urban pluvial flood-prone areas is necessary but the application of two-dimensional hydrodynamic models is limited to small areas. Data-driven models have been showing their ability to map flood susceptibility but their application in urban pluvial flooding is still rare. A flood inventory (4333 flooded locations) and 11 factors which potentially indicate an increased hazard for pluvial flooding were used to implement convolutional neural network (CNN), artificial neural network (ANN), random forest (RF) and support vector machine (SVM) to: (1) Map flood susceptibility in Berlin at 30, 10, 5, and 2 m spatial resolutions. (2) Evaluate the trained models' transferability in space. (3) Estimate the most useful factors for flood susceptibility mapping. The models' performance was validated using the Kappa, and the area under the receiver operating characteristic curve (AUC). The results indicated that all models perform very well (minimum AUC = 0.87 for the testing dataset). The RF models outperformed all other models at all spatial resolutions and the RF model at 2 m spatial resolution was superior for the present flood inventory and predictor variables. The majority of the models had a moderate performance for predictions outside the training area based on Kappa evaluation (minimum AUC = 0.8). Aspect and altitude were the most influencing factors on the image-based and point-based models respectively. Data-driven models can be a reliable tool for urban pluvial flood susceptibility mapping wherever a reliable flood inventory is available. T3 - Zweitveröffentlichungen der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe - 1297 KW - Urban pluvial flood susceptibility KW - convolutional neural network KW - deep learning KW - random forest KW - support vector machine KW - spatial resolution KW - flood predictors Y1 - 2023 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-576806 SN - 1866-8372 IS - 1297 SP - 1640 EP - 1662 ER - TY - JOUR A1 - Seleem, Omar A1 - Ayzel, Georgy A1 - Costa Tomaz de Souza, Arthur A1 - Bronstert, Axel A1 - Heistermann, Maik T1 - Towards urban flood susceptibility mapping using data-driven models in Berlin, Germany JF - Geomatics, natural hazards and risk N2 - Identifying urban pluvial flood-prone areas is necessary but the application of two-dimensional hydrodynamic models is limited to small areas. Data-driven models have been showing their ability to map flood susceptibility but their application in urban pluvial flooding is still rare. A flood inventory (4333 flooded locations) and 11 factors which potentially indicate an increased hazard for pluvial flooding were used to implement convolutional neural network (CNN), artificial neural network (ANN), random forest (RF) and support vector machine (SVM) to: (1) Map flood susceptibility in Berlin at 30, 10, 5, and 2 m spatial resolutions. (2) Evaluate the trained models' transferability in space. (3) Estimate the most useful factors for flood susceptibility mapping. The models' performance was validated using the Kappa, and the area under the receiver operating characteristic curve (AUC). The results indicated that all models perform very well (minimum AUC = 0.87 for the testing dataset). The RF models outperformed all other models at all spatial resolutions and the RF model at 2 m spatial resolution was superior for the present flood inventory and predictor variables. The majority of the models had a moderate performance for predictions outside the training area based on Kappa evaluation (minimum AUC = 0.8). Aspect and altitude were the most influencing factors on the image-based and point-based models respectively. Data-driven models can be a reliable tool for urban pluvial flood susceptibility mapping wherever a reliable flood inventory is available. KW - Urban pluvial flood susceptibility KW - convolutional neural network KW - deep KW - learning KW - random forest KW - support vector machine KW - spatial resolution; KW - flood predictors Y1 - 2022 U6 - https://doi.org/10.1080/19475705.2022.2097131 SN - 1947-5705 SN - 1947-5713 VL - 13 IS - 1 SP - 1640 EP - 1662 PB - Taylor & Francis CY - London ER - TY - JOUR A1 - Ayzel, Georgy A1 - Heistermann, Maik T1 - The effect of calibration data length on the performance of a conceptual hydrological model versus LSTM and GRU BT - a case study for six basins from the CAMELS dataset JF - Computers & geosciences : an international journal devoted to the publication of papers on all aspects of geocomputation and to the distribution of computer programs and test data sets ; an official journal of the International Association for Mathematical Geology N2 - We systematically explore the effect of calibration data length on the performance of a conceptual hydrological model, GR4H, in comparison to two Artificial Neural Network (ANN) architectures: Long Short-Term Memory Networks (LSTM) and Gated Recurrent Units (GRU), which have just recently been introduced to the field of hydrology. We implemented a case study for six river basins across the contiguous United States, with 25 years of meteorological and discharge data. Nine years were reserved for independent validation; two years were used as a