Flood risk modelling by the synergistic approach of machine learning and best-worst method in Indus Kohistan, Western Himalaya

Abstract

Flood risk mapping is essential for effective mitigation and planning. In this study, we propose a novel synergistic approach for flood risk mapping in Indus Kohistan, Pakistan, by integrating machine learning (ML) models and the best-worst method (BWM). Flood hazard assessment was carried out using random forest (RF) and extreme gradient boosting (XGBoost), while flood vulnerability was evaluated through BWM. The supporting data comprised 415 flood points, 13 flood hazard factors, and 6 flood vulnerability indicators. The RF model outperformed XGBoost, achieving an area under the curve (AUC) of 0.99, compared to 0.98 for XGBoost. Ultimately, the RF model was utilized to generate the flood hazard map. Distance to streams and elevation are key drivers of flood hazard, with relative importance values of 0.43 and 0.19, respectively. Meanwhile, population density and household density have the highest impact on flood vulnerability, contributing 40.2% and 24.7%, respectively. The flood risk map reveals that 1.4%, 4.6%, 9%, 18.4%, and 65.9% of the total land are categorized as very high-risk, high-risk, medium-risk, low-risk, and very low-risk, respectively. The proposed methodology and flood risk map results can be valuable for identifying priority measures for flood risk reduction and management.