Conference Agenda

Geodetic measurements of surface strain rate can be used to estimate seismic moment accumulation rate (Kostrov, 1974; Savage & Simpson, 1997; Ward, 1998) and thus inform seismic hazard models. Given the high-quality geodetic data available today from InSAR and GNSS, the largest uncertainty in this calculation is related to the estimation of the thickness of the seismogenic layer and its spatial variations. Most moment rate analyses use a single value of seismogenic thickness (~12 km) based on the average seismicity depth. However, along the San Andreas Fault system (SAFs), the 95% seismicity depth varies between 9.3 and 17.5 km (Lin et al., 2007) suggesting there are large spatial variations in moment accumulation rate. Moreover, locking depth estimates from geodetic data have even a larger range (5.9 – 21.5 km) and in a few cases do not agree with the seismicity depths (Smith-Konter et al., 2011).

In this study we use a 3-D earthquake cycle model, having large variations in plate thickness (Ward et al., 2021), to better understand how moment accumulation rate varies spatially. We calculate seismic moment accumulation rate using two independent methods. The modeling approach uses geodetic data to solve for slip rate and locking depth on 32 segments of the SAFs. In addition, the model includes spatial variations in the thickness of the seismogenic elastic layer based on surface heat flow and the depth to the lithosphere asthenosphere boundary (Thatcher et al., 2017). The results show large spatial variations in moment accumulation rate with high moment accumulation rates on the Carrizo segment slipping at 36 mm/yr and 6 times lower rate on the Imperial segment slipping at 44 mm/yr.

The second approach to the moment rate calculation uses a new high spatial resolution strain rate map derived from high-precision GNSS velocities (NASA MEaSUREs ESESES project, Bock et al., 2022), and velocities estimated from 7 years of integrated Sentinel-1 InSAR+GNSS time series observations (Xu et al, 2021; Guns et al., 2022). The major unknown in this approach is the seismogenic thickness and its spatial variations. A uniform seismogenic thickness of 8.4 - 10 km results in a total moment accumulation rate that matches the homogeneous earthquake cycle model. However, the data and model approaches disagree in their spatial variations. We are investigating whether the accuracy of the strain rate approach can be improved by varying the seismogenic thickness spatially. In addition, we attempt to discriminate between on-fault (32 main segments) and off-fault moment accumulation.

The InSAR Community Geodetic Model (CGM) working group, which was established as part of the Southern California Earthquake Center (SCEC), has been focused on advancing research into improving InSAR processing techniques, establishing best practices, reaching a community consensus for the best InSAR-based deformation time series and velocity model for Southern California, and exploring integration with GNSS. Our motivation is to create a set of self-consistent and well documented products (time series and velocities) over southern California, and make them easily accessible to the Earth science community through a searchable web interface.

Since its launch in 2014, the Sentinel-1 mission has provided a dataset characterized by unprecedented accessibility, spatiotemporal coverage, and cadence, making it possible to study temporally variable crustal deformation. Constraining the ground deformation associated with the earthquake cycle, long-term tectonics, hydrologic cycles, geothermal features, and other processes is crucial to understanding the seismogenic phenomena in Southern California. Our model consists of time series and velocities from four overlapping ascending and descending Sentinel-1 tracks in Southern California, spanning the time from 2015 to the Ridgecrest earthquakes in mid-2019. The model is a combination of six different line-of-sight (LOS) deformation time series and velocity solutions that were provided by groups at SIO/UC San Diego, UC Berkeley, USGS, UC Riverside, and NASA JPL. Each group has used a distinct approach to estimate the deformation time series and velocities, all of which we summarize in this presentation. We show that a combined model has an advantage over the individual solutions, as it suppresses post-processing artifacts that characterize different time series estimation techniques.

We have corrected the combined InSAR velocities for the absolute bulk plate motion, and we show that this correction improves the agreement between InSAR and GNSS velocity datasets. We have explored and compared several methods for estimating InSAR velocity and time series uncertainties. To calculate uncertainties for our consensus InSAR velocity product, we chose to apply a method that is commonly used with GNSS time series, incorporating both white and temporally correlated noise sources. Using an assumption of flicker noise, we calculate the covariance matrix for every pixel following the equations of Zhang et al. (1997) and perform a time series model inversion (including velocity and seasonal terms) to obtain the uncertainty estimate. This approach allows us to obtain a robust uncertainty estimate without requiring individual measurement uncertainties.

Our updated InSAR CGM, including the latest corrections and uncertainties as well as the tools for working with the dataset, will be made available on the SCEC website: http://moho.scec.org/cgm-viewer/. The development of the model is ongoing, and it will be updated regularly. Current research is focused on the integration of the InSAR velocities and time series with GNSS for a joint Community Geodetic Model, expanding the temporal and geographical extent of our products, and isolating deformation signals due to the 2019 Ridgecrest earthquakes.

