Mapping and Monitoring Seasonal Wetland Dynamics of Koonthankulam Ramsar Site Using Sentinel-2 Imagery and Random Forest Classification on Google Earth Engine

Abstract

Wetlands provide indispensable ecological, hydrological and socio-economic services, yet remain among the most rapidly altered ecosystems worldwide due to encroaching agriculture, urbanisation and hydrological disruption. This study presents a cloud-based, multi-temporal assessment of Koonthankulam, a Ramsar-designated wetland and Important Bird and Biodiversity Area in Tirunelveli district, Tamil Nadu, India, using Sentinel-2 Multispectral Instrument imagery processed on the Google Earth Engine (GEE) platform. Cloud-free composites were generated for the summer and monsoon seasons of three reference years (2017, 2020 and 2023) within a 1 km buffer drawn outward from the officially notified Ramsar boundary, and a supervised Random Forest classifier was used to delineate four land use/land cover (LULC) classes: water, vegetation, barren land and settlement (built-up structures were subsumed within the settlement class). In parallel, the Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI) and Normalized Difference Water Index (NDWI) were computed to characterise seasonal and interannual dynamics of vegetation vigour, built-up expansion and surface water extent. The results reveal a pronounced seasonal contrast, with vegetated cover consistently higher in the summer composites (peaking near 25 sq. km in 2017) than in the monsoon composites, reflecting the dominance of irrigated paddy cultivation around the tank during the dry months. Settlement area expanded steadily across the study period, nearly quadrupling between 2017 and 2023 (an average annual increase of approximately 24%), while barren land cover remained comparatively stable. Surface water extent fluctuated between approximately 8 and 15 sq. km across seasons and years, sustained in part by canal releases from the Manimuthar and Nanguneri systems during January and February. These findings demonstrate the utility of freely available Sentinel-2 data and cloud-based machine learning for low-cost, repeatable monitoring of small Ramsar wetlands, and provide an evidence base for habitat management and conservation planning at Koonthankulam.