(AIMS 2024) Flood inundation range prediction method based on SRR-Informer

Flood forecasting methods based on deep learning rely on a large number of observational data, and are facing serious challenges in areas with scarce data. Aiming at the problems of flood inundated range prediction in areas with scarce data, this paper proposes a flood inundated range prediction method based on spatial reduction reconstruction (SRR) and improved Deep attention neural network (Informer) to reduce the data requirements of the two core parts of water level simulation and spatial modeling. It alleviates the problem of insufficient observational data for flood inundation range prediction in areas with scarce data. This method obtains the dependency relationship of long time series data through Informer model, and inserts the built-in input selection layer to reduce the number of parameters in the model, reducing the data requirement of water level simulation. At the same time, the SRR algorithm is used to select the representative locations that are easy to be inundated in the basin, which reduces the number of locations required for spatial modeling of flood inundation range and the corresponding data requirements. The experimental results show that this method can improve the accuracy of flood inundation range prediction and speed up the efficiency of the model.