Using relief parameters in a discriminant analysis to stratify geological areas with different spatial variability of soil properties

The spatial variation of soil properties within landscapes is controlled by the soil forming factors relief, parent material, climate, organisms, and time. Although this relation is a paradigm in soil survey, it is rarely considered in the analysis of spatial variability of soil properties. Homogeneous soil units are mostly mapped according to the soil properties found close to the soil surface. Nevertheless, a large heterogeneity may occur at greater depths for soils developed in two different geological strata. This is often the case in former periglacial areas where pleistocene sediments cover older strata. The aim of this paper is to show how discriminant analysis can be used to objectively determine the soil depth at which geology changes. This information may then help to develop better soil surveys and improve the geostatistical regionalization of soil properties. The study area is a 1.5 km² soilscape 50 km north of Munich with large variability in relief and parent material. At 450 nodes of a rectangular 50 x 50 m grid, fundamental soil properties were measured for each soil horizon. Relief parameters were calculated using a Digital Elevation Model (DEM) derived from more than 4000 elevation measurements. Parent material of the soils may be sediments of the tertiary period (TS) or quaternary sediments (QS). Altogether, 86 soil horizons were classified with confidence to TS and 496 to QS. They were the training data set for a discriminant analysis to distinguish horizons of TS from horizons of QS. In addition to several relief parameters, the horizon's depth position within the soils was used as a discriminant variable. The discriminant functions classified 87% of TS and 85% of QS training data set horizons correctly by using elevation above sea level, depth, slope and upslope watershed area as independent variables. Solving the discriminant functions with respect to the boundary depth between QS and TS and applying the result to the DEM yielded a map of boundary depth for each location of the study area. At 19 locations within the study area, the predictions were validated with an independent data set. The root of the mean squared differences between the measurement and the prediction was 7.5 cm for this second data set. This is within the uncertainty of measurement. Boundary depth was finally used to devide the study area into separate areas of the two strata dependent on the depth of interest within the soil. This allowed separate variogram calculations for each stratum and depth. The resulting variograms for soil texture showed a larger spatial variability for stratum TS than for QS. Consequently, four times as many locations must be measured for stratum TS than for QS to obtain the same precision of spatial interpolation.