Animal Forage Applications
Introduction
The animal forage industry is constantly examining new methods for the analysis of nutritional and chemical composition of products. Traditional methods are often expensive and time-consuming. They are ill-suited to measure large numbers of samples and especially for determining variation within a batch of animal forage because most methods can only measure a small portion of sample. Other disadvantages include the inability of most techniques to perform real-time, on-line measurements during the manufacturing process and being unable to determine more than one parameter from a single measurement. In recent years, NIR spectroscopy has emerged as a technique that offers many advantages over traditional analytical methods. It is fast, non-invasive, and able to determine multiple parameters of interest from a single measurement. It requires no chemical or toxic solvents and can not only measure large amounts of samples but can measure a large sample amount in one reading. This is especially important when accounting for variation within a batch, as animal forage and other natural products can often show large amounts of variation. Forage particularly can show a large amount of variation among the same products due to numerous factors. These include herbage species, nutrient availability, stage of maturity, topography, and climate conditions.
The ability of NIR spectroscopy for use in measuring parameters in animal forage has been proven for some time. However, recent advances in technology have helped facilitate its use as a complete process analytical tool from the analysis of raw materials in the field to a final quality control check. Handheld spectrometers can be used for field measurements and to check incoming batches of raw materials before entering the manufacturing process. In types of forages where a manufacturing or mixing process is involved (such as the fermentation of crops to make silage), on-line instruments can be used as a real-time process control tool to optimize the manufacturing process, resulting in improved product quality, optimization of resources, and potentially significant cost savings. It can also be used to both detect and quantify adulterants in animal forage, which is a significant problem with potential health and economic consequences. An examination and review of applications using NIR spectroscopy in the animal forage industry is presented here. Reviewed topics include the use of the technique to determine the chemical components, nutritional value, and digestibility parameters of pasture and forage, characterization of biomass quality, the effects of both modeling algorithms and product variation on the use of NIR spectroscopy as a quality control tool in forage, prediction of gas production kinetics of forage, and a comparison of traditional benchtop and portable NIR spectrometers for forage analysis.
Analytes
- Crude Protein (CP)
- Crude Fat (CF)
- Neutral Detergent Fiber (NDF)
- Acid Detergent Fiber (ADF)
- Water-Soluble Carbohydrates (WSC)
- Ash
- Ether Extract (EE)
- Crude Fiber (CF)
- Milk Forage Unit (UFL)
- Meat Forage Unite (UFV)
- Cellulose
- Hemicellulose
- Total Carbohydrates
- Lignin
- Extractives
- In Vitro Organic Matter Digestibility (IVOMD)
- In Vitro True Dry Matter Digestibility (IVTD)
- Gas Production Kinetic Parameters of Forages
- Gross Energy (GE)
- Buffering Capability (BC)
Scientific References and Statistics
Determination of Forage Quality by Near-Infrared Reflectance Spectroscopy in Soybean
Soybean is an important annual legume crop for both human and animal consumption. They are rich in edible protein and oil and are a reliable source of forage during the summer, especially when other legumes have been harvested or are unavailable. In some parts of the world such as Asia, the production of high-quality forage legumes has been challenged by difficult environmental conditions, high costs, and carbon footprint issues. Soybean can offer an alternative to alfalfa and other legumes that possess acceptable energy content and nutrient digestibility for use as forage while also reducing costs and environmental impact in its harvesting. The basic quality components of forage soybean are crude protein (CP), crude fat (CF), acid detergent fiber (ADF), and neutral detergent fiber (NDF). Standard wet chemistry reference methods used to determine these parameters are time-consuming and expensive. They require the use of toxic chemicals and solvents and are ill-suited to measure both large numbers of samples and more than a single parameter of interest with one test. NIR spectroscopy offers the advantages of being fast and non-invasive. It does not require the use of toxic chemicals and solvents and little or no sample preparation is needed. The method can measure both large amounts of samples in a reasonable time and multiple chemical and physical parameters of interest from a single light measurement. In this study, NIR spectroscopy was evaluated to determine the feasibility of using it to determine CP, CF, ADF, and NDF in soybean forage.
Three hundred and fifty-three forage soybean samples were procured for the study. All samples were processed from planting to harvesting to cleaning using standard methods. After sample processing, all samples were ground in a mill and passed through a 1 mm sieve. Standard wet chemistry methods were used to determine reference values for CP, CF, ADF, and NDF. A portion of each sample was scanned with an NIR spectrometer from 400 nm to 2498 nm at 2 nm intervals. Each sample was scanned twenty times in total and the scans were averaged into a single spectrum per sample. Before chemometric modeling, various pre-processing algorithms were applied to the spectral data. The NIR spectra of some of the samples were not included in the chemometric modeling and were saved for use as an independent validation set. Partial Least Squares (PLS) calibration models were created that correlate each parameter of interest to the NIR spectra. Modeling results are shown below.
