Harty, A. A., Cow-Calf Field Specialist, SDSU Extension
Olson, K.C., Beef Extension Specialist, SDSU Extension

ABSTRACT

Fecal Near Infrared Reflectance Spectroscopy (NIRS) and the Nutrition Balance Analyzer (NUTBAL) system analysis are widely utilized in the Natural Resource Conservation Service Conservation Stewardship program; however producers in the program in South Dakota have questioned reliability of results. In 2013-2014, 7 ruminally cannulated steers were used to collect diet and fecal samples. Fecal samples were analyzed using fecal NIRS at the Grazingland Animal Nutrition Laboratory (GANLAB) in Temple, TX, while the diet samples were analyzed using wet chemistry methods at the Ruminant Nutrition Laboratory at North Dakota State University (NDSU). Following analysis, it was determined that fecal NIRS did not accurately predict crude protein (CP) and digestible organic matter (DOM) of South Dakota cattle diets. A 1:1 ratio did not exist for the regression analysis relating predicted to actual values for either CP or DOM. Additionally, the NUTBAL analysis for predicting animal performance consistently predicted lower ADG than was achieved by cattle grazing alongside the cannulated steers. These results are similar to other comparisons of NIRS/NUTBAL predictions to actual diets and cattle performance conduced in other states.

Introduction-Issue

The Conservation Stewardship Program (CSP) through the USDA Natural Resource Conservation Service (NRCS) provides tools for producers to monitor management decisions and improve conservation practices on land, livestock and water. 

Over the life of the CSP program, numerous producers have selected Animal Enhancement Activity ANM65-Monitoring nutritional status of ruminant livestock using Near Infrared Reflectance Spectroscopy (NIRS) and the Nutrition Balance Analyzer (NUTBAL) system on fecal samples (USDA NRCS Enhancement Activity 65, 2014). This enhancement utilizes NIRS analysis of fecal samples to predict dietary crude protein (CP) and digestible organic matter (DOM; an estimate of energy content of the diet). These predictions are inputs to the NUTBAL online program to predict whether the current diet is sufficient to meet cattle nutritional needs, to predict cattle performance, and to develop least cost nutritional management plans for cattle grazing forages. The NIRS/NUTBAL program was developed for NRCS by range scientists at Texas A&M University in the 1990s and has been offered to agricultural producers in various NRCS programs since then. To participate in the program, enrolled producers collect samples of fresh cattle feces from pastures of interest at important times throughout the year and send them to the Texas A&M University Grazingland Animal Nutrition Laboratory (GANLAB) at Temple, Texas for NIRS and NUTBAL analyses. Using results of these analyses, producers should be able to make management decisions such as determining the need to provide supplemental protein or energy. 

In 2012, producers from western South Dakota enrolled in this enhancement received reports from the GANLAB and questioned the accuracy of the predictions of diet quality and cattle performance. They asked for assistance from NRCS county staff in interpreting results, and the county staff in turn requested assistance from SDSU Extension. This evaluation raised additional questions regarding prediction accuracy and how well the program works with South Dakota forages. In some cases, the program was predicting daily weight losses exceeding 3 lbs, but producers were not observing losses when monitoring body condition. 

Producer concerns about predicted animal performance, as well as extreme variation in results, motivated this case study to evaluate the validity of the existing fecal NIRS and NUTBAL system for northern mixed prairie rangelands in South Dakota. A 2-year project evaluated how well fecal NIRS predictions compared to actual dietary nutrient content. Performance predicted by the NUTBAL program was also compared to actual steer performance. The hypothesis tested in this study was that the model used in the fecal NIRS program was capable of predicting diet quality values of South Dakota forages.

