Beyond Black-Box Flood Susceptibility Mapping: An Explainable and Community-Validated GIS-Machine Learning Framework

Dr Kazi Abdul Abdul Mannan , Nazia Akter

Floods are among the most destructive natural hazards, causing substantial economic losses, environmental degradation, and threats to human lives worldwide. Although Geographic Information Systems (GIS) and machine learning (ML) have significantly improved flood susceptibility mapping (FSM), many existing models remain difficult to interpret due to their black-box nature, limiting stakeholder trust and practical decision-making. This study proposes an explainable and community-validated GIS-machine learning framework that integrates ensemble learning with Explainable Artificial Intelligence (XAI) techniques, including SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations), alongside Participatory Geographic Information Systems (PGIS). Using secondary spatial and environmental datasets, the proposed framework identifies influential flood-conditioning factors, compares the predictive performance of multiple ML models, interprets global and local model behavior, and validates prediction outcomes using community-derived spatial knowledge. The integration of explainable AI and PGIS enhances transparency, credibility, and policy relevance while supporting evidence-based flood risk management. The proposed framework provides an interpretable and stakeholder-centered decision-support approach that advances sustainable flood susceptibility assessment and disaster risk reduction.