2:00pm - 2:15pmOral_15
Post-Earthquake Fold Growth Imaged in the Qaidam basin, China, With InSAR
1University of Oxford, United Kingdom; 2Now at Université Grenoble Alpes, Université Savoie Mont Blanc, CNRS, IRD, IFSTTAR, ISTerre, Grenoble, France
Questions regarding the development of folds, the role of volumetric deformation, and their relationship with earthquakes within a fault and fold system remain unanswered. Estimating fault slip and earthquake hazard using surface observations requires an understanding of how shortening is accommodated along and across tectonic structures during different phases of the seismic cycle. However, measuring the small rates of inter-earthquake deformation with space-geodesy techniques in mountainous areas is still challenging. Here, we construct a 16-year timeline of surface deformation from ESA satellite radar measurements across the North Qaidam thrust system in Tibet, where three Mw 6.3 earthquakes occurred in a relatively short time interval along basement faults underlying folded sediments.
We process using NSBAS software the complete Envisat data archive along four overlapping tracks, as well as three Sentinel-1 tracks acquired in interferometric wide-swath mode including Interferograms with both short- and long-temporal baselines to avoid biases arising from the systematic loss of coherence in short-temporal baselines interferograms. Processing steps include correcting interferograms for tropospheric delays before unwrapping based on both empirical phase-elevation estimates and the use of the ERA-5 atmospheric models. After unwrapping and time series analysis, a parametric decomposition of the cumulative surface displacements into linear trends, co-seismic steps, and seasonal functions is performed to map, respectively, the changes of ground velocities, the co-seismic ground motions, and the seasonal ground displacements attributed to the freeze and thaw cycles of the water stored in the shallow layers of the ground. Long-wavelength spatial ramps for all interferograms are also iteratively re-estimated from bedrock pixels surrounding sedimentary basins affected by frost-related processes.
The analysis reveals spatio-temporal changes of post-earthquake surface displacement rates and patterns, which continue more than ten years after the seismic events. The decomposition of the Sentinel-1 ascending and descending LOS velocities into vertical and shortening post-earthquake components indicates that long-term transient uplift and shortening is in agreement with the deformation that might be expected from kinematic models of folding. Long-term uplift coincides spatially with young anticlines observed in the geomorphology, with steep gradients in the forelimbs, gentle gradients in the back-limbs, an absence of subsidence in the footwalls, and higher gradients along interpreted bedding planes. Long-term shortening is also different from the surface displacements expected for typical time-varying creep on a narrow dislocation interface and shows rates higher than the average convergence across the whole region. These findings highlight the need to integrate the contribution of non-elastic deformation processes within shortening fault and fold systems for better quantification of slip-on underlying structures and improved understanding of the wide variety of transient ground deformation that can now be detected with geodetic networks.
2:15pm - 2:30pmOral_15
Interseismic Strain Accumulation On The Main Recent Fault (Iran) From Sentinel-1 Data
1School of Earth and Environment, University of Leeds, United Kingdom; 2Department of Earth Sciences, Durham University, United Kingdon; 3School of Geosciences, University of Sydney, Australia
The Main Recent Fault (MRF) is a major right-lateral transform fault in the Zagros mountains of Iran. The fault has experienced Mw 6 earthquakes as recently as 2006 (Peyret et al., 2008), driven by 20-30 mm/yr of convergence between the Arabian and Eurasian plates (Khorrami et al., 2019). An accurate estimate of the rate of interseismic strain accumulation is critical both for estimating local seismic hazard, and for developing understanding of the overall distribution and mechanics of continental deformation across Iran, which constitutes one of the widest zones of continental convergence on a global scale. However, there has been major disagreement between previous estimates of the rate of slip on this important fault: previous estimates from regional GNSS observations and geological offsets vary between 1.6-17 mm/yr (Talebian and Jackson, 2002; Hessami et al., 2006; Authemayou et al., 2009; Alipoor et al., 2012; Khorrami et al., 2019). The GNSS data in particular are relatively sparse in the region of the MRF, but whilst the fault is favourably orientated for measurement with InSAR, no previous InSAR velocity estimates have been published to-date.
Here we use Sentinel-1 observations to make the first InSAR estimate of the interseismic slip velocity and locking depth for a ~400 km section of the MRF. We use the LiCSAR automated system to process five years (2015-2020) of Sentinel-1 SAR acquisitions for four adjacent and overlapping frames (two descending and two ascending), producing a total of ~1500 interferograms. We apply a NSBAS-type approach using LiCSBAS (Morishita et al., 2019) in order to derive satellite line-of-sight velocities, mitigating the effects of atmospheric noise using GACOS (Yu et al., 2018). We estimate north-south velocities from regional GNSS (Khorrami et al., 2019) and use this to isolate the fault-parallel and vertical velocity components from the overlapping ascending and descending InSAR frames. Finally, in order to estimate the slip-rate and locking depth for the MRF, we fit 1-D screw dislocations models (Savage and Burford, 1973) to fault-perpendicular profiles, using a Bayesian approach as described by Goodman and Weare (2010) and Hussain et al. (2016).
