Farm-scale soil patterns derived from automated terrain classification

Landform elements (LFEs) are commonly used in soil science to demark pedological boundaries and as a first indication of soil spatial variability. A novel LFE classification system known as geomorphons, has been shown to be able to overcome limitations of other automated LFE classifiers. The pattern recognition algorithm classifies the 10 most common LFEs, is computationally efficient, and is robust to changes in scale. However, due to their novelty, research into geomorphons has been limited. This study aimed to stratify the soil landscape through an aggregated geomorphon at the farm-scale (1:25 000) in the Western Cape, South Africa (33.25 degrees S and 18.20 degrees E). Twenty-four geomorphons were created at different resolutions and their association with soil classes were compared. The best fitting geomorphon was aggregated into a 5-unit system corresponding to the South African national resource inventory. The aggregation was based on a decision tree corresponding to soil type. The 5-unit system was evaluated on how well the system stratified soil associations, soil lightness, soil electrical conductivity (EC), soil organic carbon, effective rooting depth (ERD), depth to lithology, gravel, sand, silt, and clay. The prediction potential was compared between the original geomorphon, the aggregated geomorphon, and a manually delineated LFE system. It was found that the aggregated geomorphon stratified all soil attributes except EC. Additionally, the aggregated geomorphon predicted 6 out of 9 soil properties with the greatest accuracy (RMSE). This study shows that aggregating geomorphons can stratify the soil landscape even at the farm-scale and can be used as an initial indication of the soil spatial variability. This has implications in resource poor areas where an additional soil survey is not feasible or can be used to aid in the disaggregation of existing soil-terrain datasets.