Using process models to improve spatial analysis

This paper describes a method of improving spatial analyses by using a process model to define the sampling window. This method allows the sample to adapt to changing conditions as they occur in the dataset, rather than applying the same geometric shape to all locations. Such a sampling method can be used to reduce the noise in the sample, and thus generate more sensible results. The general approach may be applied to other processes that influence or control the distribution of spatial variables, provided the processes are known and can be modelled. The method also enables exploration of the degree to which a spatial variable is controlled by an assumed driving process. In this study the sampling windows for each location are defined using uphill and downhill watersheds, and are applied to geochemical variables across a 1 100 km² area in Weipa, Queensland, Australia. The utility of the approach is assessed using variograms and the Getis-Ord G_i^* statistic. Results indicate an improvement over omnidirectional and wedge-shaped sampling, with the most improvement where the variable is highly mobile in solution. These areas are considered to be under modern hydrological control. Most errors in the example are attributed to the effect of other landscape processes, such as aeolian transport and marine incursions.