Evaluation of the CNCPS v6.5 for predicting metabolizable energy and protein allowable milk in sugarcane based diets

Wednesday, July 23, 2014
Exhibit Hall AB (Kansas City Convention Center)
Edgar A Collao-Saenz , Universidade Federal de Goiás, Jatai-GO, Brazil
Andreas Foskolos , Cornell University, Ithaca, NY
Ryan J Higgs , Cornell University, Ithaca, NY
Marcos N Pereira , Universidade Federal de Lavras, Lavras, Brazil
Michael E Van Amburgh , Cornell University, Ithaca, NY
Abstract Text:

Sugarcane is a high energy yielding crop, and is an alternative feed for dairy cows in tropical regions. This experiment evaluated the sensitivity of CNCPSv6.5 for predicting milk yield in sugarcane based diets. Data for evaluation were obtained from 13 published experiments, representing 50 treatments. Metabolizable energy (ME) and protein (MP) allowable milk were predicted based on reported DMI and diet composition. An algorithm was used to adjust the nutrient composition of individual ingredients from commercial laboratory databases when feed chemistry data were incomplete.  The correlation coefficients between observed and predicted milk yield were based on BLUP (R2BLUP) and model predictions using the mean study effect (R2MP). When milk yield was predicted based on the first limiting nutrient, either ME or MP, the correlation coefficient generated with R2BLUP was 0.985 and with R2MP was 0.81. With ME predicted milk yield, the R2BLUP correlation coefficient was 0.989 and for R2MPit was 0.81, and when the predictor was MP they were 0.92 and 0.67, respectively. The Bayesian Information Criterion was 219 for MP or ME, 110 for ME, and 127 for MP. The Mean Square Prediction Error (MSPE) using MP to predict milk yield was 5.4, and it was 10.8 when ME or MP or ME were predictors. When the MSPE was partitioned, 0.22%, 0.01% and 3.6% of the error was due to mean bias for the MP or ME, ME, or MP predicted milk yield, respectively, 35.41%, 32.94% and 30.72% was due to systematic bias, and 64.37%, 67.05% and 65.68% was due to random variation. Concordance Correlation Coefficients were computed to account for the accuracy and precision of the predictions, values were: 0.79 for MP or ME, 0.83 for ME, and 0.75 for MP. Using CNCPSv.6.5 to predict milk yield responses in sugarcane based diets was reliable. In the available data sets, the prediction of ME allowable milk yield was better than MP and this most likely reflects differences in actual rates of digestion and library values along with ingredient bias.

Keywords: dairy cattle, model evaluation, tropical feed