<abstract xmlns="http://www.w3.org/1999/xhtml">

This article presents the application of weighted least squares (WLS) extrapolation and vector autoregressive (VAR) modeling in polar motion prediction. A piecewise weighting function is developed for the least squares (LS) adjustment in consideration of the effect of intervals between observation and prediction epochs on WLS extrapolation. Furthermore, the VAR technique is used to simultaneously model and predict the residuals of <italic>x</italic>p, <italic>y</italic>p pole coordinates for WLS misfit. The simultaneous predictions of <italic>x</italic>p, <italic>y</italic>p pole coordinates are subsequently computed by the combination of WLS extrapolation of harmonic models for the linear trend, Chandler and annual wobbles, and VAR stochastic prediction of the residuals (WLS+VAR). The 365-day-ahead <italic>x</italic>p, <italic>y</italic>p predictions are compared with those generated by LS extrapolation+univariate AR prediction and LS extrapolation+VAR modeling. It is shown that the <italic>x</italic>p, <italic>y</italic>p predictions based on WLS+VAR taking into consideration both the interval effect and correlation between <italic>x</italic>p and <italic>y</italic>p outperform those generated by two others. The accuracies of the <italic>x</italic>p predictions are 13.97 mas, 18.47 mas, and 20.52 mas, respectively for the 150-, 270-, and 365-day horizon in terms of the mean absolute error statistics, 36%, 24.8%, and 33.5% higher than LS+AR, respectively. For the <italic>y</italic>p predictions, the 150-, 270-, and 365-day accuracies are 15.41 mas, 21.17 mas, and 21.82 mas respectively, 27.4%, 11.9%, and 21.8% higher than LS+AR respectively. Moreover, the absolute differences of the WLS+VAR predictions and observations are smaller than the differences from LS+VAR and LS+AR, which is practically important to practical and scientific users, although the improvement in accuracies is no more than 10% relative to LS+VAR. The further comparison with the predictions submitted to the 1st Earth Orientation Parameters Prediction Comparison Campaign (1st EOP PCC) shows that while the accuracy of the predictions within 30 days is comparable with that by the most accurate prediction techniques including neural networks and LS+AR participating in the campaign for <italic>x</italic>p, <italic>y</italic>p pole coordinates, the accuracy of the predictions up to 365 days into the future are better than accuracies by the other techniques except best LS+AR used in the EOP PCC. It is therefore concluded that the medium- and long-term prediction accuracy of polar motion can be improved by modeling <italic>x</italic>p, <italic>y</italic>p pole coordinates together.
</abstract>

This article presents the application of weighted least squares (WLS) extrapolation and vector autoregressive (VAR) modeling in polar motion prediction. A piecewise weighting function is developed for the least squares (LS) adjustment in consideration of the effect of intervals between observation and prediction epochs on WLS extrapolation. Furthermore, the VAR technique is used to simultaneously model and predict the residuals of xp, yp pole coordinates for WLS misfit. The simultaneous predictions of xp, yp pole coordinates are subsequently computed by the combination of WLS extrapolation of harmonic models for the linear trend, Chandler and annual wobbles, and VAR stochastic prediction of the residuals (WLS+VAR). The 365-day-ahead xp, yp predictions are compared with those generated by LS extrapolation+univariate AR prediction and LS extrapolation+VAR modeling. It is shown that the xp, yp predictions based on WLS+VAR taking into consideration both the interval effect and correlation between xp and yp outperform those generated by two others. The accuracies of the xp predictions are 13.97 mas, 18.47 mas, and 20.52 mas, respectively for the 150-, 270-, and 365-day horizon in terms of the mean absolute error statistics, 36%, 24.8%, and 33.5% higher than LS+AR, respectively. For the yp predictions, the 150-, 270-, and 365-day accuracies are 15.41 mas, 21.17 mas, and 21.82 mas respectively, 27.4%, 11.9%, and 21.8% higher than LS+AR respectively. Moreover, the absolute differences of the WLS+VAR predictions and observations are smaller than the differences from LS+VAR and LS+AR, which is practically important to practical and scientific users, although the improvement in accuracies is no more than 10% relative to LS+VAR. The further comparison with the predictions submitted to the 1st Earth Orientation Parameters Prediction Comparison Campaign (1st EOP PCC) shows that while the accuracy of the predictions within 30 days is comparable with that by the most accurate prediction techniques including neural networks and LS+AR participating in the campaign for xp, yp pole coordinates, the accuracy of the predictions up to 365 days into the future are better than accuracies by the other techniques except best LS+AR used in the EOP PCC. It is therefore concluded that the medium- and long-term prediction accuracy of polar motion can be improved by modeling xp, yp pole coordinates together.