warm-up period, one year for each of the calibration and validation periods, respectively; from the remaining 14 years, we sampled increasing amounts of data for model calibration, and found pronounced differences in model performance. While GR4H required less data to converge, LSTM and GRU caught up at a remarkable rate, considering their number of parameters. Also, LSTM and GRU exhibited the higher calibration instability in comparison to GR4H. These findings confirm the potential of modern deep-learning architectures in rainfall runoff modelling, but also highlight the noticeable differences between them in regard to the effect of calibration data length. KW - Artificial neural networks KW - Calibration KW - Deep learning KW - Rainfall-runoff KW - modelling Y1 - 2021 U6 - https://doi.org/10.1016/j.cageo.2021.104708 SN - 0098-3004 SN - 1873-7803 VL - 149 PB - Elsevier CY - Amsterdam ER - TY - THES A1 - Ayzel, Georgy T1 - Advancing radar-based precipitation nowcasting T1 - Fortschritte bei der radarbasierten Niederschlagsvorhersage BT - an open benchmark and the potential of deep learning BT - ein offener Benchmark und das Potenzial von Deep Learning N2 - Precipitation forecasting has an important place in everyday life – during the day we may have tens of small talks discussing the likelihood that it will rain this evening or weekend. Should you take an umbrella for a walk? Or should you invite your friends for a barbecue? It will certainly depend on what your weather application shows. While for years people were guided by the precipitation forecasts issued for a particular region or city several times a day, the widespread availability of weather radars allowed us to obtain forecasts at much higher spatiotemporal resolution of minutes in time and hundreds of meters in space. Hence, radar-based precipitation nowcasting, that is, very-short-range forecasting (typically up to 1–3 h), has become an essential technique, also in various professional application contexts, e.g., early warning, sewage control, or agriculture. There are two major components comprising a system for precipitation nowcasting: radar-based precipitation estimates, and models to extrapolate that precipitation to the imminent future. While acknowledging the fundamental importance of radar-based precipitation retrieval for precipitation nowcasts, this thesis focuses only on the model development: the establishment of open and competitive benchmark models, the investigation of the potential of deep learning, and the development of procedures for nowcast errors diagnosis and isolation that can guide model development. The present landscape of computational models for precipitation nowcasting still struggles with the availability of open software implementations that could serve as benchmarks for measuring progress. Focusing on this gap, we have developed and extensively benchmarked a stack of models based on different optical flow algorithms for the tracking step and a set of parsimonious extrapolation procedures based on image warping and advection. We demonstrate that these models provide skillful predictions comparable with or even superior to state-of-the-art operational software. We distribute the corresponding set of models as a software library, rainymotion, which is written in the Python programming language and openly available at GitHub (https://github.com/hydrogo/rainymotion). That way, the library acts as a tool for providing fast, open, and transparent solutions that could serve as a benchmark for further model development and hypothesis testing. One of the promising directions for model development is to challenge the potential of deep learning – a subfield of machine learning that refers to artificial neural networks with deep architectures, which may consist of many computational layers. Deep learning showed promising results in many fields of computer science, such as image and speech recognition, or natural language processing, where it started to dramatically outperform reference methods. The high benefit of using "big data" for training is among the main reasons for that. Hence, the emerging interest in deep learning in atmospheric sciences is also caused and concerted with the increasing availability of data – both observational and model-based. The large archives of weather radar data provide a solid basis for investigation of deep learning potential in precipitation nowcasting: one year of national 5-min composites for Germany comprises around 85 billion data points. To this aim, we present RainNet, a deep convolutional neural network for radar-based precipitation nowcasting. RainNet was trained to predict continuous precipitation intensities at a lead time of 5 min, 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 900 km x 900 km and has a resolution of 1 km in space and 5 min in time. Independent verification experiments were carried out on 11 summer precipitation events from 2016 to 2017. In these experiments, RainNet was applied recursively in order to achieve lead times of up to 1 h. In the verification experiments, trivial Eulerian persistence and a conventional model based on optical flow served as benchmarks. The