Transforming nutritional monitoring of free-grazing herbivores: four decades of progress in fecal near-infrared reflectance spectroscopy

1 Introduction

Free-grazing herbivores, including both livestock and wildlife, are essential sources of food and provide significant economic, social, and environmental services to millions of people worldwide (Gordon et al., 2004; Thornton, 2010). However, the productivity of these free-grazing herbivores is increasingly challenged by climate change, shifts in land use, scarcity of feed and water resources, and the prevalence of disease (Norbury and Sanson, 1992; Dumont, 1997). In these systems, herbivores depend on native forage resources to meet their nutritional requirements, yet sustainable forage production has become a critical concern. Climate change, for example, is expected to heighten the vulnerability of both domestic and wild herbivores by reducing forage quality and quantity (Hidosa and Guyo, 2017) and impairing animal performance (Pavlů et al., 2006).

Low-quality forage is often associated with poor animal productivity (Coleman and Moore, 2003), increased migration risk, and elevated morbidity and mortality (Jackson et al., 2021). Moreover, when herbivores graze on low-quality rangelands, overgrazing often occurs, leading to degradation of natural resources (Mills et al., 2020). Understanding the quality and dietary attributes of forages consumed by herbivores is therefore crucial not only for effective nutritional management of grazing herbivores but also for the sustainable conservation of grazing lands, which are being exacerbated by rapid climate change.

Historically, forage quality has been assessed using various methods, including wet chemistry analysis of forage samples (Ball et al., 2001; Van Saun, 2006), visual evaluation of standing forage (Niamir-Fuller, 1998; Ball et al., 2001), and visual or physical assessment of animal body condition (Pearson and Ouassat, 2000). However, these approaches have notable limitations, particularly when applied in free-grazing environments. For example, wet-chemistry analysis requires hand-collected representative forage samples, which may not accurately reflect the actual diet selected by herbivores (Undersander et al., 1993). Forage composition and nutritional value vary widely across space and time (Norbury and Sanson, 1992). Herbivores exhibit selective grazing behaviors (Dumont, 1997), making it extremely difficult to replicate their diet in sampling, including plant species selection and bite size. In addition, herbivores’ dietary preferences can shift with physiological state, age, grazing experience (Dumont, 1997), plant community composition (Norbury and Sanson, 1992), and the presence of competitors or predators (Weterings et al., 2018).

Visual assessment of forage characteristics, including color, leafiness, maturity, and odor (Ball et al., 2001), is subjective and prone to bias. Climate change is expected to increase spatial and temporal variation in forage quantity and quality by shifting plant species composition, phenology, and the relative abundance of C3 and C4 species (Muzzo et al., 2025). These shifts can alter crude protein concentration, fiber fractions (ADF and NDF), lignification, digestibility, and concentrations of plant secondary compounds, thereby increasing nutritional heterogeneity across grazing landscapes and seasons (Martins-Noguerol et al., 2023; Muzzo et al., 2025). Consequently, obtaining representative forage samples for conventional wet-chemistry analyses is more challenging.

Similarly, body condition scoring relies on visual inspection and palpation of specific anatomical regions to assess body fat deposition (Pearson and Ouassat, 2000). Because it primarily reflects past nutritional status (Lyons, 1990), its interpretation may be influenced by factors such as workload, disease, or parasite burden (Kidane, 2005). These limitations make the process difficult to objectively quantify forage intake and diet quality in free-grazing herbivores.

To address these challenges, researchers have developed a low-cost alternative tool using fecal near-infrared reflectance spectroscopy (fecal-NIRS), an analytical technique that indirectly predicts forage quality from fecal samples (Brooks et al., 1984; Lyons and Stuth, 1992; Defourneaux et al., 2025). The core assumption underlying fecal-NIRS is that herbivore feces contain chemical bonds and functional groups (e.g., C-H, O-H, and N-H) derived from undigested residues and metabolic end products. These bonds generate NIR spectral information correlated with dietary chemical composition and digestibility (Landau et al., 2016; Parra-Forero et al., 2023). Fecal-NIRS models are developed by establishing mathematical relationships between dietary chemical composition and fecal spectral signatures (Landau et al., 2006). This relationship is based on the premise that fecal spectra encode information about diet composition and digestive processes through metabolic residues and undigested dietary constituents excreted from the gastrointestinal tract (Lyons and Stuth, 1992).

Since its introduction in the early 1980s, fecal near-infrared reflectance spectroscopy (fecal-NIRS) has evolved from a novel method for estimating dietary CP and DOM in grazing elk to a widely used technology to assess diet quality, digestibility, and feed intake across a broad range of domestic and wild herbivore species. Advances in spectroscopic instrumentation, calibration development, spectral preprocessing, and multivariate data analysis have improved predictive accuracy and expanded the range of applications (Matamura et al., 2024; Andueza et al., 2026). As a result, fecal-NIRS is now used to support nutritional monitoring, precision livestock management, rangeland assessment, and wildlife conservation. These developments have established fecal-NIRS as a valuable, noninvasive tool for evaluating the nutritional status of free-grazing herbivores. Building on these technological developments, researchers have developed fecal-NIRS calibration models for a wide range of domestic and wild herbivore species (Brooks et al., 1984; Lyons, 1990; Lyons and Stuth, 1992; Kidane et al., 2008; Jarque-Bascuñana et al., 2021). Calibration datasets have been generated using diverse approaches, including in vivo, in vitro, and in situ diet-fecal pairings, leading to differences in predictive accuracy and model robustness across studies. Despite these advances, a comprehensive synthesis of methodological progress remains limited.

Therefore, the objective of this review is to synthesize four decades of advances in fecal-NIRS technology, evaluate its effectiveness in predicting dietary and nutritional attributes across herbivore species, and identify emerging opportunities and challenges for future applications.

2 Literature search and selection

Literature for this review was identified through searches of Web of Science, Scopus, Google Scholar, CAB Abstracts, and PubMed. Search terms included combinations of “fecal NIRS,” “faecal NIRS,” “fecal near-infrared reflectance spectroscopy,” “diet quality,” “crude protein,” “digestibility,” “intake,” “forage quality,” “grazing livestock,” “small ruminants,” “cattle,” “sheep,” “goats,” “equids,” and “wildlife.” Studies published between 1984 and 2025 were included. Additional references were identified through citation tracking of relevant articles and reviews. Priority was given to peer-reviewed studies reporting calibration or validation statistics for predicting diet quality, nutrient composition, digestibility, or intake using fecal-NIRS. Conference abstracts, duplicate publications, and studies lacking sufficient methodological or calibration information were excluded. The final literature set encompassed domestic and wild herbivore species across diverse geographic regions and production systems. Early development of fecal-NIRS was driven largely by research groups at Texas A&M University and CSIRO Australia. As a result, studies from these institutions comprise a substantial portion of the foundational literature reviewed in this article.

