Session G1.3

G1.3 | New developments in mathematical methods in geodesy, with a focus on machine learning

New developments in mathematical methods in geodesy, with a focus on machine learning

Convener: Benedikt Soja | Co-conveners: Kyriakos BalidakisECSECS, Mattia Crespi, Christian Gerhards, Maria KaselimiECSECS, Randa NatrasECSECS, Michael Schmidt

Orals

| Thu, 27 Apr, 08:30–10:15 (CEST)

Room -2.47/48

Posters on site

| Attendance Fri, 28 Apr, 08:30–10:15 (CEST)

Hall X2

Posters virtual

| Attendance Fri, 28 Apr, 08:30–10:15 (CEST)

vHall GMPV/G/GD/SM

This session aims to showcase novel mathematical methods in geodesy, including enhancements of conventional statistical approaches as well as the application of machine learning techniques. In this regard, two areas that have seen significant developments in recent years are the analysis of geodetic time series and potential field data.

Modern satellite missions measuring the Earth's gravity and magnetic fields such as GRACE-FO and SWARM are continuing to provide data with ever improving accuracy and resolution. Hence, there continues to be a need to develop new methods of analysis, at the global and local scales, and especially on their interface. Furthermore, space geodetic techniques, including GNSS, deliver time series describing changes in the Earth system, such as the surface geometry, sea level change variations or fluctuations in the Earth's orientation. Geodetic observation systems usually measure the integral effect, whereas the aim is typically to understand the individual contributions of the Earth’s sub-components. In general, the amount of data from geodetic observation techniques has increased significantly in past decades. Innovative approaches are required to efficiently handle and harness the vast amount of geodetic data available nowadays for scientific purposes.

We invite contributions that address new mathematical developments in the analysis of potential field data and geodetic time series, and the application of machine learning techniques in general. Improved potential field data analysis may result from the application of wavelets, radial basis functions, Slepian functions, splines, spherical cap harmonics, etc. Time series analysis could benefit from new developments in the area of time-frequency analysis, detection of features of the spatio-temporal variability of signals, as well as signal separation techniques. The application of machine learning shows significant potential for automated processing of geodetic data, pattern and anomaly detection, combination and extraction of information from multiple inhomogeneous data sets, feature selection and sensitivity analysis, super-sampling of geodetic data, and improvements of large-scale simulations. Especially encouraged are contributions that discuss the uncertainty quantification, interpretability and explainability of results from machine learning algorithms, as well as the integration of physical modeling into data-driven frameworks.