William J. Foley á Allen McIlwee á Ivan LawlerLem Aragones á Andrew P. Woolnough á Nils Berding

Ecological applications of near infrared re¯ectance spectroscopy ±a tool for rapid, cost-effective prediction of the composition of plantand animal tissues and aspects of animal performance

Received: 10 October 1997 /Accepted: 12 May 1998

Abstract Many ecological studies rely heavily onchemical analysis of plant and animal tissues. Often,there is limited time and money to perform all the re-quired analyses and this can result in less than idealsampling schemes and poor levels of replication. Nearinfrared re¯ectance spectroscopy (NIRS) can relievethese constraints because it can provide quick, non-de-structive and quantitative analyses of an enormousrange of organic constituents of plant and animal tis-sues. Near infrared spectra depend on the number andtype of CAH, NAH and OAH bonds in the materialbeing analyzed. The spectral features are then combinedwith reliable compositional or functional analyses of thematerial in a predictive statistical model. This model isthen used to predict the composition of new or unknownsamples. NIRS can be used to analyze some speci®c el-ements (indirectly ± e.g., N as protein) or well-de®nedcompounds (e.g., starch) or more complex, poorly de-®ned attributes of substances (e.g., ®ber, animal foodintake) have also been successfully modeled with NIRStechnology. The accuracy and precision of the referencevalues for the calibration data set in part determines thequality of the predictions made by NIRS. However,NIRS analyses are often more precise than standardlaboratory assays. The use of NIRS is not restricted tothe simple determination of quantities of known com-

pounds, but can also be used to discriminate betweencomplex mixtures and to identify important compoundsa�ecting attributes of interest. Near infrared re¯ectancespectroscopy is widely accepted for compositional andfunctional analyses in agriculture and manufacturingbut its utility has not yet been recognized by the majorityof ecologists conducting similar analyses. This paperaims to stimulate interest in NIRS and to illustrate someof the enormous variety of uses to which it can be put.We emphasize that care must be taken in the calibrationstage to prevent propagation of poor analytical workthrough NIRS, but, used properly, NIRS o�ers ecolo-gists enormous analytical power.

Key words Near infrared re¯ectance spectroscopy áLeaf chemistry á Plant-herbivore interactions áFood intake á Decomposition

Introduction

The analysis of nutrient concentrations in samples ofplant and animal tissues is an integral part of manyecological studies. These analyses (e.g., total nitrogen,carbohydrates and lipids) are frequently so time-con-suming and expensive that sampling strategies adoptedare less than ideal. For example, variations of nutrientswithin individuals of a plant species often may be animportant component of animal foraging that is ignoredin many studies simply because the extra analytical workcannot be accommodated.

Another typical problem for the analyst involves thelarge amount of sample required for some analyses;these may be di�cult to obtain (e.g., from seedlingplants). Alternatively, the analyst may have to bulkreplicate samples to get su�cient material for all neces-sary analyses and again compromise an ideal experi-mental design.

Often ecologists are forced to use a range of indirectmeasures in place of a more suitable direct measure of

Oecologia (1998) 116:293±305 Ó Springer-Verlag 1998

W.J. Foley (&) á I. LawlerDivision of Botany and Zoology,Australia National University, Canberra 0200, Australia(e-mail: william.foley@anu.edu.au, Fax: +61-26-2495573)

A. McIlwee á A.P. WoolnoughDepartment of Zoology and Tropical Ecology,James Cook University, Townsville 4811, Australia

L. AragonesDepartment of Tropical Environment Studies,James Cook University, Townsville 4811, Australia

N. BerdingBureau of Sugar Experiment Stations, P.O. Box 122,Gordonvale 4865, Australia

the attribute of interest. This situation has often arisenin studies of the foraging behavior of folivorous pri-mates and marsupials (Waterman et al. 1980). Typically,investigators measure the concentrations of a diverserange of elements and attributes such as ``®ber'' whenwhat is actually needed is an estimate of relative foodintake by the animal (Poppi 1996; A. McIlwee, I. Lawlerand W.J. Foley unpublished work).

Near infrared re¯ectance spectroscopy (NIRS) o�ersecologists enormous ¯exibility in meeting these sorts ofchallenges. The analysis of multiple constituents of planttissues can be made rapidly in a single operation usingNIRS technology (up to 150 samples can be processed ina day by a single operator) and with minimal samplepreparation (Marten et al. 1989). The use of NIRS islow-cost (savings of at least 80% of normal laboratorycosts are easily achievable; Aragones 1997) and amena-ble to small sample sizes (multiple analyses can be ob-tained on as little as 0.2 g of material; Aragones 1997).Finally, NIRS analysis is non-destructive and producesno chemical wastes, which should be an importantconsideration for all ecologists.

Use of NIRS has been adopted enthusiastically bymany agricultural and manufacturing industries. Thoseconcerned with evaluating and improving the nutritivevalue of feedstu�s for livestock are some of the biggestusers (see Shenk and Westerhaus 1994). Some NIRSmethods have been established for more than 20 yearsand have achieved ``o�cial'' status (e.g., AOAC 1990).Much of the early use of NIRS centered on exploitingthe technique as a time and cost-e�ective method ofnutrient analysis. However, in recent years, there havebeen many innovative uses of NIRS and it is becomingan increasingly accepted tool for routine analysis inmany research areas.

Many of the analytical techniques used in ecology arebased on those used in agricultural industries, but NIRShas been little used in ecological studies (but see Brookset al. 1984; Jo�re et al. 1992; Malley et al. 1996; Nilssonet al. 1996 for notable exceptions). NIRS o�ers anexcellent opportunity for ecologists to escape the con-straints of laboratory analyses on the design of theirinvestigations and to apply many laboratory-basedmeasurements to free-ranging animals.

In this article we summarize some recent studies usingNIRS in ways that we believe will be useful to ecologists.Our aim is to stimulate interest amongst ecologists inNIRS, to bring together a scattered literature, and towiden the applications of the methodology. We are lessconcerned with describing detailed methodologies forapplying the technique to di�erent materials that mightbe of interest to experienced users of the technique.In our own work (nutritional ecology of herbivores),we have found that NIRS has allowed us to focus ourresearch on landscape-scale questions and to applylaboratory ®ndings to the ®eld. Perforce, many of theexamples we cite are related to our own interests, but westress that our belief is that the potential applications ofNIRS are enormous.

Basis of the technique

The composition of plant or animal tissues is ultimatelyre¯ected in the types of bonds between the atoms orgroups of atoms (functional groups) that make up thosetissues and information about these functional groupscan be sought through many di�erent forms of spec-troscopy. When a sample of organic material is irradi-ated, the bonds continually vibrate, which causesstretching and bending. This in turn causes a type ofwave motion within the bond at a frequency that ischaracteristic of the functional group. The frequencies ofthe incident light that match the frequencies of the vi-brational waves are absorbed whereas other frequenciesare re¯ected or transmitted. The process is the same asthat which allows us to see colored objects that interceptwhite light (Van Kempen and Jackson 1996).

Near-infrared radiation (750±2500 nm) (Fig. 1) isabsorbed mainly by CAH, NAH and OAH bonds (Os-borne et al. 1993) which are the primary constituents ofthe organic compounds of plant and animal tissues. Thechemical constituents of the tissue determine the natureand number of bonds present and therefore the wave-lengths and amount of light that is absorbed. Therefore,the spectrum of light that is re¯ected from the samplecontains detail on the chemical composition of thatmaterial (Shenk et al. 1992; Shenk andWesterhaus 1994).

