Comparing predictive ability of laser-induced breakdown spectroscopy to visible near-infrared spectroscopy for soil property determination. (April 2017)

• Record Type: Journal Article
• Authors: Knadel, Maria; Gislum, René; Hermansen, Cecilie; Peng, Yi; Moldrup, Per; de Jonge, Lis W.; Greve, Mogens H.
• Is Part Of: Biosystems engineering. Volume 156(2017)
• Page Start: 157
• Page End: 172
• Publication Date: 2017-04
• Subjects: LIBS -- vis–NIRS -- Soil organic carbon -- Soil particle size fractions; LIBS Laser-Induced Breakdown Spectroscopy -- LV Latent Variable -- MIR Mid-infrared -- NIRS Near-Infrared Spectroscopy -- PLS Partial Least Squares Regression -- RMSECV Root Mean Square Error of Cross-validation -- RMSEP Root Mean Square Error of Prediction -- RPIQ Ratio of Performance to Inter-quartile range -- SOC Soil Organic Carbon (%) -- VIS visible; Bioengineering -- Periodicals; Agricultural engineering -- Periodicals; Biological systems -- Periodicals; Génie rural -- Périodiques; Systèmes biologiques -- Périodiques; 631
• DOI: 10.1016/j.biosystemseng.2017.01.007
• Languages: English
• ISSNs: 1537-5110

Abstract

Abstract : Soil organic carbon (SOC) and particle size fractions have a practical value for agronomy and the environment. Thus, alternative techniques to replace the expensive conventional analyses of soil are needed. Visible near-infrared reflectance spectroscopy (vis–NIRS) has already shown potential for becoming an alternative method for soil analysis since it is faster and cheaper than conventional methods. Laser-induced breakdown spectroscopy (LIBS) is another cost-effective technique with potential for rapid analysis of elements present in the soil. In this study, the feasibility of using LIBS to determine SOC, clay, silt and sand contents of Danish agricultural soils was tested and compared with the vis–NIRS method. First, country-scale Partial Least Squares (PLS) regression models on soils collected across Denmark ( N = 78) were built and validated using independent field samples ( N = 54). Secondly, the country-scale calibration data set was spiked with 14 representative samples from the fields and validated with the 54 field samples. Generated country-scale LIBS models exhibited similar and not significantly different (p > 0.05) results to vis–NIRS for all soil properties except a significantly higher (p = 0.0305) predictive ability for sand. Spiking improved the accuracy of most of the LIBS and vis–NIRS models, indicating the importance of similarities between the calibration and the validation data sets. No significant differences (p > 0.05) were found between the LIBS and vis–NIRS spiked country-scale models. Lower prediction errors for most properties were obtained using LIBS, rendering it an equally good or even a more accurate technique for soil properties determination than the well-established vis–NIRS method. Highlights: The feasibility of using LIBS for soil properties determination was tested. The predictive abilities of LIBS and vis–NIRS were further compared. Generally lower prediction errors were obtained using LIBS than vis–NIRS. Few significant differences in predictive abilities of LIBS and vis–NIRS were found.