Improving the Spatial Prediction of Sand Content in Forest Soils Using a Multivariate Geostatistical Analysis of LiDAR and Hyperspectral Data

Soil sand particles play a crucial role in soil erosion because they are more susceptible to being de-tached and transported by erosive forces than silt and clay particles. Therefore, in soil erosion as-sessment and mitigation, it is crucial to model and predict soil sand particles at unsampled locations using appropriate methods. The study was aimed to evaluate the ability of a multivariate approach based on non-stationary geostatistics to merge LiDAR and visible-near infrared (Vis-NIR) diffuse reflectance data with laboratory analyses to produce high-resolution maps of soil sand content. Remotely sensed, high-resolution LiDAR-derived topographic attributes can be used as auxiliary variables to estimate soil textural particle-size fractions. The proposed approach was compared with the commonly used univariate approach of ordinary kriging to evaluate the contribution of auxiliary variables. Soil samples (0-0.20 m depth) were collected at 135 locations within a 139 ha forest catchment with granitic parent material and subordinately alluvial deposits, where soils classified as Typic Xerumbrepts and Ultic Haploxeralf crop out. A number of linear trend models coupled with different auxiliary variables were compared. The best model for predicting sand content was the one with elevation derived from LIDAR data as the only auxiliary variable. Although the improvement in estimation over the univariate model was rather marginal, the proposed approach proved very flexible and scalable to include any type of auxiliary variable. The application of LiDAR data is ex-pected to expand as it allows the high-resolution prediction of soil properties, generally insufficiently sampled, at different spatial scales.