Prioritizing Agricultural Flood Mitigation: A GeoAI-Driven Assessment of Susceptibility, Crop Exposure, and Socioeconomic Vulnerability

Mirza Md Tasnim Mukarram , Mirza Waleed, Quazi Umme Rukiya, Marc Linderman, Ibrahim Demir

Flooding is a recurrent hazard across the U.S. Midwest, yet frameworks integrating flood hazard, agricultural exposure, and socioeconomic vulnerability at the county scale remain limited. This study presents a GeoAI-driven agricultural flood risk assessment for all 99 counties in Iowa, combining machine learning-derived flood susceptibility with crop-specific economic exposure and socioeconomic vulnerability within a hazard–exposure–vulnerability (HEV) framework. Using ten 30m resolution conditioning factors, a LightGBM model was optimized and interpreted via SHAP, identifying elevation, rainfall frequency, and river proximity as dominant susceptibility drivers (ROC-AUC = 0.95, F1-score = 0.90). This susceptibility surface—classifying 13.3% of Iowa as high and very high susceptibility—was integrated with 2022 USDA Census data representing $17.6 billion in combined corn and soybean value, alongside a five-indicator socioeconomic vulnerability index. The integrated HEV analysis identified Black Hawk County as possessing the highest compound agricultural risk, while Monona County recorded the highest scenario-based annualized loss. Statewide, annualized agricultural flood losses are estimated at $527.7 million, equivalent to 3.04% of gross crop revenue. Furthermore, spatial hotspot analysis revealed a persistent risk cluster within the Cedar River watershed, allowing for the classification of 48 counties as Priority 1 intervention areas. By translating technical susceptibility mapping into actionable economic and spatial risk metrics, this scalable framework provides critical decision-support for agricultural disaster risk reduction, targeted climate adaptation, and resource allocation in intensive farming regions.