Improving the measurement of live weight and body condition score in sheep : a thesis presented in fulfilment of the requirements for the degree of Doctor of Philosophy in Animal Science, Massey University, Turitea, Palmerston North, New Zealand

Abstract

Liveweight (LW) and body condition score (BCS) are important performance indicators in sheep management, providing a basis for decision making. Therefore, accurate measurement of these traits is imperative. The overall aims of this thesis were to 1) Determine the factors affecting the rate of LW loss of fasting sheep, 2) derive equations to predict LW and LW change over a short time period (1 to 8 hours), 3) evaluate the factors affecting the relationship between ewe LW and BCS, and 4) derive equations for predicting ewe BCS. In the LW studies, lambs were offered three herbage availability levels (Low, Medium and High) in autumn or winter. Similarly, mixed-aged ewes at different physiological states were offered two herbage levels (Low or High). These studies were conducted in two stages: A) calibration stage and B) validation stage. Equations to predict without delay LW were developed at the calibration stage and validated on data collected from independent ewes from different farms. The rate of ewe LW loss was influenced by herbage type and availability, and season. Further, in pregnant ewes, liveweight loss was influenced by the stage of pregnancy, but not pregnancy-rank. Applying correction equations improved the prediction accuracy of without delay LW estimates up to 55% and 69% in ewe lambs and mixed aged ewes compared with using the delayed weights, respectively. For the BCS studies, LW and BCS data of ewes were collected at regular times of the annual production cycle until they were six years of age. Using a ewe’s LW and BCS records to predict their current BCS using a linear model gave moderately accurate estimates. A different dataset, which included foetal- and fleece weight-adjusted LW and height at withers was then used. It was found that equations combining LW, LW change and previous BCS explained more variability in current BCS and were more accurate than LW-alone based models but the addition of adjusted LW and height at withers gave no further benefit to the BCS prediction models. Applying machine learning classification algorithms such as extreme gradient boosted trees and Random forest on a 3-point BCS scale achieved very good BCS prediction accuracies (> 85%). These combined findings provide useful prediction equations that could be incorporated into weighing systems, which along with EID would improve sheep production by aiding management decision making.