Spatial data clustering using neighbourhood analysis

Abstract. Several different spatial clustering algorithms have been implemented to group geospatial sensor-based measurements of soil attributes into a set of relatively homogeneous management zones. Although they allow multidimensional data analysis, complexity and frequently occurring discontinuities of so called “management zones” make this technology less appealing to the potential user. With the neighbourhood analysis approach presented in this paper, the primary goal was to delineate field areas of any shape that stand out from the rest of the field in terms of a measured soil attribute. To illustrate this approach, popular apparent soil electrical conductivity (EC a ) data have been used to delineate continuous areas of the field that unify measurements with the greatest deviation from the average field conditions. The algorithm is based on cyclic evaluation of the mean squared error when grouping field locations according to the set of neighbourhood constraints.