A data - driven multichannel prediction of celestial pole offsets

• Autores: Ligas, M.; Michalczak, M.; Kudrys, J.; Belda, S.; Ferrándiz, J. M.; Karbon, M.; Modiri, S.
• Año de publicación: 2025
• Idioma: inglés
• Tipo de recurso: documento de conferencia
• Materia: Ciencias Astronómicas; Celestial Pole Offsets (CPO) prediction; Dynamic Mode Decomposition (DMD); Multichannel time series analysis
• Institución: Universidad Nacional de La Plata

In this contribution, we introduce a methodology aimed at improving the accuracy of Celestial Pole Offsets (CPO; dX, dY) predictions, with a particular focus on short-term forecasts (up to 30 days). The prediction algorithm is tailored for the simultaneous analysis of multichannel data, meaning the data collected from multiple sources (e.g., several sensors measuring the same parameter; here CPO time series provided by different institutions). We use IERS EOP final data, along with data published by JPL, as the input for the prediction procedure. The core of the prediction algorithm is based on the principle of Dynamic Mode Decomposition (DMD), but due to the multidimensional character of the algorithm all operations are tensor-based. The prediction procedure is consistent since it does not depend on external data to fill any latency gaps in the IERS and JPL products. Instead, this is managed within the prediction routine by extending the forecast horizon to include both the gap-filling and proper forecast horizons. As a result, the methodology is fully operational and well-suited for real-time applications. We evaluated this approach against the results obtained within the course of the 2nd EOPPCC as well as in 100 successive yearly trials covering 10 years (01.01.2014 01.01.2024) to assess prediction capabilities of the newly introduced methodology in a long run. The latter mentioned was performed for series consistent with different reference frames, i.e., IERS EOP 14 C04 and IERS EOP 20 C04. The results indicate that the method is at the forefront of current CPO forecasting methods.
Facultad de Ciencias Astronómicas y Geofísicas