3 Near-infrared reflectance spectroscopy: principles, calibration, and model development

Near-infrared reflectance spectroscopy (NIRS) is a non-destructive analytical technique that estimates the chemical composition of biological materials by measuring their interactions with near-infrared radiation (Shenk et al., 2001; Manley, 2014). The near-infrared region spans wavelengths from approximately 770 to 2500 nm; however, most agricultural and animal science applications focus on the 1100 to 2500 nm range (Deaville & Flinn, 2000; Skoog et al., 1998). Absorption in this region results from overtone and combination vibrations of molecular bonds, which originate from fundamental vibrations in the mid-infrared region (Pasikatan et al., 2001).

The analytical foundation of NIRS is based on the interactions between near-infrared radiation and specific molecular bonds. Rather than directly quantifying individual compounds, NIRS predicts chemical composition by analyzing the absorption characteristics of functional groups, primarily C-H, O-H, N-H, and C=O bonds. These bonds are prevalent in proteins, carbohydrates, lipids, fiber, and water (Norris et al., 1976; Van Kempen et al., 1996; Shenk et al., 2001). Exposure to NIR radiation induces stretching, bending, and wagging vibrations, producing characteristic absorption features that indicate chemical composition. For instance, O-H bonds in water absorb near 1450 and 1940 nm, while aliphatic C-H bonds absorb near 1210, 1400, 1725, and 2310 nm. Carbohydrates and structural polysaccharides display O-H absorptions between 1600 and 2100 nm, and protein-related N-H absorptions occur near 2055 and 2180 nm (Van Kempen et al., 1996; Workman and Shenk, 2004). Due to the abundance of these functional groups in forage and fecal materials, NIRS can predict crude protein, digestibility, fiber fractions, moisture, and energy value.

NIRS functions as a secondary or indirect analytical method that requires prior calibration. Calibration involves combining spectral measurements with reference values obtained from conventional laboratory analyses (Kneeboneb & Dryden, 2015; Kho et al., 2023; Andueza et al., 2026). Common chemometric methods include multiple linear regression (MLR), principal component regression (PCR), and partial least squares regression (PLSR) (Faber and Kowalski, 1997; Nikolich et al., 2001). Prior to calibration, spectra are mathematically preprocessed to reduce light scattering, particle-size effects, baseline shifts, and instrument noise (Yakubu et al., 2022; Hossain et al., 2024). Frequently used preprocessing techniques include first- and second-derivative transformations, smoothing, standard normal variate (SNV) correction, and detrending (Duckworth, 2004; Yakubu et al., 2022). The resulting calibration equations are then applied to predict values for unknown samples. Prediction accuracy depends on the quality of reference data, the representativeness of the dataset, and the model’s ability to capture biological and spectral variation (Westerhaus et al., 2004; Dixon and Coates, 2009; Xu et al., 2024). Calibration models are evaluated using the coefficient of determination (R²), standard error of calibration (SEC), and standard error of cross-validation (SECV). Models with high R² values and low prediction errors are generally preferred for routine application (Stuth et al., 2003; Landau et al., 2006). Although models with lower predictive accuracy may be unsuitable for precise quantification, they can still detect relative differences and monitor changes in chemical composition.

4 Fecal-NIRS: indirect analytical technique for forage quality assessment

The first fecal-NIRS calibration models were developed at the University of Alaska’s Agricultural Experiment Station to predict the diet quality of elk (Cervus canadensis) (Brooks et al., 1984). This technique uses spectral information from animal feces to infer the chemical and nutritional attributes of the animal’s diet (Brooks et al., 1984; Lyons et al., 1995; Stuth et al., 2003; Agnew et al., 2004). Fecal-NIRS models are constructed by regressing fecal spectral data against the chemical composition of diets determined by wet-chemistry analysis of paired diet-fecal samples (Figure 1). Figure 1 summarizes the principal procedures involved in developing, validating, and maintaining fecal-NIRS calibration models. Diet-fecal calibration equations are generated by relating dietary reference measurements to fecal spectral data obtained from in vivo or in vitro studies. Model development typically involves spectral preprocessing, chemometric calibration, validation, and quality-control procedures, with periodic recalibration and refinement using new datasets to maintain predictive accuracy and robustness.

Over the past four decades, several livestock nutritionists, including Lyons and Stuth (1992); Leite and Stuth (1995); Bruno-Soares et al. (1998); Ossiya (1999); Coates (2000); Gibbs et al. (2002); Awuma (2003); Landau et al. (2004); Li et al. (2007); Kidane (2005); Kidane et al. (2008); Villamide et al. (2009), Kneebone and Dryden (2014), Schiborra et al. (2015); Tolleson and Angerer (2021), Noel et al. (2022), Paternostre et al. (2021); Ikoyi and Younge (2022), and Molle et al. (2023), and wildlife nutritionists, such as Keating (2005); Showers et al. (2006); Rothman et al. (2009); Janova et al. (2015); Ikoyi and Younge (2022); Guerra et al. (2024), have developed models for numerous domestic and wild herbivores across diverse production systems and ecological regions. These fecal-NIRS models were developed to predict key dietary components, including CP, DOM, DMI, DMD, and OMI.

The following sections examine the evolution and performance of fecal-NIRS calibration models across diverse herbivore species and ecological regions. They also assess the model’s effectiveness in predicting key dietary and nutritional attributes and discuss its strengths, limitations, and emerging applications in both domestic and wild herbivores.

