Understanding Catchment Impacts: Enhanced Water Level Prediction with LSTM and Explainable AI

Flooding is a significant global challenge, impacting millions annually and causing extensive damage to property and infrastructure. This growing threat is exacerbated by climate change and shifting monsoon patterns. Sri Lanka, with its complex geography of central mountains and extensive river basins, faces increasing flood risks due to these climatic changes. To address these pressing issues, the study introduces an LSTM-based approach integrated with explainable AI to enhance flood prediction capabilities. The methodology involves a systematic approach to data collection and structuring, including a context-based pre-processing method tailored to Sri Lanka's specific geographical characteristics. The research presents a comprehensive modeling approach for predicting water levels and incorporates explainable AI techniques to assess the impact of each catchment area on prediction outcomes, aiming to provide greater transparency in the prediction process and a clearer understanding of the model's decision-making. The model achieved a Mean Squared Error (MSE) of 0.000431 and a Root Mean Square Error (RMSE) of 0.0208, indicating relatively low prediction errors and suggesting that the predictions closely approximate the observed water levels. The study effectively assessed the impact of each catchment area and identified the Rathnapura catchment as the primary contributor to flooding in the targeted region of the Kalu Ganga River. This research is crucial for improving flood management strategies and mitigating the risks associated with flooding in vulnerable regions.