Exposure to ambient PM_2.5 (fine particulate matter) can cause adverse effects on human health. China has been experiencing dramatic changes in air pollution over the past two decades. Statistically deriving ground-level PM_2.5 from satellite aerosol optical depth (AOD) has been an emerging attempt to provide such PM_2.5 data for environmental monitoring and PM_2.5-related epidemiologic study. However, current countrywide datasets in China have generally lower accuracies with lower spatiotemporal resolutions because surface PM_2.5 level was rarely recorded in historical years (i.e., preceding 2013). This study aimed to reconstruct daily ambient PM_2.5 concentrations from 2000 to 2018 over China at a fine scale of 1 km using advanced satellite datasets and ground measurements. Taking advantage of the newly released Multi-Angle Implementation of Atmospheric Correction (MAIAC) 1-km AOD dataset, we developed a novel statistical strategy by establishing an advanced spatiotemporal model relying on adaptive model structures with linear and non-linear predictors. The estimates in historical years were validated against surface observations using a strict leave-one-year-out cross-validation (CV) technique. The overall daily leave-one-year-out CV R² and root-mean-square-deviation values were 0.59 and 27.18 μg/m³, respectively. The resultant monthly (R² = 0.74) and yearly (0.77) mean predictions were highly consistent with surface measurements. The national PM_2.5 levels experienced a rapid increase in 2001–2007 and significantly declined between 2013 and 2018. Most of the discernable decreasing trends occurred in eastern and southern areas, while air quality in western China changed slightly in the recent two decades. Our model can deliver reliable historical PM_2.5 estimates in China at a finer spatiotemporal resolution than previous approaches, which could advance epidemiologic studies on the health impacts of both short- and long-term exposure to PM_2.5 at both a large and a fine scale in China.