5:15 PM - 7:15 PM
[SCG45-P26] Strain rate field estimation in Southeastern Europe for evaluation of relationship with seismicity
Keywords:Strain rate field, GNSS, basis function expansion method, ABIC
Strain rate field estimation is important to understand crustal deformation. Recent studies have compared geodetic moment calculated from strain rates with seismic moment calculated from historical large earthquakes to investigate their usefulness for seismic hazard assessment (e.g., Donniol et al., 2024; Nishimura, 2022). The crustal deformation rate in most areas of Europe is so small that the signal-to-noise ratio is very low. There are also areas of large deformation, such as Italy and the Southern Balkans. These areas of large and small deformation are adjacent to each other. In addition, the velocity uncertainties vary widely depending on each station. Here, we estimate the strain rate field in Southeastern Europe considering spatial smoothness of strain rate and incorporating velocity uncertainties to compare them with seismicity quantitatively.
We use GNSS velocity solutions combining ten previously published velocity fields with outliers removed (Piña-Valdés et al., 2022). We apply the basis function expansion method (Okazaki et al., 2021) to horizontal secular velocities to estimate strain rate fields in Southeastern Europe (15 km-intervals of basis functions). This method can objectively optimize the spatial smoothness of the strain rate field using ABIC. We newly incorporate velocity uncertainties into the error matrix with an unknown scale factor of the variance.
Large extension is estimated in Italy and the southern Balkans, consistent with previous results (Piña-Valdes et al., 2022). In Central and Eastern Europe, extension and compression perturbations of the order of tens of nanostrain/yr are estimated at scales of tens of kilometers. However, these perturbations are smaller than the estimation error, suggesting that spatial smoothness is optimized to explain large localized deformation areas (or areas where stations are densely distributed). These results may stem from the fact that only one hyperparameter for the estimated region determines spatial smoothness, suggesting the need to vary spatial smoothness from place to place.
To incorporate spatial variations in spatial smoothness into strain rate field estimation, we divide the estimated region with overlap. 10-fold cross-validation is used to evaluate appropriate area division. As a result, the strain rate field in Central and Eastern Europe is optimized to a much smoother result compared with the estimated result without area division. In addition, we found not only the strain rate field itself but also the estimation error of strain rates highly depends on how we divide regions. These results suggest that plausible prior information and error matrix are important to estimate a plausible strain rate field.
We use GNSS velocity solutions combining ten previously published velocity fields with outliers removed (Piña-Valdés et al., 2022). We apply the basis function expansion method (Okazaki et al., 2021) to horizontal secular velocities to estimate strain rate fields in Southeastern Europe (15 km-intervals of basis functions). This method can objectively optimize the spatial smoothness of the strain rate field using ABIC. We newly incorporate velocity uncertainties into the error matrix with an unknown scale factor of the variance.
Large extension is estimated in Italy and the southern Balkans, consistent with previous results (Piña-Valdes et al., 2022). In Central and Eastern Europe, extension and compression perturbations of the order of tens of nanostrain/yr are estimated at scales of tens of kilometers. However, these perturbations are smaller than the estimation error, suggesting that spatial smoothness is optimized to explain large localized deformation areas (or areas where stations are densely distributed). These results may stem from the fact that only one hyperparameter for the estimated region determines spatial smoothness, suggesting the need to vary spatial smoothness from place to place.
To incorporate spatial variations in spatial smoothness into strain rate field estimation, we divide the estimated region with overlap. 10-fold cross-validation is used to evaluate appropriate area division. As a result, the strain rate field in Central and Eastern Europe is optimized to a much smoother result compared with the estimated result without area division. In addition, we found not only the strain rate field itself but also the estimation error of strain rates highly depends on how we divide regions. These results suggest that plausible prior information and error matrix are important to estimate a plausible strain rate field.