Short term EOP forecasting: insights from the 2nd EOP PCC and its ML-oriented follow-up

This work was presented during the TIGER Symposium in Geodesy (28 September – 1 October 2026, Gävle, Sweden). Abstract: Reliable measurements of Earth Orientation Parameters (EOP) are crucial for numerous operational applications including spacecraft navigation, astronomical observations, and satellite-based navigation systems on Earth. Many of these applications require EOP values in real time; however, obtaining them immediately is not possible because observational data from different measurement techniques must first undergo processing, which introduces delays. To overcome this limitation, several organizations worldwide, including the International Earth Rotation and Reference Systems Service (IERS), produce and distribute EOP forecasts. Over the past decade, substantial progress has been achieved in the processing of EOP observations, the development of new forecasting approaches, and the understanding of how variations in the distribution of surface fluids influence EOP variability. The number of research groups engaged in EOP prediction has also grown, with different teams employing diverse input datasets, prediction techniques, and forecast horizons. In recent years, particularly rapid development has been observed in forecasting methods based on machine learning (ML) algorithms. The Second EOP Prediction Comparison Campaign (2nd EOP PCC), organized under the auspices of the IERS, rigorously compared the accuracy of various internationally available EOP prediction algorithms. Based on additional data collected during the currently running post-operational phase of the campaign, as well as sub-campaign dedicated ML-based prediction approaches, we will discuss emerging opportunities to further improve EOP forecasts. This can be achieved by combining enhanced predictions of geophysical excitations with advanced forecasting algorithms leveraging data science and artificial intelligence methods. We will present the current levels of accuracy attainable in EOP predictions and highlight the most effective forecasting approaches. Challenges and future prospects for achieving more reliable and timely forecasts will also be addressed. Special emphasis will be placed on the potential of machine learning techniques, although our discussion will not be limited to these methods alone.