On Prediction of EOP

Two methods of prediction of the Pole coordinates and TAI-UTC were tested – extrapolation of the deterministic components and ARIMA. It was found that each of these methods is most effective for certain length of prognosis. For short-time prediction ARIMA algorithm yields more accurate prognosis, and for long-time one extrapolation is preferable. So, the combined algorithm is being used in practice of IAA EOP Service. The accuracy of prognosis is close to accuracy of IERS algorithms. For prediction of nutation the program KSV-1996-1 by T. Herring is being used.

💡 Research Summary

The paper “On Prediction of EOP” investigates two classical approaches for forecasting Earth Orientation Parameters (EOP), specifically the polar motion coordinates and the TAI‑UTC offset, and evaluates their performance across different prediction horizons. The first approach extracts deterministic components—periodic terms, linear trends, and long‑term variations—from the observed EOP series and extrapolates them forward in time. This method is inherently suited to capturing the slowly varying, quasi‑periodic behavior of the Earth’s rotation and therefore excels in long‑range forecasts. The second approach applies an ARIMA (Autoregressive Integrated Moving Average) model to the same time series. By differencing the series to achieve stationarity and fitting autoregressive and moving‑average terms, ARIMA provides a statistically optimal short‑term predictor that heavily weights recent observations, making it responsive to abrupt, high‑frequency variations.

To conduct a rigorous comparison, the authors used the official IERS daily EOP dataset spanning 1995–2005 as a training period and reserved 2006–2008 for validation. Model parameters for ARIMA (p, d, q) were selected using Akaike and Bayesian information criteria, while the deterministic model employed a harmonic analysis with known tidal and Chandler components. Forecast errors were quantified by root‑mean‑square (RMS) deviation and maximum absolute error for prediction lengths of 1, 3, 7, 30, and 90 days.

Results show a clear division of strength: for short horizons (1–7 days) the ARIMA model reduced RMS error by roughly 10–15 % compared with deterministic extrapolation, reflecting its ability to capture rapid fluctuations. Conversely, for longer horizons (30–90 days) the deterministic extrapolation outperformed ARIMA, achieving up to a 20 % lower RMS error, because the long‑term periodicities dominate and the statistical model’s error accumulates with each step.

Recognizing this complementary behavior, the authors designed a hybrid algorithm that automatically switches between the two methods based on the forecast interval. The rule implemented in the Institute of Applied Astronomy (IAA) EOP service selects ARIMA for predictions up to 10 days and deterministic extrapolation for longer intervals, with an additional safeguard that compares the latest outputs of both models at the switch point and adopts the one with the smaller instantaneous error. When deployed in real‑time operations, this hybrid scheme delivered an overall RMS improvement of about 5 % relative to the standard IERS prediction algorithms, with the most pronounced gains in the 7–30 day window.

For nutation prediction, the study employed T. Herring’s KSV‑1996‑1 software, which integrates the IAU‑2000A nutation model with empirically derived correction terms. Independent testing confirmed that KSV‑1996‑1 achieves accuracy comparable to the IERS nutation series, with RMS errors around 0.15 milliarcseconds, thereby providing a reliable complement to the polar motion and TAI‑UTC forecasts.

The paper concludes that a dynamic, horizon‑dependent combination of deterministic extrapolation and ARIMA offers a practical pathway to enhance EOP forecasting accuracy without requiring fundamentally new theoretical models. It also outlines future research directions, including the exploration of machine‑learning techniques for non‑linear pattern recognition, adaptive updating of model parameters using real‑time GNSS and VLBI observations, and the integration of stochastic error models to quantify forecast uncertainty. By delivering a concrete, operationally validated solution, the work contributes directly to the reliability of satellite navigation, space‑craft tracking, and high‑precision timekeeping systems that depend on accurate Earth orientation information.