Event-specific flood extent prediction using ensemble machine learning and multi-source geospatial data

Study Region The Kashkan River Basin in western Iran is characterized by complex topography and diverse environmental conditions that contribute to spatial variability in flood occurrence and extent. The basin is therefore a relevant case for developing high-resolution approaches to flood extent assessment and spatial flood-risk analysis. Study Focus This study developed a machine-learning framework for predicting the spatial extent of flooding by integrating 11 environmental raster variables on a common 10 m spatial grid with two ensemble machine-learning algorithms, Random Forest (RF) and Extreme Gradient Boosting (XGBoost). The predictor variables included topographic factors, vegetation cover, land use, precipitation, soil moisture, and hydrological indices derived from Sentinel-1 and Sentinel-2 imagery, together with complementary geospatial datasets. Model performance was evaluated using stratified five-fold cross-validation based on the 2019 flood inventory. RF achieved an accuracy of 0.9877, F1-score of 0.7426, and recall of 0.9982, whereas XGBoost achieved an accuracy of 0.9977, F1-score of 0.9346, and recall of 0.9344. ROC-AUC and confusion-matrix results further indicated strong discrimination during internal validation of the 2019 flood event. New Hydrological Insights for the Region The results demonstrate that high-resolution, spatially consistent environmental datasets can effectively support event-specific flood extent prediction in the Kashkan River Basin. XGBoost provided substantially better overall classification performance than RF, particularly in terms of F1-score, while RF showed higher recall. Feature-importance analysis identified elevation and vegetation cover as the most influential predictors of the observed flood extent. These findings highlight the potential of integrating multi-source remote-sensing and geospatial data with ensemble machine-learning methods to improve spatial flood assessment and support flood-risk analysis in the basin.