| CP (%) | R2 = 0.922 | SEP = 0.912 |
| CF (%) | R2 = 0.942 | SEP = 0.537 |
| NDF (%) | R2 = 0.848 | SEP = 2.053 |
| ADF (%) | R2 = 0.749 | SEP = 1.557 |
Models were validated using both cross-validation and by using the NIR spectra of the independent validation set to predict the parameters of interest. Correlation coefficient values were high and prediction error was low for both CP and CF. Good agreement was shown between the reference values and predictions using the chemometric models for both parameters. NIR spectroscopy is a proven method for protein and fat and the results shown here are comparable to numerous other studies using many types of food and agricultural products. Results were not as good for NDF and ADF and this is also comparable to other studies where these four parameters were analyzed using NIR spectroscopy. Despite this, the calibrations can be considered good enough for screening purposes. The models could potentially be improved by using a larger sample set with a wider range of values for the fiber parameters.
Development and validation of near-infrared spectroscopy for the prediction of forage quality parameters in Lolium multiflorum
| CP (% DM) | R2 = 0.99 | RMSECV = 0.68 |
| NDF (% DM) | R2 = 0.94 | RMSECV = 2.23 |
| ADF (% DM) | R2 = 0.92 | RMSECV = 1.68 |
| WSC (% DM) | R2 = 0.88 | RMSECV = 3.11 |
Modeling results were excellent and proved the feasibility of the application for all four parameters of interest, especially for CP. The independent validation set prediction values were in good agreement with the reference values for all parameters. This study showed that using NIR spectroscopy to predict values for CP, NDF, ADF, and WSC could provide a fast, non-invasive method for use in ryegrass breeding programs. Benefits could include facilitating the monitoring of the nutritional dynamics in the forage development stage and identifying the optimal utilization period of forage grasses.
Use of FT-NIRS for determination of chemical components and nutritional value of natural pasture
The nutritional value of pasture forage can vary greatly based on multiple factors, including herbage species, nutrient availability, stage of maturity, topography, and climate conditions. The continuous compositional change of natural pastures presents inherent challenges when analyzing the nutritional and chemical composition of animal forage. Traditional wet chemistry methods are time-consuming, expensive, and require the use of toxic chemicals and solvents. They can also only measure one parameter of interest per test. These shortcomings are amplified when performing quality control on natural products that often have a large amount of variability in their composition. NIR spectroscopy offers the advantages of being fast and non-invasive. Little to no sample preparation is required and once calibration models are created that correlate the NIR spectra with chemical and nutritional parameters of interest, multiple parameters can be measured from a single light reading. In this study, a FT-NIR spectrometer was used with pasture samples in Tuscany to assess the feasibility of using NIR spectroscopy to determine the nutritional value and chemical composition of natural pastures.
One hundred and five samples were procured from pastures in Tuscany for the study. Multiple mountain and hill areas were used to collect the samples with the goal of incorporating as much variability as possible in the parameters of interest. In total, thirteen different species were collected among the samples. A portion of each sample was dried and ground for wet chemistry reference tests. Tests were performed to determine reference values for the following chemical parameters: dry matter (DM), crude protein (CP), ash, ether extract (EE), crude fiber (CF), acid detergent fiber (ADF), neutral detergent fiber (NDF), and acid detergent lignin (ADL). The nutritional value parameters Milk Forage Unit (UFL) and Meat Forage Unit (UFV) were also determined. These terms basically translate to an energy value of forage for milk and meat production in cows.
The remaining sample portions were scanned using an FT-NIR spectrometer from 9999 cm-1 to 4000 cm-1 at 16 cm-1 resolution. Thirty-two scans were collected per reading and averaged into a single spectrum. This process was repeated three times for each sample and the three spectra per sample were further averaged into one spectrum. Before chemometric modeling, various pre-processing algorithms were applied to the spectral data. Partial Least Squares (PLS) models were created that correlate each parameter of interest to the NIR spectra. Two separate approaches were used for model validation. The first used 80% of the samples for the calibration models and the remaining 20% as an independent validation set. The second used cross-validation, which used the entire sample set but removed one sample and created separate models to predict the reference value for the removed sample. Results are shown below.
