5:15 PM - 6:45 PM
[MTT37-P08] Noise characteristics in GNSS time series based on robust spectral analysis: (2) application to GEONET F5 solution
Keywords:GNSS, GEONET, Spetral Analysis, Observation Noise
GNSS is an observation system to measure the locations of observation stations by receiving radio waves transmitted from the satellites. Due to this observation principle, raw GNSS data contain various information from satellites to receivers such as the ionosphere, atmosphere, surrounding environment close to the antennas, as well as the motion of the Earth. Thus, pre-processing of raw GNSS data is required to estimate the target products by modeling the other factors using analysis software.
The Geospatial Information Authority of Japan provides the pre-processed daily coordinates at the GEONET stations as the F5 solution (Takamatsu et al. 2023). Because of its convenient use, the F5 solution is frequently used for a variety of geodetic, seismic, and volcanic analyses related to crustal deformation. To precisely extract the transient deformation due to tectonic or volcanic processes, it is important to understand the spatio-temporal noise characteristics and to accurately model them. Nonetheless, the comprehensive analysis of spatio-temporal noise characteristics in the F5 solution has been rarely published.
Takabatake and Yano (2023) recently proposed a method to robustly model the observed characteristics in the frequency domain. They found that, by minimizing the spectral Renyi divergence, regression to the observed spectral becomes more stable against the presence of outliers, which reduces the effect due to pre-processing of the observations (see the accompanying abstract, Yano et al. in this session). Following this study, we apply their robust spectral estimation method to the GEONET F5 solution.
We used the F5 solution from 1996 to 2023, which was then divided into four years overlapping 0.5 years to calculate power spectral densities (PSD). We then modeled each PSD as the sum of the two Brune’s spectral models (Brune, 1970) representing both high-frequency and low-frequency components. In this model, the amplitudes, cut-off frequencies, and slopes of the spectra for both frequencies are estimated by the proposed method.
Our preliminary result indicates that, for low-frequency bands, the modeled spectral shows the colored noise with the slopes of ~-4.0–-1.5, while those for high-frequency bands are close to the pink to white noise with slopes of ~-1.5–0.0. In addition, the amplitudes of high-frequency component showed a temporally gradual reduction until ~2003. This may correspond to the improvement of the accuracy of the GNSS orbits provided by Internaltional GNSS service (Griffiths and Ray, 2009). We will investigate the characteristics of the estimated parameters in the spectral model in more detail.
The Geospatial Information Authority of Japan provides the pre-processed daily coordinates at the GEONET stations as the F5 solution (Takamatsu et al. 2023). Because of its convenient use, the F5 solution is frequently used for a variety of geodetic, seismic, and volcanic analyses related to crustal deformation. To precisely extract the transient deformation due to tectonic or volcanic processes, it is important to understand the spatio-temporal noise characteristics and to accurately model them. Nonetheless, the comprehensive analysis of spatio-temporal noise characteristics in the F5 solution has been rarely published.
Takabatake and Yano (2023) recently proposed a method to robustly model the observed characteristics in the frequency domain. They found that, by minimizing the spectral Renyi divergence, regression to the observed spectral becomes more stable against the presence of outliers, which reduces the effect due to pre-processing of the observations (see the accompanying abstract, Yano et al. in this session). Following this study, we apply their robust spectral estimation method to the GEONET F5 solution.
We used the F5 solution from 1996 to 2023, which was then divided into four years overlapping 0.5 years to calculate power spectral densities (PSD). We then modeled each PSD as the sum of the two Brune’s spectral models (Brune, 1970) representing both high-frequency and low-frequency components. In this model, the amplitudes, cut-off frequencies, and slopes of the spectra for both frequencies are estimated by the proposed method.
Our preliminary result indicates that, for low-frequency bands, the modeled spectral shows the colored noise with the slopes of ~-4.0–-1.5, while those for high-frequency bands are close to the pink to white noise with slopes of ~-1.5–0.0. In addition, the amplitudes of high-frequency component showed a temporally gradual reduction until ~2003. This may correspond to the improvement of the accuracy of the GNSS orbits provided by Internaltional GNSS service (Griffiths and Ray, 2009). We will investigate the characteristics of the estimated parameters in the spectral model in more detail.