Time series autoregressive integrated moving average modeling of test-day milk yields of dairy ewes

Monthly test-day milk yields of 1200 dairy Sarda ewes were analyzed by time -series methods. Autocorrelation functions were calculated for lactations w ithin parity classes and altitude of location of flocks. Spectral analysis of the successions of data was developed by Fourier transformation, and dif ferent Box-Jenkins autoregressive integrated moving average models were fit ted. The separation of deterministic and stochastic components highlighted the autoregressive feature of milk production pattern. The forecasting powe r of autoregressive integrated moving average models was tested by predicti ng total milk production for a standardized lactation length of 225 d from only a few test-day records. Results indicated a greater forecasting capaci ty in comparison with standard methods and suggested further development of time-series analysis for studying lactation curves with more sophisticated methods, such as wavelet decomposition and neural network models.