Field applications of near infrared spectroscopy for wildlife ecology.

INTRODUCTION
Near infrared spectroscopy (NIRS) has been used in many applications. The earliest involved predicting moisture and protein content in grains. Bench-top instruments and main-frame computers made the technique available as a research tool, but not until the personal computer became available did NIRS gain acceptance for industry and field applications. The advent of portable instruments and tablet computers have made in situ spectroscopy a reality for the field biologist. Most NIRS work to date has involved the “traditional” process of collecting spectra and reference information on the same sample and developing a “direct” calibration with which to predict constituents of interest in “unknown” samples. Variations include “indirect” calibrations; pairing spectra from feces with reference chemistry on the corresponding diet. Qualitative applications such as determining age, sex or other physiological characteristics have been developed. We have applied spectroscopy in a variety of analytical situations with wildlife species, or domestic species as a surrogate for wildlife. Selected results from this work are presented here.
EXPERIMENTAL
We conducted direct chemical constituent NIRS calibrations of plant material consumed by Giant Panda (Ailuropoda melanoleuca), bighorn sheep (Ovis canadensis) and domestic livestock. In addition, we conducted pen-feeding calibration trials to determine forage diet quality via fecal NIRS in white-tailed deer (Odocoileus virginianus), elk (Cervus canadensis), cattle (Bos spp.), sheep (Ovis aries) and goats (Capra hircus). We subsequently applied these calibrations in field studies with bison (Bison bison), yak (Bos grunniens), red deer (Cervus elaphus) fallow deer (Dama dama), mule deer (Odocoileus hemionus) and pronghorn antelope (Antilocapra americana). Additional experiments were conducted to discriminate fecal samples from animals differing in species, age, sex or physiological status. Spectral measurements have also been collected directly from live animals differing in hair or skin characteristics. Calibrations have been primarily achieved using partial least squares multivariate regression techniques.
RESULTS AND DISCUSSION
Calibration performance statistics (R2; standard error of cross validation [SECV]) for percent crude protein (CP) have ranged from 0.64;0.36 in native US range plants consumed by bighorn sheep, to 0.98;0.34 in native US range plants consumed by livestock. Similar results in four species of bamboo were 0.95;0.35. Fecal NIRS calibration statistics (R2;SECV) for domestic livestock and ungulate wildlife diet CP are typically 0.9;1.0. Mahalanobis distance values for ungulate wildlife fecal samples predicted with surrogate livestock species have ranged from <1.0 to > 10.0 and are primarily influenced by similarity in plant species of the respective diets. Discriminant identifications of age, sex, and species in ungulates ranged from ~50% to > 90% successful, depending on species. Similar results have been observed for discriminant calibration predictions of pregnancy or parasite status in livestock via fecal NIRS. Determination of three hair fiber diameter groups was 80% successful in live goats; coat color (80%) and burn scar status (60%) was discriminated in live cattle. Field portable NIRS instruments offer great possibility for the field biologist to increase the spatial and temporal scope of experimental designs. Spectra based inferences must be carefully evaluated within the context of overall habitat and species of interest characteristics.