High-resolution continuous soil classification using morphological soil profile descriptions

Soil grid data were gathered from 156 points in the 30-ha Muizen forest (Ranst, Belgium). At each grid point, soil profiles were examined morphologically by augering to 120-cm depth. In the laboratory, pH(KCl) was determined on samples from every horizon. To allow numerical analyses, all the morphological attributes were given ordinal scores. The analysis consisted of two parts. First, the master horizons were split up into subtypes using Principal Components Analysis and a non-hierarchical clustering technique. This was necessary to overcome the problem of the anisotropy of the soil profiles, which makes it impossible to pool the data of all the horizons and analyse them together. Next, the distinguished horizon subtypes were used as input for the continuous soil profile classification with the ‘fuzzy k-means with extragrades’ algorithm. Five different soil classes plus an extragrade class were distinguished. The distinguished soil classes exhibited a fair degree of spatial autocorrelation and correlated well with the Belgian Soil Map. The technique developed ensures the compatibility with national or global soil classification systems based on diagnostic horizons and properties on the one hand and the production of high-resolution soil classes for local use on the other. Furthermore, the developed technique allows reanalysis and optimisation of data from previous surveys.