Premium
Visible–Near‐Infrared Spectroscopy Can Predict the Clay/Organic Carbon and Mineral Fines/Organic Carbon Ratios
Author(s) -
Hermansen Cecilie,
Knadel Maria,
Moldrup Per,
Greve Mogens H.,
Gislum René,
Jonge Lis W.
Publication year - 2016
Publication title -
soil science society of america journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.836
H-Index - 168
eISSN - 1435-0661
pISSN - 0361-5995
DOI - 10.2136/sssaj2016.05.0159
Subject(s) - silt , total organic carbon , soil carbon , clay minerals , mineralogy , soil test , carbon fibers , soil texture , chemistry , analytical chemistry (journal) , soil science , environmental chemistry , soil water , materials science , geology , paleontology , composite number , composite material
Core Ideas Visible–near‐infrared spectroscopy is a fast and indirect method for soil analysis. The ratios of clay and (clay + fine silt) to organic C relate to soil functions. We predicted the clay and (clay + fine silt) to organic C ratios successfully. The ratios of mineral fines (<0.02 mm, clay + fine silt) to organic carbon (OC), consisting of the n ‐ratio (i.e., the clay/OC ratio) and m ‐ratio (i.e., the fines/OC ratio) have recently been used to analyze and predict soil functional properties such as tilth conditions, clay dispersibility, degree of preferential flow, water repellency, and chemical adsorption. Conventional texture and OC measurements are time consuming and expensive, and visible–near‐infrared (vis‐NIR) spectroscopy may provide a fast and inexpensive alternative for obtaining the n ‐ and m ‐ratios. In this study, a total of 480 soil samples from seven Danish and one Greenlandic fields, with a large textural range (clay: 0.027–0.355 kg kg −1 ; OC: 0.011–0.084 kg kg −1 ; n ‐ratio: 0.49–16.80; m ‐ratio: 1.46–32.14), were analyzed for texture and OC and subsequently scanned with a vis‐NIR spectrometer from 400 to 2500 nm. The spectral data were correlated to reference values of the n ‐ratio, m ‐ratio, clay, fine silt, fines, and OC with partial least squares regression. The vis‐NIR models were developed on a regional dataset comprising the 480 soil samples divided into calibration and validation subsets. Further, we tested vis‐NIR models developed on the individual eight fields using full cross‐validation. Validation results from the regional models showed high predictive abilities with a root mean square error of prediction (RMSEP) of 0.64 and R 2 of 0.97 for the n ‐ratio and RMSEP = 1.43 and R 2 of 0.97 for the m ‐ratio. The regional clay, fine silt, fines, and OC models also yielded successful predictions ( R 2 = 0.88–0.95). The higher prediction accuracy for the n ‐ and m ‐ratios compared with predictions of basic soil properties was confirmed from the field‐specific models. Our results suggest vis‐NIR spectroscopy as a precise, easily applicable, and fast method for simultaneously obtaining an ensemble of key parameters for soil structure and health.