| DM | R2 = 0.993 | RMSEP = 0.973 |
| CP | R2 = 0.992 | RMSEP = 0.592 |
| Ash | R2 = 0.965 | RMSEP = 0.33 |
| EE | R2 = 0.982 | RMSEP = 0.137 |
| CF | R2 = 0.960 | RMSEP = 1.470 |
| ADF | R2 = 0.988 | RMSEP = 1.140 |
| NDF | R2 = 0.990 | RMSEP = 1.330 |
| ADL | R2 = 0.983 | RMSEP = 0.500 |
| UFL | R2 = 0.825 | RMSEP = 0.101 |
| UFV | R2 = 0.812 | RMSEP = 0.112 |
Modeling results were excellent for all chemical parameters. All correlation coefficients were above 0.96. Both validation methods showed good agreement with the reference values and predictions made from the NIR spectra and calibration models. In the case of UFL and UFV, the predictive capability of the models was much lower. There are a number of possible reasons for this. The calculation for these parameters required an estimation of organic matter digestibility from the ADF values. This estimation could have introduced error into the reference method. Because these parameters are not a direct chemical or physical parameter and thus do not have a direct absorption in the near-infrared region of the spectrum, the calculation is an indirect result of a mathematical derivation. While indirect correlations can be used in NIR spectroscopy, they must be carefully examined and validated before real-time use. A larger sample set with more variability may improve results for UFL and UFV. The results of this study proved the feasibility of using NIR spectroscopy to measure chemical composition of pasture forage.
Investigating the Impact of Biomass Quality on Near-Infrared Models for Switchgrass Feedstocks
NIR spectroscopy is a proven method for the characterization of biomass composition. It has been used for fast and non-invasive analysis of perennial herbaceous species in various bioenergy conversion processes. These include classification of plants, compositional properties of switchgrass, and projected performance parameters such as ethanol yield. Switchgrass is a major feedstock for potential use in bioenergy and thermochemical conversion processes, especially in the Southeastern United States. Efficient storage of switchgrass is a major challenge for large facilities. Dry matter loss can lead to degradation and consumption of carbohydrates during storage from fungal growth and oxidation. Manufacturers are constantly working on new methods to improve storage conditions of switchgrass. While NIR spectroscopy can be used to measure chemical and physical quality control parameters in switchgrass, there has been minimal study on how the storage of switchgrass and subsequent chemical changes can affect the NIR spectra and the predictive capabilities of calibration models. In this study, NIR spectroscopy was used to examine the effects of storage time on both the NIR spectra and calibration models that are used to predict the chemical composition of switchgrass.
Before the study began, a design of experiment was used to optimize the incorporation of variability in the samples based on multiple factors, including storage type, sample particle size, and storage time. It was determined that using samples that varied in other factors with the following storage time values would help optimize the calibration models: 0 days, 75 days, 150 days, and 225 days. In total, one hundred and thirty samples were procured for the study. A portion of each sample was used to determine reference values for each of the following parameters: cellulose, hemicellulose, total Carbohydrates, lignin, extractives, and ash. After the reference values were collected, all samples were ground and scanned from 350 nm to 2500 nm at 4 nm spectral resolution using an NIR spectrometer. Forty scans were collected per reading and averaged into one spectrum. Before chemometric modeling, various pre-processing algorithms were performed on the NIR spectra. Principle Component Analysis (PCA) was used to determine grouping patterns in the samples based on storage time and other factors. Partial Least Squares (PLS) calibration models were created to correlate the NIR spectra to the parameters of interest. Separate groups of models were made for samples that were not stored, individual models for each separate storage time, and models that incorporated all samples. In order to assess the effectiveness of the predictive capabilities of the models for predicting samples with storage times were not incorporated into the calibration, predictions were performed in multiple ways to see how the error varied based on the type of sample predicted and the samples incorporated into the model.
The statistics and various conclusions shown from the study were extensive but can be summarized with a few conclusions. PCA showed a clear grouping between samples that were not stored and stored (regardless of the storage time). The PLS model for no storage time samples showed good correlation but when the NIR spectra of samples that had been stored were used for validation predictions, the results were poor. Prediction results were better for both samples that were stored and not stored using the models that were created from all sample types. The main conclusion that can be drawn from the results of this study is that the storage of switchgrass samples does have a profound effect on the chemical composition and thus the NIR spectra as well. In order to use NIR spectroscopy as an effective method for the chemical composition determination of biomass, the selection of appropriate calibration samples that incorporate all possible sources of variability (especially storage time) and the subsequent creation of calibration models must be carefully examined and validated before use in a real-time setting.
Use of Near Infrared Reflectance (NIR) Spectroscopy to Predict Chemical Composition of Forages in Broad-Based Calibration Models
NIR spectroscopy has shown immense potential as an alternative to traditional wet chemistry methods for quality control analysis of forage crops. The ability to test a large number of samples in a relatively short period of time can result in substantial improvements in the efficiency of the breeding process. However, there are many factors that can affect the nutritive value of forage crops. Climatic conditions, stage of plant maturity, and species & origin type are some of the most affecting factors. Calibration models that are used to correlate NIR spectra to chemical and physical parameters of interest can be affected by sources of variation in natural products like forage. One approach is to create universal models that incorporate NIR spectra of samples that contain all these sources of variability. Another is to create local models that are specific to the different variation factors, such as separate models for different forage species. In this study, a wide range of forages from Uruguay were collected from field experiments for the purpose of assessing the effects of variation on chemometric models that correlate NIR spectra to quality control parameters.
