Detection of geodynamic anomalies in GNSS time series using machine learning methods
І. Brusak, Kamil Maciuk, Oleh Haidus · GEODYNAMICS · 2025
One of the applied geodetic tasks in geodynamics is the detection of anomalous deviations in GNSS time series, which may indicate deformations of the Earth's surface caused by various geophysical phenomena. It is important to note that geodynamic anomalies may be of a local nature, manifesting at a single GNSS station, or of a regional nature, occurring simultaneously across a group of GNSS time series. The objective of this article is to develop a method for detecting geodynamic anomalies in GNSS time series using machine learning algorithms. The method has been implemented in the Python environment and allows for the semi-automated analysis of large datasets. Among the machine learning methods, the Isolation Forest algorithm was selected for this study. The research provides a detailed step-by-step description of the program’s operation and its stages, enabling the analysis of both individual time series for identifying local anomalies and groups of time series for detecting concurrent regional geodynamic anomalies. The developed method was tested on data from 37 GNSS stations of the GeoTerrace network located in western Ukraine. As a result, seven distinct groups of horizontal and vertical anomalies were identified. One of the detected anomalies was established to correspond with previously investigated vertical crustal deformations caused by non-tidal atmospheric loading in December 2019. The study presents maps of the spatial distribution of the detected group height anomalies in November 2022 and January 2013. Some anomalies observed at certain GNSS stations are of unknown origin and may be due to unidentified local geodynamic factors or measurement errors. In addition to its relevance for geophysicists and geologists in detecting collective geodynamic anomalies, the proposed method also demonstrates potential for use in structural health monitoring of large engineering constructuctions using data from GNSS station networks.