Integrating remote sensing with swarm intelligence and artificial intelligence for modelling wetland habitat vulnerability in pursuance of damming

The current study aimed to investigate the vulnerability state of wetland habitat as a result of damming. Wetland habitat vulnerability state (WHVS) models for pre and post-dam periods were built to investigate the impact, and the difference was assessed. Sixteen hydrological, land composition and water quality parameters were used for modelling WHVS. Swarm intelligence optimised machine learning algorithms such as SVM (Support Vector Machine), ANN (Artificial Neural Network), bagging, radial basis (RBF) and M5P model tree were developed. The models' efficiency was evaluated using statistical methods such as the Receiver operating characteristics (ROC) curve. According to the machine learning models, 8.13–14.58% of the area in the wetland fringe area, small patches, and edges was under the very high vulnerable wetland habitat status in the pre-dam period. During the post-dam period, the region covered by fringes and small and medium-core wetlands increased to 21.23–50.58%. The PSO-RBF model was found to be the best representative model. This study provides a large database of wetland habitat conditions, which could aid policymakers in developing wetland conservation and restoration plans. [Display omitted] •Sixteen hydrological, surface composition and water quality parameters were prepared.•Water quality parameters were prepared by integrating remote sensing and field data.•SVM, ANN, Bagging, RBF and M5P models have been optimised by PSO for obtaining WHVS.•ROC curve was used to validate the WHVS models.•8.13–14.58% of area in the wetland predicted as very high vulnerable wetland habitat status in the pre-dam period.