A total of six hundred and fifty forage samples were procured for the study. The samples represented a wide range of species, origin, and growth stages and were collected over five consecutive seasons. A total of fourteen different species were collected. Tall fescue was by far the species with the most samples at two hundred. All other species had a sample number between twenty and fifty. All samples were dried and ground before standard reference tests were performed. The following quality control parameters were determined: dry matter (DM), crude protein (CP), ash, in vitro organic matter digestibility (IVOMD), acid detergent fiber (ADF), and neutral detergent fiber (NDF).
After reference tests were performed, a portion of each sample was scanned with a NIR spectrometer. Samples were scanned from 400 nm to 2500 nm at 2 nm intervals. Thirty-two scans were collected per reading and averaged into one spectrum. Before chemometric modeling, various pre-processing algorithms were applied to the spectral data. Partial Least Squares (PLS) regression models were created that correlate the NIR spectra to the reference values. A set of universal models was created first that incorporated all samples. Subsequently, local models were created for both different species and for samples from different growing seasons. The local models were assessed by comparing the statistical values with the universal model and seeing if prediction results were both more accurate than the universal model. Predictions were also done using the local models on different sample types (such as predicting a different species type using models made using only fescue samples). A summary of results is shown below:
| Universal Models (% DM basis) | ||
| DM | R2 = 0.95 | SEP = 0.96 |
| CP | R2 = 0.95 | SEP = 1.8 |
| Ash | R2 = 0.95 | SEP = 2.2 |
| ADF | R2 = 0.95 | SEP = 3.8 |
| NDF | R2 = 0.95 | SEP = 8.2 |
| IVOMD | R2 = 0.95 | SEP = 5.7 |
Statistical analysis of the local models is not shown but the following summary is presented. Prediction results using the universal models are comparable with similar studies. Results were worse for NDF than the other parameters and one possible reason for this is error in the reference method. Based on results of other studies, it is suggested that using in vivo OMD instead of in vitro may improve correlation and prediction values. In general, the local models showed similar results to the universal model when predicting samples of the same type. When using a local model from one species to predict samples from a different species, DM, Ash, and CP showed little difference in prediction results compared to using the universal models. ADF, NDF, and IVOMD showed considerably more prediction error. When using a local model from one season to predict samples from a different season, DM, CP, ADF, and IVOMD showed little difference in prediction results compared to using the universal models. Results were adequate for Ash and poor for NDF. The results from this study showed great potential to use NIR spectroscopy for quality control analysis of forage but also warrant further study to examine the effects of sample variation in calibration models.
Prediction of Gas Production Kinetic Parameters of Forages by Chemical Composition and Near Infrared Reflectance Spectroscopy
In vivo digestibility predicts animal response to dietary intake but is insufficient for describing the dynamics of nutrient supply. Methods for determining this parameter are very expensive and ill-suited for testing large numbers of samples. In vitro and in situ methods have been used in nutrition studies for ruminants, which do provide information on fermentation kinetics. Still, these studies are very much animal dependent and require the acquisition of rumen fluid from the animals as well as the determination of relevant quality control parameters in the forage using traditional wet chemistry methods. NIR spectroscopy is a proven method for the measurement of quality control parameters in animal forage, such as dry matter, ash, crude protein, and fiber & lignin values. However, analysis on the use of NIR spectroscopy for evaluating fermentation kinetics and gas production parameters in forages is very limited. In this study, NIR spectroscopy was evaluated to determine the potential of using it to predict in vitro gas production parameters of botanically complex herbage and forage samples.
A total of ninety-four herbage samples harvested from meadows in mountainous regions of Northwest Spain were procured for the study. Samples were harvested both after the spring primary growth and after the summer & autumn secondary re-growth. The different harvesting dates resulted in samples at very different stages of maturity. Numerous different species types were obtained. All samples were oven dried and ground before chemical analysis, scanning with an NIR spectrometer, and the determination of in vitro gas production.
Standard reference tests were used to determine values for dry matter (DM), ash, crude protein (CP), neutral detergent fiber (NDF), acid detergent lignin (ADL), and acid detergent insoluble N (ADIN). Based on CP and NDF content, the samples were divided into two sets to obtain similar means and standards for these two parameters. Sixty-two samples were used as a calibration set and the remaining thirty-two were saved for validation. A standard technique was used to create in vitro gas production. Three sheep were fit with a rumen cannula and fed alfalfa hay. Rumen fluid was obtained prior to morning feed. After inoculation, gas production was measured at the following hour intervals after inoculation: three, six, nine, twelve, sixteen, twenty-one, twenty-six, thirty-one, thirty-six, forty-eight, sixty, seventy-two, ninety-six, one hundred and twenty, and one hundred and forty-four. This process was done twice for each sample. After the final gas production measurement, contents were filtered and oven dried to estimate DM loss at one hundred and forty-four hours. Based on the obtained data, gas production profiles were created to estimate the following fermentation parameters: GP (gas production at each incubation time measured in ml/500 nm DM, A (asymptotic gas production in ml), c (fractional rate of fermentation in /h), and L (lag time before fermentation started in h). From these values, the defined rumen passage rate (EDk) can be estimated.