Fig. 1 The electromagnetic spec-trum showing the position ofnear-IR

294

However, the peaks in an NIR spectrum are notdistinct or sharp because they consist of overtones andcombinations from primary absorptions in the mid-infrared region and also because some of the light isscattered (Shenk and Westerhaus 1993b). Consequently,there are few if any regions of the NIR spectrum whereabsorbance can be due to only one type of functionalgroup and direct interpretation of the spectral absor-bances is very di�cult for complex mixtures (Fig. 2).Given the limited number of functional groups presentin plant and animal tissues (due to limited numbers ofmetabolic pathways) an analyst could assume that ab-sorbance in the spectral region characteristic of theamide bond of a protein is, in fact, largely due to theprotein in the tissues. However, one could never becertain about this (see Fig. 2) and so NIRS relies onapplying statistical tests to that assumption. In otherwords, the analyst derives a statistical model that teststhe intensity of the relationship between a particularabsorbance and an independent laboratory assay ofprotein content in a range of di�erent plant or animaltissues. Since a di�erent group of plant or animal tissuemight contain di�erent functional groups that also ab-sorb in the spectral region of interest, the soundness ofthe relationship between spectral absorption and labo-ratory assays should be assessed for each type of tissuesof interest. Near infrared analysis is thus an indirect orsecondary method that estimates chemical compositionby comparing spectra with samples of known composi-tion (Shenk et al. 1992; Shenk and Westerhaus 1994).This procedure is known as calibration.

In practice, the analyst develops a multivariate sta-tistical model to describe the relationship between theNIR spectral absorbances and the chemical componentsor characteristic of interest (Shenk and Westerhaus1993b). The statistical model is then used to predict thecomposition of unknown samples that are part of thesame population. Samples which fall outside the popu-lation can be analyzed by traditional means and iterat-ively included in a new model (Shenk et al. 1992; Shenk

and Westerhaus 1994). These procedures are discussedin more detail below.

Mechanics of NIR spectroscopyand sample presentation ± importance of particle size

A NIR spectrophotometer consists of a light source, ameans of selecting particular wavelengths (NIR spec-troscopists traditionally use wavelengths (nm) ratherthan wave-numbers as favoured by those working in themid IR) within the spectrum, a detector for collectingthe re¯ected radiation and a signal and data processingcomputer (Fig. 3). The sample (which in most currentapplications is dried and ground ± but NIRS can be usedwith any other sample form ± see section below onanalysis of fresh forages), is usually packed manuallyinto a sample cup which has an optical grade quartzglass cover on one side. Care is required to ensure thatthe sample is spread evenly and packed to a consistentdegree of compression. The sample cups can be smallring cups of between 20 and 50 mm diameter or alter-natively larger cells that hold up to 50 g of coarsermaterial. These are then placed in the spectrometer andthe re¯ectance spectra collected. In some applications,®ber optics are used to bring the light to a remotelypresented sample and the information is either con-ducted back along the optic cable to the detector andmicroprocessors, or captured by remotely placed detec-tors and conveyed electronically to the microprocessor.

Particle size of the sample has a major e�ect on theNIR re¯ectance spectrum (Casler and Shenk 1985;Windham 1987). Increasing particle size results in anincreased apparent path length for the incident light andso the measure of re¯ectance also increases (Fig. 4).Consistent particle size between samples is thereforeimportant and analysts using NIRS techniques withdried, ground samples need to be aware of the perfor-mance of their laboratory grinders (Shenk and West-erhaus 1993b). The ®neness of the particles is lessimportant than the distribution of sizes and manygrinders simply cannot produce a uniform particle size.Depending on the nature of the samples being ground,analysts may have to regularly replace impellers andother grinding parts more frequently than expected toensure that this optimum is maintained.

Calibration procedures

The spectroscopic and chemometric principles used indeveloping calibration models are beyond the scope ofthis short review. Nonetheless it is appropriate to pro-vide a broad outline of approaches which are mosttypically taken by analysts so that readers can appreciatethe strengths and limitations of the whole technique.Those interested in speci®c procedures should consultsome of the many excellent references on this subject(e.g., Shenk and Westerhaus 1991a,b, 1993; Smith and

Fig. 2 Broad and overlapping peaks attributable to di�erent constit-uents are characteristic of a near infrared re¯ectance spectrum(adapted from Osborne et al. 1993; Givens et al. 1997)

295

Flinn 1991; Osborne et al. 1993; Baker et al. 1994).Figure 5 summarizes the most common steps taken todevelop calibrations for plant and animal tissues (Shenkand Westerhaus 1993b).

The essence of any calibration procedure is to ensurethat the range of spectral variation found in the wholepopulation is represented in the samples selected foranalysis for calibration development. Generally, theentire population is ranked in terms of distance from theaverage spectrum. Prior to ranking, a variety of math-ematical treatments are applied to the spectra to em-phasize only those data that are relevant. Particle sizeand structure of a sample can cause spectral di�erencesthat are unrelated to chemical composition. Thesetransformations usually include scatter correction thatmay use standard normal variate procedures andmathematical derivative transformations.

Once the population has been structured in this way,samples with either extreme spectra or those that havevery similar spectra are eliminated so that those re-maining represent a de®ned degree of spectral variation.Various algorithms have been developed for doing this.The CENTER and SELECT algorithms of Shenk andWesterhaus (1991a) have proved among the most pop-

ular for forage scientists but a variety of other methodshave been developed (Shenk and Westerhaus 1991b).

In theory selection of samples that cover the range ofspectral variation in the data set should be su�cient butwe believe that most ecologists would be more com-fortable if the calibration set also covered the range oftaxonomic variation or included all types of treatmentsapplied in ®eld experiments. For example, in our workon herbivore foraging in tropical savannas, a number ofcommon grass species were not included in the originalcalibration set chosen by the SELECT algorithm. Con-sequently, we expanded the set by selecting samples

Fig. 3 Components of mono-chromator-based near-IR spec-trometer

Fig. 4 The e�ect of particle size on spectral absorbance in the near-IRregion (adapted from Shenk et al. 1992)

Fig. 5 Alternative strategies for the development of calibrationequations needed for the routine analysis of unknown plant oranimal samples

296

from our collection of those particular plant species(A.P. Woolnough and W.J. Foley, unpublished work).

Once su�cient samples have been selected, the tra-ditional compositional analyses are performed on thissubset. These may include measures such as concentra-tion of nitrogen or protein, carbohydrates, lipids or evenindividual fatty acids and amino acids. However, as wedescribe in more detail below, other more integrativeattributes may be measured such as the digestibility andintake of food by herbivores, grain yield during milling(Welsh et al. 1996), susceptibility to insect attack, orpulp yield in the paper industry. Given the importanceof these assays or measures, it is vital that the analyst becon®dent of the degree of accuracy and precision ofthese analyses (see Conventional laboratory assays).

The actual model is then constructed by developing aregression equation between the spectral absorbancesand the traditional laboratory analyses (Fig. 6) (Shenkand Westerhaus 1991c, 1993a, b). This involves any of anumber of multivariate regression procedures includingmultiple linear regression (MLR), principal componentsregression (PCR) and partial least squares (PLS) re-gression. Although many early studies were performedwith MLR, this approach uses only a few wavelengthswhereas the other methods use the full spectrum.Greater computing power is one reason why a modi®-cation of the PLS procedures (MPLS) is one of the mostpopular regression procedures (Shenk and Westerhaus1993b) but there are other even more sophisticated ap-proaches available involving neural networks andwavelet theory. All these options are provided in variousstatistical packages but the analyst needs a clear view ofthe purpose of the calibration before choosing oneof these methods and should operate within the boundsof standard practices (Anon 1995).

Validation of the calibration can be done in severalways. Previously the most common method used in-volved an independent set of samples for which com-positional data was also available. In such cases, thecalibrations derived from a ``calibration data set'' couldbe tested by predicting values in a ``validation data set'',the relationship between these being assessed by simple

regression analysis. More recently, this approach hasbeen superseded by a procedure called cross validation,a form of Monte Carlo simulation, in which the popu-lation is arbitrarily divided into a small number ofgroups and a prediction is made of the values for onegroup based on calibrations developed from the re-maining groups. In turn, predictions are made for allgroups with the average of predictions for all groups(Shenk and Westerhaus 1993a, b). Depending on theamount of computing power available, these ``groups''may contain as few as one sample. This procedure isuseful because all available chemical analyses for allindividuals can be used to determine the calibrationmodel without the need to maintain separate validationand calibration sets (Shenk and Westerhaus 1993a, b).Nonetheless, some analysts still maintain separate vali-dation sets for verifying these models.

