Soil-Landscape Modeling of Coastal California Hillslopes Using Terrestrial Lidar.

Digital elevation models (DEMs) are the dominant input to spatially explicit digital soil mapping (DSM) efforts due to their increasing availability and the tight coupling between topography and soil variability. Accurate characterization of this coupling is dependent on DEM spatial resolution and soil sampling density, both of which may limit analyses. For example, DEM resolution may be too coarse to accurately reflect scale-dependent soil properties yet downscaling introduces artifactual uncertainty unrelated to deterministic or stochastic soil processes. We tackle these limitations through a DSM effort that couples moderately high density soil sampling with a very fine scale terrestrial lidar dataset (20 cm) implemented in a semiarid rolling hillslope domain where terrain variables change rapidly but smoothly over short distances. Our guiding hypothesis is that in this diffusion-dominated landscape, soil thickness is readily predicted by continuous terrain attributes coupled with catenary hillslope segmentation. We choose soil thickness as our keystone dependent variable for its geomorphic and hydrologic significance, and its tendency to be a primary input to synthetic ecosystem models. In defining catenary hillslope position we adapt a logical rule-set approach that parses common terrain derivatives of curvature and specific catchment area into discrete landform elements (LE). Variograms and curvature-area plots are used to distill domain-scale terrain thresholds from short range order noise characteristic of very fine-scale spatial data. The revealed spatial thresholds are used to condition LE rule-set inputs, rendering a catenary LE map that leverages the robustness of fine-scale terrain data to create a generalized interpretation of soil geomorphic domains. Preliminary regressions show that continuous terrain variables alone (curvature, specific catchment area) only partially explain soil thickness, and only in a subset of soils. For example, curvature explains 40% of soil thickness variance at <300 cm at scales up to 20 m, while soil thickness >300 cm shows no clear relation to curvature. Further efforts will be aimed at refining the regression model by integrating the spatially-constrained, generalized LE map classes, as well as simulating DEM error for uncertainty analysis.