Assessment of flood risk using remote sensing data and machine learning models in Upper Jhelum Sub-catchment, India

The assessment and mapping of flood risk are significant for effective disaster mitigation and management strategies. The study makes an attempt to assess flood risk within the watersheds of the Upper Jhelum sub-catchment by utilizing remote sensing data and machine learning models. Site specific flood inducing factors, flood breeding environmental factors and element at risk were utilized to evaluate flood risk in the Sub-catchment. Multilayer perceptron (MLP) and random forest (RF) and their ensembles of stacking were utilized for assessing the flood risk. Area of the Sub-catchment was found under low risk followed by high and moderate risk. Of all the watersheds, largest area under high risk was found in 1E1D3 watershed (Vishav) followed by 1E1D2 watershed (Rembaira) and 1EID4 watershed (Lidder). Largest area under moderate flood risk was found in the 1E1D3 (Vishav) watershed followed by 1E1D4 (Lidder) and 1E1D7 (Sandran) watersheds. Largest area under low flood risk was found in the 1E1D4 (Lidder) watershed followed by 1E1D6 (Bringi) and 1E1D3 (Vishav). The ensemble stacking with random forest (S-RF) recorded the highest performance under the receiver operating characteristics ROC curve (AUC=0.920). The findings may help policymakers and stakeholders in formulating effective flood risk management strategies.