Medium- and Long-Term Prediction of Polar Motion Using Weighted Least Squares Extrapolation and Vector Autoregressive Modeling

Artificial Satellites

This work is licensed under the Creative Commons Attribution 4.0 International License.

This article presents the application of weighted least squares (WLS) extrapolation and vector autoregressive (VAR) modeling in polar motion prediction. A...

Prediction technology	x_p					y_p
Prediction technology	10 d	30 d	150 d	270 d	365 d	10 d	30 d	150 d	270 d	365 d
LS+AR	11.28	29.31	67.22	74.84	77.03	9.76	19.66	64.78	73.86	84.96
WLS+VAR (equal weights)	10.63	21.22	51.89	65.71	69.16	8.65	14.16	47.69	71.76	76.66
WLS+VAR (piecewise weights)	11.07	23.57	46.11	61.05	64.70	9.17	13.24	47.29	63.55	70.80

Prediction technology	x_p					y_p
Prediction technology	10 d	30 d	150 d	270 d	365 d	10 d	30 d	150 d	270 d	365 d
LS+AR	2.93	7.71	21.83	24.57	30.86	1.94	5.43	21.90	24.04	27.90
WLS+VAR (equal weights)	2.90	7.15	15.22	20.35	22.69	1.79	4.27	15.91	22.88	23.40
WLS+VAR (piecewise weights)	2.87	7.03	13.97	18.47	20.52	1.81	4.41	15.41	21.17	21.82

Prediction technology	x_p					y_p
Prediction technology	10 d	30 d	150 d	270 d	365 d	10 d	30 d	150 d	270 d	365 d
LS+AR (Kosek 051)	3.56	9.52	21.16	23.85	19.78	2.01	6.08	25.65	21.86	28.91
LS+AR (Kalarus 061)	3.33	8.78	19.36	17.46	14.50	1.83	5.01	21.28	25.10	21.32
AR (Zotov 091)	4.17	9.31	—	—	—	2.30	5.64	—	—	—
Neural networks (Zotov 093)	4.21	11.51	—	—	—	2.14	6.08	—	—	—
WLS+VAR (piecewise weights)	3.49	9.20	21.13	29.67	28.04	1.89	5.84	23.18	27.58	32.55

Medium- and Long-Term Prediction of Polar Motion Using Weighted Least Squares Extrapolation and Vector Autoregressive Modeling

Pubblicato online: 18 lug 2023

Pagine: 42 - 55

Ricevuto: 05 gen 2023

Accettato: 17 mag 2023

DOI: https://doi.org/10.2478/arsa-2023-0004

Parole chiave
polar motion, prediction model, weighted least squares, vector autoregressive

© 2023 Yu Lei et al., published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Figure 1.

Figure 2.

Figure 3.

Maximum absolute differences (in mas) between the pole coordinate observations and their predictions up to different prediction ranges

Mean absolute errors (in mas) for different prediction ranges by the LS+AR and WLS+VAR

Mean absolute errors (in mas) for different prediction ranges by the WLS+VAR and participants in the 1st EOP PCC

Medium- and Long-Term Prediction of Polar Motion Using Weighted Least Squares Extrapolation and Vector Autoregressive Modeling

Pubblicato online: 18 lug 2023

Pagine: 42 - 55

Ricevuto: 05 gen 2023

Accettato: 17 mag 2023

DOI: https://doi.org/10.2478/arsa-2023-0004

Parole chiavepolar motion, prediction model, weighted least squares, vector autoregressive

© 2023 Yu Lei et al., published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Figure 1.

Figure 2.

Figure 3.

Maximum absolute differences (in mas) between the pole coordinate observations and their predictions up to different prediction ranges

Mean absolute errors (in mas) for different prediction ranges by the LS+AR and WLS+VAR

Mean absolute errors (in mas) for different prediction ranges by the WLS+VAR and participants in the 1st EOP PCC

Parole chiave
polar motion, prediction model, weighted least squares, vector autoregressive