Enhanced Celestial Pole Offset forecast via combination of different data sources

This study introduces a methodology designed to enhance the accuracy of Celestial Pole Offset (dX, dY) prediction, with a focus on a short-term forecast horizon (up to 30-days). IERS EOP final data as well as those published by JPL are used as input for the  prediction algorithms. The prediction procedure is consistent, in the sense that, it does not rely on any external data to fill any latency gaps in the final IERS product. This is handled within the prediction routine itself by enlarging the forecast horizon to the gap filling horizon and proper forecast horizon. In this way, the presented methodology is ready to use under operational settings what makes it well suited for real time applications. Such an approach enables also to asses prediction capabilities of the methods in offline experiments whilst maintaining the operational settings. JPL CPO data serves as supplementary series for prediction and adjusting using Deming regression to align it  with IERS CPO values (attempt to assess fixed and proportional biases between series). The prediction strategy applies also the Whittaker-Henderson smoother to IERS CPO series, which after smoothing is treated as an additional source of information in the prediction process. Separate predictions based on JPL, IERS and smoothed IERS series are also averaged in different combinations giving rise to ensemble data-based prediction model. In this way we show that the overpredictive and underpredictive characteristics of specific input data, even with the application of a single prediction method, can result in a more precise and accurate final forecast. The presented approach was tested against the results obtained within the course of the 2^nd EOPPCC, as well as other contemporary studies. This presentation includes also a comparison of performance of the method in reference to different series, i.e., IERS EOP 14 C04 and IERS EOP 20 C04.