Use of time series analysis and Kalman filters in dairy management

The objective of this study was to evaluate the use of time series analyses and Kalman filters in dairy management. Chapter I provides a short introduction on the scope of the dissertation. In chapter II, a literature review on the current state of the field of time series analysis and Kalman filters in the dairy industry is presented. The review compiles 29 studies, and evaluates the time series analysis techniques used and their application in agriculture. Most commonly used with the electronic milk meters available are rolling average models to indicate deviations in milk yield. In the literature the use of a Kalman filter is described in experimental models to indicate mastitis based on multiple variables.; In chapter III an exponentially smoothed average model, an autoregressive integrated moving average (ARIMA) model, a univariate Kalman filter and a multivariate Kalman filter were compared in their ability to monitor and predict bulk tank somatic cell count (BTSCC) values for dairies in Texas. With the model variances used in the study the Kalman filter classified too many variables as too high or too low. However the Kalman filter could easily be adapted to include more variables, as demonstrated in the study. Inclusion of extra variables decreases the number of observations classified incorrectly.; In chapter IV an expectation maximization (EM) algorithm was developed to calculate maximum likelihood estimation for system matrices in a Kalman Filter. Data with weekly milk weights, bulk tank somatic cell count (BTSCC) and monthly dairy herd improvement association (DHIA) somatic cell count (SCC) was used. Three models were compared in their ability to predict individual cow SCC. The first model includes BTSCC and SCC, the second model includes MILK and SCC, and the third model includes BTSCC, MILK and SCC. Including BTSCC in the model increased the ability of the models to predict individual cow SCC. When a mastitis event occurred during lactations, the mean square error (MSE) of prediction for milk increased significantly.