LAUSR

Abstract Knowledge of the spatial distribution of populations at finer spatial scales is of significant value and fundamental to many applications such as environmental change, urbanization, regional planning, public health, and disaster management. However, detailed assessment of the population distribution data of countries that have large populations (such as China) and significant variation in distribution requires improved data processing methods and spatialization models. This paper described the construction of a novel population spatialization method by combining land use/cover data and night-light data. Based on the analysis of data characteristics, the method used partial correlation analysis and geographically weighted regression to improve the distribution accuracy and reduce regional errors. China's census data for the years 1990, 2000, and 2010 were assessed. The results showed that the method was better at population spatialization than methods that use only night-light data or land use/cover data and global linear regression. Evaluation of overall accuracies revealed that the coefficient of correlation R-square was >0.90 and increased by >0.13 in the years 1990, 2000, and 2010. Moreover, the local R-square of over 90% of the samples (counties) was higher than the adjusted R-square of the general linear regression model. Furthermore, the gridded population density datasets obtained by this method can be used to analyse spatial-temporal patterns of population density and provide population distribution information with increased accuracy and precision compared to conventional models.