<p>Floods are among the most frequent, destructive natural hazards globally, with increasing intensity in arid and semi-arid regions due to climate variability and land-use change. In Tunisia’s Jeffara basin, flash floods remain poorly anticipated, challenging local resilience and adaptive water governance. This study addresses the limitations of conventional flood susceptibility mapping approaches, particularly their underrepresentation of geological controls and spatial variability. It aims to develop a robust, knowledge-driven, framework that improves the ranking of flood-controlling factors and enhances the spatial prediction of flood-prone zones, especially under data-scarce conditions. A multi-criteria methodology was designed, integrating Remote Sensing, Geographic Information Systems, Geostatistics, and Multivariate Analysis. Eight conditioning factors were considered, encompassing, topographic, hydrological, and geological characteristics. Unlike common approaches, geological and structural features were not merely treated as data inputs, but integrated as explanatory drivers, enabling improved representations of subsurface heterogeneity. Weights were assigned using the Analytical Hierarchy Process, with a consistency ratio of 0.01 confirming methodological soundness. The resulting flood susceptibility model indicates that 44% of the study area is exposed to high/very high flood susceptibility. Its predictive performance was excellent, achieving a 96% score based on the area under the receiver operating characteristic (AUC-ROC) curve. Reliability was further confirmed through sensitivity analysis, and validated by flood maps, for 2018 and 2024 events, generated using Sentinel-1 imagery processed on Google Earth Engine platform. These outcomes highlight the operational relevance of knowledge-driven, geologically informed models in supporting adaptive planning and flood mitigation strategies, including nature-based solutions such as Jessour and Tabias.</p>