Hydro-Informatic Modeling for Flood Prediction Through Explainable AI to Interpret Water Dynamics in Bangladesh Perspective

Bangladesh's low-lying geography and climatic sensitivity are making Sylhet in particular more vulnerable to flooding. The Surma and Kushiara rivers spilling their banks is the primary source of the floods in Sylhet. These rivers flow into Bangladesh with more monsoon water from their source in northeastern India. A detailed study of upstream causes is necessary to develop effective mitigation techniques, particularly about the rainfall patterns in India's northeastern regions. To address this issue, we provide an approach that investigates the intricate link between rainfall patterns in Northeastern India and flood dynamics in Sylhet by combining Explainable AI (SHAP) with cutting-edge deep learning models. An important addition is the comparison of Long Short-Term Memory Recurrent Neural Networks (LSTM-RNN) and Feedforward Neural Networks (FNN), which are critical for interpreting time-series data in flood preparedness. LSTM-RNN shows strong temporal sensitivity with an R-squared of 0.94; FNN, on the other hand, is more precise with an enhanced R-squared of 0.96, a bigger Root Mean Squared Error of 0.112, and a smaller Mean Squared Error of 0.02.