Obtaining population estimates in non-census reporting zones: An evaluation of the 3-class dasymetric method

Interpolating population data between incompatible spatial zones is an important task in many GIS applications. This paper investigates whether regional regression models between population and land cover outperform a global approach, and whether the 3-class dasymetric method improves upon the binary dasymetric approach. In the experiments conducted, regional regressions resulted in better areal interpolation, but also highlighted spatial non-stationarity in the relationship between population and land cover. The benefits of a 3-class dasymetric model over a binary model were inconclusive. However, it is suggested that greater flexibility in model calibration to more fully incorporate spatial non-stationarity could improve 3-class dasymetric performance. Accurate urban residential density mapping is also important since the 3-class dasymetric method seems less robust than the binary approach to land cover classification error.