In traditional spatial statistics, the spatial dependence between two arbitrary sites on a plane is expressed in terms of the Euclidean distance between the two sites. However, in actual cities or regions, infrastructure networks such as railway networks cause spatial interaction and thus strongly affect the spatial dependence. We propose a new spatial prediction method, based on a primal spatial statistical method called Kriging, that explicitly considers such dependence. We first divide the entire plane into Voronoi regions and configure a network node in each region to create a hypothetical network. Then, in order to investigate the spatial interaction enforced by the network, we estimate the shortest-path distance by considering a transportation network and present the covariance function of that distance. Next, we combine the covariance functions based on both the shortest-path distance and the Euclidean distance. Finally, we assess the effectiveness of the proposed method by applying it to a case study of land price prediction.
