Mapping urban wetlands at 10-m resolution using Sentinel-1/2 on GEE: an object-based hierarchical classification framework for International Wetland Cities

The highly fragmented and complex typology of urban wetlands poses substantial challenges for achieving high spatial resolution and detailed mapping using remote sensing. This study employs Sentinel-1/2 imagery and the Google Earth Engine (GEE) platform to develop an object-based hierarchical wetland classification (OHWC) framework supported by high-quality training samples, producing 10-m wetland distribution maps for five periods from 2016 to 2024 in Changshu (China), Changnyeong (Republic of Korea), and Saint Omer (France). Spatiotemporal patterns of wetland dynamics were analyzed. The results show that (1) average overall accuracies for Changshu, Changnyeong, and Saint Omer reached 94.84%, 90.9%, and 90.7%, respectively, outperforming existing wetland datasets; (2) wetlands in Changshu first expanded then declined, with a net loss of 20.42 km², whereas those in Changnyeong grew continuously, with a net gain of 9.58 km², and those in Saint Omer showed a fluctuating upward trend, with a net increase of 1.04 km²; and (3) changes in inland wetlands were the main drivers of wetland pattern evolution across the three cities. The proposed framework demonstrates strong generalizability and provides a valuable reference for detailed urban wetland mapping, as well as scientific support for wetland restoration and implementation of the Sustainable Development Goals (SDGs).