Optimal Number of Terrain-Based Clusters for Knowledge-Based Inference Digital Soil Mapping.

With recent advancements in technology, digital soil mapping (DSM) has become an alternative for conventional soil surveys due to lower costs and shorter time required for mapping. Soil mapping using the knowledge-base approach is a type of predictive soil mapping that is an efficient and consistent way for predicting soil properties. In the knowledge-based approach, understanding the soil-environment relationship is critical for predicting and developing continuous soil property maps. Calculation of terrain attributes and subsequent clustering of environmental attributes is the first stage in this DSM methodology. To identify repeatable patterns in landscapes it is necessary to numerically relate common terrain attribute values to each other. Selection of the optimum number of clusters has a tremendous effect on the final map products. Most of the time selection of the optimal number of clusters is based on visual evaluation rather than statistical evaluation. In this study, we used principle component analysis (PCA), analysis of variance (ANOVA), and histogram distribution for the determination of the optimal number of terrain attribute clusters. We used the Agronomy Center for Research and Education (ACRE) as a test area. ACRE consists of a 4.6 km² area in Tippecanoe Co., IN and is on a low relief, gently rolling Wisconsinan age till plain. To determine the optimal number of clusters, we ran three different iterations of unsupervised classification in SAGA GIS software using the topographic wetness index (TWI) and topographic position index (TPI) as inputs and either 3, 5, or 10 clusters as outputs. Chemical and physical data from 176 georeferenced point observations were then assigned to the appropriate cluster.  All three of the statistical methods (PCA, ANOVA, and histogram distribution) showed that 5 clusters best capture soil variation based on topographic differences within the study area.