Conventional laboratory assays ± what NIRShas revealed

Most scientists who have embarked on studies usingNIRS have found that conventional analyses involve asmuch work as the NIRS technique in providing an as-sessment of the calibration model. Because NIRS is socritically dependent on the quality of the data used tobuild predictive models, analysts often need to checkand evaluate their assays in more detail than previously,as unknown sources of error become identi®ed (Wind-ham et al. 1988). NIRS can be used, to some degree, toprovide an independent check of laboratory values. Forexample, in our laboratory we were con®dent fromearlier work that we could predict total nitrogen in plantmaterial with a high degree of accuracy, so when weobtained a set of samples and data from another labo-ratory that resulted in a poor calibration, we ®rstquestioned the accuracy of the original analyses. Re-analysis for total nitrogen in our laboratory with blindduplicates resulted in a much better calibration thatdoubled the precision of the NIRS analysis.

Some debate exists as to whether NIRS calibrationscan be more accurate than the underlying referencemethod on which they are based. NIRS predictionscontain both the underlying laboratory errors and NIRinstrument errors, so it is unlikely that this could be thecase. However, what may be most important is the kindof laboratory errors that are made. During the devel-opment of calibration models, most software packageswill highlight erroneous mixes between spectra andlaboratory values which can be of use in helping toidentify potential sources of error. It is also possible, intheory, that unbiased but randomly inaccurate sourcesof error may be averaged out in large calibration sets.Whether a calibration is used for analysis depends verymuch on the discretion of the analyst. However, inpractice, a good NIRS method should have a predictionerror which is close to the standard error of the labo-ratory reference data.

Fig. 6 Fictitious example illustrating how spectral information iscorrelated with analytical information in order to develop acalibration equation (adapted from Van Kempen and Jackson 1996)

297

Several studies have highlighted the fact that inter-assay variation can also be a potential problem whentrying to construct calibration equations (e.g., Shenket al. 1992; Shenk and Westerhaus 1994). This variabilitymay need to be assessed by re-analyzing series of blindduplicate samples on successive occasions. Exchangeof samples between laboratories conducting the sameanalyses bene®ts all parties.

Linking wavelengths to chemical constituents

Although the statistical models built on NIRS data areessentially empirical, users must realize that the wave-lengths used in the equations relate to the underlyingchemical composition. Ecologists will have greater con-®dence in NIRS if they can see that the wavelengths usedin calibration models are those characteristic of thecomponents of interest. For example, several studieshave shown that the prediction of crude protein intemperate grasses depend on absorbances at 2120±2160 nm. The spectral region between 2100 and2200 nm corresponds to NAH stretching in amidebonds (Shenk and Westerhaus 1993b) and so wave-lengths of 2120±2160 nm should be suitable for esti-mating crude protein. Not surprisingly, manycalibration equations for crude protein emphasizewavelengths in this region. Plant secondary metabolites(PSMs) such as phenolics are expected to have a majorabsorbance around 1650 nm (Flinn et al. 1996) and thiswavelength has been used to separate forage samplescontaining high and low concentrations of phenolicconstituents (Fig. 7; Flinn et al. 1996) but again wecaution that lignin could also absorb in this region andso the 1650 nm spectral region is not necessarily diag-nostic for phenolics. Calibration is required to makequantitative estimates.

Assignment of wavelengths for attributes such as di-etary ®ber, total digestibility, or food intake to particularchemical compounds is not so easy because these entitiesare either poorly de®ned chemically, depend on a greatnumber of individual compounds or vary from forage toforage. One would expect however, that wavelengths for

plant ®ber estimation should correspond to regionswhere CAH bonds absorb and this is what has beenfound (Clark and Lamb 1991). Assigning wavelengths tocomponents of interest remains an important researchgoal for many using NIR spectroscopy, particularly inagricultural industries because of the extensive databasesthat have already been built to describe aspects of plantquality on which producers are paid.

Ecological applications

The main purpose of this section is to highlight some ofthe recent innovative studies using NIRS that have po-tential applications to ecologists. From these studies, wesynthesize implications for its possible use in the future.We conclude the review with a list of new challengeswhich NIRS might help to overcome. However, westress that in many of these ``new areas'' much workneeds to be done in testing the limitations and potentialof NIRS before the technique can be universally adoptedin these areas of research. Despite this, we are hopefulthat given time and the growing enthusiasm and supportfor NIRS, that this will happen in most, if not all, of the®elds mentioned below.

In assessing the potential use of NIRS in each of theseareas, it is important to bear in mind that the mainbene®t of NIRS lies not as a replacement tool for tra-ditional analyses, but rather as a complementary toolwhich can be used alongside conventional analyses toimprove the e�ciency and cost of studies, or alterna-tively to explore areas which are impossible usingalternative means. As with the application of any tech-nology into new areas, there is a potential danger toover-use or misapply NIRS and any predictions that aremade should be carefully evaluated before they areapplied. The following examples illustrate some of thecurrent and emerging applications of NIRS in the ®eldof ecology.

Resolution of complex mixtures

Measuring the botanical composition of di�erent plantcommunities is integral to many ecological studies, es-pecially those concerned with foraging in grasslands. Inthe past this information was obtained either by hand-sorting clipped quadrats, or by a variety of visualestimation. Hand sorting is slow and laborious andalthough visual estimation is rapid, there may be manyunquanti®able errors (Coleman et al. 1985).

Shenk et al. (1979) originally showed that NIRS couldbe used to determine the proportion of legumes in le-gume-grass mixtures to an accuracy of about �10% andseveral studies have attempted to expand this concept tomore complex pastures (e.g., Coleman et al. 1990; Pitmanet al. 1991; Atkinson et al. 1996). Arti®cial mixtures of

Fig. 7 Example showing the allocation of wavelengths to speci®cchemical bonds (taken from Flinn et al. 1996)

298

three to four di�erent components were accurately pre-dicted by Coleman et al. (1990), but subsequent workshowed that the predictive equations were too sensitive tothe source of material used to make up the mixtures.Single sources of each component led to acceptable cali-brations whereas calibrations involving many sources ofeach component, as would occur with natural pastures,were not su�ciently robust for routine use. Nonetheless,acceptable calibrations have been generated from pasturesamples that were meticulously separated by hand(Petersen et al. 1987; GarcõÂ a-Criado et al. 1991).

NIRS has been used to assess the botanical diversity ofgrassland habitats by comparing the spectral similarity or``composition'' between samples (Hill et al. 1988). Forexample, Atkinson et al. (1996) used NIRS to monitorchanges in the composition of vegetation samples acrosstime and space, which could be traced back to hand-sorted component species. In this study, NIRS was usedas a discriminant tool which could be used to groupsamples at varying levels of detail, in order to identify theoverall similarity in composition between samples (seesection on Emerging applications for more detail).

A second example of how NIRS can be used toresolve di�erences in complex mixtures is a study byDowney and Boussion (1996) who tested the ability ofNIRS to discriminate between di�erent blends of co�ee.They obtained a 95% successful resolution based ondi�erences in spectra alone. These di�erences were pre-sumed to be predominantly due to variation in theconcentration of ca�eine or another related alkaloid.

Overall, it seems that NIRS has the potential to saveresearchers substantial time and e�ort in monitoring themake-up of mixtures. From past studies, it is evidentthat calibrations for narrow or closed populations canbe expected to yield more precise information thanpredictions across multiple landscapes and a range ofenvironmental conditions. However, before this pros-pect is adopted, a greater e�ort needs to be devoted tounderstanding the factors that a�ect the robustness ofthe predictive equations in individual situations.