Large earthquakes deform surrounding rocks and, for the largest of them, we can derive fault displacements at the ground surface using InSAR. The release of the stored elastic deformation during earthquakes results in surface displacement patterns that vary in space around the principal fault plane. For blind earthquakes, where the rupture plane does not reach the ground surface, the observed displacement field appears largely smooth as it is continuous around the seismic source. However, for a number of coseismic interferograms from M~6 blind normal fault earthquakes, we observe lineaments that disrupt the otherwise smooth displacement field, indicating localized strain. This localized and inelastic strain is small, mostly involving less than 2 cm of shallow fault slip, but is mappable for kilometers along pre-existing tectonic faults, apparently not involved in the causative rupture. Rather these faults are thought to have been activated in response to slip at depth on the primary fault. The resulting surficial fractures appear to have been accommodated by pre-existing zones of weaknesses which are inherently weaker than the surrounding rocks. Such secondary fault activation, which herein we call synseismic slip, has previously been reported for numerous earthquakes, including the strike-slip Mw6.4 and Mw7.1 2019 Ridgecrest and the Mw6.2 and Mw7.0 2016 Kumamoto earthquake sequences.

Here, we record examples of similar synseismic slip but for smaller (i.e. M~6) normal fault earthquakes that result in subtle (< 5 cm) but measurable slip at the ground surface. Synseismic fault displacements are here detected for the 2021 Tyrnavos sequence in central Greece, the 2021 Arkalochori earthquake in Crete (Greece) and Tibetan earthquakes in Asia in 2020. To detect such small displacement changes, we used Sentinel-1 interferometric wide-swath SAR acquisitions, processed in high resolution, and compared the observed surface strain patterns with the modeled surface strain caused by the mainshocks. We can show that indeed the type of synseismic slip is controlled by the mainshock strain regimes: synseismic normal slip is commonly observed in zones of dilatation, reverse faulting in compressional zones and slip reversals are observed along individual faults where the strain field changes (from compression to extension and vice versa).

In summary, our work suggests that synseismic activation of faults during large-magnitude earthquakes may be more common than previously thought and also shows that the type of synseismic slip detected at the ground surface is controlled by the local stress field resulting from the mainshock. Further, while our observations confirm textbook knowledge, they provide a basis for exploring a number of key questions such as why are some faults synseismically activated and others not? what can this tell us about relative fault weakness (or stress state)? at which depth is synseismic slip accommodated? does it compensate some of the observed shallow slip deficit and/or does it play a significant role in the seismic cycle of the fault that slipped syn-seismically?

The Ridgecrest earthquake pair consists of an M_w6.4 foreshock and an M_w7.1 mainshock that ruptured a set of orthogonal faults in the Eastern California Shear Zone. Despite being the most well-studied Californian earthquakes in history, little is known of the Ridgecrest early post-seismic stage. Resolving the post-seismic deformation is challenging due to the sparseness of the geodetic network in that area, and due to the orientation of the Mw7.1 rupture being subparallel to the available satellite orbits. Consequently, post seismic slip on that segment has not been well resolved by conventional across-track SAR interferometry. To overcome this difficulty, we take advantage of the Terrain Observation with Progressive Scan (TOPS) imaging mode of the Sentinel-1 radar satellites that enables along-track interferometry in areas of burst overlaps (BOI).

The BOI is applied to strips that are measured twice in each satellite pass, once in a forward and once in a backward looking direction. Since the two looking geometries differ by about 1°, the along-track displacement can be retrieved from the phase difference between the forward- and backward-looking interferograms in the overlap regions. The method is sensitive mainly to the azimuthal component of the deformation field, sub-parallel to the direction of the major co- and post-seismic motions.

We show results from 2 years following the M_w7.1 mainshock. We find that the surface displacement field sub-parallel to the fault is asymmetric with respect to the fault trace, with the maximum displacement ~10 km NE of the M_w7.1 fault termination. The BOI and available GNSS data match almost perfectly, while the InSAR and the horizontal GNSS projected to the range direction show up to 3 cm of discrepancy, due to the vertical component near the fault trace. Elastic inversion of the BOI and GNSS data suggest that post-seismic slip is distributed among the main fault and secondary fault/s in the northeastern sector of the fault.

The Ridgecrest inferred post-seismic slip seems to terminate at the Garlock fault in the south and to decay toward the Coso geothermal range in the north. The BOI displacement decay rates become higher close to the geothermal range. A similar trend of decay is observed for the seismicity rate, implying a possible effect of the higher ductility in the geothermal area.

