Water Level Forecasting Using Spatiotemporal Attention-Based Long Short-Term Memory Network

Bangladesh is in the floodplains of Ganges, Brahmaputra, and Meghna River delta, crisscrossed by an intricate web rivers. Although country highly prone to flooding, use state-of-the-art deep learning models predicting river water levels aid flood forecasting underexplored. Deep attention-based have shown high potential for accurately floods over space time. The present study aims develop a long short-term memory (LSTM) network its architectures predict rivers Bangladesh. developed this incorporated gauge-based level data 7 days prediction at Dhaka Sylhet stations. This five models: artificial neural (ANN), LSTM, spatial attention LSTM (SALSTM), temporal (TALSTM), spatiotemporal (STALSTM). multiple imputation chained equations (MICE) method was applied address missing time series analysis. results showed that both together increases predictive performance model, which outperforms other models. STALSTM-based system, study, could inform management plans elsewhere.