4.1 Fecal-NIRS for predicting crude protein and digestible organic matter in ruminants

Brooks et al. (1984) were the first to develop fecal-NIRS calibration equations for elk, producing highly accurate models for CP (R² = 0.99; SEC = 0.55) and DOM (R² = 0.88; SEC = 2.23). Building on this pioneering work, Lyons and Stuth (1992) developed fecal-NIRS models for cattle in Texas using extrusa samples collected from esophageal-fistulated steers and in vitro digestibility assays. The resulting models achieved high predictive accuracy for both CP and DOM across two study sites. Subsequent validation using samples from additional in vivo feeding trials further confirmed model performance, yielding an R² of 0.98 for CP and 0.87 for DOM. These findings demonstrated that fecal spectral signatures reliably capture variation in dietary quality, supporting the use of fecal-NIRS as a practical, noninvasive alternative to labor-intensive dietary assessment methods.

More recent cattle studies have continued to demonstrate strong predictive performance for dietary CP. Landau et al. (2016) reported a cross-validated R² (R²cv) of 0.89 and a SEcv of 13 g/kg DM for dietary CP in grazing beef cattle. Similarly, Johnson et al. (2017) reported R² and R²cv values of 0.90 and 0.88, respectively. Collectively, these studies indicate that protein-related dietary attributes are among the most accurately predicted variables using fecal-NIRS, likely because dietary nitrogen is strongly reflected in fecal chemical composition and can be measured using highly standardized laboratory reference procedures.

Following its successful application in cattle, fecal-NIRS was extended to small ruminants. Leite and Stuth (1995) applied the technology to estimate CP and DOM in goats grazing subtropical woodlands in Texas and reported high predictive accuracy for both traits. Additional validation trials yielded similar results, confirming the robustness of the calibration models. In the Middle East (Israel), Landau et al. (2004) evaluated goats grazing Mediterranean rangelands using controlled in vivo feeding studies and reported high predictive accuracy for both CP (R² = 0.98; SEC = 0.50) and DOM (R² = 0.98; SEC = 2.00). Comparable advances were reported in sheep. Li et al. (2007) developed fecal-NIRS models to predict CP and DOM using in vivo feeding trials and achieved R² values of 0.95 and 0.80, respectively, when Partial Least Squares regression was applied. Model performance declined slightly when Stepwise Regression was used, suggesting that calibration accuracy may be influenced by a model’s ability to capture complex relationships between spectral signals and dietary constituents.

Beyond North America and the Middle East, fecal-NIRS has been evaluated extensively in African grazing systems. Ossiya (1999) demonstrated high predictive accuracy for CP in free-ranging cattle and sheep across Ethiopia, Niger, and Nigeria. Building on this work, Awuma (2003) evaluated calibration models across five African countries and three ruminant species. Crude protein models consistently achieved high predictive performance across cattle, sheep, and goats, whereas DOM predictions exhibited greater variability. These findings suggest that CP can be reliably predicted across diverse production environments and species, whereas digestibility-related traits remain more challenging to model because they are influenced by multiple dietary, physiological, and microbial factors.

Australian studies have produced similar results. Coates, (2000, 2004) reported excellent predictive performance for CP and DOM in free-grazing and stall-fed cattle, while Gibbs et al. (2002) developed calibration equations for dietary CP and DMD in cattle fed forage-based diets with and without supplementation. Although CP prediction remained highly accurate, digestibility-related traits showed greater prediction error. Mahipala et al. (2010) further demonstrated the utility of fecal-NIRS for predicting dietary CP in browse-containing sheep diets, reporting calibration models with R² values exceeding 0.80. Together, these studies demonstrate that fecal-NIRS can maintain predictive accuracy across a wide range of forage resources and management systems.

In Europe, Bruno-Soares et al. (1998) reported high predictive accuracy for dietary CP in sheep fed mixed forage diets. More recently, in South America (Brazil), Dutra et al. (2026) developed bi-species fecal-NIRS calibration models for cattle and sheep, achieving high predictive accuracy for dietary CP (R² = 0.92; RPE = 8.91%). These findings indicate that calibration models may become increasingly transferable across ruminant species, potentially reducing the need to develop separate calibrations for individual species and production systems. Recent work in Italy has further expanded the applications of fecal-NIRS in small ruminants. Danese et al. (2025) evaluated the combined use of fecal-NIRS and internal markers, including undigested neutral detergent fiber (NDF), organic matter-corrected uNDF (uNDFom), and acid-insoluble ash (AIA), to estimate nutrient digestibility in dairy sheep and goats. However, prediction accuracies were generally moderate to low (maximum R² ≈ 0.63), indicating that current applications are better suited for nutritional screening than precise quantification of total-tract digestibility (Danese et al., 2025). Similarly, Guerra et al. (2024) demonstrated the potential of fecal-NIRS for predicting fecal composition and digestibility traits in lactating buffaloes, further expanding its applications in dairy production systems.

Studies conducted across North America, Africa, Australia, Europe, the Middle East, and South America demonstrate that fecal-NIRS consistently predicts CP with high accuracy across ruminant species and production environments (Table 1). Characteristics of the corresponding calibration datasets are presented in Table 2. In contrast, predictions of digestibility-related attributes such as DOM and DMD are generally more variable. Taken together, the available evidence supports the broad applicability of fecal-NIRS for assessing diet quality in ruminants. The consistently strong performance of CP models across species and environments suggests that protein-related attributes are robust calibration targets, whereas the greater variability observed for digestibility-related traits likely reflects the complex biological processes governing nutrient digestion and utilization. Consequently, CP remains the most reliably predicted dietary attribute and serves as a benchmark for evaluating the performance of fecal-NIRS calibration models.

Table 1. Summary of calibration statistics of crude protein (CP) and digestible organic matter (DOM) prediction models developed for various domestic and wildlife herbivores.

Some studies developed multiple CP calibration equations; therefore, fewer models appear in Table 1 than are discussed in the text”.

R², Coefficient of determination; SEC, Standard Error of Calibration; CP, Crude Protein; DOM, Digestible Organic Matter; NR, not reported, because the information was presented individually. To ensure a meaningful comparison of calibration performance across studies, only reports providing parameter-specific coefficients of determination (R²) and standard errors of calibration (SEC) were summarized. Studies that did not report individual calibration statistics for the predicted variables or alternative performance metrics were excluded from the table but are discussed in the text.

Table 2. Characteristics of calibration datasets used for developing fecal-NIRS models, including species, sample size, diet-fecal pairs, and diet composition.

CP, crude protein; DOM, digestible organic matter; NR, not reported.

^a Values in parentheses following animal numbers indicate sex and/or animal class or source of population when reported.

^b Values in parentheses following diet-fecal pairs indicate CP and DOM sample sizes, respectively, or calibration (cal.) and validation (val.) sample sizes when reported and after outliers removed.

