A Multi-Index Classification Approach for Wetland Mapping Between 1986-2022 in the Australian Tropical Savanna Using Google Earth Engine

Northern Australia still presents preserved wetland ecosystems but faces escalating threats due to development. To help wetland monitoring, we developed a wetland mapping framework utilizing a multi-index classification approach through Google Earth Engine for tropical savannas in Northern Australia from 1986 to 2022. The methods involve generating annual image mosaics combining Landsat images, spectral indices, and topographic metrics, training samples based on fieldwork information and image interpretation, and random forest classifier. To address the seasonal variations, we used a dynamic threshold segmentation based on the Otsu method. The classification of six main wetland classes demonstrated high accuracy (89%), with results indicating stability in floodplain areas but a reduction in wet areas during the dry season. The framework provides an effective tool for ongoing monitoring, crucial for mitigating the impacts of human activities and climate change on Northern Australia's wetlands, addressing a significant gap in wetland mapping in the Northern Territory.