A Hybrid CNN-LSTM approach for multi-step discharge forecasting with satellite data

Papis Wongchaisuwat¹Konlawat Yomwilai²WANDEE THAISIAM^2*

• ¹Department of Industrial Engineering, Kasetsart University, Bangkok, Thailand
• ²Department of Water Resources Engineering, Kasetsart University, Bangkok, Thailand

An innovative hybrid forecasting model was developed to predict hourly river discharges up to 24 hours in advance. The proposed model combined the spatial analysis capabilities of convolutional neural networks with the temporal sequence modeling strengths of long short-term memory networks. It was designed to extract spatio-temporal patterns from satellite precipitation data while incorporating historical discharge trends to improve forecasting accuracy. The model was developed and internally validated on a 3 day lag dataset before being tested on an external test set with a relatively shorter 30 minute lag to simulate real-time conditions. The findings provided a scalable solution to improve discharge forecasting and support water resource management, particularly for upstream, unmeasured catchments in data-scarce regions. The hybrid model's predictive capabilities and its generalization across datasets were confirmed based on the evaluation metrics (correlation coefficient, Nash-Sutcliffe efficiency, root mean square error, and mean absolute error). Notably, our proposed model maintained a performance level well within acceptable error limits, even using the 24 hour prediction horizon.