Seasonal effects can be separated from other challenges in the pig environment using time series analysis

The pig environment can be quantified through the mean performance of a contemporary group (CG), adjusted for systematic and genetic effects (Li and Hermesch 2016). The objective of this study was to use time series analysis to decompose CG estimates into the seasonal, long-term trend and residual components generally observed in time series data. It was hypothesised that seasonal effects can be partitioned from the other environmental challenges that are simultaneously captured in CG estimates of average daily gain.

Production records from 1999 to 2013 were available from a commercial herd of Large White pigs located in Queensland, Australia (n = 31 230). Bodyweight averaged 90.9 ± 9.9 kg (mean ± s.d.), measured at an average age of 127.9 ± 5.1 days. Defined by birth month, there were 167 CG with an average size of 187 pigs. Using ASReml (Gilmour et al. 2009), CG estimates for average daily gain were derived using linear mixed models, fitting sex as a fixed effect, and additive genetic effect, CG and common litter environment as random effects. An additional model was evaluated to account for minimum monthly temperatures of test month (MinT; data from www.bom.gov.au) using splines (model described in Guy et al. (2017)). The CG estimates from each model were decomposed using the ‘stl’ function in R (v3.3.2, R Foundation, Vienna, Austria).

The CG estimates ranged from –67 to +55 g/d, and –52.0 to +37.9 g/d when adjusted for MinT. Figure 1 shows these estimates decomposed into seasonal, trend and residual components. The seasonal contribution accounted for between –26 to +21 g/d. Pigs born in April were born in the best growing environment, while October was the most challenging environment. Even though the CG estimates adjusted for MinT had a smaller range than unadjusted estimates, a seasonal component was still extracted, ranging from –11.9 to +10.5 g/d. This demonstrates that temperature, represented by MinT, accounts for some, but not all, seasonality for this Queensland herd. However, this is may vary depending on herd location.

Fig. 1.  Contemporary group (CG) estimates (— unadjusted and - - - adjusted for minimum monthly temperatures), decomposed into seasonal, trend and residual components using time series analysis.

The trend component ranged from –19 to +14 g/d, and described the gradual changes in environmental conditions over time. The residual component ranged from –33 to + 33 g/d, and can be interpreted as irregular, short-term perturbations. Although there is possible confounding, the trend and residual components together can be seen as a measure of environmental challenges other than seasonal effects, including infection challenges. While decomposition may depend on parameter choice, different model parameters were explored and produced similar results. Therefore, decomposing environmental variability through time series analysis indicates that selection for improved robustness is partly for improved response to seasonal fluctuations, and partly for other environmental challenges, which may need to be considered separately for genetic improvement of traits such as disease resilience.