A Deep Learning-Based Multi-model Ensemble Method for Hydrological Forecasting.

Efficient, robust, and accurate forecasting is crucial in early flood warning, flood hazard prevention, and water resources management. This research builds an integrated deep learning model (MCR-BiLSTM) integrating the predictive merits of Convolutional Neural Network (CNN) and Bi-directional Long Short-Term Memory (BiLSTM) Network to design and evaluate a flood forecasting model to forecast the future occurrence of flood events. The model can reflect the influence of rainfall’s temporal and spatial distribution on future water level changes by extracting spatial-temporal features of hydrological processes with CNN enhanced by Residual Neural Units (ResNet) and BiLSTM, respectively. Moreover, this model adopts a self-attention mechanism to establish the long-distance dependence relationship between input and output sequences. The performance of the proposed model is validated against 2 different rainfall datasets in flood-prone regions in China which faces flood-driven devastations almost annually. The results illustrate that the integrated deep learning forecast model provides stability and more accurate forecasting results than a single model providing a new idea for hydrological forecasting.