Prediction of dry matter intake throughout lactation in a dynamic model of dairy cow performance.

In the dynamic modeling of dairy cow performance over a full lactation, the difference between net energy intake and net energy used for maintenance, growth, and output in milk accumulates in body reserves. A simple dynamic model of net energy balance was constructed to select, out of some common dry matter intake (DMI) prediction equations, the one that resulted in a minimum cumulative bias in body energy deposition. Dry matter intake was predicted using the Cornell Net Carbohydrate and Protein System, Agricultural Research Council, or National Research Council (NRC) DMI equations from body weight (BW) and predicted fat-corrected milk yield. The instantaneous BW of cows at progressive weeks of lactation was simulated as the numerical integral of the BW change obtained from the predicted net energy balance. Predicted DMI and BW from each DMI equation, using either of 2 equations to describe maintenance energy expenditures, were compared statistically against observed data from 21 herd average published full lactation data sets. All DMI equations underpredicted BW and DMI, but the NRC DMI equation resulted in the minimum cumulative error in predicted BW and DMI. As a general solution to prevent predicted BW from deviating substantially over time from the observed BW, a lipostatic feedback mechanism was integrated into the NRC DMI equation as a 2-parameter linear function of the relative size of simulated body reserves and week of lactation. Residual sum of squares was reduced on average by 52% for BW predictions and by 41% for DMI predictions by inclusion of the negative feedback with parameters taken from the average of all 21 least squares fits. Similarly, root mean square prediction error (%) was reduced by 30% on average for BW predictions and by 23% for DMI predictions. Inclusion of a feedback of energy reserves onto predicted DMI, simulating lipostatic regulation of BW, solved the problem of final BW deviation within a dynamic model and improved its DMI prediction to a satisfactory level.