Models for predicting NDFap and iNDF in tropical conditions

Abstract

Our objective was to develop and evaluate equations for estimating neutral detergent fiber corrected for ash and protein residue (NDF_ap) and indigestible neutral detergent fiber (_iNDF). All feeds data were obtained from information cataloged in the Brazilian Tables of Feed Composition for Cattle (CQBAL 4.0). In addition, information from ruminal digesta, omasal digesta and feces derived from 14 studies was used, which included the chemical characteristics studied in this approach. Eight subsets of data were constructed to predict: NDF_ap of roughages, concentrates, ruminal digesta, omasal digesta, and feces; _iNDF of roughages, concentrates and feces. Within each data subset, 80 % of the information was used for model development and 20 % for model evaluation. To predict the NDF_ap the existence of a linear relationship between the NDF_ap and NDF contents was assumed, and simple linear regression equations were fitted. A Pearson correlation analysis and Stepwise selection process was performed between _iNDF and the potential predictor variables. The evaluation of the equations was carried out using the Model Evaluation System Software. The subset data represented a wide variation in NDF_ap content of roughage (237.9–777 g/kg), concentrates (27.2–600 g/kg), rumen digesta (341.7–673.9 g/kg), omasal digesta (19.1–662.6 g/kg) and feces (243.1–721.8 g/kg); and in the _iNDF content of roughage (80–394.7 g/kg), concentrates (2.4–370 g/kg) and feces (72.7–624.6 g/kg). The linear relationships between the observed and predicted values of NDF_ap for roughages, concentrates, ruminal digesta, omasal digesta and feces presented an intercept not different from zero (P ≥ 0.19) and a slope not different from one (P ≥ 0.13), and high values of R² and CCC. All equations generated to predict roughage _iNDF were suitable, presenting P ≥ 0.08 for intercept equal to zero and/or slope equal to one in the plot of the relationship between observed and predicted values. For _iNDF of concentrates and feces, the power function was the only equation that showed good predictive performance (P ≥ 0.21 for intercept and slope). The models developed in this study will provide support in estimating the energy value of feeds, especially in conditions where direct assessments are limited.