^c Same n indicates that identical diet-fecal pairs were used for the development of both CP and DOM calibration equations.

^d Actual calibration sample sizes may differ from the total number of diet-fecal pairs due to outlier exclusion during calibration development.

^e Diet composition is reported according to the original studies. Species names are provided when available; otherwise, diets are described using the terminology reported by the original authors.

4.2 Fecal-NIRS models for predicting digestibility

Fecal-NIRS has also been used to predict forage digestibility, an important indicator of forage quality and animal performance. Lyons (1990) was among the first to develop fecal-NIRS calibration equations for predicting in vivo OMD of forages consumed by cattle, reporting a high R² of 0.95 and an SEC of 2.05. Subsequent studies expanded these applications across different forage types, animal species, and production systems. For example, Bruno-Soares et al. (1998) developed models to predict DMD and OMD of green cereal crops consumed by sheep and reported R² values of 0.86 and 0.88, respectively. In cattle, Lovett et al. (2004) evaluated the ability of fecal-NIRS to predict the in vitro digestibility of maize silage and reported a lower R² of 0.60, illustrating the greater variability often observed for digestibility-related traits. Similar findings were reported by Boval et al. (2004), who developed fecal-NIRS models to predict OMD in cattle fed two grass species. Model performance was moderate (R² = 0.72; SEC = 2.10), which the authors attributed to substantial inter-animal variation during the feeding trials. Collectively, these studies indicate that digestibility-related traits are generally more difficult to predict than dietary chemical constituents because animal-specific factors can introduce additional variation into calibration datasets. More recently, Dutra et al. (2026) reported satisfactory accuracy for predicting organic matter digestibility using bi-species calibration models developed for cattle and sheep (R² = 0.82; RPE = 5.02%). These results suggest that continued improvements in calibration development may enhance the prediction of digestibility-related traits.

Overall, prediction accuracy for digestibility measures such as DMD and OMD has generally ranged from moderate to high, although performance remains lower and more variable than that reported for CP. Despite these limitations, the available evidence supports the use of fecal-NIRS as a practical tool for estimating forage digestibility in grazing ruminants.

4.3 Fecal-NIRS models for predicting dry matter intake

Fecal-NIRS has been applied to estimate DMI and nutrient intake in ruminants consuming a wide range of diets and forage types. Early studies reported variable prediction accuracy. For example, Bruno-Soares et al. (1998) developed a calibration equation to predict DMI in cattle and reported relatively low predictive performance (R² = 0.41; SEC = 6.05 g/kg BW⁰·⁷⁵). Similarly, Boval et al. (2004) developed a fecal-NIRS model to predict OMI in cattle and achieved modest accuracy (R² = 0.61; SEC = 4.62 g/kg BW^0.75). In contrast, Agnew et al. (2004) reported moderate predictive performance for DMI of grazed forages, achieving an R² of 0.76 and an SEC of 0.37.

More recent studies have demonstrated substantial improvements in predicting intake-related traits. Peters et al. (2023) developed fecal-NIRS calibration models for predicting DM and OM intake in beef cattle and reported R² values of 0.89 for both variables, with corresponding SEC values of 0.92 and 0.89, respectively. Likewise, Molle et al. (2023) demonstrated that fecal-NIRS can accurately estimate forage intake and diet composition in Sarda dairy ewes under both confinement and grazing conditions. Under confined feeding conditions, DMI (herbage and hay) was predicted with high accuracy (R² = 0.90 and 0.97, respectively), highlighting the potential of fecal-NIRS for monitoring nutritional status and feeding behavior in pasture-based production systems.

Despite these advances, intake-related traits remain more difficult to predict than dietary composition and digestibility attributes. For example, Dutra et al. (2026) reported that a bi-species fecal-NIRS calibration model developed for cattle and sheep predicted OMI with relatively low accuracy (R² = 0.68; RPE = 22.15%). Intake is influenced by numerous factors, including age, physiological status, health, production stage, and feeding behavior, all of which contribute additional variation to calibration datasets. Consequently, prediction accuracy for DMI and OMI tends to be more variable than that reported for CP and other dietary quality attributes.

Overall, the available evidence indicates that fecal-NIRS can be used to estimate feed intake with moderate to high accuracy. However, model performance remains more variable than that observed for dietary composition traits, highlighting the need for larger and more diverse calibration datasets to improve prediction robustness.

4.4 Fecal-NIRS models for non-ruminant herbivores

Fecal-NIRS has also been applied to non-ruminant herbivores, including poultry, pigs, and equids. Smith et al. (2001) developed calibration equations for chickens and reported strong predictive accuracy for dietary CP (R² = 0.88; SECV = 0.185). However, digestibility-related attributes could not be predicted reliably. In pigs, fecal-NIRS has shown promise in assessing nutrient utilization. Schiborra et al. (2015) reported acceptable accuracy for predicting apparent CP digestibility (RPD = 2.4; SECV = 0.024), while Nirea et al. (2018), Noel et al. (2022), and Paternostre et al. (2021) further demonstrated the potential of fecal-NIRS for evaluating dietary and nutritional characteristics in swine production systems.

Applications in hindgut fermenter equids remain comparatively limited. Kidane et al. (2008) developed fecal-NIRS calibration equations to predict CP and DOM in equine diets using in vivo feeding trials with donkeys. The resulting models achieved high predictive accuracy for CP (R² = 0.97; SEC = 1.79) and moderate accuracy for DOM (R² = 0.87; SEC = 2.16). Similarly, Ikoyi and Younge (2022) developed fecal-NIRS equations to predict organic matter, neutral detergent fiber, and mineral concentrations in horse diets, reporting generally strong predictive performance. Furthermore, Martuzzi et al. (2024) demonstrated that fecal-derived measurements can serve as noninvasive indicators of forage utilization and digestive processes in horses through the evaluation of forage digestibility and fecal particle size distribution.

Results from equine studies indicate that fecal-NIRS can successfully predict forage quality in hindgut fermenters, although prediction precision is generally lower than that reported for ruminants. These differences likely reflect fundamental differences in digestive physiology. Unlike ruminants, hindgut fermenters rely on microbial fermentation in the cecum and colon (Kidane, 2005), which can alter fecal chemical composition through microbial transformation of dietary nutrients. This process may weaken the relationship between dietary composition and fecal nutrient concentrations, reducing calibration accuracy.