Determination of plant nutrients

By far the widest use of NIRS has been in the mea-surement of nutritional components in animal feeds(e.g., Norris et al. 1976; Brown et al. 1990; Coleman andMurray 1993). Components measured include total ni-trogen, moisture, ®ber (neutral and acid detergent ®berand acid lignin), starch, individual sugars, amino acidsand plant tannins. To give but one example of the po-tential time and cost bene®ts of using NIRS to analyzesuch components, Jin et al. (1994) used NIRS tosimultaneously measure concentrations of sucrose, glu-cose, fructose, citric acid, malic acid and vitamin C inintact strawberries. NIRS gave a similar accuracy toconventional HPLC techniques, but with the advantageof near-instantaneous, non-destructive and chemical-free analysis. The prediction of crude protein in grain

remains the most common application of NIRS in ag-ricultural industries but plant nutrients and carbohy-drate fractions have been successfully predicted in arange of di�erent shrub and tree leaves as well (Meuretet al. 1993; Martin and Aber 1994).

Plant secondary metabolites and anti-nutritionalcomponents

Components of several PSMs such as the hydroxylgroups of plant phenolics, have signi®cant absorbancesin the NIR region and successful calibrations have beendeveloped for these constituents. For example, Wind-ham et al. (1988) predicted the content of ``total tannin''(Folin-Denis assay of 50% methanol extractives) with acoe�cient of determination of 0.91 in Lespedeza cunea-ta. Given the errors of conventional laboratory assaysfor these poorly de®ned compounds these levels of ac-curacy may well be the best achievable. However, thereis no reason to suspect that other types of tannin ana-lyses would not yield calibrations that are satisfactoryfor most ecological studies (Roberts et al. 1993).

Less work has been directed at other PSMs. Clarket al. (1987a) developed a robust calibration to predicttotal alkaloids in larkspur (Delphinium occidentale) andlupine (Lupinus leucophyllus). Both morphine and nico-tine have been predicted with a high degree of precisionin poppies and tobacco (McClure and Williamson 1988),and glucosinolate concentrations have been reliablymeasured in rapeseed (Mika et al. 1997). In many ofthese studies, examination of spectral characteristics hasrevealed strong relationships between functional groupsand speci®c wavelengths, potentially a valuable means ofscreening samples for the presence of these groups.

The wide range of plant types and chemistries in-volved in these examples gives con®dence that NIRS canbe used successfully to analyze large numbers of samplesfor PSMs. However, all these examples are of organiccompounds and the ability of NIRS to predict func-tional aspects of PSMs such as astringency or protein-precipitating capacity is unknown.

Given the clear potential of NIRS to investigate thechemical composition of plants, we believe the techniquehas a wide application for investigating many ecologicalquestions. Examples where NIRS might be particularlyuseful is in understanding the e�ects of resource avail-ability on plant growth and chemical defense, the role ofnutrition in in¯uencing the population dynamics ofherbivores and in understanding the speci®c foragingpatterns of animals in relation to the nutritional hetero-geneity and complexity of their surroundings.

Mineral analysis

The estimation of mineral elements by NIRS is usuallydependent on the occurrence of those elements in eitherorganic or hydrated molecules (Clark et al. 1987b;Vasquez de Aldana et al. 1995). For example, key

299

wavelengths used in detecting Mg were similar to thepeaks of the chlorophyll spectrum and Ca peaks weresimilar to those of pure calcium pectate which may be acomponent of cell walls. Mineral analysis studies havesuggested that K, P, Mg, Ca, S, Al and possibly Si existin forms detectable by NIRS at least in some grasses andlegumes (Clark et al. 1987b, 1989; Saiga et al. 1989).

Despite the reports of many satisfactory calibrationsfor mineral elements, it is di�cult to apply their resultsto other plants because the form of organic molecule-mineral compounds can vary seasonally or among spe-cies and genera. This may lead to unstable calibrations,inconsistent results and di�culty in expanding calibra-tions beyond a well characterized population. For ex-ample, phosphorus in plants exists mainly in organicforms frequently as phytate, phospholipids and nucleicacids. The proportion of total P in this form varies be-tween closely related cereals and forages and also sea-sonally (De Boever et al. 1994). Calibrations obtainedfor total P in forages were considered acceptable byClark et al. (1987b) but were not successful in similarstudies by Vasquez de Aldana et al. (1995). Cautionneeds to be exercised when using NIRS to analyzemineral elements. However, the technique can be usefulfor broad ®rst approximations and for selecting samplesfor more accurate analyses.

Soil analysis

Due to the large inorganic fraction of material in soil,NIRS has not been as widely used for soil analysis(Meyer 1989). However from the few studies for whichthere are data, it is evident that at least for carbon andnitrogen, calibration equations in size-speci®c soil frac-tions can yield an accuracy similar to that recorded forplant tissues (Morra et al. 1991).

A potential application of NIRS for soil biologists isits usefulness in selecting soil samples from populationsto maximize the variation in particular soil properties ina minimum subset of samples from the population(Stenberg et al. 1995). This is a powerful application ofNIRS, as it allows one to select samples over the fullrange of variation in any one or combination of pa-rameters. In this way, extreme or peripheral samplesmay be chosen which might otherwise fail to be repre-sented in a sample set for analysis. In addition, a speci®cdistribution of samples can be chosen in reference to aparticular parameter. Soil parameters used by Stenberget al. (1995) included clay content, cation exchange ca-pacity, base saturation, and pH . Cost savings of 70%were achieved when the technique was compared to themost appropriate sampling strategy involving only wetchemistry analysis.

Prediction of functional attributes

NIRS has enormous potential as a holistic tool for in-vestigating natural systems. Although reductionist ap-

proaches are useful in understanding how individualcomponents of a system function, what is often moredesirable is an understanding of how these connectedparts function as a whole. For example, at times theprediction of chemical composition may be the ultimateaim of the analyst. However, in many situations ecolo-gists are interested in more complex attributes and areconstrained to use compositional measures as proxies forthese more complex phenomena. Examples of complexattributes which have been examined using NIRS includefeatures such as the susceptibility of plants to insect attack(Rutherford and Van Staden 1996), the yield of pulp andbiological degradability of sawlogs (Wright et al. 1990;Ho�meyer and Pedersen 1995) and the prediction of thenutritional quality of wild foods for free-ranging herbi-vores (A. McIlwee, I. Lawler and W.J. Foley, unpub-lished work).

Understanding the organization of complex systemsis an important objective of ecology. However, becausethere is an almost unlimited variety of ways that bioticand abiotic components can be assembled into complexecological systems, our understanding of the structureand dynamics of these systems has remained inherentlylimited due to traditional approaches that have thus fardominated ecology (Brown 1997). Many systems areinherently non-linear, which limits our ability to makeecological predictions. However, much time and e�ortcan be saved by looking at these systems from a holisticviewpoint and deciding whether they are in fact capableof being modeled. NIRS o�ers the potential for a shifttowards whole-system empirical modeling in severalareas of ecology. This methodology can still be usedalongside more formal experimental approaches togenerate new insights into how complex systems func-tion.

We see the prediction of functional attributes as beingone of the most important applications of NIRS inecology. Any attribute that is suspected of being in¯u-enced by the composition of plant or animal materialcould potentially be modeled in this way. As well asaspects of animal performance, NIRS could also be usedto measure attributes of plant performance, such asinnate mean relative growth rate (see Cornelissen et al.1996). Although this is partly an empirical approach, itis similar to the many successful bioclimatic models thathave been used to predict the limits of animal distribu-tion. Below, we provide a range of examples whichillustrate the power of NIRS.