Our results show an interseismic slip velocity of 3±2 mm/yr below a locking depth of 20 km. The locking depth is poorly constrained because of the presence of incoherence and large subsidence signals close to the fault trace, and because the slip rate is close to the current sensing limit of Sentinel-1 time series (2-3 mm/yr). These results show that the MRF is an important major crustal structure that shows clear localisation of strain at depth and which accommodates a significant portion of the relative motion between Arabia and Eurasia. However, the overall slip rate is towards the lower end of the range of previous estimates, supporting previous GNSS-derived rates but which is inconsistent with longer-term estimates.
2:30pm - 2:45pmOral_15
InSAR-Based Earthquake Slip Inversion Facilitated by Statistical Hypothesis Testing
School of Surveying and Geospatial Engineering, University of Tehran, Iran, Islamic Republic of
Estimation of coseismic slip distributions constrained by geodetic data is an important step in seismic analyses and earhquake cycle studies. Obtained slip distributions can be exploited to infer various characteristics of earthquakes such as rupture geometry, stress-strain distribution on adjacent faults, and consequently hazard assessment for future tremors. The mathematical model of slip inversion is often ill-conditioned and requires some sort of a priori assumption or regularization to achieve a stable and reliable solution. For this purpose, there are different a priori models and assumptions. The most common assumption is the smoothness of the slip behavior, which is imposed on the solution by the minimization of the second spatial derivative (or Laplacian) of the slip. More recently, some other assumptions have been also introduced such as slip self-similarity behavior that is incorporated in the slip inversion by a self-similar autocorrelation function (e.g., the Von Karman or exponential correlation function). In this regard, there are two challenging issues as follows.
- The ill-conditioned nature of the slip inversion implies that we can get similar misfit (between data and the model) for different a-priori assumptions. The main question is then which model/assumption is the best given the statistical characteristics of the data? For example, if we apply both Laplacian and Von Karman regularization on a dataset and both provide similar misfit, how can we judge which model is better?
- Another property of the regularized slip estimates is that they are biased. Consequently, the estimate of each slip patch has a relationship to many (or even all) other patches. This can result in a spurious leakage of the slip to the non-slipping patches. This effect is more significant in the methods that use the self-similarity slip assumption. So to obtain reliable results, it is important to know beforehand what is the real fault rupture area (i.e., which patches have a significant slip) and to avoid the inclusion of the non-slipping patches in the slip inversion.
In this study, we try to answer both above questions using the concept of statistical hypothesis testing and by exploiting the knowledge about the statistical properties of geodetic data (in particular InSAR data). For the first problem, we develop a hypothesis-testing scheme based on the probability distribution of the norm of the misfit vector and also the characteristics of the a priori assumption. The model/assumption that have a higher p-value is then selected as the best model. The main property of the proposed approach is that the test not only depends on the misfit and the data statistics but also it accounts for the structure of the a priori assumption. In this way, the proposed test is capable to discriminate between two models with similar misfit.
For the second problem, we propose to apply the parameter significance test (PST) on each patch in an iterative manner. We first estimate the slip and its precision (variance) for each patch. Then using the PST we decide whether the estimated slip is significant or not. If not, the patch will be removed from the rupture area and the procedure will be iterated until all the estimated patches have a significant slip. Note that in the both proposed tests, the data noise structure (or covariance matrix) is required. In case of InSAR data, the noise covariace matrix is obtained based on statistical analysis of data over a stable area far from the fault. The performance of the proposed tests is validated by synthetic study, follows by application of the method on the 2015 Mw 8.3 Chile earthquake using Sentinel-1 InSAR data.
2:45pm - 3:00pmOral_15
Interseismic Strain Accumulation And Fault Creeping Behavior Detection At Shallow Crust Along Big Faults Over Tibetan Plateau Region By Parallel Processing Of Sentinel-1 InSAR Time-Series Data
Institute of Geology, China Earthquake Administration, China, People's Republic of
We process Sentinel-1 Synthetic Aperture Radar (SAR) data accumulated in the past ~5 years to study the crustal deformation of the Tibetan Plateau region. In particular, the interseismic behaviors of big faults are considered for strain accumulation or release by creeping activities, such as the Altyn Tagh fault (ATF) at the northern boundary, the Haiyuan fault (HF) at the northeastern boundary, the Kunlun fault (KF) in northern Tibet, and the Xianshuihe-Xiaojiang fault system (XXF) in the eastern or southeastern Tibet regions. As reviewed by Harries (2017), some big faults over the world in different tectonic environments are found to be creeping. Among these creeping faults, the strike-slip, rather than dip-sip faults, are more popular. It is not clear yet the role of creeping faults or fault sections on seismic hazards due to limited samples and related observations. Based on ample SAR data acquired in the past few years, we now have high-quality geodesy data for fault creeping detections. By analyzing the interseismic Spatio-temporal behaviors of Tibetan faults, we expect to find locations of creeping fault sections, analyze its relationship to locked parts with strain accumulation, and reveal its seismic hazards.
