A local polycategorical approach to areal interpolation. (November 2015)

• Record Type: Journal Article
• Authors: Lin, Jie; Cromley, Robert G.
• Is Part Of: Computers, environment and urban systems. Volume 54(2015)
• Page Start: 23
• Page End: 31
• Publication Date: 2015-11
• Subjects: Areal interpolation -- Local statistical -- Geographically weighted regression -- Quantile regression; City planning -- Data processing -- Periodicals; Regional planning -- Data processing -- Periodicals; 303.4834
• DOI: 10.1016/j.compenvurbsys.2015.05.007
• Languages: English
• ISSNs: 0198-9715

Abstract

Highlights: Novel local polycategorical method was proposed to solve areal interpolation problem. Geographic and statistical proximity filtering windows were presented. The proposed method provided comparable results to other local methods. Abstract: Areal interpolation is a technique used to transfer attribute information from source zones with known values to target zones with unknown values. This paper presents and describes a new polycategorical method that integrates positive aspects of both geographically weighted regression (GWR)-based and quantile regression (QR)-based interpolators for solving areal interpolation problems. Two different types of neighborhoods for selecting observations used to estimate ancillary control densities are presented: one that is spatially based and one that is statistically based. The new polycategorical methods are evaluated against a number of existing methods – areal weighting, pycnophylactic, binary dasymetric, intelligent dasymetric mapping, and GWR using test data from the 2010 census population, the National Land Cover Database 2006 (NLCD2006) and the Topologically Integrated Geographic Encoding and Reference (TIGER) line graph files. The evaluations include several overall error measurement indices as well as maps of the spatial distribution of the error associated with selected methods. Results suggest that with appropriate land cover categories and neighborhoods, the new polycategorical methods provide comparable results to local regression models but with much less computation complexity.