4.5 Fecal-NIRS models for monitoring the nutritional quality of wild herbivores

Fecal-NIRS has also been used to estimate diet quality in wild herbivores, including both ruminants and non-ruminants. Early studies in cervids demonstrated strong predictive performance for dietary CP and digestibility. For example, Showers et al. (2006) developed calibration equations for white-tailed deer fed mixed diets containing 50 forage species and reported R² values of 0.94 for CP and 0.89 for DOM. Similarly, Keating (2005) reported R² values of 0.95 for CP and 0.80 for DOM in Rocky Mountain elk fed a diverse range of grasses, forbs, and browse species. More recently, Tellado et al. (2015) developed and validated fecal Fourier-transform near-infrared spectroscopy (fecal-FT-NIRS) models for red deer (Cervus elaphus hispanicus) and fallow deer (Dama dama), reporting R² values ranging from 0.80 to 0.99 for organic matter digestibility.

Applications have also been extended to non-ruminant wildlife species. Rothman et al. (2009) developed CP calibration equations for mountain gorillas (Gorilla gorilla beringei) and reported high predictive accuracy (R² = 0.97; SEC = 0.48). In cecum-fermenting rabbits, Villamide et al. (2009) reported R² values of 0.84 for both CP and DMD prediction. Similarly, Janova et al. (2015) developed CP prediction models for the common vole (Microtus arvalis) and reported an R² of 0.85. These findings demonstrate that fecal-NIRS can be applied successfully across diverse non-ruminant herbivore species despite differences in digestive physiology.

Overall, studies on wildlife species demonstrate that fecal-NIRS can effectively predict protein- and digestibility-related components of herbivore diets. However, prediction precision varies considerably among species, as reflected by the wide range of reported SEC values. This variability likely reflects differences in digestive physiology, diet diversity, sample size, and fecal sampling conditions, highlighting the need for species-specific validation and expanded calibration datasets.

5 Discussion

5.1 Performance of fecal-NIRS models across herbivore species

This review demonstrates that fecal-NIRS has evolved from a novel method for estimating forage diet quality in grazing herbivores into a widely applicable tool for monitoring nutritional status across diverse domestic and wild herbivore species. Since its introduction in the early 1980s, technology has advanced considerably, with calibration models now developed for a wide range of species, production systems, and ecological regions.

This expansion of taxonomic and geographic scope builds on the earlier synthesis by Dixon and Coates (2009), which focused primarily on domestic ruminants in temperate and tropical production systems. The evidence reviewed here shows that fecal-NIRS has moved beyond its original livestock-centered application to become a versatile framework for assessing diet quality in domestic ruminants, hindgut fermenters, and free-ranging wildlife. Early studies primarily focused on predicting dietary attributes such as CP, digestibility, and intake (Lovett et al., 2004; Roberts et al., 2004; Agnew et al., 2004; Boval et al., 2004; Dixon and Coates, 2009). More recent efforts have extended model development to non-ruminant species, including equines, pigs, poultry, and wildlife. Of the 28 crude protein (CP) calibration models summarized in Table 1, most were developed for domestic ruminants (60.7%), including cattle (25.0%), sheep (21.4%), and goats (14.3%). Non-ruminants and wild herbivores accounted for 21.4% and 17.9% of the models, respectively. These studies demonstrate the broad applicability of fecal-NIRS across diverse herbivore species and production systems. Nevertheless, the limited number of models for non-ruminants and wild herbivores reflects the historical focus of fecal-NIRS research on domestic ruminants. This imbalance likely arises from the greater need for nutritional monitoring in free-grazing and pasture-based ruminant systems, where direct measurement of diet quality and intake is difficult. In contrast, pigs and poultry are typically raised under controlled feeding conditions with well-defined grain-based diets, reducing the need for indirect nutritional assessment via fecal-NIRS.

Model performance was typically evaluated using R² and SEC. Both metrics varied across herbivore species, dietary attributes, and experimental conditions. Fecal-NIRS models consistently exhibited high predictive accuracy for CP, with R² values ranging from 0.85 to 0.99 and average values of 0.96, 0.94, and 0.92 for domestic ruminants, equines, and wild herbivores, respectively. Fecal-NIRS models developed for poultry exhibited slightly lower accuracy (R² = 0.88) than those developed for ruminants. This reduced accuracy likely reflects uric acid’s influence on fecal nitrogen concentrations, which can weaken the relationship between dietary protein and fecal nitrogen and increase prediction error (Villamide et al., 2009). In general, SEC values for CP prediction remained within acceptable limits as defined by Stuth et al. (2003), indicating reliable estimation of dietary protein content. Recent work by Xu et al. (2024), which reported high predictive accuracy for fecal nitrogen-related components, further supports the robustness of CP prediction. The strong performance of CP prediction models likely reflects the accuracy and consistency of laboratory nitrogen determination methods, particularly the Kjeldahl procedure (Ball et al., 2001), which enhances the relationship between fecal nitrogen concentration and dietary CP.

Although predicting dietary CP has consistently achieved high accuracy, considerably greater challenges remain in predicting animal response variables such as DOM, DMD, and DMI. Models developed for these variables generally exhibited lower predictive accuracy (R² = 0.60-0.93) and greater variability than those developed for dietary chemical composition. This reduced performance likely reflects the complexity of digestive and metabolic processes, as well as physiological differences among animals. The prediction of intake and digestibility is complicated by factors such as age, physiological status, production stage, health, and feeding behavior, which increase variation in calibration datasets. Peters et al. (2025) similarly reported that, while intake can be predicted with moderate to high accuracy, prediction errors remain larger than those typically observed for dietary composition. Recent studies demonstrate continued improvement in these applications. Peters et al. (2023) developed fecal-NIRS models capable of predicting DMI and OMI in beef cattle with relatively high accuracy, while Xu et al. (2024) reported strong predictive performance for fecal chemical composition in dairy cattle. More recently, Peters et al. (2025) demonstrated that fecal-NIRS predicted nutrient digestibility with greater accuracy and precision than the internal markers acid detergent lignin (ADL) and undigestible neutral detergent fiber (uNDFom), highlighting its potential as a robust tool for assessing digestibility in forage-based beef production systems. Likewise, Tolleson et al. (2025) showed that fecal-NIRS can be effectively applied under grazing conditions, particularly when calibration models are developed using region-specific datasets.

