Plausibility test of conceptual soil maps using relief parameters

The motivation for this article results from the fact that conceptual soil maps show oftentimes inaccuracies with regard to soil unit boundaries or misfits between original paper and actual soil-related information. Using the example of a German conceptual soil map (CSM), we introduce a procedure which could be considered as a framework for testing the terrain-related plausibility applied within a genetic based soil-ordering system. Framework means that all tests and the underlying methods can be adapted to specific targets. The procedure enables both reproducible integration of expert knowledge and application of statistically sound methods. M of the German Federal State of Saxony-Anhalt was tested regarding the plausibility of colluvial and fluvial process domains. The plausibility test consists of four steps and was exemplified on a study area of 100 km2. First, basic relief parameters were combined to the explaining relief parameters Floodplain Index (FPI) and Mass Balance Index (MBI) enabling a classification of process domains by relative descriptions. Second, relief parameters and aggregated CSM soil units were integrated to soil-terrain objects (STO) executing a region-growing segmentation algorithm. In the third step, the one-dimensional MBI or FPI feature space of STO entities were clustered by using the K-means algorithm. The fourth step comprises the expert-based selection of reference clusters (RC) representing colluvial and fluvial process domains accepted as being true. Then, empirical cumulative distribution functions (ECDF) of RC and remaining soil unit-related STO clusters were compared by a traditional goodness-of-fit test whose suitability for estimation of terrain-related CSM plausibility is shown. Finally, the resulting ECDF distances were visualized. sting procedure could also be used for the supervised selection of appropriate samples for automatic classification algorithms. The data integration approach is generally suitable for adopting existing data in computer-based systems.