Abstract
The use of deep learning methods for fluvial flood forecasting is rapidly gaining traction, offering a promising solution to the challenges associated with accurate yet time-consuming numerical models. This paper presents two physics-inspired deep learning approaches specifically designed for fluvial flood forecasting, each embracing different learning principles: centralized and federated learning. The centralized model utilizes an Encoder-Decoder technique to handle input data of varying types and scales, while the federated model is based on a node-link graph with a seq2seq internal model. Both models are enhanced with a probabilistic forecasting head to account for the inherent uncertainty in streamflow forecasts. The objective of these approaches is to address the limitations of traditional numerical models while leveraging the potential of deep learning to improve the speed, accuracy, and scalability of flood forecasting. To validate their effectiveness, the models were tested across different use cases. The findings from the federated learning approach emphasize the importance of catchment clustering before model utilization and demonstrate the models’ ability to generalize effectively in catchments with similar properties. On the other hand, the results of the centralized method highlight the model’s reliance on the test set falling within the data range of the training set (Average NSE and KGE for the sixth hour ahead of 0.88 and 0.78, respectively). To address this limitation, the paper suggests the development of a method for the future, such as leveraging a numerical model or using Generative Adversarial Networks, to generate highly extreme events, particularly in the context of a changing climate. The models are implemented in a flexible operational framework based on open standards, ensuring their adaptability and usability in various settings.
Zusammenfassung
Dieser Beitrag präsentiert zwei Deep-Learning-Ansätze für die Vorhersage von Hochwassern: zentrales und föderiertes Lernen mit probabilistischem Vorhersagekopf. Das zentrale Modell verwendet eine Encoder-Decoder-Technik, um Eingabedaten unterschiedlicher Typen und Skalen zu verarbeiten, während das federated Modell auf einem Knoten-Link-Graphen mit einem seq2seq-Internen Modell basiert. Die Modelle überwinden Einschränkungen traditioneller Modelle und verbessern Geschwindigkeit, und Skalierbarkeit. Validierung in verschiedenen Anwendungsfällen zeigt die Effektivität der Methoden. Methoden wie die Nutzung von Generative Adversarial Networks werden vorgeschlagen, um extreme Ereignisse zu generieren. Die Modelle sind flexibel und benutzerfreundlich in einem operativen Framework implementiert.
About the authors
Dr.-Ing. LL.M Pat. Ing. Divas Karimanzira is currently vice group leader of the Maritime group at the Fraunhofer IOSB in Ilmenau. He did his Ph.D in Automation and Control at the University of Ilmenau in Germany. He has published several research articles in the field of flood modelling.
Lucas Richter is a trained meteorologist and has been working at the Fraunhofer IOSB-AST in the Department of Cognitive Energy Systems since 2020. His main topics are data mining, time series analysis, data engineering, time series forecasting, and optimization. In his doctoral thesis, he is dealing with the energy consumption forecasting of distributed and non-stationary time series.
Dr.-Ing. Desiree Hilbring has a degree in Geodesy. In her research at Fraunhofer IOSB, she dealt with system architectures based on open geodata standards, which have been applied for the development of information systems in in the environmental sector or in crisis management. In recent years, the main focus has been on the application and integration of AI in corresponding systems.
M.Sc. Michaela Lödige ist wissenschaftliche Mitarbeiterin im Geschäftsfeld Future City Solutions am Fraunhofer IOSB-INA. In ihrer aktuellen Tätigkeit unterstützt sie die wissenschaftliche Begleitung und Umsetzung von Smart City Projekten zur Planung und Gestaltung moderner Lebensräume. Ein Schwerpunkt ihrer Forschung ist insbesondere der Einsatz von IoT-Systemen in der Hochwasser- und Starkregenvorsorge. 2019 schloss sie den Masterstudiengang Raumplanung an der Technischen Universität Dortmund mit dem Schwerpunkt auf die Freiraum- und Umweltplanung ab und beschäftigte sich in der Lehre und Forschung an der Technischen Universität Dortmund insbesondere mit innovativen Möglichketen der digitalen Transformation in der Stadt- und Raumplanung.
Jonathan Vogl is currently working at the Institut für Industriebetriebslehre und Industrielle Produktion (IIP) of the Karlsruhe Intsitute of Technology as a reaearch scientist. In the Energy Demand & Mobility group. He specializes in software architecture and development.
Acknowledgment
We are very grateful to three anonymous reviewers for their detailed and constructive suggestions and critics that helped to reshape the approach behind the article. We would further like to thank all the colleagues involved in the PrognoSF Technology Innovation Programmme (TEP) for their assistance.
-
Research ethics: Not applicable.
-
Author contributions: Michaela Lödige: Data; Jonathan Vogl and Desiree Hilbring: Operational framework; Lucas Richter: Federated Learning; Divas Karimanzira: All analysis and central learning model.
-
Competing interests: The authors declare not competing interest.
-
Research funding: None declared.
-
Data availability: Not applicable.
