New publications by Kiani Shahvandi et al. (2024)

Predicting Earth Orientation Parameters (EOPs) is important for a host of geoscientific applications, including satellite and space navigation, as well as orientation of deep space telescopes. Among EOPs, dUT1 and its rate LOD are among the most challenging to predict. dUT1 is a measure of the difference between universal time and its counterpart coordinated universal time. dUT1 varies on all measurable timescales and is the cumulative of the differences between actual duration of a day and its nominal 86,400 seconds, i.e., LOD (length of day). As such, any process that affects the rotational speed of the Earth will induce LOD variations and thus, dUT1. Hence, the mass redistribution and motion of the geofluids will drive LOD and dUT1. Any method to predict dUT1 and LOD should consider these effects to the best possible extent. Kiani Shahvandi, et al. (2024a,b) designed two algorithms for the prediction of LOD and dUT1, respectively. They used atmospheric, oceanic, hydrological, and sea-level angular momentum data with robust, probabilistic deep learning architectures, namely, Bayesian Hamiltonian Monte Carlo Autoencoders, and Laplacian deep ensembles. The authors showed that they can enhance the prediction accuracy of LOD and dUT1 considerably, thus setting a new baseline and state-of-the-art for the task of LOD and dUT1 prediction.

In a separate study, Kiani Shahvandi, et al. (2024c) focused on explaining the causes of the long-period fluctuations in the length of day. LOD variations derived from lunar occultation and eclipse records contain two main features: (1) a secular trend, and (2) decadal and millennial fluctuations. The causes of this secular trend were an enigma until Kiani Shahvandi et al. (2024d) explained this secular trend based on a combination of lunar tidal and glacial isostatic adjustment. However, the causes of long-period fluctuations are more ambiguous. These fluctuations have magnitudes of around 3 to 4 ms and occur on decadal and millennial timescales. Several candidates exist that might be able to explain these fluctuations, including climatic oscillations and dynamics of the Earth's core. Kiani Shahvandi et al. (2024c) investigate the causes of aforementioned decadal and millennial fluctuations. They tested two possible sources: climatic effects and core dynamics. They concluded that climatic effects do not have enough power to drive these long-period fluctuations. On the other hand, core processes result in azimuthal flows in the Earth's core that could explain these fluctuations. The authors used the methodology of Bayesian Physics-Informed Neural Networks (BPINNs) in the study, which was previously used by Kiani Shahvandi et al. (2024e) to explain the causes of long-period polar motion. The authors used BPINNs and trained them on geomagnetic data and constrained them to satisfy geophysical models of core dynamics. Using BPINNs, they independently reconstruct LOD variations in the past three millennia and compare them with the observed fluctuations. The authors show—for the first time—the good match between observed and reconstructed values, thus explaining the origin of these fluctuations to be the fluid motion at the top of the Earth's core (here, fluid being the molten iron).