Modeling of daily body weights and body weight changes of Nordic Red cows.

Increased availability of automated weighing systems have made it possible to record massive amounts of body weight (BW) data in a short time. If the BW measurement is unbiased, the changes in BW reflect the energy status of the cow and can be used for management or breeding purposes. The usefulness of the BW data depends on the reliability of the measures. The noise in BW measurements can be smoothed by fitting a parametric or time series model into the BW measurements. This study examined the accuracy of different models to predict BW of the cows based on daily BW measurements and investigated the usefulness of modeling in increasing the value of BW measurements as management and breeding tools. Data included daily BW measurements, production, and intake from 230 Nordic Red dairy cows. The BW of the cows was recorded twice a day on their return from milking. In total, the data included 50,594 daily observations with 98,418 BW measurements. A clear diurnal change was present in the BW of the cows even if they had feed available 24 h. The daily average BW were used in the modeling. Five different models were tested: (1) a cow-wise fixed second-order polynomial regression model (FiX) including the exponential Wilmink term, (2) a random regression model with fixed and random animal lactation stage functions (MiX), (3) MiX with 13 periods of weighing added (PER), (4) natural cubic smoothing splines with 8 equally spaced knots (SPk8), and (5) spline model with no restriction on knots but a smoothing parameter corresponding to a fit of 5 degrees of freedom (SPdf5). In the original measured BW data, the within-animal variation was 6.4% of the total variance. Modeling decreased the within animal variation to levels of 2.9 to 5.1%. The smallest day-to-day variation and thereafter highest day-to-day repeatabilities were with PER and MiX models. The usability of modeled BW as energy balance (EB) indicator were evaluated by estimating relationships between EB, or EB indicators, and modeled BW change. In all cases the modeling increased the correlation and thus the reliability of the BW measurements. From all of the tested models, the best predictive value was attained by the random regression model with fixed and random animal lactation stage functions. Based on results, modeling of BW significantly increases the usefulness of BW as an EB predictor and management indicator.