We analyzed ground-level ozone variations and exposure hotspots across China (2000–2020) at multiple scales using daily 1 km ozone data derived from satellite-sourced land-surface temperature data via a machine-learning hindcasting model. The dataset was validated using cross-validation and external measurements. A non-monotonic trend with regional and seasonal variations emerged, with turning points around 2008 and 2015. Ozone levels >100 μg m‾³ shifted from June to May, while levels >160 μg m‾³ expanded in the North China Plain.