Numerical Clustering of Soil Series Using Profile Morphological Attributes for Potato

Abstract

Potato fertilization response models have been developed for 46 soil series in the province of Quebec, Canada. This study aimed to create a set of representative soil classes based on morphological data so that they reflect suitable soil properties for growing potato. Data of modal soil profiles of soil series contain morphological attributes from master horizons (including bedrock) with diagnoses indicating the absence (0), weak expression (0.5) or presence (1) of specific properties (pedogenetic features), and particle-size distribution. A distance matrix was calculated to represent the dissimilarity between the soil profiles. Using multidimensional scaling technique, soil profiles distributed in a feature space were clustered using the fuzzy k-means with extragrades algorithm to allow expressing soil groups as continuous variables, hence facilitating modeling. The dissimilarity measure between soil profiles computed using soil descriptions (e.g., color, pH, and C content) at experimental sites showed that genetic horizon indices can be used as a basis to compare and allocate soil profiles to existing classes. In conclusion, numerical clustering provided a quantitative basis to integrate soil profile descriptions into crop response models.