U.S. Environmental Protection Agency


Date of this Version



Geoderma 230–231 (2014) 29–40


This article is a U.S. government work, and is not subject to copyright in the United States.


Quantifying the spatial distribution of soil properties is essential for ecological and environmental modeling at the landscape scale. Terrain attributes are among the primary covariates in soil-landscape models due to their control on energy and mass fluxes,which in turn control the spatial distribution of soil properties and processes. While numerous studies have demonstrated the importance of terrain attributes for predicting landscape-scale soil variability, considerable uncertainty exists as to the scale-dependency of light detection and ranging (LiDAR) derived terrain attributes on the accuracy of soil-landscape model predictions. Thirty five pedons were sampled by genetic horizon in a 2300 ha forested watershed and three soil properties (clay, sum of bases, and total carbon), representing dominant pedogenic processes within the watershed were analyzed. Soil properties were used as dependent variables and terrain attributes, calculated from LiDAR derived DEMs of various grid resolutions (ranging from 5 to 50 m) and neighborhood extents (ranging from 15 to 350 m), were used as predictor variables in ordinary least-squares (OLS) regression models. Results from this study show that model predictions exhibit a strong scale-dependency, with percent clay, sum of bases, and total carbon having the highest R2-adj and lowest root mean square error (RMSE) at coarse neighborhood extents (i.e., 150 to 300 m) both between soil variables and across soil depths. Furthermore, in certain instances grid resolution was also shown to affect soil–terrain correlations, although to a lesser degree than neighborhood extent. In many cases fine to moderate scale grid resolutions (i.e.,b30 m) more accurately represented terrain features, resulting in higher correlations to soil properties at fixed neighborhood extents relative to course grid resolutions. Additionally, these results show that fine scale topographic information (i.e., 1 to 5 m) does not necessarily provide a stronger predictor of soil spatial variability relative to moderate scale information. This study provides a robust framework for investigating pedogeomorphological processes on a landscape scale through examination of the scale dependency of modeled terrain attributes in quantitative soil-landscape modeling.