Although most fecal-NIRS models have been developed for livestock species, growing evidence indicates that the technology can also be applied successfully to wild herbivores, albeit with greater calibration challenges. Models developed for wildlife often perform comparably to those developed for domestic species; however, they generally exhibit higher SEC values. These higher errors likely reflect challenges associated with collecting representative diet-fecal pairs under natural grazing conditions. Additional sources of variation include diverse diets, heterogeneous landscapes (Dumont, 1997), plant secondary metabolites, fecal weathering, and individual animal differences (Morgan et al., 2021; Stuth et al., 2003). Recently, Čupić et al. (2021) demonstrated that fecal nutrient composition in red deer was influenced not only by diet quality but also by season, pasture availability, and animal-related factors such as age and body weight. According to Čupić et al. (2021), seasonal effects accounted for up to 21% of the observed variability, highlighting the importance of considering environmental and physiological factors when interpreting fecal-NIRS predictions in free-ranging herbivores. Similarly, Čupić et al. (2023) reported significant species-specific differences in fecal nitrogen and fiber concentrations among three Saharan gazelle species that were maintained on similar diets. Consequently, wildlife calibration models often require broader datasets to adequately capture species-, environmental-, and diet-related sources of variation. The improved performance reported by Tolleson et al. (2025) further highlights the importance of region-specific approaches in minimizing prediction error under grazing conditions.

Beyond these species-specific applications, recent fecal-NIRS research has expanded the scope of prediction targets beyond CP and DOM to include individual nutrient intake, nutrient digestibility, and diet nutrient composition (Parra-Forero et al., 2023; Guerra et al., 2024; Dutra et al., 2026). Consequently, parameter-specific calibration statistics such as R² and SEC were often not reported, and model performance was instead assessed using cross-validation metrics, including R²cv, SECV, and RPE. This shift reflects the growing role of fecal-NIRS in precision nutritional management and its transformation from a nutritional assessment tool into a platform for monitoring nutrient intake and utilization, as well as grazing management, thereby supporting data-driven decision-making across livestock production systems.

Despite these advances, the predictive performance of fecal-NIRS remains influenced by many interacting factors, including diet composition, sample collection methods, animal characteristics, instrument performance, spectral preprocessing, and calibration methodology (Morgan et al., 2021; Xu et al., 2024). Statistical approaches such as partial least squares regression (PLS), principal component regression (PCR), and stepwise regression (SWR) can also affect model accuracy (Li et al., 2007; Morgan et al., 2021). One of the major challenges limiting broader adoption is the transferability of calibration models across species, production systems, ecological regions, and dietary environments. Many models are developed under specific environmental and biological conditions and may lose predictive accuracy when applied beyond their original calibration domain. This limitation underscores the importance of developing robust calibration datasets that capture a wide range of dietary, environmental, and animal-related variation.

Despite these limitations, fecal-NIRS remains a valuable and practical tool for monitoring the nutritional status of free-grazing herbivores when supported by robust calibration datasets. The technology provides an efficient means of detecting seasonal changes in forage quality and nutritional status, allowing producers and land managers to make timely management decisions. Information derived from fecal-NIRS can support improved supplementation strategies, enhance nutrient utilization efficiency, and optimize animal productivity in extensive grazing systems. Furthermore, the emergence of portable NIRS devices has expanded opportunities to move beyond laboratory-based analyses and support real-time nutritional monitoring under field conditions.

Looking ahead, advances in precision livestock technologies offer new opportunities to enhance fecal-NIRS. Emerging tools such as wireless sensor networks, RFID systems, and automated intake-monitoring technologies can generate high-quality animal-level data under both confinement and grazing conditions (Berckmans, 2014; Tedeschi et al., 2021). Integration of these technologies with fecal-NIRS has the potential to improve prediction accuracy, broaden its use across species and production systems, and accelerate adoption of precision nutritional management strategies. Because calibration performance remains sensitive to environmental and biological variation, continued model refinement and periodic recalibration will be essential for maintaining predictive accuracy across diverse grazing environments. Recent evidence suggests that calibration models may become increasingly transferable across species. For example, Dutra et al. (2026) demonstrated that bi-species calibration models developed for cattle and sheep can reliably predict dietary CP and digestibility. Although further validation with larger, more diverse multispecies datasets is needed, these results underscore the potential for broadly applicable calibration models that could reduce the costs and logistical challenges of maintaining species-specific calibrations.

5.2 Key challenges in calibration development, validation, and model transferability

A common misconception in fecal-NIRS applications is the belief that a single calibration model can be universally applied. However, this assumption does not hold across varying animal species, grazing environments, or production systems. Evidence from the literature indicates that calibration robustness depends on the representativeness of the calibration dataset (Dixon and Coates, 2009). Models developed under limited dietary or environmental conditions often perform poorly when applied outside their original calibration domain. Including samples that capture substantial nutritional and spectral variation improves both model robustness and transferability (Andueza et al., 2026). Therefore, future calibration efforts should focus on datasets that encompass a broad spectrum of forage types, seasons, geographic regions, management systems, and animal classes.

Ensuring biological diversity within the calibration population is also essential. The inherent heterogeneity of fecal samples increases spectral complexity and may limit the effectiveness of universal calibration models (Andueza et al., 2026). Biological variation associated with animal age, sex, breed, and physiological status can alter the chemical composition of feces (Hossain et al., 2024). These factors may reduce the accuracy of NIRS predictions unless they are adequately represented during calibration development (Hossain et al., 2024). In contrast, calibration datasets derived from a limited number of animals may not capture this variability, even when dietary diversity is present. For instance, repeatedly sampling feces from the same animals under different dietary treatments increases dietary variation but does not substantially enhance biological variation. Therefore, calibration development should incorporate both dietary and animal diversity. Including biologically meaningful outliers can broaden calibration ranges and improve prediction reliability across diverse grazing environments.

