Files
Abstract
Visible/near-infrared scans on a range of soil series under longleaf pine (Pinus palustris) were calibrated with soil total carbon using different methods to assess measurement accuracy. Soil data for 900 samples were provided by the Natural Resources Conservation Service. Spectra were pre-processed using three methods; Savitzky-Golay (SG), continuum removal, and wavelets. Two multivariate algorithms, the partial least square regression and support vector machine (SVM), were used to determine the best calibration model based on the coefficient of determination (R2). The SVM algorithm combined with the SG transformation provided the best calibration and validation prediction at a R2= 0.70 for mineral soils. Adjoining soil map units that vary in slope steepness were also analyzed for that vary in slope steepness with soil property patterns at four different sites in GA. Few significant differences were observed with slope steepness at any depth (0-200 cm) for the measured variables (percent clay, C, pH). VNIR calibrations for percent clay demonstrated potential predictive value (i.e., R2 0.9) while those for C and pH cacl2, although not as strong (i.e., R2 0.67 and R2 0.54, respectively), indicated some utility for field classification or monitoring of dynamic soil properties under longleaf pine ecosystems.