Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
2024527 | Soil Biology and Biochemistry | 2015 | 10 Pages |
•Three infrared software packages reached similar estimation accuracies for soil chemical properties.•The treatment of outliers affected the results markedly.•Microbial biomass C and N were estimated using laboratory data of the chemical properties, infrared data were not required.
Middle infrared (MIR) spectroscopy (MIRS) has been used to characterize soil chemical and biological properties and to date, the accuracy for estimation of these properties has been variable. Objectives were (i) to study the usefulness of MIRS for the estimation of a series of chemical and biological properties for three different sets of soil samples using OPUS Quant 2 software, WinISI software (developed for vis-near infrared spectra) and ParLeS software and (ii) to analyse possible predictive mechanisms for the biological properties statistically. A data set of chemical and biological properties of 422 samples from three arable soils in Germany of different long-term experiments was used, and the reflectance spectra in the MIR region were recorded after shock-freezing followed by freeze-drying. Cross-validation was carried out for the entire soil sample set and for each soil sample set from the three sites. Accuracies of estimation for the chemical and biological properties depended markedly on the sample set. All three software packages tested reached similar accuracies (despite differences in the mathematical treatments), whereby WinISI and OPUS slightly outperformed ParLeS. Additionally, the treatment of outliers affected the results markedly and the coefficients of determination increased with an increasing removal of outliers. Multiple linear regressions indicated that, at least for microbial biomass C (Cmic) and N (Nmic), an estimation using pH and the contents of C, N, P, S, sand and clay was as accurate (entire set, Cmic: r2 = 0.77, Nmic: r2 = 0.72) as an estimation using MIR data (entire set, Cmic: r2 = 0.78, Nmic: r2 = 0.73). The general idea behind infrared spectroscopy is that only a part of a sample set needs to be investigated by laboratory methods and that after appropriate calibration then the determination of spectra may be sufficient for the remaining part of the sample set. This benefit of MIRS does obviously not apply for predicting soil biological properties in sample sets when multiple linear regressions using the contents of chemical and physical soil properties give estimates of similar accuracies (in the calibration and cross-validation, respectively). However, MIRS provides an effective tool to estimate spectrally active chemical and physical properties.