Insect resistance

A good example of this approach is in studies of theresistance of di�erent sugarcane cultivars to attack bystem borers. Rutherford and Van Staden (1996) devel-oped a MLR model of NIR data to predict resistance ofsugarcane cultivars to Eldana saccharina. They foundthat they could account for 54% of the variation in re-sistance but, more signi®cantly, wavelengths in theirmodel indicated that alcohols and carbonyl components

300

of the stalk surface wax correlated with an importantpart of this resistance. In this case, NIRS became notjust a tool to build an empirical model but indicated anunderlying mechanism that could be further investi-gated. This approach can save substantial time in tar-geting a research program to enhance resistance.

A second example where NIR has been used to pre-dict a poorly de®ned attribute is the variable suscepti-bility of Eucalyptus trees to insect attack. Generally thisis caused by variations in both primary and secondarychemistry of the foliage (Edwards et al. 1993) but themost important features have not yet been identi®ed.However, it was still possible to develop a model toaccount for 88% of the variation in resistance to defo-liation of individual trees based solely on the spectralcharacters of the foliage (W. Foley, P. Edwards,W. Wanjura and M. Matsuki, unpublished work).

Prediction of animal performance

Many studies postulate di�erences in the nutritionalquality of the diets consumed by animals as a basis ofdi�erent population densities, social organization, andevolutionary success. However, testing these hypothesesis limited by an inability to measure ``nutritional quali-ty'' in free-living animals (Hanley 1997). Nutritionalquality has most often been estimated indirectly bymeasuring the concentration of nutrients including ni-trogen and ®ber and PSMs such as condensed tannins.In most cases, whether tannins a�ect the intake or di-gestibility of foods in these species, or whether the crudeprotein is in a usable form or not is not known. In fact,crude measures of plant composition tells us little abouthow well a particular food will sustain an animal.

Nutritional quality is re¯ected in two parameters thatcan be measured in feeding experiments with captiveanimals; the level of food intake and the digestibility/metabolizability of that food. These estimates have beenmade by NIRS using both samples of the food beingeaten (Redshaw et al. 1986; Givens et al. 1991, 1992) oralternatively from samples of the feces excreted (e.g.,Lyons and Stuth 1992). Prediction of diet quality fromfeces samples means that it is the actual diet chosen bythe animal that is being evaluated, not that which theecologist decides is being eaten. Brooks et al. (1984)developed a number of calibrations for use with bothfood and feces samples of white-tailed deer but mostinnovations in this area have been from Stuth and co-workers (Lyons and Stuth 1992; Leite and Stuth 1994).

Use of feces samples to predict the quality of dietsingested by free-ranging herbivores means that the re-searcher need not know what the actual diet ingested bythe animal comprised. Examination of feces has alsobeen used to examine the digestive e�ciency of herbi-vores on range of natural diets through a simple com-parison of di�erence spectra between diet and feces (e.g.,Coleman and Murray 1993). Although there are severalassumptions that need to be addressed using this meth-

od, namely the e�ects of microbial residues andsloughed animal tissue, it does provide a simple andpotentially valuable means of evaluating the nutritionalvalue of foods and looking at the associative e�ects offood ingestion and nutrient assimilation.

When combined with geographic information sys-tems, estimates of the nutritional quality of herbivorediets can become powerful tools for monitoring pastoraldegradation at many scales. For example, J. W. Stuth(personal communication) has used NIRS to predict dietquality from feces samples of cattle and deer and byplotting the position of each feces sample has identi®eddi�erences in the quality of di�erent habitats to supportthese herbivores. Knowledge of the spectral stability offeces exposed to ®eld conditions adds to its power as amanagement tool (Leite and Stuth 1994).

Isotope discrimination ± water use e�ciencyand herbivore diets

Stable isotopes are used in many areas of ecology, in-cluding estimation of water use e�ciency, separation ofC4 and C3 plants and in studies of foraging and foodwebs. Modern isotope ratio mass spectrometry (IRMS)is highly accurate and precise, but it is relatively ex-pensive, laborious, and time consuming. NIRS may beable to predict relative abundances of 13C. For example,Okano et al. (1993) found that NIRS could predictabundance of 13C in a range of C3 and C4 plants towithin 97% of measured values. This degree of accuracycould be su�cient for many studies of herbivore forag-ing where the aim is to measure the proportion oftropical grass in the diet.

Clark et al. (1995) used NIRS to predict 13C dis-crimination in a range of grasses and legumes so thatthey could select genotypes with high water use e�ciency(WUE), based on the well-established relationshipbetween water use e�ciency and 13C carbon ratios inplants (Salisbury and Ross 1992). They found that NIRScould identify up to 82% of samples with low WUE, andconcluded that accuracy may be su�cient to identifypromising genotypes in the early selection stages. ClearlyNIRS is not expected to replace IRMS as a means ofdetermining isotope abundance but we believe thatNIRS may have a role in rapidly screening a largenumber of samples of plant tissues so that the moreexpensive analysis can be targeted on those samples thatare of interest.

Litter decomposition studies

Mineral cycling in forest and aquatic ecosystems hasbeen investigated using NIRS. The chemical composi-tion of leaf litter is a key factor regulating its rate ofbreakdown and the speed and non-destructive nature ofNIRS has been used by several groups. McClellan et al.(1991) showed that total N, C, and ash could be accu-

301

rately measured by NIRS as well as several componentsof plant cell walls. Following on from this, Gillian et al.(1993) showed that NIRS was able to directly predict thestage of decomposition (expressed as the percentage ofash-free litter mass remaining) without relying on indi-rect estimates based on chemical composition. Corn-elissen and Thompson (1997) provide a range ofpotential functional attributes of living leaves (for bothmonocots and dicots) which can be used as a basis forpredicting the decomposition rate of leaf litter. Suchattributes could be used by people using NIRS toexamine the impact of anthropogenic activities andenvironmental changes on decomposition processes.

Limnology and freshwater ecology

NIRS has been used to measure C, N, and P in sus-pended sediments and in seston by collecting the mate-rial on glass-®ber ®lters and then obtaining NIR spectra(Malley et al. 1993, 1996). Agreement between the NIRSmeasures and those obtained by traditional methods wasexcellent in both cases. However, the use of NIRS is notrestricted to these types of compositional assays. Kors-man et al. (1992) and Nilsson et al. (1996) have shownthat NIRS can be used to predict the past chemicalcomposition of lake water from models of the surfacesediments and current water chemistry. A record of thecurrent water chemistry is contained in the surface sed-iments and once a relationship is established betweenthese parameters, past histories can be evaluated bytaking further sediment samples. Nilsson et al. (1996)found that they could predict 83±85% of the variance intotal phosphorus and pH, respectively, and 68% of thevariance in total organic carbon. This approach couldbecome a powerful tool for monitoring past events inwater chemistry.

Biochemical markers and rates of biological activity

NIRS can be used to detect biochemical markers, suchas chitin (Roberts et al. 1994) which is used to estimatelevels of fungal infection in crops and pastures. Directestimates of fungal infection have also been made byNIRS (Roberts et al. 1988). Dysprosium (Dy), a pulse-dosed marker used to measure rates of digesta ¯ow inherbivores, has also been successfully measured usingNIRS in studies on cattle (Reeves and Glenn 1995).Although NIRS was capable of determining Dy inlabeled forages, interference by a second co-markerlimited the accuracy of predictions.

In addition to detecting markers, NIRS also has thepotential to measure rates of biological activity. For ex-amplesHall et al. (1991) usedNIRS tomeasure bloodmealsize and its disappearance rate in live disease-carryingmosquitoes, NIRS has been used as an non-intrusivemeans of monitoring changes in blood composition (Hi-nckley et al. 1995; Delpy and Cope 1997), and Palmborg

andNordgren (1993) appliedNIRS to themeasurement ofmicrobial biomass and activity of forest soils.

Analyses of high-moisture materials ± prospectsfor analysis of fresh samples

Although some of the earliest applications of NIRS in-volved analysis of fresh meat samples, most applicationscurrently involve dry and ground materials. Even thislimited sample preparation increases analysis time andcost substantially. The imbalance between the number ofapplications for dried and fresh samples discredits thereal advantages o�ered by the direct analysis of freshmaterials.

