Spatially explicit population data is critical to investigating human-nature interactions, identifying at-risk populations, and informing sustainable management and policy decisions. Existing long-term global population data are generally limited in three key ways: 1) they were estimated with simple scaling or trend extrapolation methods which are not able to capture detailed population variation spatially and temporally; 2) the rate of urbanization and the spatial patterns of settlement changes were not fully considered in their population redistribution; and 3) the spatial resolution is generally coarse. To address these limitations, we proposed a framework for large-scale spatially explicit downscaling of populations from census data and projection of future population distributions under different Shared Socio-economic Pathways (SSP) scenarios with distinctive changes in urban extent patterns. We separated the urban and rural population numbers before downscaling and included the spatial dynamic changes in urban extent during future projection. Treating urban and rural populations as distinct but interconnected entities, we constructed a random forest model to downscale historical populations and designed a gravity-based population potential model to project future population changes at the grid level. This work unleashes a new capacity for getting accurate spatially explicit demographic change with an unprecedented combination of temporal, spatial, and SSP scenario dimensions, paving the way for cross-disciplinary studies on long-term socio-environmental interactions.