Executive Summary

Malley, D.F. and P.D. Martin. 2000. Evaluation of near-infrared spectroscopy as a rapid method for soil analysis. Report to the Manitoba Rural Adaptation Council Inc. March, 46 pp. Unpublished report.

Recent advances in technology are bringing precision agriculture into reality. The use of GPS/GIS and computer-equipped farm machinery permit accurate recording of locations in fields to the nearest metre. Using this capability, weight and moisture of grain can be recorded during harvest and inorganic fertilizers can be applied at variable rates on a fine scale throughout the field to match the crop yield.

Instrumentation that will allow on-the-move measurement of crop quality (protein and oil content) during harvest is in the prototype stage. A missing link in complete precision agriculture capability is the ability to analyze the inherent fertility and properties of soil sufficiently inexpensively that soil maps can be developed that are as detailed as the crop yield maps. Preferably, the soil analytical method should operate on-site and rapidly as well.

Near-infrared spectroscopy (NIRS) is a 30-year old technology that has the capability of rapidly and non-destructively determining quantities of organic constituents in liquids, slurries, and solids. It is used globally for determination of a wide variety of constituents, composition, and functionality in agricultural products, feeds, food, forages, petrochemicals, cosmetics, polymers, waste streams, pharmaceuticals, textiles, and other materials.

The technique is based on measurement of the intensity of the absorption of near-infrared radiation (780 to 2500 nm) by a sample. Commonly, NIRS is used for quantitative measurement of constituents containing organic functional groups, such as covalent bonds O-H, C-H, N-H, C=O, and C-N. The basis of prediction of metals and inorganic constituents is less well known but may depend upon absorption of NIR light by natural organic matter, oxides, hydroxides, carbonates, or clays that bind or adsorb metals.

Near-infrared spectroscopy is not a stand-alone analytical technique. Its ability to provide rapid analyses depends on the prior preparation of mathematical calibrations used to predict constituents in unknown samples. Simply, calibration is the process of "training" the instrument to correctly predict future samples. Calibrations are developed on "training sets" of samples that represent the variability that will be encountered in future samples to be predicted.

The first step in evaluating the usefulness of NIRS for specific analyses is to develop calibrations and evaluate their success according to statistical criteria. This project utilized two sets of soil samples and analytical data from previous studies. The purposes of this project were to:

1. determine whether calibrations could be developed as successfully on field moist soil as on the dried and ground samples that are conventionally used in NIRS
2. determine whether calibrations could be improved by sub-dividing a large sample set into subgroups of samples based on common soil parent material, soil horizon, soil textural class, or soil type.

A small set of 28 samples of Red River/Osborne clay series (0-10 m deep) had been analyzed for P, S, K, Ca, Mg, Na, Fe, Mn, Ba, Co, Cu, Cr, Ni, Sr, Ti, V, and Zn. These were scanned in the field moist state and again following drying and grinding and used to compare calibration success for field moist vs dry samples. NIRS would be most useful in precision agriculture if it can be performed on field moist, i.e., "as is" samples.

Calibrations had previously been developed on the samples in the dry state for all these constituents. For Ca, Mg, Fe, Ni, Sr, and Zn, the values predicted by the NIR calibrations agreed > 95% with the values measured by chemical methods on the samples. For P, Na, K, Mn, Cu, Cr, Ti, and V, the agreement was > 90%. These calibrations are judged as highly successful. Calibrations for Co and Ba were least successful.

For the samples in the field moist state, the calibrations were generally slightly less good, but agreement between NIR-predicted and measured values was ~ 95% for % moisture, Ca, Mg, K, Mn, Cr, Ni, V, and Zn, ~90% for S, Fe, and Ti, and between 80 and 90% for P, Na, Ba, and Cu. These are excellent to good results. Calibration for Sr was not successful in the moist samples.

Close examination of the constituent data showed that most of constituents were highly correlated with one another. In this sample set, it was hypothesized that excellent calibrations could be developed for numerous constituents because of their correlation with spectrally-active carbonate or clay. Calibrations of this type, encompassing a small amount of soil variability and small geographical range, have value for environmental monitoring where the same site is visited periodically over time. They would not be generalizable to other locations. They are also of use in showing that excellent calibrations can be developed on a site-specific basis.

A set of 1000 archived, dried soil samples from the Canada-Manitoba Soil Survey had been analyzed for standard constituents, % sand, % silt, % clay, pH, cation exchange capacity, organic C, N; and for the metals, Fe, Mn, Zn, Cu, Co, Pb, Ni, Cd, Mo, Ag, V, Hg, Se, As and Cr. The samples represented surface agricultural soils over a wide geographical area in southern Manitoba. A previous study demonstrated that a number of constituents and parameters could be predicted by NIRS, including Fe, Mn, Zn, Cu, Co, Pb, Ni, Cd, Mo, V, Cr, Se, pH, % organic C, % N, cation exchange capacity, % sand, % silt, and % clay. The constituents, Ag, As, and Hg were judged not to be predictable by NIRS in soil. Nevertheless, "global" calibrations for most constituents or parameters encompassing all of the soils in agro-Manitoba were not as good as desirable for predicting soil composition in precision agriculture.

By subdividing the soils by parent material, horizons, or textural class it was possible to obtain excellent, good or useful calibrations for organic C, nitrogen, CEC, clay, sand, Fe, Zn, Cu, Ni, Co, V, Cr, and sand. Nevertheless, most of these improved calibrations were for only some of the categories within the subgroups and not for all.

In summary, NIRS shows promise as a rapid, field-portable, on-site or as a cost-effective laboratory soil analytical method. This study identified that it is possible to develop calibrations for some constituents on field moist soil. Depending on the application, the extra steps of drying and grinding may be warranted if high precision is required. Also depending on the application, tailoring calibrations to specific soil parent material categories, soil horizons, or textural classes may provide superior results to those from calibrations encompassing soils from a large geographic range.

Further work is warranted to determine how NIRS can best be used for soil analysis for the benefit of the agricultural community. Moreover, NIRS is expected to provide a unique tool for rapid assessment of organic matter quality in soil.