Predicting voluntary forage intake in cattle

A large database was compiled of forage intake observations published during the past 20 years. Inputs included a wide range of factors believed to be related to voluntary intake. An analysis was designed to pinpoint which feed and animal characteristics were most valuable in predicting voluntary intake across a range of feeding situations and to compare the ability of different models to predict intake. Results emphasized the complexity of intake prediction. A wide range was evident in the variables included in the optimal models for predicting intake within different data subsets. In many cases, we observed that ratios between feed values (e.g., forage acid detergent fiber:forage crude protein) were more useful in predicting intake than the measures themselves.


Introduction
Accurate estimation of an animal's feed variables were identified.Simple regressions intake is necessary to formulate diets or predict were run with pairs of intake and predictor performance on a particular diet.However, variables that showed significant correlacurrent intake prediction models are not suffi-tions.Stepwise multiple regressions were ciently accurate, especially when applied across conducted, both on the entire set and within varied feeds, cattle types, and supplementation selected feed and animal groupings, to evaluprograms.To address these limitations, we ate the potential for improving predictive compiled a large diverse data set of intake accuracy.Several measures of intake were observations published during the past 20 years, considered: dry matter intake (DMI), orthen identified which variables consistently ganic matter intake (OMI), and total digestexerted the greatest impact on voluntary intake.
ible organic matter intake (TDOMI).All In addition, this data set was used to evaluate intake values were expressed both as a whether models that would improve intake percentage of body weight (BW) and per prediction could be constructed from currently available data.

Experimental Procedures
the forages ranged from 1.9 (dormat prairie Once the data set was complete, statistically significant correlations between the unit of metabolic body weight (BW ).In .75addition, we also evaluated the changes in forage data were grouped by supplementation level and total intakes seen with supplementation (low, medium,or high), R values were when compared with unsupplemented cattle in lower for diets containing intermediate the same trial and eating the same forage.amounts of supplement but were higher at

Results and Discussion
Unsupplemented Cattle.We identified five forage variables that in a single-variable regression model, could explain approximately half the variation seen within this data set in intake per unit of BW :OMD, OMD:CP; and the squares .75 of values for CP, DIP (expressed as a percent of CP), and OMD:NDF.A "best fit" multiple regression utilizing OMD:CP, ADF, DIP, DIP , 2 CP , and OMD:NDF was able to account for All Cattle.Regression analysis was 2 nearly 75% of the variation observed in volun-conducted on the entire data set (including tary forage intake.From a practical viewpoint, both supplemented and unsupplemented it would be beneficial to limit predictor variables cattle) and gave results very similar to those to those available from typical feed analysis.
seen with supplemented cattle.Forage Multiple regression using simple feed analysis ADF:forage CP was the most powerful values yielded a model with CP, ADF, and NDF.
single predictive variable, but by itself, it Lignin content was not found to be a useful could account for only 30% of the observed predictor.However, this model had an R of variation in forage intake.The best multiple 2 just .41(that is, it only explained 41% of the regression developed for the complete data observed variation in intake).
This was set had an R of just .30,with virtually no increased to nearly 60% with the addition of improvement over the simple ADF:CP ADF:CP.model.Improvements were not seen when Supplemented Cattle.The best single predictor of forage intake in cattle fed supplement in conjunction with forage was the ratio between forage ADF and forage CP, which explained about 33% of the variation in forage intake.No other single-variable model had an R greater than .25.A multiple regression using 2 a combination of forage factors (NDF, CP ); 2 supplement factors (DIP, NDF, CP, % supplement in total diet); and one ratio (forage ADF:forage CP) was able to explain nearly 50% of the variation in forage intake.Subsequent work showed that the ability to predict intake of supplemented cattle depended upon the forage quality and supplementation approach.For example, predicted intake deviated more from actual intake when animals were receiving energy (i.e., grain) supplements compared with high-fiber or protein supplements.When the 2 either extreme.Similarly, intake prediction was more effective in diets based on high (>60%) or low (<45%) digestibility forages and less accurate with roughages of moderate quality.In the case of low-quality forages, the three highly significant variables in the model were forage CP , diet digestibility, 2 and forage ADF.Predictions with highquality hays were tied most closely to forage CP, forage ADF, and forage DIP. 2 the data were sorted by forage digestibility, but separate analysis of the information collected on dairy breed animals allowed development of a model that accounted for about 75% of the variation seen in that subset.Although none of these analyses generated a highly successful prediction model, the complexities of intake prediction were illustrated, and some key interrelationships were identified.In addition, several ratios between key feed characteristics, most notably forage crude protein and forage ADF levels, were identified as effective predictor variables.