Samples which contain substantial quantities ofmoisture present particular di�culties for NIRS. Thehydrogen bonds in water absorb signi®cant amounts ofNIR radiation and result in broad peaks that obscurespectral information derived from other compounds(Abrams et al. 1988). Wet samples tend to be moreheterogeneous and moisture can be expressed from thesample when it pressed against the glass window of thesample cell. A major di�culty is to generate samples forinstrument standardization that will not degrade overtime. Currently this is a serious weakness in applyingNIRS to high-moisture materials (Berding and Broth-erton 1996).

There is no single solution to these problems and eachcase has to be treated individually. The di�culties arenot insurmountable since many robust calibrations havebeen developed for moist materials including dairyproducts, ®sh, fresh, intact fruits and vegetables, meats,and sugarcane (Blosser and Reeves 1986; Mathias et al.1987). In calibration development for example, useof high absorbance spectral regions characteristic ofhigh moisture materials are best avoided because of thenon-linearity associated with these regions.

Emerging applications: challenges and prospectsfor the future use of NIRS

Use of NIRS as a discriminant tool ±chemotaxonomy and beyond

Ecologists often wish to classify elements of populationsinto groups for a range of di�erent reasons, even if it isas simple as determining whether an animal is one sex oranother. NIRS has been successfully used in this fashionto predict whether pupae are male or female (Jin et al.1995) and to predict whether individual seeds or grainshave been infected by larval insects (Ridgway andChambers 1996). In such applications, the spectral dataare often acquired by ®ber-optic probes that can be fo-cused on very small targets such as single insect larvae.

NIRS is also ®nding use as a tool for discriminatingamong products from di�erent provenances or of

302

di�erent species. The wealth of data available in NIRspectra means that the digitized spectra are particularlysuited to the variety of multivariate statistical proce-dures that are typically used in discriminant analyses.For example, Schimleck et al. (1996) acquired the NIRspectra of a range of timber samples. They were able tobuild models based on principal components analysis ofNIR spectral data that successfully discriminated be-tween woods from Eucalyptus and Pinus, between woodsof di�erent species of Eucalyptus and between woods ofthe same species of Eucalyptus grown at di�erent sites.Other similar applications have involved identifyingadulterated food products, such as contamination byfungi (Davies et al. 1987) or determining the infectionlevels of diseases such as anther smut disease in crops(Nilsson et al. 1994). These approaches can easily betransferred to ecological applications.

Discriminant analyses are very common in ecologyand typically require substantial amounts of chemical orphysical measurements to be successful. Collecting thesecan be time-consuming and expensive and we believethat many ecological applications could pro®t fromusing NIRS data as an input to discriminant procedures.

On-line processing and portable and hand-held NIRS

Most NIR spectrophotometers that are currently usedfor research are dedicated laboratory instruments whichare housed in carefully controlled conditions. However,the greatest use of NIRS is as an on-line analytical toolin industrial plants ranging from sugar mills and wheatsilos to pharmaceutical manufacturing plants (for ex-amples see Bellon and Boisde 1989; Osborne et al. 1993;Schulz and Losing 1995). In these environments the in-strumentation is exposed to much more rugged condi-tions and usually involved in the analysis of fresh, intactsamples. This provides the potential for more exposedenvironmental monitoring.

In the future we can expect NIR instruments to bebecome smaller and more portable. There are already anumber of hand-held instruments that have been used tomeasure the nitrogen concentration of the leaves of cropplants as an aid to fertilizer management. For example,Blakeney et al. (1996) successfully used a hand-heldbattery operated instrument for nitrogen measurementsin rice leaves.

In the past year, portable instruments that can scanthe whole NIR spectrum have become available. Theseinstruments use a variety of ®ber-optic probes and,coupled with the speed of advances in optics and micro-processor technology, should lead to even more robustportable instruments in the future that o�er ecologiststhe possibility of real-time analyses of plant and animalcomposition in the ®eld. In particular we could expectcloser alliances to be formed between those interested intraditional airborne remote sensing and those interestedin ®ner resolutions as well as advances in statistical

processing which will allow a better description ofspectral features.

Conclusion

Although we are very optimistic about the future ofNIRS, we conclude with the warning that NIRS cannotturn poor analysts into good ones. Used uncritically, itcan allow poor results to be propagated. Ultimately, thequality of NIRS-based predictions depends entirely onthe quality of the measurements used to generate thestatistical model and this must guide any analytical en-deavour.

NIRS o�ers several special advantages such as speedof determination, minimal or no preparation of sample,non-destructive analysis, no consumption of reagents,and low costs of analysis. However, the capital costof NIR spectrophotometers is high (ca. $US 40,000±80,000), so ecologists might ®rst consider developingcollaborations with existing users of NIRS in agricul-tural industries or research stations where the techniqueis widely used. In this way, they can evaluate the use-fulness of the method for particular samples and ana-lyses before committing to purchase. As NIRS movesforward into the future, we look forward to witnessingthe integration of the technique into applied ecology.

Acknowledgements We are grateful for the advice of many scien-tists who shared their knowledge of NIRS with us, especially Drs.Peter Flinn, Graeme Batten and members of the Australian NIRSUsers Group. We thank Foss-NIR Systems (Australia) and Lin-brook International for their support of our work.

References

Abrams SM, Shenk JS, Harpster HW (1988) Potential of near in-frared re¯ectance spectroscopy for analysis of silage composi-tion. J Dairy Sci 71:1955±1959

Anon (1995) Standard practices for infrared multivariate quanti-tative analysis (designation E1655±94). American Society forTesting and Materials, Washington

AOAC (1990) O�cial methods of analysis. Association of O�cialAnalytical Chemists, Arlington

Aragones L (1997) Dugongs and green turtles: grazers in thetropical seagrass ecosystem. PhD thesis, James Cook Univer-sity, Townsville, Australia

Atkinson MD, Jervis AP, Trueman IC (1996) The potential of nearinfrared spectroscopy for monitoring species diversity ingrassland. In: Association of Applied Biologists (eds) Vegeta-tion management in forestry amenity and conservation areas.Managing for multiple objectives. University of York, York, pp191±209

Baker CW, Givens DI, Deaville ER (1994) Prediction of organicmatter digestibility in vivo of grass by near infrared re¯ectancespectroscopy: e�ect of calibration method residual moistureand particle size. Anim Feed Sci Technol 50:17±26

Bellon V, Boisde G (1989) On-line control of agri-food: Use of anear infrared spectrometer coupled with optical ®bers. In: Sa-gaspe JP, Villeger A (eds) Agrotique 89. Proceedings of thesecond international conference. Teknea, Marseilles

303

Berding N, Brotherton GA (1996) Analysis of high-moisture ma-terial±®brated sugarcane. In: Davies AMC, Williams P (eds)Near infrared spectroscopy: the future waves. NIR Chichester,pp 648±654

Blakeney AB, Batten GD, Welsh LA (1996) Leaf nitrogen deter-mination using a portable near-infrared spectrometer. In: Da-vies AMC, Williams P (eds) Near infrared spectroscopy: thefuture waves. NIR, Chichester, pp 149±152

Blosser TH, Reeves JB (1986) Near infrared re¯ectance spectros-copy (NIRS) for predicting chemical composition of undriedfeedstu�s. J Dairy Sci 69(suppl 1):136

Brooks J III, Anderson M, Urness PJ (1984) Infrared re¯ectanceanalysis of forage quality for elk. J Wildl Manage 48:254±258

Brown JH (1997) An ecological perspective on the challenge ofcomplexity (EcoEssay series 1). National Center for EcologicalAnalysis and Synthesis, Santa Barbara

Brown WF, Moore JE, Kunkle WE, Chambliss CG, Portier KM(1990) Forage testing using near infrared re¯ectance spectros-copy. J Anim Sci 68:1416±1427

Casler MD, Shenk JS (1985) E�ect of sample grinding size onforage quality estimates of smooth bromegrass clones. Crop Sci25:167±170

