ELUCIDATION OF SOIL LANDFORM INTERRELATIONSHIPS BY CANONICAL ORDINATION ANALYSIS

This paper examines the robustness of various multivariate ordination methods applied to elucidate the relationship between soil developed on hillslope and the environment in a South Australian subcatchment. Canonical correspondence analyses were found to be less attractive than the linear methods (i.e., principal component analysis and redundancy analysis) because interrelations among soil variables and/or between soil and the landform attributes are more linear than unimodal. Among the linear methods, standard redundancy analysis following transformation of soil data gave a clearer insight into pedogenetic implications of soil-landform interactions. Gradient angle, plan convexity, profile convexity and to a lesser extent, upslope distance and area, account for much of the soil variation in the study area. Gradient influences the distribution of surficial soil particles which is suggestive of selective removal and deposition of fine materials on the mid-to lower-slopes. Higher subsoil clay percentage associated with concave slopes may be ascribed to either increased weathering rate and/or flux of fine material due to convergence of water flows. Solum depth and depth to bedrock also appeared to have controlled some of the contemporary pedo-geomorphological processes. Eluvio-illuvial processes are more pronounced in the relatively deep lower-slope pedons evidenced by cutans in their subsoil layers. Soil colours are related to the soil depth attributes, slope and upslope distance and area, probably through the influence of these attributes on soil drainage condition. The extent to which landform and parent-material related attributes (e.g., depth to bedrock) affect soil development and distribution patterns in the study area is shown quantitatively. These findings have considerably improved our understanding of soil genesis in the study area and also point to the importance of land unit delineation to design sampling patterns. The enhanced sampling design would be beneficial in terms of reduced extrapolation error and thus misclassification of soil.