To accelerate the data processing in such a large region, we utilize a high-performance computation (HPC) facility, named the Sunway TaihuLight, and parallelized computational codes for conventional (ISCE, Gamma, etc.) and time-series (StaMPS) InSAR analysis (Hooper et al., 2007). We adopt the persistent scatterer (PS) method for deformation detection, as the arid condition of the study area is quite suitable for PS identification with high enough density and the processing is easy to be parallelized by patch decomposition. Most of the processing steps work automatically, except the atmospheric filter step, in which we try with different options to minimize the atmospheric phase screen (APS) in time-series data. In particular, we reduce the APS by using the ECMWF ERA-5 model, which is proved to be useful in a statistical framework and could reduce observation uncertainties to some extent depending on InSAR coherence. Finally, we use the surrounding GPS data provided by Wang and Shen (2020) to calibrate InSAR observations in multiple orbits (descending and ascending pass geometry) and invert for high-quality 2D ground deformation and produce fault-parallel motion maps (Shen and Liu, 2020). It is also critical for using time-series GPS observations to calibrate InSAR time-series results, as long-wavelength errors and strong water vaper delays could seriously bias temporal behavior assessments of big faults, leading to unstable and wield motions inconsistent with limited cGPS observations. Besides, we also applied a more conventional while robust SBAS inversion code (Mintpy) to do InSAR time-series analysis on multi-looked SAR data, by evaluating phase closure distributions, temporal coherence, and bootstrap uncertainties of InSAR velocities, which is implemented in parallel by DASK technique. Both PS analysis in full resolution and SBAS processing in reduced resolution are carefully compared for consistency on InSAR velocity estimate.
As some consistent results, we obtain the fault creeping signals along the Haiyuan fault (such as Jolivit et al., 2012, 2013) and the Xianshuihe fault (Zhang and Wen, 2014; Allen et al., 1994; Li and Burgmann, 2020), which is already known in previous studies, but with a more clear picture of fault motion. Besides, we also find some creeping signals along different sections (western and middle) of the Altyn Tagh fault system. Though the creeping signals are only localized at a short distance (several 10s of km), compared with the >1500 km ATF, the role of the creeping sections may have some implications for the evolution of the ATF and its seismic hazards at different sections. By invert for interseismic strain accumulation along locked sections and slip distribution on creeping sections along those faults, we infer that the creeping behaviors of big faults could be related to nearby big events within very localized parts, and most parts of those faults are actually still in locked status, and accumulated strains could be released in future earthquakes.
3:00pm - 3:15pmOral_15
Seasonal Deformation in South Iceland from InSAR – Influence on Earthquake Activity?
King Abdullah University of Science and Technology (KAUST), Saudi Arabia
A surprisingly large portion of large historical earthquakes in South Iceland have occurred in Spring and early Summer, with far less events during the wintertime. Statistical testing shows that this occurrence pattern is very unlikely due to chance, indicating that some seasonal process is influencing the timing of the earthquakes. We use Sentinel-1 time-series analysis to study and quantify the seasonal deformation in the area. Using snow-free images acquired from early Summer to Fall, the results show that the center of Iceland is strongly uplifting over time, most likely in response to reducing load of the major ice caps in Iceland. In addition, we find clear seasonal deformation with fast uplift during Summer, in response to snow melting, while in Winter the deformation reverses to subsidence. These results are in accord with continuous GPS measurements that show similar deformation patterns. Also, the results indicate stronger seasonal variations in the eastern part of the South Iceland seismic zone (~10 mm), closer to significant snow loading, than in its western part (~5 mm). This leads to differential seasonal motions across the seismic area (east-west distance about 60-70 km) that cause spatio-temporal subsurface stress changes that may influence the timing of earthquakes. We assess these seasonal variations by modeling the load-induced stress changes through a single year both in elastic homogenous and in layered half-space models. For this, we account for snow and glacier loading, as well as atmospheric and ocean load variations. Our results show that snow loading puts the seismic zone into compression during the late winter and snow melting during Spring and Summer in turn leads to relaxation of the fault zone compression. The peak in earthquake occurrence is in May and June, or soon after seasonal unloading starts. Therefore, the earthquake rate appears to correlate better with maximum unloading rate rather than the peak of the unloading in the Fall. The periodic load-induced stresses on seismogenic faults in South Iceland are however small (~1 kPa) and appear to only mildly modulate the rapid tectonic stressing rate in the area of ~20 kPa/year. This suggests that other mechanisms may be more influential in controlling the timing of large earthquakes in South Iceland.