A Geographically Weighed Regression-based Approach for the Change of Support Problem

Differences in spatial supports (shape, size, etc.) among spatial data often complicate spatial data analyses, and spatial support conversions, such as aggregation, disaggregation, and interpolation, are often applied to cope with them. However, in general, spatial support conversions present the following problems: (i) how to convert data accurately? and (ii) how to reduce biases caused by the conversions? These problems have been discussed in the field of geography. On the other hand, recently, they have also been discussed in geostatistics under the framework of the change of support problem (COSP). Sophisticated methods for COSP, or for problems (i) and (ii), have been proposed in geostatistics. However, the discussion of COSP is still only within geostatistics. Thus, this study extends the geographically weighted regression (GWR), a geographical method for capturing spatial heterogeneity, to a method that addresses COSP. The constructed method can be used for both (i) spatial unit (support) conversion and (ii) reducing the biases due to aggregation (conversion). In addition, it gives local trend parameters in arbitrary spatial units. The effectiveness of the method is verified using a simulation study.