latter is available in the previously developed rainymotion library. RainNet significantly outperformed the benchmark models at all lead times up to 60 min for the routine verification metrics mean absolute error (MAE) and critical success index (CSI) at intensity thresholds of 0.125, 1, and 5 mm/h. However, rainymotion turned out to be superior in predicting the exceedance of higher intensity thresholds (here 10 and 15 mm/h). The limited ability of RainNet to predict high 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 5 min, an analysis of power spectral density confirmed a significant loss of spectral power at length scales of 16 km and below. Obviously, RainNet had learned an optimal level of smoothing to produce a nowcast at 5 min 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 5 min, 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 on model development for precipitation nowcasting, including an adjustment of the loss function for model training, model training for longer lead times, and the prediction of threshold exceedance. The model development together with the verification experiments for both conventional and deep learning model predictions also revealed the need to better understand the source of forecast errors. Understanding the dominant sources of error in specific situations should help in guiding further model improvement. The total error of a precipitation nowcast consists of an error in the predicted location of a precipitation feature and an error in the change of precipitation intensity over lead time. So far, verification measures did not allow to isolate the location error, making it difficult to specifically improve nowcast models with regard to location prediction. To fill this gap, we introduced a framework to directly quantify the location error. To that end, we detect and track scale-invariant precipitation features (corners) in radar images. We then consider these observed tracks as the true reference in order to evaluate the performance (or, inversely, the error) of any model that aims to predict the future location of a precipitation feature. Hence, the location error of a forecast at any lead time ahead of the forecast time corresponds to the Euclidean distance between the observed and the predicted feature location at the corresponding lead time. Based on this framework, we carried out a benchmarking case study using one year worth of weather radar composites of the DWD. We evaluated the performance of four extrapolation models, two of which are based on the linear extrapolation of corner motion; and the remaining two are based on the Dense Inverse Search (DIS) method: motion vectors obtained from DIS are used to predict feature locations by linear and Semi-Lagrangian extrapolation. For all competing models, the mean location error exceeds a distance of 5 km after 60 min, and 10 km after 110 min. At least 25% of all forecasts exceed an error of 5 km after 50 min, and of 10 km after 90 min. Even for the best models in our experiment, at least 5 percent of the forecasts will have a location error of more than 10 km after 45 min. When we relate such errors to application scenarios that are typically suggested for precipitation nowcasting, e.g., early warning, it becomes obvious that location errors matter: the order of magnitude of these errors is about the same as the typical extent of a convective cell. Hence, the uncertainty of precipitation nowcasts at such length scales – just as a result of locational errors – can be substantial already at lead times of less than 1 h. Being able to quantify the location error should hence guide any model development that is targeted towards its minimization. To that aim, we also consider the high potential of using deep learning architectures specific to the assimilation of sequential (track) data. Last but not least, the thesis demonstrates the benefits of a general movement towards open science for model development in the field of precipitation nowcasting. All the presented models and frameworks are distributed as open repositories, thus enhancing transparency and reproducibility of the methodological approach. Furthermore, they are readily available to be used for further research studies, as well as for practical applications. N2 - Niederschlagsvorhersagen haben einen wichtigen Platz in unserem täglichen Leben. Und die breite Abdeckung mit Niederschlagsradaren ermöglicht es uns, den Niederschlag mit einer viel höheren räumlich-zeitlichen Auflösung vorherzusagen (Minuten in der Zeit, Hunderte von Metern im Raum). Solche radargestützten Niederschlagsvorhersagen mit sehr kurzem Vorhersagehorizont (1–3 Stunden) nennt man auch "Niederschlagsnowcasting." Sie sind in verschiedenen Anwendungsbereichen (z.B. in der Frühwarnung, der Stadtentwässerung sowie in der Landwirtschaft) zu einer wichtigen Technologie geworden. Eine erhebliche Schwierigkeit in Modellentwicklung zum Niederschlagsnowcastings ist jedoch die Verfügbarkeit offener Softwarewerkzeuge und Implementierungen, die als Benchmark für den Entwicklungsfortschritt auf diesem Gebiet dienen können. Um diese Lücke zu schließen, haben wir eine Gruppe von Modellen auf