Estimating Soil Texture from a Limited Region of the Visible/Near-Infrared Spectrum

摘要

Soil particle size is an attribute of fundamental importance when defining soil horizons. Proximal soil sensors can facilitate the acquisition of a larger amount of soil data using a faster and less laborious technique. Thus, the objective of this study is to evaluate the capacity of a limited spectral acquisition region (325-1075 nm) for estimating soil texture. Soil samples were collected in the southwest part of Marombas river watershed located near the center of Santa Catarina State, south of Brazil. A total of 42 soil profiles were sampled according to the GlobalSoilMap specification. A dataset of 166 samples was used for model calibration and another set of 71 samples was used for model validation. Diffuse reflectance spectroscopy of sieved samples (2 mm) was collected with a spectrometer FieldSpecHandHeld II (ASD Inc.). Savitzky-Golay second derivatives were calculated and used in partial least-squares regression modeling. Calibration and validation datasets showed statistically similar meanand variance. The root-mean-square error of prediction for sand, silt, and clay content is 5.47, 5.18, and 5.39 g 100 g-1, respectively. The R2 for validation is 0.30, 0.59, and 0.69 for the same attributes. Partitioning the model by depth did not improve the predictions significantly. The results show that estimating soil texture from a limited spectral region is promising and can contribute toward the development of cheaper spectrometers or infrared cameras that can be used for digital soil morphometries.