Hybrid ensemble machine learning algorithms for landscape ecological vulnerability assessment to riverbank erosion

Riverbank erosion is a catastrophic geomorphological hazard that poses severe ecological and socio-economic challenges across densely populated floodplains. This study advances a machine learning (ML) framework that integrates individual and bagging-classifier approaches using random forest (RF), multilayer perceptron (MLP) and bagging classifiers to assess landscape ecological vulnerability (LEV) to riverbank erosion. The site-specific environmental, climatic, geomorphological and ecological parameters were selected by employing a multicollinearity test. The effectiveness of model performance was validated through accuracy, precision, recall, F1-score and area under curve (AUC). The results demonstrated that the bagging ensemble model outperformed individual classifiers, achieving the highest AUC of 0.97, accuracy of 0.91, precision of 0.93, recall of 0.92 and F1-score of 0.95. The results show that the largest area was found under the very high-vulnerability zone (27.6%) followed by the high (25.8%), very low (21.1%), moderate (12.9%) and low (12.6%) zones. Vulnerability zones were most concentrated in the middle and lower reaches of the study area particularly in the states of Bihar and West Bengal. Sensitivity analysis revealed that elevation, rainfall, soil type and geomorphology were the most influential factors. Removal of elevation resulted in the largest reduction in model performance. The findings provide the immediate need for targeted land management interventions and offer a scalable ML framework to inform sustainable planning, disaster risk reduction and ecological conservation in dynamic riverine environments.