 

Methods

South Dakota State University, North Dakota State University (NDSU), United States Department of Agriculture-Agricultural Research Service (USDA-ARS), and Sitting Bull College conducted a grazing study on native rangeland in north-central South Dakota during the summers of 2013 and 2014 (Olson, et al., 2016). Vegetation on the site was typical of northern mixed prairie rangeland. It was comprised of a diverse mixture of native species dominated by western wheatgrass (Pascopyrum smithii Rydb.), green needlegrass (Nassella viridula Trin.), needle-and-thread (Hesperostipa comata Trin. and Rupr), blue grama (Bouteloua gracilis Willd. Ex Kunth), buffalograss (Bouteloua dactyloides Nutt.), and sedges (Carex spp.). In this study, 7 ruminally cannulated steers were used to collect diet samples to determine nutrient content (Lesperance et al., 1960; Olson, 1991). Grazing ruminants are highly selective so their diets are always nutritionally superior to clipped forage samples. Therefore, diet sample collection using cannulated animals is considered the best research tool available for evaluating grazing livestock diets. All steers were cannulated as yearlings in 2013 and used in both years. This study provided an opportunity to compare fecal NIRS predictions of nutrient content to actual diets. Fecal samples were collected from the rectum of cannulated steers at the same time diet samples were collected. Diet and fecal sampling was conducted monthly beginning in June and ending in August of each year. Diet and fecal samples were frozen immediately after being collected. Diet samples were analyzed at the Ruminant Nutrition Laboratory at NDSU to determine CP content and in vitro organic matter digestibility (IVOMD, an estimate of energy content of the diet that is synonymous with DOM). Fecal samples were sent to the GANLAB in Texas for fecal NIRS analysis and generation of the NUTBAL report.

To determine if fecal NIRS and NUTBAL provided an accurate and reliable prediction of actual CP, IVOMD/DOM, and steer average daily gain (ADG), regression analysis was used to statistically evaluate the predictive relationship between the results from the fecal NIRS and NUTBAL report with actual diets and steer performance. The regression relationship between years was examined to detect a year effect using repeated measures analysis in the Mixed Procedure of SAS (SAS Institute, Cary, NC). Actual value (CP, IVOMD, ADG) was the dependent variable. The corresponding fecal NIRS predicted value, year, and their interaction were independent variables. If year and interaction were not significant, then data were pooled across years and the simple regression of actual on predicted was evaluated using the regression procedure of SAS. If year or the interaction were significant, then a separate linear regression was developed for each year. Within each linear regression, the r² value was evaluated to determine how much of the variation in the relationship between fecal NIRS predictions and actual values could be explained. The r² value can range from 0 to 1, with 0 meaning there is no relationship and 1 meaning there is a perfect fit. For fecal NIRS predictions to be considered accurate and useful, a 1:1 relationship between predicted and actual values should exist. The regression line should have a slope of 1 (i.e. the actual value and the fecal NIRS prediction would be the same without adjustment) and the intercept of the regression line should be 0 (i.e. 0 should be predicted when 0 is the actual value). A hypothesis test was constructed to test if slope was different from one. The test of the significance of the intercept estimate was used to evaluate if it was different from zero.

 

Results

Crude protein. The relationship between predicted and actual CP was statistically similar across years (P > 0.05), meaning that all data could be combined into one regression analysis (Figure 1). This outcome means the predictive relationship had consistent value across years and should have similar predictive value in the future. The r² for the regression equation was 0.78, meaning 78% of the variation in actual dietary CP could be explained by the predicted fecal NIRS values. The predictive relationship is reasonably strong. The regression slope was 0.70, which was not statistically similar to 1 (P < 0.001). The intercept was 4.1, which was not statistically similar to 0 (P < 0.001). Thus, there was not a 1:1 relationship between actual and predicted values for CP. For example, if fecal NIRS predicts dietary CP of 9.5%, one cannot assume that equates to actual dietary CP of 9.5%. In this example, the actual CP value from the diet sample would be 10.76% after adjusting the predicted value using the regression equation. Thus, any other attempted recommendations would be cumbersome because they would require applying the regression equation to the predicted values to obtain accurate estimates of actual dietary CP. For the remaining NUTBAL predictions and nutritional management recommendations to be valid, this regression relationship would need to be 1:1.

