Method development to extract spatial association structure from soil polygon maps

Abstract

Existing soil information systems contain mainly qualitative data on soilscapes, however, quantitative data would be necessary to more effectively guide digital soil mapping efforts. Detailed analysis of small scale overview maps offers the most appropriate way to delineate soilscapes where they are available. In our study, the genetic soil map of Hungary have been used which displays the most complete representation of the Hungarian Soil Classification System. Our goal was to analyse spatial association structure based on the boundary segments between soil polygons. We transformed the polygons into lines. The features of each line segment were the names (or codes) of the soil polygons on both sides. After omission soils with low representation (less than three polygons) and boundaries beside state border, forests and cities, 69 soil units were retained. We calculated a similarity matrix among soil types based on logarithm of ratios between existing segment lengths and theoretical segment lengths. The theoretical lengths were calculated with a Chi-squared calculation by using sums of lengths in rows and columns in the 69 × 69 matrix. The similarity matrix was converted into dissimilarity matrix to distinguish between complete dissimilarity (missing values) and complete similarity (main diagonal). Dissimilarity matrix was clustered and represented in a form of dendrogram both in original form and after dimension reduction with multidimensional scaling method. Our method has resulted a promising approach for delineating soilscapes in presence of overview soil maps. The study resulted fuzzy soilscapes with broad transition zones. The method should be refined by using variable sized moving window method and by combining boundary data with terrain, geology etc.

Keywords

• soilscape quantification
• genetic soil map of hungary
• boundary segment based chi-squared calculation
• hierarchical clustering
• multidimensional scaling