High-precision earth rotation parameter forecasting: multiscale computational models and multivariate data fusion

Earth Rotation Parameters (ERPs, including polar motion and length-of-day in this study) serve as the essential link of celestial-terrestrial reference frames. This connection and their high-precision prediction are crucial for modernized satellite navigation systems. However, high-precision ERP predictions are challenging because of their complex variations and long latency (several days to months). In this study, we developed an advanced Hankel Spectrum Analysis (HSA) + Auto-regression (AR) model, fused geophysical Effective Angular Momentum (EAM) excitation and ERP rapid solutions, and improved ERP predictions in short-middle-term ERP predictions. The major contributions of this work are: (1) Long-period superposition of interannual to multidecadal oscillations rather than simple linear trends. (2) Time-variant harmonic modulation using Hankel-z parameters instead of assuming fixed periods for Chandler, Annual and Semi-annual wobbles. (3) AR stochastic process integrates the entire ERP rapid solutions instead of the previous 30-day rapid filling, and EAM residuals converted to the ERP domain instead of the excitation domain. In real-time ERP predictions, our predictive results greatly surpass the benchmark from the LS-based (Least-squares), IERS Bulletin A and mostly classical methods. These results demonstrate effectiveness of HSA + AR methodology, which can be enhanced with more multi-data/models (e.g. EAM forecasts and machine learning) in future.