GIS and remote sensing-based hybrid AHP–machine learning framework for flood susceptibility and risk assessment in the Wolaita Zone, Southern Ethiopia

Abstract

Flooding is a major natural hazard with significant social, economic, and environmental impacts, particularly in data-scarce regions. This study applied GIS and remote sensing–based framework integrating multi-criteria decision analysis via the analytical hierarchy process with ensemble machine-learning models to assess flood susceptibility and risk in the Wolaita Zone. Eleven flood-conditioning factors, including elevation, slope, rainfall, distance to rivers, drainage density, TWI, LULC, NDVI, soil type, geology, and curvature, were first reclassified to a common scale. These factors were then weighted using AHP and used as inputs for both standalone AHP and hybrid machine-learning models. Spatial analysis using the three models shows that moderate flood susceptibility dominates most of the area, while high-susceptibility zones are concentrated in lowlands and along river corridors. Flood risk mapping indicates that approximately 60% of the study area is at moderate risk and 31% at high risk, with very low, low, and very high-risk areas covering minimal portions. The hybrid AHP–RF model demonstrated superior predictive capability, achieving an AUC value of 0.957. These findings demonstrate that the integration of expert-derived AHP weights with machine-learning models produces accurate, high-resolution flood susceptibility and risk maps, offering a robust framework for flood mitigation and land-use planning.