der Grundlage verschiedener Tracking- und Extrapolationsverfahren entwickelt und systematisch verglichen. Es konnte gezeigt werden, dass die Vorhersagen dieser einen Skill haben, der sich mit dem Skill operationeller Vorhersagesysteme messen kann, teils sogar überlegen sind. Diese Benchmark-Modelle sind nun in Form der quelloffenen Software-Bibliothek rainymotion allgemein verfügbar (https://github.com/hydrogo/rainymotion). Eine der vielversprechenden Perspektiven für die weitere Modellentwicklung besteht in der Untersuchung des Potenzials von "Deep Learning" – einem Teilgebiet des maschinellen Lernens, das sich auf künstliche neuronale Netze mit sog. "tiefen Architekturen" bezieht, die aus einer Vielzahl von Schichten (computational layers) bestehen können. Im Rahmen dieser Arbeit wurde daher RainNet entwickelt: ein Tiefes Neuronales Netz für radargestütztes Niederschlags-Nowcasting. RainNet wurde zunächst zur Vorhersage der Niederschlagsintensität mit einem Vorhersagehorizont von fünf Minuten trainiert. Als Datengrundlage dazu dienten mehrere Jahre qualitätskontrollierter Radarkompositprodukte des Deutschen Wetterdienstes (DWD). RainNet übertraf die verfügbaren Benchmark-Modelle für Vorhersagezeiten bis zu 60 min in Bezug auf den Mittleren Absoluten Fehler (MAE) und den Critical Success Index (CSI) für Intensitätsschwellenwerte von 0.125, 1 und 5 mm/h. Allerdings erwies sich das das Benchmark-Modell aus dem Softwarepaket rainymotion bei der Vorhersage der Überschreitung höherer Intensitätsschwellen (10 und 15 mm/h) als überlegen. Die eingeschränkte Fähigkeit von RainNet zur Vorhersage hoher Niederschlagsintensitäten ist eine unerwünschte Eigenschaft, die wir auf ein hohes Maß an räumlicher Glättung durch das Modell zurückführen. Im Kontext der Frühwarnung ist die Glättung besonders ungünstig, da ausgeprägte Merkmale von Starkniederschlägen bei längeren Vorlaufzeiten tendenziell verloren gehen. In dieser Arbeit werden daher mehrere Optionen vorgeschlagen, um dieses Problem in der zukünftigen Forschung zur Modellentwicklung anzugehen. Ein weiterer Beitrag dieser Arbeit liegt in der Quantifizierung einer spezifischen Fehlerquelle von Niederschlagsnowcasts. Der Gesamtfehler eines Nowcasts besteht aus einem Fehler in der vorhergesagten Lage eines Niederschlagsfeatures (Ortsfehler) sowie einem Fehler in der Änderung der Intensität eines Features über die Vorhersagezeit (Intensitätsfehler). Herkömmliche Verifikationsmaße waren bislang nicht in der Lage, das Ausmaß des Ortsfehlers zu isolieren. Um diese Lücke zu füllen, haben wir einen Ansatz zur direkten Quantifizierung des Ortsfehlers entwickelt. Mit Hilfe dieses Ansatzes wurde wir Benchmarking-Experiment auf Grundlage eines fünfminütigen DWD Radarkompositprodukts für das komplette Jahr 2016 umgesetzt. In diesem Experiment wurden vier Nowcasting-Modelle aus der rainymotion-Softwarebibliothek verwendet im Hinblick auf den Ortsfehler der Vorhersage verglichen. Die Ergebnisse zeigen, dass für alle konkurrierenden Modelle die Ortsfehler von Bedeutung sind: die Größenordnung dieser Fehler entspricht etwa der typischen Ausdehnung einer konvektiven Zelle oder einer mittelgroßen Stadt (5–10 km). Insgesamt zeigt diese Arbeit die Vorteile eines "Open Science"-Ansatzes für die Modellentwicklung im Bereich der Niederschlagsnowcastings. Alle vorgestellten Modelle und Modellsysteme stehen als offene, gut dokumentierte Repositorien zusammen mit entsprechenden offenen Datensätzen öffentlich zu Verfügung für, was die Transparenz und Reproduzierbarkeit des methodischen Ansatzes, aber auch die Anwendbarkeit in der Praxis erhöht. KW - Weather radar KW - nowcasting KW - optical flow KW - deep learning KW - Wetterradar KW - Deep Learning KW - Nowcasting KW - Optischer Fluss Y1 - 2021 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-504267 ER - TY - JOUR A1 - Ayzel, Georgy T1 - Deep neural networks in hydrology BT - the new generation of universal and efficient models BT - новое поколение универсальных и эффективных моделей JF - Vestnik of Saint Petersburg University. Earth Sciences N2 - For around a decade, deep learning - the sub-field of machine learning that refers to artificial neural networks comprised of many computational layers - modifies the landscape of statistical model development in many research areas, such as image classification, machine translation, and speech recognition. Geoscientific disciplines in general and the field of hydrology in particular, also do not stand aside from this movement. Recently, the proliferation of modern deep learning-based techniques and methods has been actively gaining popularity for solving a wide range of hydrological problems: modeling and forecasting of river runoff, hydrological model parameters regionalization, assessment of available water resources. identification of the main drivers of the recent change in water balance components. This growing popularity of deep neural networks is primarily due to their high universality and efficiency. The presented qualities, together with the rapidly growing amount of accumulated environmental information, as well as increasing availability of computing facilities and resources, allow us