References
[1] H. Tabari, “Climate change impact on flood and extreme precipitation increases with water availability,” Sci. Rep., vol. 10, pp. 1–10, 2020, https://doi.org/10.1038/s41598-020-70816-2.Search in Google Scholar PubMed PubMed Central
[2] European Union, “Directive 2007/60/EC of the European counil and European parliment of 23 October 2007 on the assessment and management of flood risks,” Off. J. Eur. Union, pp. 27–34, 2007.Search in Google Scholar
[3] K. Khosravi, et al.., “Convolutional neural network approach for spatial prediction of flood hazard at national scale of Iran,” J. Hydrol., vol. 591, p. 125552, 2020, https://doi.org/10.1016/j.jhydrol.2020.125552.Search in Google Scholar
[4] Zhongrun and I. Demir, “Fully distributed rainfall-runoff modeling using spatial-temporal graph neural network,” 2022, https://doi.org/10.31223/X57P74.Search in Google Scholar
[5] M. Sit, B. Z. Demiray, Z. Xiang, G. J. Ewing, and Y. Sermet, “A comprehensive review of deep learning applications in hydrology and water resources,” Water Sci. Technol., vol. 82, no. 12, pp. 2635–2670, 2020. https://doi.org/10.2166/wst.2020.369.Search in Google Scholar PubMed
[6] F. Kratzert, D. Klotz, S. Hochreiter, and G. S. Nearing, “A note on leveraging synergy in multiple meteorological data sets with deep learning for rainfall–runoff modeling,” Hydrol. Earth Syst. Sci., vol. 25, no. 5, pp. 2685–2703, 2021. https://doi.org/10.5194/hess-25-2685-2021.Search in Google Scholar
[7] J. Zhang and H. Yan, “Long short-term components neural network model with data augmentation for daily runoff forecasting,” J. Hydrol., vol. 617, p. 128853, 2023, https://doi.org/10.1016/j.jhydrol.2022.128853.Search in Google Scholar
[8] P. Weng, Y. Tian, Y. Liu, and Y. Zheng, “Time-series generative adversarial networks for flood forecasting,” J. Hydrol., vol. 622, p. 129702, 2023, https://doi.org/10.1016/j.jhydrol.2023.129702.Search in Google Scholar
[9] D. Karimanzira, L. Ritzau, and K. Emde, “Catchment area multi-streamflow multiple hours ahead forecast based on deep learning,” Trans. Mach. Learn. Artif. Intell., vol. 10, no. 5, pp. 15–29, 2022. https://doi.org/10.14738/tmlai.105.13049.Search in Google Scholar
[10] M. Castangia, et al.., “Transformer neural networks for interpretable flood forecasting,” Environ. Model. Software, vol. 160, p. 105581, 2023, https://doi.org/10.1016/j.envsoft.2022.105581.Search in Google Scholar
[11] S. Gao, et al.., “Short-term runoff prediction with GRU and LSTM networks without requiring time step optimization during sample generation,” J. Hydrol., vol. 589, p. 125188, 2020, https://doi.org/10.1016/j.jhydrol.2020.125188.Search in Google Scholar
[12] G. Liu, et al.., “Assessing spatial connectivity effects on dailystreamflow forecasting using Bayesian-based graph neural network,” Sci. Total Environ., vol. 855, p. 158968, 2023, https://doi.org/10.1016/j.scitotenv.2022.158968.Search in Google Scholar PubMed
[13] G. Zhou, Z. Guo, S. Sun, and Q. Jin, “CNN-BiGRU-AM neural network for AI applications in shale oil production prediction,” Appl. Energy, vol. 344, p. 121249, 2023, https://doi.org/10.1016/j.apenergy.2023.121249.Search in Google Scholar
[14] T. Zhao, Q. J. Wang, J. C. Bennett, D. E. Robertson, Q. Shao, and J. Zhao, “Quantifying predictive uncertainty of streamflow forecasts based on a Bayesian joint probability model,” J. Hydrol., vol. 528, pp. 329–340, 2015, https://doi.org/10.1016/j.jhydrol.2015.06.043,Search in Google Scholar
[15] Y. Pokhrel, et al.., “Global terrestrial water storage and drought severity under climate change,” Nat. Clim. Change, vol. 11, pp. 226–233, 2021, https://doi.org/10.1038/s41558-020-00972-w.Search in Google Scholar
[16] M. H. Ramos, S. J. Van Andel, and F. Pappenberger, “Do probabilistic forecasts lead to better decisions?” Hydrol. Earth Syst. Sci., vol. 17, no. 6, pp. 2219–2232, 2013. https://doi.org/10.5194/hess-17-2219-2013.Search in Google Scholar
[17] F. Ghobadi and D. Kang, “Multi-step ahead probabilistic forecasting of daily streamflow using bayesian deep learning: a multiple case study,” Water, vol. 14, p. 3672, 2022, https://doi.org/10.3390/w14223672.Search in Google Scholar
[18] G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, “Physics-informed machine learning,” Nat. Rev. Phys., vol. 3, pp. 422–440, 2021, https://doi.org/10.1038/s42254-021-00314-5.Search in Google Scholar
[19] M. Kahle, M. Kempf, B. Martin, and R. Glaser, “Classifying the 2021 ‘Ahrtal’ flood event using hermeneutic interpretation, natural language processing, and instrumental data analyses,” Environ. Res. Commun., vol. 4, no. 5, p. 051002, 2022. https://doi.org/10.1088/2515-7620/ac6657.Search in Google Scholar
© 2024 Walter de Gruyter GmbH, Berlin/Boston