Predictive Soil Mapping to Improve the Physical Basis of Distributed Ecohydrological Models in Arid Environments
Colby W. Brungard, Mikayla J. Allan
Spatial patterns in soil properties such as particle size and soil depth significantly affect hydrological and ecological processes. Finely spatially resolved information about the spatial distribution of soil properties is needed for hydrological and ecohydrological modeling. This information is not currently provided by existing small-scale soil maps. This research uses geostatistical methods to interpolate soil depth as well as sand and clay concentrations at four harmonized depth increments (0-5, 5-15, 15-30, and 30-60 cm) within a single alluvial landform surrounding a small, heavily instrumented watershed at the Jornada Experimental Range in southern New Mexico. Soil depth and sand and clay concentration observations were obtained from two sampling campaigns. Each variable was analyzed for anisotropy and statistically significant relationships with nine terrain variables to account for non-stationarity. Spherical, circular, and exponential variogram models were fitted to all sand and clay concentrations and soil depth and compared using root-mean-square-error (RMSE) derived from leave-one-out cross validation. RMSE ranged between 4.8 and 5.9% for sand and between 1.3 and 1.9% for clay. RMSE for soil depth was 37.7 cm. In general, sand had a shorter range of spatial autocorrelation and a smaller nugget than did clay at all depths. The range of spatial autocorrelation for sand was between 150 and 225 m, while clay had a much more variable range of values between 90 and 3206 m. In general, nugget values were relatively low because of the sampling design that had a minimum distance of 3 m, which appears to have captured most of the small-scale variability. Spatial prediction was done using Kriging with External Drift. Uncertainty in sand and clay concentration predictions were low while the uncertainty of soil depth predictions was greater. Interpolated variables and the associated prediction uncertainty will be used to improve the parameterization of future ecohydrological modeling applications.
Keywords: Soils, ecohydrological models, soil maps, Jornada Experimental Range, spatial patterns