O uso de machine learning para detecção de descontinuidades nas séries temporais das coordenadas da RBMC

The use of machine learning for detecting descontinuities in RBMC station coordinate time series

Abstract

The GNSS observations collected by RBMC allow estimating the daily coordinates of its stations with millimeter precision, and consequently robust time series capable of providing the determination of geodynamic displacements such as the movements of lithospheric plates, as well as the effects of terrestrial, oceanic and atmospheric tides. However, discontinuities present in the series may indicate changes in the station's reference coordinates, and need to be considered when significant. Events such as antenna changes and earthquakes are the main sources of discontinuities in time series. In this work, we evaluated the ability of machine learning algorithms to automatically identify discontinuities caused by antenna changes. Five methods were evaluated, and Random Forest presented the best result with an F1-Score of 0.78 and the correct identification of 77.5% of the discontinuities recorded with values greater than or equal to 1 cm. This study showed that machine learning is capable of classifying patterns from RBMC coordinate time series, but its quality depends on adequate data processing, as well as the representativeness of the events to be modeled.