A core principle of predictive modeling is that model performance is contingent upon the quality of input data. Accordingly, fecal-NIRS calibration performance relies on the quality and representativeness of dietary and fecal reference samples used during model development (Paternostre et al., 2021). Fecal samples must be collected immediately after voiding and rapidly dried or frozen to preserve their chemical composition. Prolonged exposure in grazing environments allows sunlight, rainfall, wind, and microbial activity to alter fecal chemistry and modify spectral characteristics. Additional factors such as moisture loss, microbial degradation, oxidation, and nutrient leaching can further contribute to spectral variation prior to sample collection (Fortatos et al., 2023). Therefore, field-collected samples require meticulous handling to minimize these effects. Sample preparation procedures, including drying, grinding, particle-size standardization, and moisture control, significantly influence spectral quality and calibration accuracy. Variability in sample collection, preparation, and laboratory protocols introduces further sources of error and increases inter-study variation. This variability constrains the comparability, reproducibility, and transferability of calibration models. These challenges are especially pronounced when models are applied across different regions, species, production systems, or environmental conditions. In such contexts, inconsistencies in sampling and processing protocols may substantially reduce prediction accuracy.

Beyond biological variability, spectral data processing also plays a critical role in model performance. Common preprocessing methods include SNV, MSC, detrending, Savitzky-Golay derivatives, and BSC (Hossain et al., 2024). These techniques reduce spectral artifacts and improve calibration performance (Yakubu et al., 2022). The choice of preprocessing method depends on sample characteristics and the complexity of the target parameter. However, Andueza et al. (2026) cautioned that mathematical preprocessing cannot compensate for poorly designed calibration datasets, as dataset quality remains the primary determinant of calibration robustness and transferability. Furthermore, Hossain et al. (2024) noted that standardized sample preparation is essential because variations in particle size, moisture content, and sample homogeneity can negatively impact spectral quality and calibration reliability. Reinforcing the importance of these methodological considerations, Xu et al. (2024) reported that calibration performance benefits from appropriate spectral preprocessing, representative datasets, and independent external validation.

Another significant challenge is the limited use of independent external validation datasets. Many recent studies rely on cross-validation, in which a portion of the calibration dataset is withheld for model evaluation. Although cross-validation reduces analytical costs and labor, it does not fully assess model performance under independent field conditions. External validation provides a more rigorous assessment of predictive accuracy and practical utility. Recent research has shown that predictive performance is often substantially lower when models are tested on truly independent datasets than when using internal validation approaches (Andueza et al., 2026). Consequently, studies that omit external validation may overestimate model robustness and transferability.

Variability in dataset representativeness, reference method accuracy, spectral preprocessing techniques, validation procedures, and biological and population characteristics contributes to inconsistencies among studies. Collectively, these factors explain why some calibrations maintain predictive accuracy, whereas others decline substantially. Future research should emphasize standardized calibration protocols, diverse and representative datasets, spectral optimization, and rigorous external validation across animal classes and grazing environments. These measures will improve model robustness, transferability, and practical utility.

5.3 Prospectus for portable handheld NIRS instruments

Recent advances in sensor technology and data-processing capabilities have broadened the use of handheld near-infrared spectroscopy (NIRS) devices for rapid feed-quality assessment (Modroño et al., 2017; Bec et al., 2020). A growing range of portable and field-deployable NIR spectrometers is now available for livestock nutrition applications, including the DLP NIRscan Nano, ASD LabSpec 4, SCiO, and PHAZIR 1624 (Yakubu et al., 2022; Hossain et al., 2024). These instruments improve accessibility, reduce analytical costs, and provide real-time decision support, making them attractive tools for precision livestock management and field-based nutritional monitoring (Bec et al., 2020). Although their use in fecal-NIRS remains limited, handheld spectrometers offer considerable potential to advance nutritional assessment in free-grazing livestock systems. Importantly, although benchtop NIRS systems remain the standard for most laboratory applications, Yakubu et al. (2022) found that handheld devices often perform comparably to benchtop instruments on key feed and dairy quality traits. This finding suggests that handheld NIRS systems could support rapid in-field nutritional monitoring and facilitate broader adoption of fecal-NIRS across diverse grazing environments and production systems. Further research is needed to evaluate the accuracy, robustness, and calibration transferability of portable NIRS systems for diet-fecal calibrations across diverse species, forage conditions, and grazing environments.

5.4 Future prospectus for unprocessed fecal samples

Most fecal-NIRS studies have relied on dried and ground fecal samples to minimize spectral variation caused by moisture content and particle-size differences. However, advances in portable spectrometers have increased interest in using unprocessed feces for rapid field applications. Although relatively few studies have examined unprocessed or minimally processed samples, the reported results demonstrate considerable potential for in-field monitoring. For example, Allen et al. (2010) evaluated unprocessed cattle feces scanned in plastic bags with a portable spectrometer, and Becaccia et al. (2015) developed calibration models from unprocessed pig feces for nutritional and emissions-related attributes. Similarly, Preece et al. (2013) used portable visible-NIRS to predict the heating value of fresh cattle manure. Collectively, these studies demonstrate that unprocessed fecal samples can be used to predict diet quality indicators (CP and DM), fiber fractions (NDF and ADF), and health-related biomarkers. This capability may enhance the practicality of handheld NIRS spectrometers for field-based applications by reducing sample preparation requirements, processing time, and labor costs. However, Fortatos et al. (2023) cautioned that unprocessed fecal samples exhibit greater physical heterogeneity, including particles of varying sizes and colors, as well as larger void spaces. These characteristics increase light scattering and deviations from linearity, thereby reducing calibration accuracy. Thus, further research is needed to evaluate model robustness, calibration transferability, and environmental influences before unprocessed fecal samples can be routinely applied in field-based fecal-NIRS systems.

Despite the limited number of studies evaluating handheld near-infrared spectroscopy (NIRS) instruments with unprocessed fecal samples, these emerging approaches show considerable potential for field-based nutritional monitoring. The development of portable NIRS devices, combined with the effective use of unprocessed or minimally processed fecal samples, represents an important step toward real-time nutritional assessment in grazing livestock. Further research is needed to strengthen calibration robustness, standardize field protocols, and validate model performance across diverse production environments. These advances could facilitate the integration of fecal-NIRS with other precision livestock technologies and support future Nutritional Health Index (NHI) frameworks for monitoring nutrition and health in free-grazing livestock and wildlife herbivores.

