Premium
Complete Soil Texture is Accurately Predicted by Visible Near‐Infrared Spectroscopy
Author(s) -
Hermansen Cecilie,
Knadel Maria,
Moldrup Per,
Greve Mogens H.,
Karup Dan,
Jonge Lis W.
Publication year - 2017
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/sssaj2017.02.0066
Subject(s) - soil texture , silt , particle size distribution , soil science , texture (cosmology) , soil water , diffuse reflectance infrared fourier transform , particle size , near infrared spectroscopy , particle (ecology) , environmental science , soil test , mineralogy , materials science , geology , chemistry , physics , optics , computer science , artificial intelligence , image (mathematics) , paleontology , biochemistry , oceanography , photocatalysis , catalysis
Core Ideas Two PSC models are fitted to detailed measurements of clay, silt, and sand fractions. Both models well describe the PSCs of a broad soil data base. Within and between field variations in PSC and OM are well predicted by vis‐NIRS. The Fredlund model performs slightly better in data‐fitting and vis‐NIRS predicted PSCs. New vis‐NIRS concept enables soil type classification in any texture system worldwide. The particle‐size curve (PSC) defines the continuous size distribution of mineral particles <2 mm. It is used for soil classification and to derive functional soil parameters such as the soil–water characteristic (SWC) curve, soil hydraulic properties, and gas transport properties. Conventional methods for measuring texture are time‐consuming and most methods only provide discrete particle‐size intervals. The Rosin–Rammler and Fredlund functions enable a continuous description of the size distribution of mineral particles using two and three fitting parameters, respectively. Visible near‐infrared diffuse reflectance spectroscopy (vis‐NIRS) is a time‐saving and well‐known alternative soil analysis method. In this study vis‐NIRS was used to indirectly obtain PSCs by predicting the fitting parameters of the Rosin–Rammler (α R , β R ) and Fredlund ( α F , n F , and m F ) functions. A total of 431 soil samples from 7 agricultural fields in Denmark and Greenland were analyzed for soil texture (clay: 0.028–0.426 kg kg ‐1 ) and organic matter (OM) content (0.018–0.143 kg kg ‐1 ). The Rosin–Rammler and Fredlund functions were fitted to the PSCs. Soil diffuse reflectance was measured from 400 to 2500 nm with a spectrometer. The important spectral regions for correlating α R , β R , α F , n F , m F , and OM to spectra were selected using forward interval partial least squares (iPLS) regression on a calibration set. The soil spectra showed high correlation to PSC function parameters and OM content for the validation set. Further, vis‐NIRS cross‐validation models for the fitting parameters of the Rosin–Rammler and Fredlund functions were built on all samples and used as input for the PSCs, generating RMSE values of 4.2 and 3.5%, respectively. Both PSC functions convincingly covered the PSC variation within fields, although the Fredlund function performed slightly better. From one vis‐NIRS scanning the complete texture comprising the PSC and the OM content was successfully characterized.