AUTHOR=Shi Kunpeng , Ding Hao , Chen Tao , Zou Chuanyi TITLE=The numerical prediction of the Earth’s polar motion based on an advanced multivariate algorithm JOURNAL=Frontiers in Astronomy and Space Sciences VOLUME=10 YEAR=2023 URL=https://www.frontiersin.org/journals/astronomy-and-space-sciences/articles/10.3389/fspas.2023.1158138 DOI=10.3389/fspas.2023.1158138 ISSN=2296-987X ABSTRACT=

Since there are complicated changes in the polar motion (PM) from sub-annual to decadal, precisely predicting it is challenging. Here, we provide an advanced multivariate algorithm by combining an iterative oblique singular spectrum analysis (IOSSA) with pseudo data (IOSSApd) and consider more periodic and quasi-periodic signals, especially long-period oscillations (Ding et al., Geophys. Res. Lett., 2019, 46, 13765–13774) and multi-frequency Chandler wobble (Pan, International Journal of Geosciences, 2012, 3, 930–951), than previous studies. The IOSSA in oblique coordinates, due to its weak separability conditions, has a better separation performance than general singular spectrum analysis (SSA), and the IOSSApd approach further solved the shift problem. Upon using the IOSSApd method, the PM data can be separated into deterministic and stochastic components, extrapolated by the multiple-harmonic (MH) and autoregressive integrated moving average (ARIMA) models, respectively. Based on the IERS EOPC04 PM series, we produced multiple sets of PM predictions with a 1-year leading time and reported the IERS Bulletin A predictions as a comparison. For 90-day leading time predictions, the mean absolute errors (MAEs) of the x- and y-components were 7.69 and 5.12 mas, respectively, while the corresponding MAEs obtained by IERS Bulletin A were 9.45 and 5.69 mas, respectively. For up to 360 days, our algorithm obtains the MAEs of PM slowly accumulating to 12.98 mas on average, far better than the 19.14 mas for Bulletin A’s predictions (also significantly superior to the corresponding results given by previous studies). The prediction performance in the middle- and long-term prediction is further compared against the general SSA predictor. By virtue of weak periodic error, our results show that combining the IOSSApd + MH + ARIMA models improved the prediction success rate up to 75.39% and 69.58% for the x- and y-component, respectively.