| Sumario: | The available global or national land use and land cover (LULC) maps are not explicitly focused on wetland management. However, developing a LULC focused on wetlands has become crucial to mitigate the challenges related to water resource management, specifically where anthropogenic stress is high. This study addresses the paucity of tailored LULC classification methodologies for such dynamic peri-urban wetland ecosystems, taking the case of the East Kolkata Wetlands (EKW), a Ramsar site recognized for its community-managed wastewater-fed aquaculture system, threaten by rapid urban expansion and hydrological alteration. The multi-temporal Sentinel-2 MSI imagery (February–May 2025) and a Random Forest (RF) classifier (with an 80:20 training/testing ratio) within Google Earth Engine (GEE) were used. The study developed a hierarchical seven-stage sequential LULC classification significant to wetland management: (1) water vs. non-water; (2) vegetation vs. non-vegetation; (3) built-up vs. fallow land/pond; (4) aquatic vs. terrestrial vegetation; (5) agricultural vs. other vegetation; (6) fallow land vs. fallow pond; and (7) landfill sub-classes (active/closed dump). Various spectral indices related to soil, vegetation, and water quality, with FABDEM data, were used to build the RF model. Thus, the hierarchical framework proposed improves the class separability over conventional RF approaches, producing a detailed LULC map (with an overall accuracy of 89.29%) that distinguishes between aquaculture ponds, marshes, agricultural areas, fallow lands, built-up areas, and landfill sites. These results provide a granular, operationally relevant insights for effective resource management of wetland ecosystems and the sustainable governance of urban wetlands.
|
|---|