EPOSAR [1] is a scientific service of the EPOS (European Plate Observing System) Research Infrastructure [2], developed by CNR-IREA, that provides co-seismic displacement maps at global scale. In particular, following the occurrence of an earthquake of a) magnitude greater and b) depth shorter than selected thresholds, EPOSAR automatically retrieves and process all the Copernicus Sentinel-1 data necessary to generate all the possible DInSAR co-seismic maps within a monthly time window, so that the earthquake can be analyzed from different satellite paths. EPOSAR exploits the P-SBAS [3,4] processing chain (up to the interferogram generation step) to generate the DInSAR products, which is deployed in a Cloud Computing environment hosted by AWS.

Clearly, only those earthquakes that occur on Land or significantly close to Land (i.e. the induced surface displacement is expected to be detected on land) are considered. To do this, by exploiting the moment tensors provided by public catalogs (USGS, INGV, Global CMT project), EPOSAR relies on a forward modelling procedure that generates the predicted co-seismic displacement field, used by the P-SBAS algorithm to optimize some of the DInSAR processing steps. This also allows to optimize the extension of the investigated area and to reduce the processing time by effectively exploiting the available computing resources.

The EPOSAR service is currently operative and the generated DInSAR products are freely available to the scientific community through the EPOS infrastructure [5]. Up to February 2023, EPOSAR catalog contains about 15000 products among wrapped interferograms, displacement maps and spatial coherence. This amount of DInSAR products is relevant to 552 earthquakes that occurred around the globe from 2015 to 2023 and that respect the service constraints of magnitude, depth and land coverage.

In this work we present a processing chain we implemented to retrieve, in a completely automatic way, the seismic source with distributed slip starting from the DInSAR co-seismic displacement maps generated through the EPOSAR service.

The implemented processing chain is, as said, fully automatic and acts in cascade to the EPOSAR service, with the aim to: reveal the seismic source at the occurrence of every new event detectable through DInSAR and provide a complete database of sources that includes all the earthquakes occurred since the launch of Sentinel-1 satellites and that can be observed by them.

The procedure starts from the DInSAR data, produced by the EPOSAR service, and a focal mechanism automatically retrieved from several catalogs (USGS, Global CMT, INGV-TDMT). First, a non-linear inversion is implemented with a coarse and a refined stages, to get a robust and well centered, uniform slip solution. The so retrieved source is then extended and subdivided into small elements to get the slip distribution via linear inversion. For every single step, a number of algorithms, based on two decades of experience in modeling at INGV, were implemented to face the large number of options and conditions usually handled by an expert user: image selection, setup and iterative update of the input parameters, definition of the regularization strength, detection of specific conditions (point-source, poorly constraining data, etc.). Moreover, with the availability of new DInSAR data, the model is also automatically updated, always balancing the contribution from ascending and descending acquisitions.

The developed tool is designed to deploy a service aimed at providing a quick and reliable automatic fault model solution and it has been tested and validated on hundred events, which are characterized by different magnitudes, rupture mechanisms and locations (see Figure 1). The main algorithm aspects and performance, and the capabilities arising with the availability of a complete and homogeneous database of DInSAR-based source models (updated scaling factors, systematic bias, etc.) will be discussed at the conference. We also remark that such huge database of displacement maps and source models can be exploited to train Artificial Intelligence algorithms (convolutional neural networks, for instance) that are aimed at automatically identifying ground deformation patterns in noisy interferograms.

Finally, while been already under pre-operation, our tool will be soon operative and integrated within the EPOS infrastructure, thus allowing the user community to access the generated results and benefit from quick and reliable products on the source mechanisms of the more significant seismic events.

This work is supported by the 2022-2024 IREA-CNR and Italian Civil Protection Department agreement, and by the H2020 EPOS-SP (GA 871121) and Geo-INQUIRE (GA 101058518) projects.

References

1. Monterroso et al. (2020) “A Global Archive of Coseismic DInSAR Products Obtained Through Unsupervised Sentinel-1 Data Processing,” Remote Sens., vol. 12, no. 3189, pp. 1–21. https://doi.org/10.3390/rs12193189

2. EPOS web site: https://www.epos-eu.org/

3. Casu et al. (2014) “SBAS-DInSAR Parallel Processing for Deformation Time Series Computation”, IEEE JSTARS, doi: 10.1109/JSTARS.2014.2322671

4. Manunta et al. (2019) “The Parallel SBAS Approach for Sentinel-1 Interferometric Wide Swath Deformation Time-Series Generation: Algorithm Description and Products Quality Assessment”, IEEE Trans. Geosci. Remote Sens., doi: 10.1109/TGRS.2019.2904912

5. EPOS Data Portal: https://www.ics-c.epos-eu.org/