Comparing the effects of an NLCD-derived dasymetric refinement on estimation accuracies for multiple areal interpolation methods

Comparability among population data enumerated within different time periods may be complicated by changing enumeration boundaries over time. Areal interpolation methods are commonly used to solve such zoning incompatibilities, but are frequently based on the questionable assumption of homogeneous population density within the different zones. To achieve more accurate estimates, land cover or other ancillary data may be used to better characterize the underlying source zone population density surface prior to areal interpolation. Although dasymetric techniques such as these are well documented, their effectiveness across different areal interpolation methods are not well established. This research compares the accuracy of a number of areal interpolation methods used to support temporal analysis of population data, and evaluates the effect of dasymetric mapping on interpolation accuracy. Our findings demonstrate that dasymetric refinement noticeably improves interpolation accuracy for the areal weighting, pycnophylactic, and target density weighting (TDW) methods of areal interpolation. A fourth method in which land cover densities are inherently incorporated, the expectation–maximization algorithm (EM), performs equally well. Our results show that the dasymetrically refined TDW method outperforms other areal interpolation methods in most instances, but suggest that the EM algorithm may be preferred as the interval between enumeration periods grows large.