Integrated Wetland Classification Using Spectral and Numerical Data for Sustainable Ecosystem Management

Abstract— Wetlands are crucial ecosystems that provide a range of essential services, including water filtration, flood regulation, carbon sequestration, and biodiversity support. The classification of wetland types is key to understanding their distribution, health, and response to disturbances. In the initial step, Random Forest and Gradient Boosting are used to find significant traits and model complex interactions in the numerical data. This allows for good initial classification of wetland kinds. In the second stage, highresolution thermal, SAR, and UAV acquired spectral pictures are used to improve classification. Pixel-wise semantic segmentation is carried out using advanced deep learning techniques like UNet and its variants. The proposed Synergy Net integrates the images and numerical data improves classification accuracy, allowing for the exact identification of wetland categories such as “Marine” and “Riverine” with improved accuracy. Without accurate wetland classification, it becomes increasingly difficult to monitor changes, implement effective conservation strategies, and mitigate the impacts of wetland loss on local and global scales.