Predicting Soil Carbon and Nitrogen Concentrations and Pasture Root Densities from Proximally Sensed Soil Spectral Reflectance

Abstract

A modified soil probe for a portable spectroradiometer (ASD FieldSpecPro, Boulder, CO) was developed to acquire reflectance spectra (350–2,500 nm) from flat-sectioned horizontal (H method) soil surfaces of soil cores or from the vertical side (V method) of cylindrical soil cores. The spectra have been used to successfully predict soil carbon (C) and nitrogen (N) concentrations and root density. Partial least squares regression (PLSR) of the first derivative of the 5 nm space spectral data from method H against laboratory determined soil C and N concentrations produced calibrations that allowed quantitative estimates of C and N concentrations in unknown Pumice, Allophanic, and Tephric Recent soil samples (for C: R 2 validation = 0.76, RPD = 1.97; for N: R 2 validation = 0.84, RPD = 2.45). Compared to the H method, spectra acquired by the V method gave slightly more accurate predictions of soil C and N concentrations in Fluvial Recent soil (for C: R 2cross-validation (cv) = 0.95 and 0.97, RPD = 4.45 and 5.80; for N: R 2cross-validation = 0.94 and 0.96, RPD = 4.25 and 5.17, where the two values are for the H and V methods, respectively). Spectra acquired by the V method from drier soils in May produced a calibration against soil C and N concentrations that was capable of accurately predicting the soil C and N concentrations from spectra collected from wetter soils in November (C: R 2 validation = 0.97 and RPD = 3.43; for N: R 2 validation = 0.95 and RPD = 3.44). This indicates that a calibration dataset can have temporal robustness, which may reduce the number of calibrations that have to be performed. The root density predictions from spectra acquired by the H method were more accurate if soil types were separated into Allophanic soil (RPD = 2.42; R 2 cross-validation = 0.83) and Fluvial Recent soil (RPD = 1.99; R 2 cross-validation = 0.75).