Multi-Source Environmental Data Flood Forecasting System Based on ConvLSTM Networks

Abstract

To lessen or avoid fatalities, injuries, and property damage brought on by heavy rainfall-induced floods, improve the precision and timeliness of flood early warning notifications. In this project, develop a sophisticated flood forecasting system using deep learning and a variety of environmental data. To investigate the upper reaches of the X River, a new system has been set up in collaboration with the departments of hydrology, meteorology, and remote sensing. For high-resolution, real-time flood forecasting, spatial-temporal features are learned using a ConvLSTM-Attention neural network. With an average Nash-Sutcliffe Efficiency of 0.84 and an average Root Mean Square Error (RMSE) of 0.18 meters, the system can outperform conventional hydrodynamic and other machine learning models, according on studies conducted over the last five years and 14 flood events. In certain locations, the model has been able to provide an early warning of peak flood more than four hours in advance, and it works well in all Basin areas and meteorological conditions. The aforementioned analysis indicates that the suggested course is workable and will be the foundation for developing a functional emergency response and flood early warning system.