Spatio-temporal characteristics in the GEONET F5 solution in the frequency domain estimated based on the robust spectral analysis

Spatio-temporal characteristics in the Global Navigation Satellite System (GNSS) time series provide fundamental insights into data properties. Owing to its ease of use, the F5 solution, which is the most frequently used dataset from the GNSS Earth Observation Network System (GEONET), a nationwide GNSS network in Japan, is frequently employed in a wide range of geodetic, seismic, and volcanic analyses related to crustal deformation. However, comprehensive analyses of the spatiotemporal characteristics of F5 solutions have rarely been reported. Therefore, we aimed to model the spatio-temporal characteristics of these time series in the frequency domain over a few decades. The power spectral densities were estimated from each sliding four-year time window for each component and station and modeled using the two terms representing low- and high-frequency contributions. The model parameters were robustly estimated against the presence of outliers by maximizing the spectral Rényi divergence. The model parameters in the low-frequency components generally corresponded to the occurrence of transient deformations, such as postseismic movements, slow slip events, and movements related to volcanic phenomena. In contrast, the parameters in the high-frequency components were related to the positioning strategy itself, such as improvements in the accuracy of the GNSS orbits until 2003, and might be related to the deactivation of Selective Availability in 2000. The frequency dependence of the spectra indicated temporally correlated observation noise, even in the high-frequency components. All the estimated model parameters are publicly available and can be utilized for various research applications, including the generation of synthetic time series with realistic noise, which is particularly useful in machine learning studies, modeling or correcting transient deformation, discovering previously undetected slow slip events, and detecting anomalies related to the local environment of stations. A robust estimation method for spectral modeling can be applied to any GNSS time series, and it provides essential baseline information on the properties of the observed data.