Accurate information on the spatial and temporal dynamics of surface water is essential for monitoring freshwater ecosystems. However, existing satellite-based studies are often limited by coarse spatio-temporal resolution or constraints in their underlying classification algorithms. Here, we evaluate the Dynamic World land cover model for mapping inland water across globally diverse conditions using Sentinel-2 imagery. Cross-validation demonstrates that Dynamic World enables accurate (F1=0.90±0.04) and computationally efficient water mapping, with performance further improved by combining it with the Normalized Difference Water Index (NDWI; F1=0.91±0.04), particularly for smaller water bodies and challenging conditions such as terrain shadows. Building on these results, we propose a computationally efficient and scalable workflow for mapping surface water dynamics from full annual Sentinel-2 time series. Our approach derives monthly and annual Water Occurrence Frequency (WOF) maps through automated processing on Google Earth Engine with advanced cloud and terrain-shadow filtering. Applied to the Rhône-Mediterranean region in France — a hydro-morphologically complex mountainous region — the WOF product captures 21 % more stable water surface and 71 % more river network length than Landsat-based datasets. It better represents ephemeral water bodies and smaller streams, while significantly increasing precision over DSWE, the only other global 10-meter WOF product, which relies solely on the DW Top-1 label. We also assessed global applicability, revealing that cloud cover substantially reduces valid observations for several months in tropical and mountainous regions, impacting the information density in WOF maps. The enhanced spatio-temporal detail of the method unlocks opportunities for characterizing inland water dynamics, including river geomorphology, floodplain connectivity, and wetland ecosystem functioning. The workflow is publicly available for use in other studies.