This study introduced an enhanced space–time extremely randomized trees (STET) approach to improve the 1 km resolution ground-level PM_2.5 estimates across China using the remote sensing technology. The STET model shows high accuracy and strong predictive power and appears to outperform most models reported by previous studies. Thus, it is of great importance for future air pollution studies at medium- or small-scale areas and will be applied to generate the historical PM_2.5 dataset across China.