All herbage samples were scanned from 1100 nm to 2500 nm at 2 nm intervals using a NIR spectrometer. Each sample was scanned twice in duplicate repacking, resulting in four NIR spectra per sample. These four spectra were averaged into a single spectrum per sample. Before chemometric modeling, various pre-processing algorithms were applied to the spectral data. Partial Least Squares (PLS) calibration models were created that correlated both the physical parameters and gas production parameters to the NIR spectra. Equations were created for the estimation of in vitro gas production parameters.
While the statistical analysis was too extensive to be documented fully here, some conclusions can be drawn based on the prediction results from using the NIR spectra of the validation set samples. Good correlation was shown for most chemical parameters of the samples. Some error was shown for NDF and this is likely because of differing botanical composition of the samples. There was good agreement between the correlation of the chemical parameters and gas production parameters with the exception of lag time (L). The chemometric models for gas production parameters were considered acceptable for estimation of the parameters, despite many potential sources of experimental error. Using chemometric modeling for prediction of gas volume at each incubation time did not improve the accuracy of the gas production parameter predictions. Overall, the results of this study demonstrated great potential for the use of NIR spectroscopy to create gas production and fermentation kinetic profiles for forage, but more work and study are necessary before using this method in a real-time setting.
Use of LOCAL algorithm with near infrared spectroscopy in forage resources for grazing systems in Colombia
The advantages of NIR spectroscopy over traditional wet chemistry methods for analysis of animal forage are proven and vast. The method is fast, non-invasive, and does not require the use of toxic chemicals and solvents. Little to no sample preparation is required and unlike wet chemistry, NIR spectroscopy is well-suited to both measure large amounts of samples in a reasonable amount of time and can determine multiple physical and chemical parameters from a single light reading. However, there are inherent challenges in the creation of calibration models that correlate the NIR spectra to physical and chemical parameters of interest. While NIR spectroscopy is a proven method for measuring quality control parameters in forage, agricultural products often have variation based on other factors that create differences in NIR spectra that are not related to the parameters of interest. There are multiple approaches that can help account for such variation. Creating universal models that contain samples that account for all sources of variation is one approach. Another is creating local models that are specific to species, origin, maturity time, and other factors that can affect NIR spectra. A more systematic approach is the use of various algorithms for chemometric modeling that can improve results. One such algorithm is known as LOCAL, which is designed to select specific samples in the calibration models that have NIR spectra which resemble the spectra of the samples being analyzed. In this study, the GLOBAL Partial Least Squares and LOCAL modeling algorithm were compared to determine which approach was better suited for the prediction of quality control parameters in forage using NIR spectroscopy.
Forage grasses and legumes play an important role in feeding systems for meat and milk cattle in many countries, especially those where production of cereal grains is limited and alternative feeds are required. Colombia is one such country and different environmental conditions greatly affect the nutritional composition of forages. Seasonal variations can have a strong effect as well. Wet chemistry methods are ill-suited to analyze large numbers of samples that can provide information on the dynamics of necessary decision making to optimize farm strategy to maximize milk and meat production in cattle that are fed forage. In order to assess the use of NIR spectroscopy for this purpose, two thousand and twenty samples were collected over a course of five years. Three separate geographic regions were characterized and samples were collected during both dry and rainy seasons during separate stages of growth and regrowth. Three groups of samples were collected: one thousand four-hundred and eighteen grass forage, three hundred and twenty legume forage, and two hundred and eighty-two other types of forage plants. All samples were dried in an oven and ground. Standard reference tests were performed for the following parameters: dry matter (DM), crude protein (CP), crude ash (CA), neutral detergent fiber (NDF), acid detergent fiber (ADF), and acid detergent lignin (ADL). After reference testing, all samples were scanned from 400 nm to 2498 nm at 2 nm intervals using a NIR spectrometer. Duplicate spectra were collected for each sample and averaged into one spectrum. Before chemometric modeling, various pre-processing algorithms were applied to the spectral data. Multiple model sets were created using all samples and specific models for each sample type using both the GLOBAL and LOCAL algorithms. Only the results for the entire database for all samples are shown below. However, the improvement in the results for the LOCAL selective algorithm using all samples is generally indicative of the improvement for the specific sample type models as well.
