Nonstationarity in regression-based spatial interpolation models

The existence of nonstationarity, or spatial variability in geographical relationships, is a topic that has received some attention in the geographical literature in recent years. Its effect in regression-based spatial interpolation methods, however, remains an open research question. In order to explore this question, the paper describes a general regression model which can be used to derive a number of standard interpolation methods, and also a new approach capable of accommodating geographical nonstationarity. The approach proposed, it will be seen, is a synthesis of concepts derived from the methods of kriging and geographically weighted regression (GWR). Application to a case study suggests that potentially useful performance improvements may result from using the new approach.