Intelligent estimation of spatially distributed soil physical properties

Spatial analysis of soil samples is often times not possible when measurements are limited in number or clustered. To obviate potential problems, we propose a new approach based on the self-organizing map (SOM) technique. This approach exploits underlying nonlinear relation of the steady-state geomorphic concave–convex nature of hillslopes (from hilltop to bottom of the valley) to spatially limited soil textural data. The topographic features are extracted from Shuttle Radar Topographic Mission elevation data; whereas soil textural (clay, silt, and sand) and hydraulic data were collected in 29 spatially random locations (50 to 75cm depth). In contrast to traditional principal component analysis, the SOM identifies relations among relief features, such as, slope, horizontal curvature and vertical curvature. Stochastic cross-validation indicates that the SOM is unbiased and provides a way to measure the magnitude of prediction uncertainty for all variables. The SOM cross-component plots of the soil texture reveals higher clay proportions at concave areas with convergent hydrological flux and lower proportions for convex areas with divergent flux. The sand ratio has an opposite pattern with higher values near the ridge and lower values near the valley. Silt has a trend similar to sand, although less pronounced. The relation between soil texture and concave–convex hillslope features reveals that subsurface weathering and transport is an important process that changed from loss-to-gain at the rectilinear hillslope point. These results illustrate that the SOM can be used to capture and predict nonlinear hillslope relations among relief, soil texture, and hydraulic conductivity data. ► We model soil texture and hydraulic conductivity. ► Nonlinear relations between soil texture and hillslope features are quantified. ► The self-organizing map identifies weathering relations among relief features. ► Component plots reveal higher clay content at concave areas and lower at convex areas.