Using Visible and Near Infrared Diffuse Reflectance Spectroscopy to Characterize and Classify Soil Profiles

Visible and near infrared diffuse reflectance spectroscopy (VisNIR-DRS) is a method being investigated for quantifying soil properties and mapping soil profiles. Because a VisNIR-DRS system mounted in a soil penetrometer is now commercially available for scanning soil profiles in situ, methodologies for using scans to map soils and quantify soil properties are needed. The overall goal of this research is to investigate methodologies for collecting and analyzing VisNIR-DRS scans of intact soil profiles to identify soil series. Methodologies tested include scanning at variable versus uniform moistures, using individual versus averaged spectra, boosting an intact spectral library with local samples, and comparing quantitative and categorical classifications of soil series. Thirty-two soil cores from two fields, representing three soil series, were extracted and scanned every 2.5 cm from the soil surface to 1.5 m or to the depth of parent material at variable field moist conditions and at uniform moist condition. Laboratory analyses for clay, sand, and silt were performed on each horizon. Soil series were classified using partial least squares regression (PLS) and linear discriminant analysis (LDA). A Central Texas intact spectral library (n=70 intact cores) was used for PLS modeling, alone and boosted with the two fields. Because whole-field independent validation was used, relative percent difference (RPD) values were used to compare model performance. Wetting soils to uniform moisture prior to scanning improved prediction accuracy of total clay and RPD improved by 53 percent. Averaging side-by-side scans of the same soil profile improved prediction accuracy of RPD by 10 percent. When creating calibration models, boosting a library with local samples improved prediction accuracy of clay content by 80 and 34 percent for the two fields. Principal component plots provided insight on the spectral similarities between these datasets. Overall, using PLS alone performed the same as LDA at predicting soil series. Most importantly, results of this project reiterate the importance of fully-independent calibration and validation for assessing the true potential of VisNIR-DRS. Using VisNIR-DRS is an effective way for in situ characterization and classification of soil properties.