Global gridded population projection via spatial econometric model considering rank-size rule

Abstract Global gridded population (GGP) projections that are consistent with shared socioeconomic pathways are widely used. However, existing approaches for generating future GGPs have several limitations, including difficulties in incorporating covariates, a lack of consideration of city-level interactions, and insufficient integration of the empirical rule observed in the distribution of city sizes across years. This study attempts to address this gap by introducing a multiscale GGP projection model consisting of city- and grid-level models. The city-level model comprises a spatial econometric model that models city growth and decline, alongside a rank-size model that introduces the empirical rules as macro-level constraints. A random forest was employed for the grid-level model. For validation, the model was trained on Japanese Grid Square Statistics (approximately 1 km2), based on population census data from 2000 to 2010, and its forecast accuracy was evaluated against 2020 data. The results show a root mean squared percentage error of 1.17% between the observed and predicted values. Moreover, to assess the model’s applicability for projecting GGPs in the distant future, the 2070 forecast was compared with that from a standard cohort component model. The results indicate that the spatial patterns produced by the two models are similar, with an RMSPE of 1.89%. This RMSPE value was better than that of the existing representative studies.