Dynamic surface water fraction (DSWF): Global surface water fraction mapping at 10-meter spatial resolution with Sentinel-2 imagery in Google Earth Engine

• Sentinel-2-based global dynamic surface water fraction (DSWF) mapping is proposed. • DSWF uses a self-trained regression model with Dynamic World as ancillary data. • DSWF is calculated on-demand in response to areas of interest and time in GEE. • Mean RMSE of DSWF is 0.090 through global validation. Mapping surface water at high spatiotemporal resolution is critical for managing water resources and mitigating disasters. Current global surface water datasets are typically generated at a relatively coarse monthly temporal resolution and 30-meter spatial resolution. Moreover, the mixed pixel problem further limits their ability to precisely map small water bodies. This study produced the global Dynamic Surface Water Fraction (DSWF) mapping by combining Sentinel-2 imagery and Dynamic World dataset using Google Earth Engine (GEE). Different from the analysis-ready wall-to-wall global datasets, DSWF is generated on-demand online in response to user-defined areas of interest and time. DSWF explored sub-pixel surface water fraction information at 10-meter spatial resolution, enabling the precise representation of fine-scale spatial features of surface water and minimizing the blurring artifacts commonly associated with conventional hard classification. The accuracy of DSWF was evaluated across 113 validation tiles, demonstrating an overall root mean squared error (RMSE) of 0.090 and mean absolute error (MAE) of 0.021 when validated against both pure and mixed pixels. In comparison to the 30-meter Landsat-based global datasets, DSWF provides more accurate spatial distributions of surface water, with particular effectiveness for small ponds and narrow rivers.