A method combining FTIR-ATR and Raman spectroscopy to determine soil organic matter: Improvement of prediction accuracy using competitive adaptive reweighted sampling (CARS)

• Published in: Computers and electronics in agriculture Vol. 191; p. 106549
• Main Authors: Xing, Zhe, Du, Changwen, Shen, Yazhen, Ma, Fei, Zhou, Jianmin
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.12.2021; Elsevier BV
• Subjects: Adaptive sampling; agriculture; Algorithms; CARS algorithm; Chemical analysis; Data integration; electronics; Fourier transforms; FTIR-ATR spectroscopy; Infrared reflection; Infrared spectroscopy; Least squares method; Organic matter; Performance prediction; prediction; Prediction models; Raman spectra; Raman spectroscopy; Soil fertility; Soil improvement; soil organic matter; Soil properties; SOM prediction; Spectrum analysis; Variable selection; CARS algorithm; FTIR-ATR spectroscopy; Raman spectroscopy; SOM prediction; Variable selection
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2021.106549

[Display omitted] •Combination of Raman and ATR spectroscopies to predict the content of SOM.•Reduction of RMSEP of prediction model with the application of CARS algorithm.•Improved performance of models based on fused spectra with selected variables. Determination of soil organic matter (SOM) is extremely important for diagnosing the fertility status of agricultural soils. Thus, fast and efficient approaches are needed to aid soil fertilization assessment. In this work, the method proposed is based on the combination of mid-infrared attenuated total reflection (FTIR-ATR) and dispersive Raman spectroscopy, as a rapid and nondestructive alternative to traditional chemical analysis. The ability of both two individual and the fused spectroscopy in SOM prediction was tested. Partial least squares regression (PLSR) was used to construct predictive models to correlate soil spectra with SOM content. Simple data fusion was accomplished by concatenating the principal components of the two spectra. The predictive performance was not essentially improved, and even decreased for the fused ATR-Raman spectra based on the simple fusion strategy. Better results were obtained with the advanced method of data fusion, which is, concatenating the selected variables of the two spectroscopic techniques after the step of variable selection by competitive adaptive reweighted sampling (CARS). The results showed that the RMSEP of the prediction model was decreased using both the individual and fused spectra data, combining with the CARS algorithm. Models based on fused spectra data with selected variables had the best performance in accuracy of SOM prediction. Therefore, the fused technology of ATR and Raman spectroscopy is a promising approach to predict soil properties, such as SOM, with the advantage of simple preparation of soil samples.