Flood susceptibility mapping in El Niño Phenomenom integrating multitemporal radar analysis, GIS and machine learning techniques, Piura River basin, Peru

Edwin Badillo-Rivera^1*Ramiro Santiago²Ivan Poma²Teodosio Chávez²Antonio Arroyo-Paz³Andrea Aucahuasi-Almidon⁴Edilberto Hinostroza⁵Eric Segura⁶Luz Eyzaguirre⁷Hairo León⁸Paul Virú-Vásquez⁵

• ¹Faculty of Environmental Engineering and Natural Resources, National University of Callao, Callao, Peru
• ²Faculty of Geological, Mining and Metallurgical Engineering, Universidad Nacional de Ingenieria, Rimac, Peru
• ³Faculty of Engineering, Universidad Tecnológica del Perú, Arequipa, Peru
• ⁴Faculty of Engineering, Universidad Nacional Amazonica de Madre de Dios, Puerto Maldonado, Peru
• ⁵Faculty of Environmental Engineering and Natural Resources, Universidad Nacional del Callao, Callao District, Peru
• ⁶Instituto Nacional de Defensa Civil, San Isidro, Peru
• ⁷Faculty of Petroleum, Natural Gas and Petrochemical Engineering, Universidad Nacional de Ingenieria, Rimac, Peru
• ⁸Research Center for Environmental Earth Science and Technology, Universidad Nacional Santiago Antunez de Mayolo, Huaraz, Peru

Floods represent the most frequent natural hazard, causing major impacts on people and severe economic and environmental losses worldwide. These events are exacerbated by extreme climatic phenomena, such as the 2017 Coastal El Niño, the most intense in the past century, with the Piura region of Peru most affected. Flood susceptibility models (FSM) are essential for mitigating impacts through land-use planning, policy formulation, and resilience building. This study developed FSM in northern Peru using a hybrid methodology integrating optical and radar remote sensing (RS), GIS, and machine learning (ML). Sentinel-1 data mapped flood extent with the Normalized Difference Flood Index (NDFI), while flood susceptibility was modeled with ten topographic variables, NDVI, geology, and geomorphology. Multicollinearity issues were addressed with Principal Component Analysis (PCA), selecting four components explaining 75.4% of variance. Six FSMs were generated using Support Vector Machines (SVM) and Random Forest (RF) with Quantiles (q), Frequency Ratio (FR), and Weights of Evidence (WoE). Validation used AUC, F1-score, Accuracy, and cross-validation. The RF-WoE ensemble achieved highest performance (AUC = 0.988 training, >0.907 validation), outperforming SVM. SHAP analysis confirmed geology, geomorphology, and aspect as key predictors. Susceptibility maps identified the lower Piura basin as most vulnerable during the 2017 event due to morphology and land use. This study demonstrates a robust, interpretable hybrid methodology combining PCA, ML, and SHAP analysis, offering valuable tools for decision-makers to strengthen resilience, reduce vulnerability, and integrate flood susceptibility into land-use planning for sustainable management.