Improvement Of Polar Motion Prediction

Abstract:Accurate prediction of Earth Orientation Parameters (EOP) is of great scientific and practical importance. EOP is necessary for the precise transformations between the international celestial and terrestrial reference frames. The growing demands by spacecraft tracking and navigation for EOP have prompted research on EOP prediction.EOP are obtained mainly through VLBI, SLR, GPS and other related techniques to jointly survey and process data. Due to complexity of the measurement model and data processing, it is impossible to attain timely measurements of EOP. Therefore, numerous related studies have been conducted to improve the model’s accuracy. At present, the precession and nutation model, the Universal time-Coordinated Universal time (UT1-UTC) model and the Length of Day (LOD) model have yielded practical prediction results for EOP. However, ideal results for the prediction of polar motion (PM) have not been achieved because of its complex activated mechanism that includes the gravitation of the sun and moon, the atmosphere and the ocean. As an important part of EOP, PM prediction is worthy of in-depth study.This paper emphasis on the improvement of PM prediction which includes two aspects. On one hand, improvement of Least squares support vector machine (LS-SVM) model which is the latest model among neural network models. On the other hand, improvement of the physical model. The atmospheric and oceanic excitation source are introduced into the models.The main contents are as follows:(1)LS-SVM model is applied to polar motion prediction. Application of non-linear model is a better choice because of the complex non-linear factors in EOP. LS-SVM model is a new machine learning model. And it can process data well that contain the non-linear factors. The result shows that this model is both feasible and effective.(2)The prediction accuracy is limited by single models. While GM (1,1) model is pretty simple and effective, and easy to program. This model is widely used in many fields. In this paper, the combined model based on LS-SVM and GM(1,1) model is used to forecast the residual series of polar motion. The result shows that this model is both feasible and effective in the ultra short-term prediction.(3)Empirical Mode Decomposition (EMD) is applied to the short-term prediction which is limited by the high-frequency signals. EMD is used to analyze the non-linear and non-stationary signals. This method is different from the traditional signal analysis methods, and the decomposition is data-driven and self-adaptive. This paper applies EMD to decompose PM series. Firstly, removing the high-frequency signals from the PM series, then the combined model of least squares (LS) extrapolation and LS-SVM model is used to predict PM without the high-frequency signals from one to thirty days in the future. The result shows that the prediction accuracy is improved obviously.(4)Considering the strong interrelation between AAM or OAM and PM, χ1and χ2series of AAM or OAM are converted into PM domain, and the PM series are excited by AAM or OAM. This excited series are added into the prediction model. The result shows that the prediction accuracy is improved.AAM or OAM series are vectors. We look them as vector superposition as the joint angular momentum. PM is excited by the joint angular momentum. The result shows that the prediction accuracy is improved.However, in terms of the excitation source and the joining ways, there are no clear conclusions. It shows the complexity of PM excited to some extent.