to speak about deep neural networks as a new generation of mathematical models designed to, if not to replace existing solutions, but significantly enrich the field of geophysical processes modeling. This paper provides a brief overview of the current state of the field of development and application of deep neural networks in hydrology. Also in the following study, the qualitative long-term forecast regarding the development of deep learning technology for managing the corresponding hydrological modeling challenges is provided based on the use of "Gartner Hype Curve", which in the general details describes a life cycle of modern technologies. N2 - В течение последнего десятилетия глубокое обучение - область машинного обучения, относящаяся к искусственным нейронным сетям, состоящим из множества вычислительных слоев, - изменяет ландшафт развития статистических моделей во многих областях исследований, таких как классификация изображений, машинный перевод, распознавание речи. Географические науки, а также входящая в их состав область исследования гидрологии суши, не стоят в стороне от этого движения. В последнее время применение современных технологий и методов глубокого обучения активно набирает популярность для решения широкого спектра гидрологических задач: моделирования и прогнозирования речного стока, районирования модельных параметров, оценки располагаемых водных ресурсов, идентификации факторов, влияющих на современные изменения водного режима. Такой рост популярности глубоких нейронных сетей продиктован прежде всего их высокой универсальностью и эффективностью. Представленные качества в совокупности с быстрорастущим количеством накопленной информации о состоянии окружающей среды, а также ростом доступности вычислительных средств и ресурсов, позволяют говорить о глубоких нейронных сетях как о новом поколении математических моделей, призванных если не заменить существующие решения, то значительно обогатить область моделирования геофизических процессов. В данной работе представлен краткий обзор текущего состояния области разработки и применения глубоких нейронных сетей в гидрологии. Также в работе предложен качественный долгосрочный прогноз развития технологии глубокого обучения для решения задач гидрологического моделирования на основе использования «кривой ажиотажа Гартнера», в общих чертах описывающей жизненный цикл современных технологий. T2 - Глубокие нейронные сети в гидрологии KW - deep neural networks KW - deep learning KW - machine learning KW - hydrology KW - modeling KW - глубокие нейронные сети KW - глубокое обучение KW - машинное обучение KW - гидрология KW - моделирование Y1 - 2021 U6 - https://doi.org/10.21638/spbu07.2021.101 SN - 2541-9668 SN - 2587-585X VL - 66 IS - 1 SP - 5 EP - 18 PB - Univ. Press CY - St. Petersburg ER - TY - JOUR A1 - Costa Tomaz de Souza, Arthur A1 - Ayzel, Georgy A1 - Heistermann, Maik T1 - Quantifying the location error of precipitation nowcasts JF - Advances in meteorology N2 - In precipitation nowcasting, it is common to track the motion of precipitation in a sequence of weather radar images and to extrapolate this motion into the future. The total error of such a prediction consists of an error in the predicted location of a precipitation feature and an error in the change of precipitation intensity over lead time. So far, verification measures did not allow isolating the extent of location errors, making it difficult to specifically improve nowcast models with regard to location prediction. In this paper, we introduce a framework to directly quantify the location error. To that end, we detect and track scale-invariant precipitation features (corners) in radar images. We then consider these observed tracks as the true reference in order to evaluate the performance (or, inversely, the error) of any model that aims to predict the future location of a precipitation feature. Hence, the location error of a forecast at any lead time Delta t ahead of the forecast time t corresponds to the Euclidean distance between the observed and the predicted feature locations at t + Delta t. Based on this framework, we carried out a benchmarking case study using one year worth of weather radar composites of the German Weather Service. We evaluated the performance of four extrapolation models, two of which are based on the linear extrapolation of corner motion from t - 1 to t (LK-Lin1) and t - 4 to t (LK-Lin4) and the other two are based on the Dense Inverse Search (DIS) method: motion vectors obtained from DIS are used to predict feature locations by linear (DIS-Lin1) and Semi-Lagrangian extrapolation (DIS-Rot1). Of those four models, DIS-Lin1 and LK-Lin4 turned out to be the most skillful with regard to the prediction of feature location, while we also found that the model skill dramatically depends on the sinuosity of the observed tracks. The dataset of 376,125 detected feature tracks in 2016 is openly available to foster the improvement of location prediction in extrapolation-based nowcasting models. KW - inuosity Y1 - 2020 U6 - https://doi.org/10.1155/2020/8841913 SN - 1687-9309 SN - 1687-9317 VL - 2020 PB - Hindawi CY - London 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 -