| GLOBAL | ||
| DM % | R2 = 0.65 | SEP = 1.23 |
| CP (% DM) | R2 = 0.98 | SEP = 0.99 |
| CA (% DM) | R2 = 0.78 | SEP = 1.08 |
| NDF (% DM) | R2 = 0.91 | SEP = 3.21 |
| ADF (% DM) | R2 = 0.87 | SEP = 2.62 |
| ADL (% DM) | R2 = 0.78 | SEP = 0.96 |
| LOCAL | ||
| DM % | R2 = 0.81 | SEP = 0.92 |
| CP (% DM) | R2 = 0.99 | SEP = 0.87 |
| CA (% DM) | R2 = 0.86 | SEP = 0.89 |
| NDF (% DM) | R2 = 0.95 | SEP = 2.40 |
| ADF (% DM) | R2 = 0.95 | SEP = 1.71 |
| ADL (% DM) | R2 = 0.88 | SEP = 0.71 |
The results clearly show that the LOCAL algorithm with selective sample selection in the calibration models improved results. It offers the advantage of being robust enough to create a new calibration equation for each sample, making it unnecessary to identify the particular sample type that needs to be analyzed. Results were relatively poor for DM in all model types but this is likely due to changes in moisture in the samples over a period of time and a small range of values. DM and moisture are proven parameters that can be measured using NIR spectroscopy because water is highly absorbing of near-infrared light. Protein, ash, and fiber measurements all showed good correlation and low prediction error. The results of this study demonstrate the enormous potential of using NIR spectroscopy for forage quality analysis in a way that minimizes the effects of variation in sample type and environmental conditions in different types of forage samples.
Performance of two handheld NIR spectrometers to quantify crude protein of composite animal forage and feedstuff
NIR spectroscopy is a proven method for the prediction of Crude Protein (CP) in animal forage. It offers numerous advantages over traditional wet chemistry methods. It is fast and non-invasive with little to no sample preparation. Unlike wet chemistry, NIR spectroscopy does not require the use of toxic chemicals or solvents. Traditionally, the use of NIR spectroscopy is done with a benchtop laboratory instrument in a controlled environment. In recent years, technological advances in optics and electronics have enabled the development of portable and handheld NIR spectrometers. They are designed for ease of operation and have reduced space and energy requirements. Portable instruments vary in many ways, including cost, size, weight, power requirements, durability, accuracy, and overall performance. The ideal choice for a portable or handheld spectrometer can vary greatly based on the particular application. Instrument portability can greatly aid model calibration development, especially for natural products as the ability to move an instrument to various locations enhances the scope of samples that can be tested. There is a need to assess different types of portable NIR spectrometers for different applications in order to select the instrument that is most suited for a specific application. In this study, two different kinds of portable NIR spectrometers were evaluated and compared for the purpose of using them to measure CP in animal forage. One instrument was a portable handheld spectrometer and the other was a smartphone spectrometer.
The handheld spectrometer has a wavelength range of 780 nm to 2500 nm, an interval of 1 nm, and weighs approximately 2.5 kg. The smartphone spectrometer has a wavelength range of 900 nm to 1700 nm, an interval of 4 nm, and weighs 136 g. One hundred and forty-seven forage and feed samples were procured for the study. Sample types included sweet bran, corn silage, corn stalks, and three different kinds of corn distiller grains. All samples were dried, ground, and CP was determined using a standard reference method. A remaining portion of each sample was packaged in 0.08 mm thick polypropylene bags. Each sample was scanned with both spectrometers over the full wavelength range. Fifty scans were collected per reading and averaged into one spectrum. Two spectra were collected per sample with the second spectrum collected after flipping the bag over. After visual examination of the spectra from both instruments, the wavelength ranges were truncated for both sets of data. 780 nm to 2500 nm was used for the handheld spectrometer and 940 nm to 1660 nm. This was done to reduce the noise in the NIR spectra. Before chemometric modeling, various pre-processing algorithms were applied to the spectral data. The data was divided into two sets. One hundred and twenty samples were used in a calibration set for the chemometric models and the remaining twenty-seven samples were set aside for use as a validation set. Partial Least Squares (PLS) calibration models were created using the data from both spectrometers to correlate the NIR spectra to CP. Modeling results are shown below.
| Handheld | ||
| CP | R2 = 0.96 | RMSEP = 2.22% |
| Smartphone | ||
| CP | R2 = 0.97 | RMSEP = 2.05% |
The results for both types of spectrometers were comparable and slightly better for the smartphone instrument, although the differences in correlation coefficient and RMSEP are small enough to be statistically insignificant. Calibration models often have more error and lower performance with a reduced wavelength range and interval, but this was not the case when comparing the two instruments in this study. The potential was demonstrated for using portable NIR spectrometers as a tool for monitoring CP in animal feed and forage. These instruments can help improve the control and management of animal feeding programs. One factor that must be considered when analyzing the results is that the samples used in this study were both dry and ground, which controlled both moisture and particle size. In order to use this application in a real-time setting, the sensitivity of differences in moisture and particle size in animal feed and forage samples must be examined.