5.5 New frontiers in fecal-NIRS: integration with precision livestock technologies and the nutritional health index

The future of fecal-NIRS lies in its integration with precision livestock technologies, expanding its role from diet assessment to comprehensive nutritional monitoring. By combining fecal-NIRS with RFID/EID systems, GPS tracking, remote sensing, decision-support platforms, and advanced sensors, nutritional information can be interpreted within a broader management context. RFID/EID systems enable individual-animal nutritional profiling by linking fecal-NIRS predictions to production, reproductive, and health records (Tzanidakis et al., 2023). These data can inform precision grazing management. Likewise, decision-support tools such as NutBal can translate diet-quality estimates into recommendations for supplementation and grazing management (Tolleson et al., 2014; Tolleson et al., 2025).

GPS-based monitoring further enhances interpretation by linking dietary quality estimates to animal movement, habitat use, and forage selection patterns (Hassan-Vásquez et al., 2022; Nyamuryekung'e et al., 2023). Similarly, remote sensing technologies can characterize forage biomass, vegetation composition, and nutritional heterogeneity across grazing landscapes (Shahi et al., 2025; Dias and Borba, 2026; Pinto et al., 2026). Together, these technologies strengthen the spatial and ecological context for fecal-NIRS predictions.

Recent advances in handheld near-infrared spectroscopy (NIRS) further expand fecal-NIRS applications by enabling rapid, field-based nutritional assessment. Portable instruments have achieved prediction accuracies comparable to those of benchtop spectrometers for several key nutritional attributes (Rukundo et al., 2021; Yakubu et al., 2022; Yamada et al., 2024). Although moisture variability and sample heterogeneity can affect some predictions, handheld devices offer greater portability, lower costs, and more frequent monitoring. Their adoption could reduce dependence on centralized laboratories and enable near-real-time assessment of the nutritional status of grazing herbivores. Consequently, handheld systems may substantially improve the scalability of fecal-NIRS applications.

More importantly, the convergence of fecal-NIRS with these technologies could create opportunities for integrated nutritional assessment systems. Such systems can combine dietary, behavioral, physiological, and production information within a unified decision-support framework. One promising example is the Nutritional Health Index (NHI) proposed below.

5.5.1 Nutritional health index: a conceptual framework for integrated nutritional monitoring

We propose a conceptual Nutritional Health Index (NHI) that integrates fecal-NIRS-derived estimates of diet quality with behavioral, physiological, and production metrics (Figure 2). Diet quality is derived from fecal-NIRS predictions; grazing behavior from GPS monitoring; rumination activity from wearable sensors; rumen function from intraruminal sensing technologies; and animal performance from indicators such as body weight, body condition score, growth rate, reproductive performance, and milk production.

By integrating these complementary data streams, the NHI provides a more comprehensive assessment of nutritional status than diet quality alone. The index could also serve as an early-warning indicator of nutritional deficiencies, forage limitations, digestive disturbances, and declining animal performance. Integration with automated sensing technologies would enhance real-time monitoring under commercial grazing conditions. Ultimately, the NHI could function as a quantitative decision-support tool for animal-specific nutritional management, improving resource-use efficiency, productivity, animal welfare, and the sustainability of grazing herbivore systems.

6 Implications

Effective management of free-grazing herbivores requires timely information on their nutritional status, which is largely determined by the quality and quantity of nutrients obtained from grazed forages. Access to accurate and timely information on forage quality is therefore essential for making informed management decisions, including determining stocking rates, designing supplementation programs, and optimizing herd composition and size.

Meeting these information needs under free-grazing conditions has historically been challenging. Fecal-NIRS has substantially improved nutritional assessment by providing rapid, noninvasive, and reliable estimates of diet quality and nutritional status. Over the past four decades, the technique has demonstrated particularly strong predictive accuracy for dietary CP and has also provided useful estimates of intake- and digestibility-related attributes. These capabilities enable producers, researchers, and land managers to make informed feeding and grazing decisions, reduce the risk of nutritional deficiencies, and improve resource-use efficiency. However, the effectiveness of fecal-NIRS depends on robust calibration models developed under conditions representative of the target species and production environment. Despite the initial costs of calibration development and instrumentation, fecal-NIRS remains a practical, safe, and scalable technology. Continued refinement of calibration models and integration with precision livestock technologies will further transform fecal-NIRS from a nutritional assessment method into a data-driven platform for precision nutritional management of domestic and wild grazing herbivores.

7 Conclusions

Over the past four decades, fecal-NIRS has evolved from an experimental technique into a widely used tool for assessing diet quality and nutritional status in free-grazing herbivores. This review evaluated the effectiveness of fecal-NIRS in predicting key dietary and nutritional attributes, including CP, DMI, DOM, DMD, and OMD. Across diverse ecological regions and herbivore species, fecal-NIRS models consistently showed high accuracy in predicting dietary CP for both ruminant and non-ruminant herbivores. In contrast, predictions of digestibility and intake were less accurate and more variable, reflecting differences in digestive physiology, diet composition, environmental conditions, and sampling procedures. Although most studies have focused on livestock species, existing research demonstrates considerable potential to extend fecal-NIRS applications to wild herbivores. Broader adoption across species such as bison, giraffe, and zebra could enhance nutritional and ecological monitoring in systems where wildlife and livestock share grazing resources.

Overall, fecal-NIRS is a robust, scalable, and increasingly valuable technology for assessing diet quality in both domestic and wild herbivores. Future progress will depend on improving predictions of complex traits such as intake and digestibility, enhancing calibration transferability across species and ecological regions, and integrating fecal-NIRS with precision livestock technologies. Advances in portable NIRS and the use of unprocessed fecal samples are creating new opportunities for rapid nutritional assessment under grazing conditions. Future integration with GPS, RFID, rumen boluses, wearable sensors, and remote sensing technologies may facilitate the development of Nutritional Health Index systems for precision livestock management. Together, these advances will further transform fecal-NIRS into a platform for precision nutritional management and data-driven decision-making in grazing systems.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Giovanni Buonaiuto, Parmigiano Reggiano Consortium, Italy

Reviewed by: Pichad Khejornsart, Kasetsart University, Thailand

Tommaso Danese, University of Parma, Italy

Footnotes

BCS, body condition score; CP, Crude protein; DM, Dry matter; DMI, Dry matter intake; DMD, dry matter digestibility; DOM, digestible organic matter; NDF, neutral detergent fiber; NIRS, near-infrared reflectance spectroscopy; OMD, organic matter digestibility; OMI, organic matter intake; SEC, standard error of calibration; SECV, standard error of cross validation; R², coefficient of determination; R²cv, coefficient of determination cross validation; RPE relative prediction error; SEP, standard error of prediction.