Evaluating the effects of rapid urbanization on the encroachment of the east Kolkata Wetland ecosystem: a remote sensing and hybrid machine learning approach

The East Kolkata Wetland (EKW) is a Ramsar wetland near the Kolkata megacity. EKW is a vital wetland ecosystem that provides several direct and indirect ecosystem services to the Kolkata metropolitan area. The rapid expansion of urban development near the EKW is exerting significant stress on the EKW, resulting in the fastest rate of wetland degradation in recent decades. A practical method for assessing the danger of wetland conversion is essential for conserving this ecosystem. This study aims to analyze the potential impact of EKW change utilizing four advanced data-driven machine learning (ML) models: random forest (RF), support vector machine (SVM), artificial neural network (ANN), and gradient boosting machine (GBM), to monitor the dynamics of change in wetland ecological systems employing Landsat TM and OLI data from 2006–2010, 2011–2015, and 2016–2022. The research innovation applies ML models to water quality, urban heat islands, groundwater potential, and storm surge vulnerabilities, which have not been previously investigated in wetland conversion risk assessment. The benefit of using four ML models lies in their ability to surpass the constraints of assumption-driven traditional approaches for assessing wetland threats related to land surface temperature (LST) in the context of urban expansion. Five components have been identified, categorized as the bare soil index (BI), normalised difference water index (NDWI), normalised difference vegetation index (NDVI), built-up index (BU), and LST. The findings of the ML model demonstrate that around 50–70% of the region has undergone significant changes owing to the increase in urban land use and land cover (LULC). A comparison of these four approaches is also conducted to determine the most accurate prediction technique for this research field. Relative importance (RI) is used to verify the models’ outcomes statistically, and all of these approaches indicate a more accurate prediction model for this research with an average accuracy of 83% and RMSE of 2.95. Four ML models effectively represent the geographical pattern of encroachment and decrease of EKW caused by urban expansion. Results from RF, SVM, ANN, and GBM models, along with related maps, can help land-use planners and policy analysts pinpoint where LULC and LST changes are occurring in urban areas, evaluate the success of programs to protect wetlands, draft legislation to halt the deterioration of wetlands and implement sustainable practices for the ecosystem's long-term sustainability.