Population-density estimation using regression and area-to-point residual kriging

Census population data are associated with several analytical and cartographic problems. Regression models using remote-sensing covariates have been examined to estimate urban population density, but the performance may not be satisfactory. This paper describes a kriging-based areal interpolation method, namely area-to-point residual kriging, which can be used to disaggregate the residuals remaining from regression. Compared with conventional cokriging, the area-to-point residual kriging is much simpler in that only a semivariogram model for the point residuals is required, as opposed to a set of auto- and cross-semivariogram models involving the dependent variable and all the covariates. In addition, area-to-point residual kriging explicitly accounts for any scale differences between source data and target values. The method is illustrated by disaggregating population from census units to the land-use zones within them. Comparative results for regression with and without area-to-point residual kriging show that area-to-point residual kriging can substantially improve interpolation accuracy.