Mapping wetland habitat health in moribund deltaic India using machine learning and deep learning algorithms

Researchers have increasingly integrated machine learning (ML) and deep learning (DL) algorithms to forecast the risk, vulnerability, and susceptibility of various geo-environmental challenges. However, to the best of our knowledge, there is a dearth of studies that have employed DL to predict the health status of wetland habitats, and none have explored a comparative analysis between ML and DL models in this context. This study aims to fill this gap by focusing on the development of wetland habitat health status using both ML and DL models, seeking to determine whether DL models exhibit superior predictability compared to ML models. The assessment of wetland habitat health status reveals that smaller fringe wetlands situated away from main rivers tend to be identified as poor habitats. The transition from phase II to III is marked by a substantial reduction in wetland area, decreasing from 438.76 km² to 235.68 km² across different habitat zones, underscoring the significant loss of wetland areas. The observed 43–46 % decline in very poor and poor habitat areas from phase II to III lends credibility to the predictive capabilities of the models. Notably, among the applied ML and DL models, XGB from the ML category and DNB from the DL category have demonstrated superior performance. In all instances, DL models outperformed ML models, suggesting that deep learning algorithms hold promise for evaluating wetland habitat health status. The mapping and modelling of wetland habitat health status at a spatial scale are pivotal for formulating effective wetland management strategies. The identification of areas with poor and good habitat health provides valuable information for prioritized planning and targeted wetland restoration efforts.