Clark DH, Lamb RC (1991) Near infrared re¯ectance spectrosco-py ± a survey of wavelength selection to determine dry matterdigestibility. J Dairy Sci 74:2200±2205

Clark DH, Ralphs MH, Lamb RC (1987a) Total alkaloid deter-minations in larkspur and lupine with near infrared re¯ectancespectroscopy. Agron J 79:485±90

Clark DH, Mayland HF, Lamb RC (1987b) Mineral analysis offorages by near infrared re¯ectance spectroscopy. Agron J79:485±490

Clark DH, Cary EE, Mayland HF (1989) Analysis of trace ele-ments in forages by near infrared re¯ectance spectroscopy.Agron J 81:91±95

Clark DH, Johnson DA, Kephart KD, Jackson NA (1995) Near-infrared re¯ectance spectroscopy estimation of 13C discrimina-tion in forages. J Range Manage 48:132±136

Coleman SW, Murray I (1993) The use of near infrared re¯ectancespectroscopy to de®ne nutrient digestion of hay by cattle. AnimFeed Sci Technol 44:237±249

Coleman SW, Barton FE II, Meyer RD (1985) The use of nearinfrared re¯ectance spectroscopy to predict species compositionof forage mixtures. Crop Sci 25:834±837

Coleman SW, Christiansen S, Shenk JS (1990) Prediction of bo-tanical composition using NIRS calibrations developed frombotanically pure samples. Crop Sci 30:202±207

Cornelissen JHC, Thompson K (1997) Functional leaf attributespredict litter decomposition rate in herbaceous plants. NewPhytol 135:109±114

Cornelissen JHC, Diez PC, Hunt R (1996) Seedling growth allo-cation and leaf attributes in a wide range of woody plant speciesand types. J Ecol 84:755±765

Davies AMC, Dennis C, Grant A, Hall MN, Robertson A (1987)Screening of tomato puree for excessive mould content by nearinfrared spectroscopy: a preliminary evaluation. J Sci FoodAgric 39:349±355

De Boever JL, Eeckhout W, Boucque ChV (1994) The possibilitiesof NIRS to predict total phosphorus phytate-phosphorus andphytase activity in vegetable feedstu�s. Neth J Agric Sci 42:357±369

Delpy DT, Cope M(1997) Quanti®cation in tissue: near infraredspectroscopy. Philos Trans R Soc Ser B: Biol Sci 352:649±659

Downey G, Boussion J (1996) Authentication of co�ee bean varietyby near infrared re¯ectance spectroscopy of dried extract. J SciFood Agric 71:41±49

Edwards PB, Wanjura WJ, Brown WV (1993) Selective herbivoryby Christmas beetles in response to intraspeci®c variation inEucalyptus terpenoids. Oecologia 95:551±557

Flinn PC, Edwards NJ, Oldham CM, McNeil DM (1996) Nearinfrared analysis of the fodder shrub tagaste (Chamaecytisusproliferus) for nutritive value and anti-nutritive factors. In:

Davies AMC, Williams P (eds) Near infrared spectroscopy: thefuture waves. NIR, Chichester, pp 576±580

Garcia-Criado B, Garcia-Criado A, Perez-Corona ME (1991)Prediction of botanical composition in grassland herbagesamples by near infrared re¯ectance spectroscopy. J Sci FoodAgric 57:507±515

Gillian D, Jo�re R, Dardenne P (1993) Predicting the stage ofdecay of decomposing leaves by near infrared re¯ectancespectroscopy. Can J For Res 23:2552±2559

Givens DI, Baker CW, Moss AR, Adamson AH (1991) A com-parison of NIRS with three in vitro techniques to predict thedigestibility in vivo of untreated and ammonia treated cerealstraws. Anim Feed Sci Technol 35:83±94

Givens DI, Baker CW, Adamson AH, Moss AR (1992) In¯uenceof growth type and season on the prediction of the metaboliz-able energy content of herbage by NIRS. Anim Feed SciTechnol 37:281±295

Givens DI, De Boever JL, Deaville ER (1997) The principles,practices and some future applications of near infrared spec-troscopy for predicting the nutritive value of foods for animalsand humans. Nutr Res Rev 10:83±114

Hall MH, Dutro SM, Klowden MJ (1991) Determination by near-infrared re¯ectance spectroscopy of mosquito (Diptera: Culic-idae) bloodmeal size. J Med Entomol 27:76±79

Hanley TA (1997) A nutritional view of understanding and com-plexity in the problem of diet selection by deer (Cervidae).Oikos 79:209±218

Hill NS, Petersen JC, Stuedmann JA, Barton II FE (1988) Pre-diction of percentage leaf in strati®ed canopies of alfalfa withnear infrared re¯ectance spectroscopy. Crop Sci 28:354±358

Hinckley KA, Fearn S, Howard BR, Henderson W (1995) Nearinfrared spectroscopy of pedal haemodynamics and oxygen-ation in normal and laminitic horses. Equine Vet J 27:465±470

Ho�meyer P, Pedersen JG (1995) Evaluation of density andstrength of Norway spruce by near infrared re¯ectance spec-troscopy. Holz Roh Werksto� 53:165±170

Jin TM, Cui HC, Jin TM, Cui HC (1994) A new method for de-termination of nutrient contents of intact strawberries ± nearinfrared spectrometry. Acta Agric Boreal Sin 9:120±123

Jin TM, Wu XQ, Pan SY, Liu L (1995) Distinguishing sex ofsilkworm pupae in cocoon by NIR with ®ber optics. In: BattenGD, Flinn PC, Welsh LA, Blakeney AB (eds) Leaping aheadwith near infrared spectroscopy. NIR, Chichester, pp 524±526

Jo�re R, Gillon D, Dardenne P, Agneessens R, Biston R (1992)The use of NIRS in litter decomposition studies. Ann Sci For49:481±488

Korsman T, Nilsson M, Ohman J, Renberg I (1992) Near infraredre¯ectance spectroscopy of sediments ± a potential method toinfer the past pH of lakes. Environ Sci Technol 26:2122±2126

Leite ER, Stuth JW (1994) Technical note: in¯uence of duration ofexposure to ®eld conditions on viability of fecal samples forNIRS analysis. J Range Manage 47:312±314

Lyons RK, Stuth JW (1992) Fecal NIRS equations for predictingdiet quality of free-ranging cattle. J Range Manage 45:238±244

Malley DF, Williams PC, Stainton MP, Hauser BW (1993) Ap-plication of near-infrared re¯ectance spectroscopy in the mea-surement of carbon nitrogen and phosphorus in seston fromoligotrophic lakes. Can J Fish Aquat Sci 50:1779±1785

Malley DF, Williams PC, Stainton MP (1996) Rapid measurementof suspended C, N and P from Precambrian shield lakes usingnear-infrared re¯ectance spectroscopy. Water Res 30:1325±1332

Marten GC, Shenk JS, Barton FE (1989) Near infrared re¯ectancespectroscopy (NIRS): analysis of forage quality 2nd edn(Handbook 643). United States Department of Agriculture,Washington

Martin ME, Aber JD (1994) Analysis of forest foliage. III. Deter-mining nitrogen lignin and cellulose in fresh leaves using near-infrared re¯ectance data. J Near Infrared Spectrosc 2:25±32

Mathias JA, Williams PC, Sobering DC (1987) The determinationof lipid and protein in freshwater ®sh using near-infrared re-¯ectance spectroscopy. Aquaculture 61:303±311

304

McClellan TM, Aber JD, Martin ME, Melillo JM, Nadelho�er KJ(1991) Determination of nitrogen lignin and cellulose content ofdecomposing leaf material by near infrared re¯ectance spec-troscopy. Can J For Res 21:1684±88

McClure WF, Williamson RE (1988) Near infrared analysis oftobacco. In: Creaser CS, Davies AMC (eds) Analytical appli-cations of spectroscopy. Royal Society of Chemistry, London,pp 109±132

