Infrared Soil Analysis Laboratory - Adelaide, South Australia

About the ISAL | ISAL strengths | Current Research & Applications | In-field Applications | Staff/Contact

History of MIR at CSIRO | What is mid infrared (MIR) spectroscopy? | Predicting soil properties | Statistical Methods & Analysis | Additional Services | Routine Analyses

ISAL multivariate statistical methods

The ISAL uses a variety of statistical methods, some of which are listed below.

Partial Least-Squares (PLS) | Artificial Neural Networks (ANN) | Principal Components Analysis (PCA)

Partial Least-Squares (PLS) analysis

The organic and mineral components of any given soil will determine many of its chemical properties - such as pH, cation exchange, lime requirement and P-buffer index.

Some of the soil physical properties such as particle size, water-holding capacity and bulk density also depend on the distribution of soil components.  Because the specific infrared signatures of these organic and mineral components are known, it is possible to predict many of the soil properties from single MIR spectra using PLS analysis.

Infrared PLS analysis has been used since the 1980s for the prediction of sample properties from their spectra. Most of the early published work used the near infrared (NIR) spectral region for the analysis of cereal, meat and forages but the method is applicable to the prediction of y-data (properties) from any multidimensional x-data (spectra), including MIR spectra. Advantages of PLS are that it is rapid, can handle co-linear data, is a ‘full-spectrum’ method, and can provide useful qualitative information.

PLS regression is a method with which the infrared spectra are reduced into latent variables, or loadings similar to principal components in PCA analysis, and scores (multipliers). These scores are then regressed with corresponding y-values, i.e. soil properties, and the PLS regressions saved as calibration models. A large number of soil properties, used to derive the calibration models, can then be predicted from the spectra of unknown samples. In principle, most soil calibrations require at least 100 samples per set, but current work suggests that calibration soil set sizes of at least 1,000 are required to account for the large variability within different soil types. Several large soil data sets from South Australia, Western Australia, New South Wales and Queensland have been used to calibrate the facility’s MIR spectrometer.

Artificial Neural Networks (ANN)

Using a combination of PLS loadings and neural networks, PLS-NN, we can combine the robustness and ease of interpretation of PLS with the non-linear capabilities of  neural networks, as shown in Janik, L.J.*, Forrester, S.T. and Rawson, A. “The prediction of soil chemical and physical properties from mid-infrared spectroscopy and combined partial least-squares regression and neural networks (PLS-NN) analysis”, Chemometrics and Intelligent Laboratory Systems, Volume 97, Issue 2, 15 July 2009, Pages 179-188. 

Principal Components Analysis (PCA)

PCA is a data compression process (i.e. a bilinear modelling process), which can be used to reduce a complex multidimensional data (e.g. spectra) into a smaller number of latent variables (principal components, PC) which reflect the underling structure of the original dataset.  The first principal component typically explains most of the variation in the dataset with further principal components being orthogonal to the preceding PC and explaining less variation in the dataset.  By plotting the PC’s in two or three dimensional data space, interrelationships between the samples and variables can be examined.

[Back to Top]

Back to CSIRO Land and Water Testing and Services page