Comparison of benchtop and handheld near-infrared spectroscopy devices to determine forage nutritive value
The development of accurate and robust calibration models is an essential step when implementing NIR spectroscopy for the measurement of chemical and physical parameters in natural products. Natural products can show variation in the NIR spectra that are due to many factors that are not related to changes in the constituents of interest. Numerous studies have been conducted that assess the development of calibration models for the analysis of animal forage. Work and study are limited for the comparison of analytical results for prediction of forage nutritive value when NIR spectrometers with different spectral ranges and resolutions are used. Such analysis is especially important when considering technological advances and the advent of the focus of NIR spectrometers shifting from benchtop and laboratory use to both process and portable instruments. Handheld NIR spectrometers have enormous potential as a tool for field analytical work that can greatly aid farm management and crop strategy. In this study, three different NIR spectrometers were used to create calibration models for quality control parameters in forage grass samples. One instrument is a benchtop instrument and the other two are handheld instruments.
A total of two hundred and ten samples were procured for the study. One hundred and thirty-eight samples were switchgrass and seventy-two were Bermuda grass. The samples were cultivated over a two-year period. Standard reference tests were performed for the following quality control parameters: crude protein (CP), acid detergent fiber (ADF), neutral detergent fiber (NDF), and in vitro true dry matter digestibility (IVTD). The three NIR spectrometers used in the study all had different wavelength ranges and intervals. The benchtop instrument has a range from 1100 nm to 2498 nm at 2 nm intervals. Handheld instrument #1 has a range from 1600 nm to 2400 nm at 8 nm intervals. Handheld instrument #2 has a range from 900 nm to 1700 nm at 5 nm intervals. Samples were scanned using all three spectrometers over the full wavelength range. Before chemometric modeling, various pre-processing algorithms were applied to the spectral data. Partial Least Squares (PLS) calibration models were for all three sets of data that correlated the NIR spectra to the reference values. Modeling results for all three spectrometers were shown below.
| Benchtop | ||
| CP | R2 = 0.99 | SEC = 5.0 g/kg |
| aNDF | R2 = 0.90 | SEC = 16.2 g/kg |
| ADF | R2 = 0.84 | SEC = 21.8 g/kg |
| IVTD | R2 = 0.99 | SEC = 16.9 g/kg |
| Handheld #1 | ||
| CP | R2 = 0.98 | SEC = 7.5 g/kg |
| aNDF | R2 = 0.89 | SEC = 17.1 g/kg |
| ADF | R2 = 0.84 | SEC = 21.6 g/kg |
| IVTD | R2 = 0.97 | SEC = 26.3 g/kg |
| Handheld #2 | ||
| CP | R2 = 0.97 | SEC = 8.6 g/kg |
| aNDF | R2 = 0.91 | SEC = 15.3 g/kg |
| ADF | R2 = 0.84 | SEC = 22.1 g/kg |
| IVTD | R2 = 0.96 | SEC = 30.8 g/kg |
The values shown above were the best results from each model using the different pre-processing algorithms and spectral transformations. Predictions were performed using both cross-validation and a data split using 75% of the data to create calibration models and the other 25% as an independent validation set. In general, the chemometric model statistics and prediction results were similar for the three different instruments. The benchtop instrument did show slightly better results when taken as a whole, but the differences were not vast especially when considering the variation in wavelength range and interval in between the three instruments. It is likely that the optimization of the calibration models from the NIR chemometric software was a significant contributor to similar modeling and predictive results. Overall, it can be concluded that any reduction in performance from handheld instruments is outweighed by the benefits presented by these spectrometers based on the results of this study. However, it is recommended that further comparative study be conducted using a larger and more varied sample set before implementing the results found here in a real-time setting.
Prediction of the chemical composition and nutritive value of lucerne (Medicago sativa L.) by Near Infrared Spectroscopy
Lucrene is a low input energy efficient crop that can improve the fertility of soil. It is a significant animal forage product in some parts of the world. Italy is one such area and lucerne is particularly significant in the cheese producing areas of Northern Italy. Because there can be significant variation in the chemical composition and nutritive value of forages based on environmental conditions and other factors, there is a need for rapid, non-invasive testing of new cultivars for determining adaptation and quality traits. NIR spectroscopy is a proven method for the prediction of chemical composition, digestibility, and other nutritional characteristics in forages in other foods. Ensilability is defined as the suitability of converting plant material into compost. Two parameters in forage which are important for good ensilability are water soluble carbohydrates (WSC) and buffering capability (BC). In this study, NIR spectroscopy was assessed to confirm the ability to use it for fast assessment of quality control parameters in lucerne that encompass a wide range of maturity and growth stages, including ensilability characteristics.
