GWC FCS10

The GWC FCS10 (Global 10m Wetland City with Fine Classification System) is the first global dynamic land cover product specifically designed to support the Ramsar Wetland City accreditation and evaluation process. The dataset provides 10m spatial resolution mapping of urban wetland and non-wetland areas, spanning the period from 2016 to 2024, with updates released biennially. The dataset adopts a fine-scale classification system consisting of 16 urban wetland sub-types and 5 non-wetland types. The overall accuracy of the initial classification reaches 94.96 ± 0.5%, while the accuracy of wetland sub-classification achieves 94.69%. Production follows a hierarchical subdivision and year-by-year optimization strategy. First, based on spectral separability among land cover types, Sentinel-1 and Sentinel-2 imagery are integrated within an object-based classification framework to extract nine initial classes: four wetland categories (marsh, swamp, mudflat, and water) and five non-wetland categories (cropland, forest, grassland, impervious surfaces, and bare land). Building upon the four wetland classes, a stepwise refinement procedure is applied—incorporating POI sample distributions, geometric characteristics, coastal vs. inland distinctions, and inundation frequency rules—to further extract wetland sub-types and to delineate the “ocean” class as a distinct non-wetland type. For each period, initial classification is conducted using a random forest classifier enhanced with feature selection and hyperparameter optimization. Results are then progressively refined through a geometry-similarity-based hierarchical inheritance method, enabling consistent temporal and spatial classification.