Jasmine Peterson*: Machine Learning and Australia’s shape
Authors: Jasmine Peterson (University of Tasmania), Matt King (University of Tasmania), Daniel Smith (CSIRO Data61), Anna Riddell (Geoscience Australia), Christopher Watson (University of Tasmania)
Title: Machine Learning and Australia’s shape
Poster abstract:
Rapid expansion of Global Navigation Satellite System (GNSS) networks has produced large volumes of high-precision geodetic data, with hundreds of stations operating nationally and tens of thousands globally. These datasets offer potential to estimate Earth’s three-dimensional motion and change over time at sub-millimetre precision, supporting applications including the contribution of vertical land motion to sea level change. Practically, undetected offsets (sudden step changes) in time series limit this. Contemporary practice relies on manual identification by expert analysts. Although accurate, this is labour-intensive, subjective, and not scalable for modern GNSS datasets. There is a strong resulting motivation to develop robust, reproducible, and accurate automated methods. Existing approaches often lack the sensitivity required to detect small offsets. Recent machine learning advances, particularly in time series segmentation, change-point and anomaly detection offer new opportunities, by learning complex temporal patterns and distinguishing subtle structural changes in noisy data. Such methods have not yet been tailored to GNSS time series characteristics, including temporally correlated noise, irregular sampling, its three coordinate components, and seasonal variability. The desire to bridge traditional geodetic analysis and machine learning techniques motivates this project. In doing so, it aims to enable robust, scalable, objective, and high-precision detection of offsets that currently require expert interpretation. This poster examines the problem and showcases initial results from the first six months of this project. This work is relevant to researchers in geodesy, geophysics, and Earth system science, as well as machine learning practitioners interested in time series segmentation, anomaly detection, and environmental data.
Keywords: Machine Learning, GNSS, Time Series, Offsets, Geodesy