This study presents a deep-learning-based forecast model for spatial and temporal prediction of pluvial flooding. The developed model can produce the flooding situation for the upcoming time steps as a sequence of flooding maps. Thus, a dynamic overview of the forthcoming flooding situation is generated to support the decision of crisis management actors. The influence of different input data, data formats, and model setups on the prediction results was investigated. Data from multiple sources were considered as follows: precipitation information, spatial information, and an overflow forecast. In addition, models with different layers and network architectures such as convolutional layers, graph convolutional layers, or generative adversarial networks (GANs) were considered and evaluated. The data required to train and test the models were generated using a coupled hydrodynamic 1D/2D model. The model setup with the inclusion of all available input variables and an architecture with graph convolutional layers presented, in general, the best results in terms of root mean square error (RMSE) and critical success index (CSI). The prediction results of the final model showed a high agreement with the simulation results of the hydrodynamic model, with drastic reductions in computation time, making this model suitable for integration into an early warning system for pluvial flooding.

This study employs a temporal fusion transformer (TFT) for predicting overflow from sewer manholes during heavy rainfall events. The TFT utilised is capable of forecasting overflow hydrographs at the manhole level and was tested on a sewer network with 975 manholes. As part of the investigations, the TFT was compared to other deep learning architectures to evaluate its predictive performance. In addition to precipitation measurements and forecasts, the issue of how the additional consideration of measurements in the sewer network as model inputs impacts forecast accuracy was investigated. A varying number of sensors and different measurement signals were compared. The results indicate high performance for the TFT compared to other model architectures like a long short-term memory (LSTM) network or a dual-stage attention-based recurrent neural network (DA-RNN). Additionally, results suggest that considering a single measuring point at the outlet of the sewer network instead of an entire measuring network yields better forecasts. One possible explanation is the high correlation between measurements, which increases model and training complexity without adding much value.

Deep learning (DL) approaches to forecast precipitation and inundation areas in the short-term forecast horizon have up until now been treated as independent research problems from the model development perspective. However, for the urban hydrology area, the coupling of these models is necessary in order to forecast the upcoming inundation area maps and is, therefore, of the utmost importance for successful flood risk management. In this paper, three deep-learning-based models are coupled in a systematic modular approach with the aim to analyze the performance of this model chain in an operative setup for urban pluvial flooding nowcast: precipitation nowcasting with an adapted version of the NowcastNet model, the forecast of manhole overflow hydrographs with a Seq2Seq model, and the generation of a spatiotemporal sequence of inundation areas in an urban catchment for the upcoming hour with an encoder–decoder model. It can be concluded that the forecast quality still largely depends on the accuracy of the precipitation nowcasting model. With the increasing development of DL models for both precipitation and flood nowcasting, the presented modular approach for model coupling enables the substitution of individual blocks for better and newer models in the model chain without jeopardizing the operation of the flooding forecast system.