Application of a New Hybrid Deep Learning Model That Considers Temporal and Feature Dependencies in Rainfall–Runoff Simulation

Runoff forecasting is important for water resource management. Although deep learning models have substantially improved the accuracy of runoff prediction, temporal and feature dependencies between rainfall–runoff time series elements not been effectively exploited. In this work, we propose a new hybrid model to predict hourly streamflow: SA-CNN-LSTM (self-attention, convolutional neural network, long short-term memory network). The advantages CNN LSTM in terms data extraction from are combined with self-attention mechanism. By considering interdependences sequence timesteps features, prediction performance enhanced. We explored Mazhou Basin, China; compared its performances LSTM, CNN, ANN (artificial network), RF (random forest), SA-LSTM, SA-CNN. Our analysis demonstrated that robust different flood magnitudes lead times; it was particularly effective within times 1–5 h. Additionally, mechanism alone, respectively, at some however, overall unstable. contrast, integrating exhibited better robustness. Overall, study considers importance then proposes improve conventional prediction.