Two cultivars of lucerne were procured over two consecutive growing seasons for the study. The herbage was harvested at a five cuts per year schedule from early spring until late fall. In order to provide large variability in the quality parameters, four to five cuts were performed at morphological stages progressing from early vegetative to late flowering. In total, three hundred and two samples were used in the study. A portion of each sample was dried in an oven and another portion was dry frozen in storage. The dried portion of each sample was used to perform standard reference tests for the following parameters: ash, crude protein (CP), neutral detergent fiber (NDF), gross energy (GE), and organic matter digestibility (OMD). The water extract from the frozen samples was used to determine values for WSC and BC. The remaining portion of each sample was scanned from 1098 nm to 2500 nm using a NIR spectrometer. An algorithm was used to assess spectral variability in the samples and based on the results, the software selected two hundred samples to be used as a calibration set and the remaining one hundred and two samples as a validation set. Various pre-processing algorithms were performed on the spectral data before chemometric modeling. Partial Least Squares (PLS) chemometric models were created that correlate the quality control parameters of interest to the NIR spectra. Results are shown below.
| Ash (% DM) | R2 = 0.98 | SEC = 0.40 |
| CP (% DM) | R2 = 0.96 | SEC = 0.55 |
| NDF (% DM) | R2 = 0.95 | SEC = 1.55 |
| OMD (%) | R2 = 0.89 | SEC = 2 |
| GE (MJ/kg DM) | R2 = 0.98 | SEC = 0.40 |
| WSC | R2 = 0.81 | Not Given |
| BC | R2 = 0.78 | Not Given |
Modeling results were excellent for the physical and chemical quality control parameters. Prediction results from the independent validation set proved the accuracy of the models. In the case of the ensilability parameters WSC and BC, correlation was lower but the results still indicate that the models were accurate enough to be used for screening purposes. A critical point in this study is the wide variation of the lucerne samples associated with different cultivars, year of cultivation, and harvest time. Considering the relatively small sample set and multiple sources of variability, the results showed that NIR spectroscopy has the ability to predict feeding and nutritive value in lucerne and provide relevant information for ensilability parameters.
The accuracy of NIRS in predicting chemical composition and fibre digestibility of hay-based total mixed rations
In Italy, eighteen percent of the total Bovine milk production is used for Parmigiano Reggiano cheese production. This cheese is one of the most traded Italian dairy products. It is manufactured from the clotting of bovine milk that is produced by cows with a Total Mixed Ration (TMR) hay-based diet without the addition of any silage. Proper nutritional requirements of TMR are crucial for the optimization of the productivity and efficiency of dairy cows. Thus, proper management requires frequent analysis of rations and necessary corrections. The variability of feed quality, particularly of forages, presents inherent challenges for farm managers. Traditional methods for analyzing forage quality are often ill-suited to measure large amounts of samples and account for variation. They often require the use of toxic chemicals and solvents with extensive sample preparation and can only measure a single chemical or physical parameter of interest with one test. NIR spectroscopy offers the advantages of being fast and non-invasive while requiring little to no sample preparation. It can test large amounts of samples in a reasonable amount of time and can determine multiple parameters of interest from a single light reading. In this study, the effectiveness of NIR spectroscopy was assessed for the prediction of the chemical composition and fiber undegradable fractions (uNDF) in TME for hay-based rations that are fed to dairy cows.
The TMR samples used in the study were sampled from different trials over a five-year period from six commercial dairy farms in Italy. All samples were collected from three distinct positions (beginning, middle, and end) in the manger after morning preparation of the TMR. The farms used a standard feeding system that fed cows every twelve hours. TMR mainly consisted of dried alfalfa and dried grass without the presence of silage. In total, two hundred and five samples were used in the study. A portion of each sample was used to perform standard wet chemistry tests to determine values for the following parameters: crude protein (CP), starch, acid detergent fiber (ADF), acid detergent lignin (ADL), ash, and fiber digestibility. The remaining portion of each sample was scanned from 900 nm to 2500 nm using a FT-NIR spectrometer. Thirty-two scans were collected per reading and averaged into one spectrum. Various pre-processing algorithms were applied to the spectral data before chemometric modeling. Partial Least Squares (PLS) chemometric models were created that correlate the parameters of interest to the NIR spectra. Results are shown below.
| Starch (% DM) | R2 = 0.87 | RMSECV = 2.83 |
| CP (% DM) | R2 = 0.80 | RMSECV = 0.826 |
| ADF (% DM) | R2 = 0.83 | RMSECV = 2.37 |
| ADL (% DM) | R2 = 0.76 | RMSECV = 0.71 |
| Ash (% DM) | R2 = 0.57 | RMSECV = 0.64 |
| uNDF (% DM) | R2 = 0.79 | RMSECV = 2.60 |
Modeling results were decent but higher correlation and lower prediction error has been shown in other similar studies. In the case of ash, there was certainly some error introduced with either the sampling, reference error, or both. NIR spectroscopy is a certified method for the measurement of ash and other studies have shown a correlation coefficient as high as 0.99. The results for other parameters were reasonable. The potential was demonstrated to use NIR spectroscopy as a cost-effective and fast method for the determination of compositional and digestibility traits in TMR. Use of this method can support nutritionists in diet and feed formulation to create nutrient profiles that can maximize conversion of the digested feed and milk output in dairy cows.