Geostatistical Approach to Spatial Data Transformation

Spatial data transformation is one of the most fundamental functions in geographic information system technology. Despite its popularity and significance, the uncertainty associated with data transformation and the potential errors involved with it draw less attention. In this article, we offer a geostatistical approach as a means of conducting spatial data transformation, which enables analysts to explicitly take into account support of the data (source) and the outputs (target). We also discuss a few means of evaluating and quantifying the quality of the outcomes. We present a brief review of linear geostatistics and variant kriging methods, followed by a case study in which a realization of the spatial process of interest is available through stochastic simulation. To mimic real-world situations, we created two data sets with different footprints (supports)-point and areal data-from the reference point values. We conducted kriging to reproduce the true surface based on the sample data derived from the reference values and the theoretical variogram parameters. Also, we assessed the differences between the predicted values and reference point values using two criteria-the reproduction of spatial structure and the coherence property. We found that data transformation tends to alter the variability of attributes and spatial continuity (spatial dependence). For example, we showed that both point kriging methods, using either point or areal data, underestimate the variability of attributes but overestimate the spatial contiguity in the empirical variogram analysis. We also showed in the case study how much errors are committed in the outputs by oversimplifying the support of source data.