Abstract
The shapes of lactation curves are affected by genetic and environmental factors, and flexible models are required to fit such curves. This study aimed to compare the effects of the Gaussian process regression model (Gaussian model) for fitting lactation curves of Saanen dairy goats versus the parametric Wood's model. In addition, we investigated the effects of environmental factors on the shape of lactation curves. Principal component analysis (PCA) detected 3 (541 lactations fitted using the Wood's model [WDS]), 5 (raw data from the WDS fitted using the Gaussian model [GWDS]), and 6 (1,032 lactation datasets fitted using the Gaussian model [GDS]) principal components (PCs). The interpretation of PC1 and PC2 in the 3 datasets was consistent, with PC1 accounting for total milk production, PC2 accounting for persistency (reflecting the difference between early and late lactation), PC3 accounting for milk yield in early lactation (WDS) and relative milk yield in mid-lactation (GWDS and GDS), and PC4 to 6 being associated with fluctuations throughout lactation. The lactation curves of 3 datasets were clustered into 2 (WDS), 2 (GWDS), and 3 (GDS) clusters based on their PC scores, and mainly differed in total milk production and persistency. The total milk production increased from the first to the third parities, but the mid-term relative milk production was highest in first-parity goats. Compared with kidding in spring, kidding in winter led to higher total milk production and persistency, and lower mid-term relative milk production. Low persistency was detected when the number of kids was ≥2. The Gaussian model is suitable for fitting daily milk yield records, with the main sources of variance in the lactation curves of Saanen goats in China being total milk yield and persistency.