Meuret M, Dardenne P, Biston R, Poty O (1993) The use of NIR inpredicting nutritive value of Mediterranean tree and shrub fo-liage. J Near Infrared Spectrosc 1:45±54

Meyer JH (1989) Simultaneous rating of soil texture organic mattertotal nitrogen and nitrogen mineralization potential by NIR.S Afr J Plant Soil 6:59±63

Mika V, Bartko A, Nerusil P, Smital F (1997) Use of NIR methodfor evaluation of rape quality Rost Vyroba 43:283±286

Morra MJ, Hall MH, Freeborn LL (1991) Carbon and nitrogenanalysis of soil fractions using near-infrared re¯ectance spec-troscopy. Soil Sci Soc Am J 55:288±291

Nilsson M, Elmqvist T, Carlsson U (1994) Use of near infraredre¯ectance spectrometry and multivariate data analysis todetect anther smut disease (Microbotryum violaceum) in Silenedioica. Phytopathology 84:764±770

Nilsson MB, Dabakk E, Korsman T, Renberg I (1996) Quantifyingrelationships between near infrared re¯ectance spectra of lakesediments and water chemistry. Environ Sci Technol 30:2586±2590

Norris KH, Barnes RF, Moore JE, Shenk JS (1976) Predictingforage quality by near-infrared re¯ectance spectroscopy. J AnimSci 43:889±897

Okano KI, Osamu I, Nokubun N, Totsuka T (1993) Determinationof 13C in plant materials by infrared absorption spectroscopyusing a simple calibrationmethod. Soil Sci PlantNutr 29:369±374

Osborne BG, Fearn T, Hindle PH (1993) Practical NIR spectros-copy with applications in food and beverage analysis. Long-man, Singapore

Palmborg C, Nordgren A (1993) Modelling microbial activity andbiomass in forest soil with substrate quality measured usingnear infrared re¯ectance spectroscopy. Soil Biol Biochem25:713±718

Petersen JC, Barton FE, Windham WR, Hoveland CS (1987)Botanical composition de®nition of tall fescue-white clovermixtures by near infrared re¯ectance spectroscopy. Crop Sci27:1077±1080

Pitman WD, Piacitelli CK, Aiken GE, Barton II FE (1991) Bo-tanical composition of tropical grass-legume pastures estimatedwith near infrared re¯ectance spectroscopy. Agron J 83:103±107

Poppi DP (1996) Predictions of food intake in ruminants fromanalyses of food composition. Aust J Agric Res 47:489±504

Redshaw ES, Mathison GW, Milligan LP, Weisenburger RD(1986) NIRS for predicting forage composition and voluntaryconsumption and digestibility in cattle and sheep. Can J AnimSci 66:103±115

Reeves JB, Glenn BP (1995) Observations on determination ofdysprosium in labelled digesta samples by near infrared spec-troscopy. J Dairy Sci 78:1962±1969

Ridgway C, Chambers J (1996) Detection of external and internalinsect infestation in wheat by near-infrared re¯ectance spec-troscopy. J Sci Food Agric 71:251±264

Roberts CA, Barton FE, Moore KJ (1988) Estimation of Acre-monium coenophialum mycelium in infected tall fescue. Agron J80:737±740

Roberts CA, Beuselinck PR, Ellersieck MR, Davis DK, McGrawRL (1993) Quanti®cation of tannins in birdsfoot trefoil germ-plasm. Crop Sci 33:675±679

Roberts CA, Marek SM, Lei W, Karr AL (1994) Determination ofchitinase activity in tall fescue by near-infrared re¯ectancespectroscopy. Crop Sci 34:1070±1073

Rutherford RS, Van Staden J (1996) Towards a rapid near-infraredtechnique for prediction of resistance to sugarcane borer Eldana

saccharina Walker (Lepidoptera Pyralidae) using stalk surfacewax. J Chem Ecol 22:681±694

Saiga S, Sasaki T, Nonaka K, Takahashi K, Watanabe M,Watanabe K (1989) Prediction of mineral concentrations oforchardgrass (Dactylis glomerata L.) with near infrared re¯ec-tance spectroscopy. J Jpn Soc Grassland Sci 35:228±233

Salisbury FB, Ross CW (1992) Plant physiology. Wadsworth,Belmont

Schimleck LR, Michell AJ, Vinden P (1996) Eucalypt wood clas-si®cation by NIR spectroscopy and principal componentsanalysis. Appita J 49:319±324

Schulz H, Losing G (1995) Quality control of essential oils. DtschLebensm Rundsch 91:239±242

Shenk JS, Westerhaus MO (1991a) Population de®nition sampleselection and calibration procedures for near infrared re¯ec-tance spectroscopy. Crop Sci 31:469±474

Shenk JS, Westerhaus MO (1991b) New standardization and cali-bration procedures for NIRS analytical systems. Crop Sci31:1694±1696

Shenk JS, Westerhaus MO (1991c) Population structuring of nearinfrared spectra and modi®ed partial least squares regression.Crop Sci 31:1548±1555

Shenk JS, Westerhaus MO (1993a) Near infrared re¯ectanceanalysis with single and multiproduct calibrations. Crop Sci33:582±584

Shenk JS, Westerhaus MO (1993b) Monograph: analysis ofagriculture and food products by near-infrared re¯ectancespectroscopy. Infrasoft, Port Matilda

Shenk JS, Westerhaus MO (1994) The application of near infraredspectroscopy (NIRS) to forage analysis. In: Fahey GC Jr,Mosser LE, Mertens DR, Collins M (eds) National conferenceon forage quality evaluation and utilization. American Societyof Agronomy, Madison, pp 406±449

Shenk JS, Westerhaus MO, Hoover MR (1979) Analysis of foragesby infrared re¯ectance. J Dairy Sci 62:807±812

Shenk JS, Workman Jr JJ, Westerhaus MO (1992) Application ofNIR spectroscopy to agricultural products. In: Burns DA,Ciurczak EW (eds) Handbook of near-infrared analysis. MarcelDekker, New York, pp 383±431

Smith KF, Flinn PC (1991) Monitoring the performance of abroad-based calibration for measuring the nutritive value oftwo independent populations of pasture using near infraredre¯ectance (NIR) spectroscopy. Aust J Exp Agric 31:205±210

Stenberg B, Nordkvist E, Salomonsson L (1995) Use of nearinfrared re¯ectance spectra of soils for objective selection ofsamples. Soil Sci 159:109±114

Van Kempen T, Jackson D (1996) NIRS may provide rapid eval-uation of amino acids. Feedstu�s 12:12±15

Vasquez de Aldana BR, Garcia-Criado B, Garcia-Criado A, Perez-Corona ME (1995) Estimation of mineral content in naturalgrasslands by near infrared re¯ectance spectroscopy. CommSoil Sci Plant Anal 26:1383±1396

Welsh LA, Blakeney AB, Reece JE, Ronalds JA (1996) Rice whole-grain milling yield by near-infrared transmittance. In: DaviesAMC, Williams P (eds) Near infrared spectroscopy: the futurewaves. NIR, Chichester, pp 601±603

Waterman PG, Mbi CN, McKey DB, Gartlan JS (1980) Africanrainforest vegetation and rumen microbes: phenolic com-pounds and nutrients as correlates of digestibility. Oecologia47:22±35

Windham WR (1987) In¯uence of grind and gravimetric techniqueon dry matter determination of forages intended for analysis bynear infrared re¯ectance spectroscopy. Crop Sci 27:773±776

Windham WR, Fales SL, Hoveland CS (1988) Analysis for tanninconcentration in Sericea lespedeza by near infrared re¯ectancespectroscopy. Crop Sci 28:705±708

Wright JA, Birkett MD, Gambino MJT (1990) Predictions of pulpyield and cellulose content from wood samples using near in-frared re¯ectance spectroscopy. Tappi J 73:164±166

305