Possible Improvement of Polar Motion Prediction using
Autocovariance Prediction Procedures

W. Kosek, D. D. McCarthy, B. Luzum


Abstract.
     The current method to predict polar coordinates x and y is
the least-squares extrapolation of an annual ellipse and a
Chandler circle fit to past observations.  Autocovariance
prediction has been applied to attempt to improve the predicted
polar motion using residuals from fits of such curves to the past
two years and six years as well as the entire data set, assuming
different starting prediction epochs.  The results show the
possibility of decreasing polar motion prediction errors by about
50% for different prediction lengths.  Autocovariance prediction
errors depend on the data length of residuals but mostly they
depend on starting prediction epochs due to irregular short
period variations in polar motion.