Global Navigation Satellite Systems (GNSS) can sense deformations of the Earth’s crust. All components, but in particular the vertical component are affected by mass loading, i.e. external forces resulting from the redistribution and changes in fluid mass. These effects include non-tidal atmospheric loading (NTAL), non-tidal ocean loading (NTOL), and hydrological loading (HYDL). If these loading effects are not compensated in the processing of space geodetic data, the obtained results will be distorted. Thus, physics-based loading models exist that can be applied to correct station positions. This study investigates if machine learning (ML) in combination with environmental variables can replace or augment the existing physics-based models via a data-driven modelling of GNSS displacements. Therefore, vertical displacements of 3553 GNSS stations in Europe are utilized to train and validate XGBoost models. Three different strategies were tested, differing in the preprocessing of the GNSS data, i.e. whether or which physics-based loading models were applied beforehand. A significant improvement was achieved for all strategies ranging from 4.4% to 22.9%. The improvement is calculated based on the root mean squared error (RMSE) reduction of the GNSS residual coordinates w.r.t. a trajectory model, accounting for a linear trend, seasonal signals, and discontinuities in the GNSS coordinate time series. In addition to evaluating the ML models, a thorough feature importance analysis based on SHapley Additive exPlanations (SHAP) is carried out to better understand the driving factors of the model output and to gain insights into what signals could still be found to enhance existing physical models.

Rapid provision of Earth Orientation Parameters (EOPs, here polar motion and dUT1) is indispensable in many geodetic applications and also for spacecraft navigation. There are, however, discrepancies between the rapid EOPs and the final EOPs that have a higher latency, but the highest accuracy. To reduce these discrepancies, we focus on a data-driven approach, present a novel method named ResLearner, and use it in the context of deep ensemble learning. Furthermore, we introduce a geophysically-constrained approach for ResLearner. We show that the most important geophysical information to improve the rapid EOPs is the effective angular momentum functions of atmosphere, ocean, land hydrology, and sea level. In addition, semi-diurnal, diurnal, and long-period tides coupled with prograde and retrograde tidal excitations are important features. The influence of some climatic indices on the prediction accuracy of dUT1 is discussed and El Ni\~{n}o Southern Oscillation is found to be influential. We developed an operational framework, providing the improved EOPs on a daily basis with a prediction window of 63 days to fully cover the latency of final EOPs. We show that under the operational conditions and using the rapid EOPs of the International Earth Rotation and Reference Systems Service (IERS) we achieve improvements as high as 60\%, thus significantly reducing the differences between rapid and final EOPs. Furthermore, we discuss how the new final series IERS 20 C04 is preferred over 14 C04. Finally, we compare against EOP hindcast experiments of European Space Agency, on which ResLearner presents comparable improvements.