Application of Outlier Detection Methods in GNSS Time Series Analysis

Abstract

In the study of determining vertical displacements of the Earth's crust, GNSS is the technology that enables the highest accuracy in displacement measurement. Moreover, with GNSS time series data, it is possible to identify patterns of displacement over time. An existing issue to address is the detection of outliers and discontinuities within the measurement series. This study investigates outlier detection methods within GNSS time series data to serve the purpose of determining vertical displacements and predicting altitude component values over time. Methods such as IQR, Z-Score, and Percentile were implemented using data from CORS stations named HYEN, QNAM, and CTHO within the VNGEONET network in Vietnam. The data from these stations were initially analyzed using Gamit/Globk software to obtain daily coordinate components of the points. Results from outlier detection and analysis with the Multiple Linear Regression Model indicate that with approximately 2% of measurements identified as outliers, displacement may vary by 0.4mm/year. The LSTM+ICA artificial intelligence model demonstrated excellent performance in prediction with QNAM and CTHO datasets. However, prediction with the LSTM+ICA model raises ongoing research questions, particularly regarding the data collected by the HYEN station.