Aerosols play an important role in climate change, and ground aerosols (e.g., fine particulate matter, abbreviated as PM2.5) are associated with a variety of health problems. Due to clouds and high reflectance conditions, satellite-derived aerosol optical depth (AOD) products usually have large percentages of missing values (e.g., on average greater than 60% for mainland China), which limits their applicability. In this study, we generated grid maps of high-resolution, daily complete AOD and ground aerosol coefficients for the large study area of mainland China from 2015 to 2018. Based on the AOD retrieved using the recent Multi-Angle Implementation of Atmospheric Correction advanced algorithm, we added a geographic zoning factor to account for variability in meteorology, and developed an adaptive method based on the improved full residual deep network (with attention layers) to impute extensively missing AOD in the whole study area consistently and reliably. Furthermore, we generated high-resolution grid maps of complete AOD and ground aerosol coefficients. Overall, compared with the original residual model, in the independent test of 20% samples, our daily models achieved an average test R2 of 0.90 (an improvement of approximately 5%) with a range of 0.75–0.97 (average test root mean square error: 0.075). This high test performance shows the validity of AOD imputation. In the evaluation using the ground AOD data from six Aerosol Robotic Network monitoring stations, our method obtained an R2 of 0.78, which further illustrated the reliability of the dataset. In addition, ground aerosol coefficients were generated to provide an improved correlation with PM2.5. With the complete AOD data and ground coefficients, we presented and interpreted their spatiotemporal variations in mainland China. This study has important implications for using satellite-derived AOD to estimate aerosol air pollutants.