 

Figure 1. Regression of actual dietary crude protein on fecal NIRS prediction of dietary crude protein to validate ability of fecal NIRS to predict actual dietary crude protein. Coefficient of variation (r2) estimates proportion of variation in actual values explained by predicted values. R² values range from 0 to 1 with those closer to 1 being better. The regression intercept should be 0 and slope should be 1 for a 1:1 relationship between predicted and actual values. Intercept and slope differ from 0 and 1 (P < 0.05), respectively.

 

In vitro organic matter digestibility. The regression relationship for IVOMD was not consistent across years (i.e. year interacted with the prediction of IVOMD [P = 0.02], indicating the regression relationship for 2013 was different from the 2014 relationship. Because results were not consistent across years, the capacity to confidently use the equations in future years is limited. Differing regression relationships are contrasted in Fig. 2. For 2013, the r² value indicated that the model explained about 56% of the variation, which was less than desirable. However, for 2013, the intercept (-7.7) was statistically similar to 0 (P = 0.60) and the slope (1.17) was statistically similar to 1 (P = 0.49), approaching a 1:1 predictive relationship. In 2014, the r² value of 0.85 was greater, but the intercept (-73.1) was substantially different from 0 (P < 0.001) and the slope (2.17) was substantially different from 1 (P < 0.001). Overall, fecal NIRS did not consistently nor adequately predict IVOMD in a 1:1 relationship.

 

Figure 2. Regression of actual dietary in vitro organic matter digestibility (IVOMD) on fecal NIRS prediction of dietary digestible organic matter (DOM) to validate ability of fecal NIRS to predict actual dietary IVOMD. Regression relationships differed among years (P < 0.05). Coefficient of variation (r2) estimates proportion of variation in actual values explained by predicted values. R² values range from 0 to 1 with those closer to 1 being better. The regression intercept should be 0 and slope should be 1 for a 1:1 relationship between predicted and actual values. Intercept and slope were similar to 0 and 1 (P > 0.05), respectively, in 2013, but differed from 0 and 1 (P < 0.05) in 2014.

 

Results with steer performance were much like those reported by producers: negative gain was predicted for conditions where cattle were actually in a positive plane of nutrition and gaining weight (Table 1). In particular, in August 2013, NUTBAL predicted average daily weight change that ranged from -3.24 lb. to + 2.48 lb. (average was -1.5 lb.) across the 7 cannulated steers that diet and fecal samples were collected from. Negative gains were predicted for 6 of the 7 head. Weight loss was predicted despite fecal NIRS predictions for the same steers of CP and DOM that were great enough to support weight gain. Actual ADG of the contemporary group of yearling steers that grazed the pastures where the diet and fecal samples were collected was 1.48 lb. during August 2013.This was 3 lb. more than the average of the NUTBAL predictions. Although NUTBAL predictions of ADG for the remainder of 2013 and all of 2014 were for positive ADG, they were different from actual ADG. Because of these obvious differences, statistical analysis was not attempted because the lack of a relationship between predicted and actual performance was so great.

 

Table 1. Average daily gain (ADG) predicted by NUTBAL compared to actual ADG of the contemperary group of steers used in this study for each month of sampling where diet samples, fecal samples, and weights of steers were collected.

Year	Month	Predicted ADG, lb	Actual ADG, lb
2013	June	2.21	2.50
2013	August	-1.46	1.48
2014	June	2.84	2.22
2014	August	1.20	1.83

 

 

Discussion

The fecal NIRS/NUTBAL system was developed in the early 1990’s.  An early criticism of NIRS/NUTBAL predictions was that they were developed using samples collected in Texas and Gulf Coast states, dominated by warm-season grasses. This was of particular concern in northern states where pastures are often dominated by cool-season grasses. A number of experiments were conducted to validate the system. It is our understanding that much of the data gained from these northern experiments was added to the data set and the NIRS models were recalibrated. However, in many of the cases, particularly South Dakota, the second year of evaluation using the recalibrated equations did not suggest improved accuracy. Results from these earlier studies, as well as a more recent study in Nebraska present similar results as the current study. 

