Explainable Machine Learning and Remote Sensing for Urban Flood Susceptibility Mapping: A Simulated GIS-Based Assessment

This empirical-style paper presents a simulated study at the intersection of Environmental Science, Geoinformatics, Urban Planning, and Machine Learning. The study addresses a current research direction in which digital and data- driven methods are being integrated with domain-specific decision systems. 4,800 simulated grid cells, including 1,056 modeled flood-positive cells, were analyzed. Twenty-five percent of the city area was reserved as a spatial hold-out zone. A simulated GIS dataset of 4,800 grid cells was created for a hypothetical rapidly urbanizing city. Each 100 m cell included elevation, slope, topographic wetness index, distance to drainage, drainage density, impervious-surface fraction, normalized difference vegetation index, antecedent rainfall, road density, and built-up intensity. Flood labels were generated from a nonlinear risk surface with added noise. The dataset was split spatially into training and hold-out zones to reduce leakage. Logistic regression, random forest, XGBoost, and a stacked ensemble were compared. Permutation importance and SHAP-like standardized contribution scores were used for model interpretation. The stacked ensemble produced the highest simulated hold-out AUROC (0.928) and F1 score (0.81), followed by XGBoost. Impervious surface, topographic wetness, distance to drainage, antecedent rainfall, and elevation were the strongest predictors. Spatial error analysis showed that false positives clustered near newly urbanized fringe zones where modeled drainage capacity was uncertain. All numerical observations, experimental assignments, outcomes, and performance values are simulated for methodological illustration; no real participants, institutions, farms, factories, hospitals, or city systems were studied. The results are intended to demonstrate a coherent research design, provide testable hypotheses, and show how a future real-world study could report evidence transparently. The paper concludes with implementation constraints, methodological limitations, and priorities for empirical validation.