Joint use of the Sentinel-1–derived Kennaugh parameters and Sentinel-2 data for temporal landcover dynamics over the Indian Sundarbans region

Rapid changes in the surrounding areas of the Indian Sundarbans substantially influence local habitats and the larger ecosystem, impacting biodiversity, water quality, and environmental stability. Obtaining real-time ground truth data might be challenging due to minimal human involvement. Therefore, evaluating the temporal dynamics in landcover is crucial in these regions. We use the Sentinel-1–derived Kennaugh matrix elements and Sentinel-2 data to classify 13 land cover types over this region. To monitor seasonal and inter-annual fluctuations, we focused on the pre-monsoon (April) and post-monsoon (October) times throughout 6 years from 2018 to 2023. Three machine learning algorithms, extreme gradient boosting (XGB), random forest, and light gradient boosting machine, are utilized for classification purposes. With an overall classification accuracy of 98% by combining optical bands with Kennaugh components, XGB outperformed the other methods in precision. In contrast, individual features resulted in an accuracy range of only 50% to 90%. This approach offers a practical solution for understanding wetland dynamics without ground truth data, making it highly adaptable and scalable for wetland monitoring.