In the mid-1990’s a team of SDSU researchers entered into an agreement with NRCS to collaboratively validate the predictive accuracy of NIRS/NUTBAL on South Dakota rangelands (Zalesksy, 1998). In 1996, significant differences between predicted and actual performance of both lactating cows and replacement heifers, as well as significant differences between NIRS predictions and esophageal extrusa samples caused them to recommend new NIRS calibration equations for northern plains rangelands. Samples from this South Dakota project and samples collected at the USDA ARS Livestock and Range Research Laboratory at Ft. Keogh (near Miles City, Montana) were used to develop new calibrations. These calibrations improved predictions, but were still significantly different from actual values, particularly in mature, low-quality forages.

In 1997, SDSU conducted additional grazing experiments with replacement heifers as well as dry and lactating cows, but comparative extrusa samples were not collected. Zalesky (1998) reported NIRS-predicted CP and DOM, and suggested seasonal profiles were consistent with expected profiles as forage matured, however interpretations could differ. In heifers, actual ADG was consistently greater than NUTBAL predicted ADG in all months. In all classes of cows (dry and lactating), NUTBAL predicted ADG was usually less than actual ADG in early summer and greater than actual ADG in late summer and fall.

Oklahoma State University (OSU) also evaluated the predictions of fecal NIRS/NUTBAL (O’Neil, 2001). Due to the program being developed in the southern plains with more warm season grasses, it was thought that it should work better in Oklahoma. However, their results were similar to the current SDSU results, with considerable differences in esophageal extrusa diet samples compared to fecal samples. In the OSU study, paired samples were collected on a monthly basis throughout the growing season from native tall grass prairie and bermudagrass pastures. O’Neill (2001) used the same regression procedures as the current study, including r² values, statistical similarity among years, and 1:1 relationship between predicted and actual values. They found that results were similar across years and among the two pasture types. However, they did not find that predictions were accurate or that r² values indicated that adequate variation in the regression relationships were explained (Table 2).

 

Table 2. Statistical results for regression of actual dietary nutrient values on fecal NIRS predictions to validate ability of fecal NIRS to predict dietary nutrients in Oklahoma tall grass prairie and bermudagrass pastures (from O'Neil, 2001). Coefficient of variation (r²) estimates proportion of variation in actual values explained by predicted values. R² values range from 0 to 1 with those closer to 1 being better. The regression intercept should be 0 and slope should be 1 for a 1:1 relationship between predicted and actual values. Intercept and slope values for crude protein (CP) and digestible organic matter (DOM) differ from 0 and 1 (P < 0.05), respectively.

	r2	Slope	Intercept
CP	0.51	0.7698	2.05731
DOM	0.32	1.2989	22.50984

 

Similar to our study, the slope and intercept values for CP were different from 1 and 0, respectively, indicating that a 1:1 relationship did not exist between actual and predicted values. For DOM, the slope was similar to 1, but the intercept differed from 0. The slope and intercept for CP for the OSU study was very similar to the current SDSU study. Additionally, the slope and intercept for DOM were within the range of the slopes and intercepts of the current study.

Because of the potential bias toward warm-season grasses, it was expected by many that NIRS/NUTBAL predictions would be more accurate for warm-season grass dominated situations such as the OSU study. However, the results were similar between Oklahoma and South Dakota, suggesting that such bias did not occur.

The most recent study evaluating the validity of the fecal NIRS/NUTBAL predictions took place in the Nebraska Sandhills from 2015-2017 at the University of Nebraska-Lincoln Gudmundsen Research Ranch (Johnston, et al., 2019). On Nebraska sandhills upland rangeland (warm-season grass dominated), comparisons were made between esophageal, hand-plucked and fecal samples. Additionally, fecal and esophageal samples were collected on subirrigated meadows (cool-season grass dominated). Samples were collected in both areas during July, September and November over the three-year period. Like our study, Nebraska results were inconsistent across forage type. Overall, considering both forage types, CP was slightly underestimated and TDN was consistently over estimated in the fecal NIRS model, ranging from 2.5-7.1 percentage points (Table 3).  Decline in diet quality as forage matured over time was not fully captured through NIRS/NUTBAL analysis.

 

Table 3. Comparison of actual crude protein (CP) and total digestible nutrients (TDN) of diet samples to fecal NIRS predictions on upland range and subirrigated meadows in the Nebraska Sandhills (from Johnston et al., 2019). DOM reported by GANLAB was converted to TDN by multiplying DOM by 1.06. Fecal NIRS prediction differs from actual nutrient content when P < 0.05.

Upland
Item	Diet	NIRS	SE	P-Value
CP
Jul	8.0	8.0	0.3	0.99
Sep	7.1	5.2	0.3	<0.01
Nov	5.4	5.2	0.3	0.92
TDN*
Jul	56.6	65.8	1.0	<0.01
Sep	46.2	64.4	1.0	<0.01
Nov	44.3	62.4	1.0	<0.01
Subirrigated Meadows
CP
Jul	10.2	9.4	0.3	0.05
Sep	9.3	9.3	0.3	0.99
Nov	8.1	5.0	0.3	<0.01
TDN
Jul	58.9	60.6	1.2	0.17
Sep	51.2	60.3	1.2	<0.01
Nov	43.9	55.8	1.2	<0.01

 

 

Conclusions

We reject our hypothesis that the model used in the fecal NIRS program was capable of predicting diet quality values of South Dakota forages. Additionally, predictions of fecal NIRS and NUTBAL were similarly inaccurate across other times and regions. There was a lack of consistency of results for CP, TDN and cattle performance that eliminated the possibility of developing an adjustment factor to apply to GANLAB reports. Under current conditions, the value of this tool to assist in making management decisions based on diet quality and cattle performance is limited. This case study provides a robust evaluation of a situation that is very important to the economic livelihood of cattle producers. Feed management recommendations based on fecal NIRS/NUTBAL are often unnecessary and would be costly if followed. Alternatively, body condition scoring and visual monitoring of fecal dry matter content can be used to monitor nutritional status of beef cattle and make feed management recommendations (Harty, 2016).

 

References

Harty, A. (2016). Monitoring nutrient status of cows. http://igrow.org/livestock/beef/monitoring-nutrient-status-of-cows/.

Johnston, B.M., Jenkins, J.D,  Musgrave, J.A., Mulliniks, J.T., Stephenson, M.B., MacDonald, J., and Stalker, L.A. (2019). Three year summary: Comparison of diets collected from esophageally fistulated cows to forage quality estimated from fecal analysis. Nebraska 2019 Beef Cattle Report (in review).

Lesperance, A.L., Bohman, V.R., and Marble, D.W. (1960). Development of techniques for evaluating grazed forage. J. Dairy Sci., 43:682-689.

Olson, K.C. (1991). Diet sample collection by esophageal fistula and rumen evacuation techniques. J. Range Manage., 44:515-519.

Olson, K.C., Schauer, C. Engel, C., Kincheloe, J.J., Brennan, J.R., and Hauptman, B.L. (2016). Effect of grazing prairie dog colonized rangeland on cattle nutrition and performance: A progress report. Rangelands, 38:29-33. doi: 10.1016/j.rala.2015.12.003.

O’Neil, A.D. (2001). Chapter 3. Accuracy of fecal NIRS profiling of beef cattle grazing bermudagrass and native tall grass prairie.  In: M.S. Thesis. Oklahoma Agricultural Experiment Station, Stillwater, OK.

USDA NRCS. (2014). Animal Enhancement Activity - ANM65 - Monitoring nutritional status of ruminant livestock using the NUTBAL system. https://www.nrcs.usda.gov/wps/PA_NRCSConsumption/download?cid=nrcseprd794207&ext=pdf.

Zalesky, D. (1998). NIRS/NUTBAL project. A progress report submitted to NRCS.

---