Earthquakes and tectonics I

Sentinel-1’s continuous observation program over all major plate boundary regions makes it well suited for earthquake studies. However, decorrelation due to large displacement gradients and limited azimuth resolution of the Terrain Observation by Progressive Scan (TOPS) data challenge acquiring measurements in the near field of many earthquake ruptures and prevent measurements of displacements in the along-track direction. Here we present how to fully exploit the coherent and incoherent information of TOPS data by using standard InSAR, split-bandwidth interferometry in range and azimuth, burst-overlap interferometry, and amplitude cross-correlation to map displacements in both the line-of-sight and the along-track directions, and from the far field to the near field.

In the applications to the 2015 Tajikistan earthquake, the 2016 Kumamoto (Japan) earthquake and the 2016 Kaikoura (New Zealand) earthquake, InSAR provides the most accurate results in the far field but is useless in the near field due to phase aliasing and decorrelation. For the first time, the split-bandwidth interferometry in range is applied to reveal the near-field LOS displacement where unwrapping of standard interferogram becomes impossible due to phase aliasing. Pixel offsets derived from amplitude cross-correlation are used to map the rupture trace, and to fill the final blank where interferometric phases are completely decorrelated. We also map the displacement in the azimuth direction along the 1.5 km wide burst-overlap belts using bust-overlap interferometry. Adding near-field and azimuth constraints increase the resolution of the slip distribution particularly in the shallow part of the crust, implying that it is essential to take advantage of all the coherent and incoherent information in the Sentinel-1 TOPS data for studying geodynamic processes that produce large displacements.

Understanding the relative importance of aseismic and seismic deformation in fold-and-thrust belts requires long time series of geodetic data which capture various stages of the earthquake cycle. Furthermore, observations need to have sufficient spatial resolution in order to capture motion on individual faults and the growth of folds.

Three generations of ESA SAR satellites (ERS1/2, Envisat and Sentinel-1a/b) cover a time period from 1991 through to the present day. The images acquired by these satellites are sufficiently high resolution to capture variation in deformation over short length scales and investigate the earthquake cycle in a fold-and-thrust belt.

We use ERS1/2, Envisat and Sentinel-1a/b SAR data to form an InSAR time series which captures 25 years of deformation in the region of Harnai, Pakistan. This time period includes the 1997 M7 Harnai earthquake which ruptured a series of faults in the fold-and-thrust belt (Nissen et. al, 2016). We use preseismic, coseismic and postseismic data to examine the role of each stage of the earthquake cycle in the evolution of this fold-and-thrust belt and place constraints on fault rheology.

Our results show a range of deformation styles. Some structures are growing at constant rates throughout the observation period with only minor postseismic rate changes, whilst others show postseismic rates 15 times faster than those before the earthquake. The greatest postseismic deformation rates are strongly correlated with short wavelength folds suggesting that these structures grow, at least in part, following large earthquakes.

Preliminary fault modelling suggests that postseismic fault slip is concentrated both up-dip and down-dip of the coseismic rupture. The degree to which this slip can be explained by the coseismic stresses offers an opportunity to constrain fault friction parameters.

These results have implications for the seismic hazard in the area as well as contributing to our understanding of the dynamics of fold and thrust belts. The results also demonstrate the value to tectonics of the long time series of deformation observations that successive ESA missions have made possible.

It is challenging to effectively extract the near-field deformation of large earthquakes using InSAR approach, especially from the short-wavelength radar sensors, such as the C-, X- band systems. The issue leads to incomplete deformation field of large earthquakes in the vital regions, which are critical for earthquake studies. Some of the well-known decorrelation effects could prevent successful phase unwrapping and signal loss, such as the geometric (or spatial), temporal, and doppler decorrelation etc. The decorrelation effects may reduce the received energy of satellite radar sensors from the back-scattering of ground targets. Hence, the incoherence (or partial of it) of radar signals will greatly reduce the signal-to-noise ratio of InSAR phase and leads to information loss in the deformation field or serious phase jumping issues in the final products. In the near-field of large earthquakes, high phase gradient may be irresolvable by traditional unwrapping methods, in addition to the decorrelation effects. In extreme situations, the complete incoherence of radar returns could occur when the phase difference of neighbor pixels exceeding ∏  radians. Except for this physical limitation of the radar systems, it is possible to extract useful information from InSAR data. Moreover, severe DEM errors (depending on baselines and topography) will also greatly influence the phase unwrapping process and introduce heavy phase jumping errors, leading to the unreliability of InSAR observations. 

In order to overcome the limitations described above, we developed a simple strategy for phase unwrapping. Due to strong phase gradient following large earthquakes, we first multi-look the InSAR phase to a lower horizontal resolution and implement conventional Goldstein filtering and unwrapping. After the step, we use a 2-D spatial filter to estimate the deformation in its first order and remove this component from the unwrapped phase. We do the same procedure in an iterative way until it is impossible to unwrap the residual phase. The final residual phase plus the filtered phase using the Goldstein filter are deemed as the total residuals for the wrapped phase. We then resample the total residuals to the original horizontal resolution of InSAR phase and removed it from the observations. Then the conventional unwrapping procedures will be able to used for extracting InSAR phase covering even the near-field of earthquake deformation area. The total residuals could also include useful information excluded from earthquake deformation, such as localized deformation of landslides, small areas of uplift or subsidence, besides the DEM error phase. For some large ground ruptured events, it is also possible to use the range offset data to simulate the InSAR phase data to estimate the first-order deformation features of earthquakes, and simplify the unwrapping process.

We had successfully applied the method to some large earthquakes, such as the 2015 Nepal earthquake, the 2015 Illapel, Chile earthquake, the 2016 Ecuador earthquake, the 2016 Central Itlay earthquake sequence, and the 2016 New Zealand earthquake (under investigating) etc . So far, we only consider the Sentinel-1 data here, due to its small temporal decorrelation effects, but it would be much easier to work with L-band data, such as ALOS-1/2. In addition, in some mountainous regions, it is also valuable to use the method to overcome the phase unwrapping issues for interseismic or postseismic InSAR phase retrieval. 

The Sentinel-1 mission offers unprecedented spatial and temporal coverage of continental areas when operating in TOPS mode, with 24 day repeat coverage over most tectonically active continental areas. In some priority areas the repeat interval between acquisitions is as little as 6 days. Such short recurrence times, along with the free availability of SLC data, raise the possibility of routine measurement of the crustal deformation due to shallow continental earthquakes globally. Such measurements provide more accurate measures of earthquake location than teleseismic methods (e.g. Weston et al., 2011, 2012), and give a similar level of location accuracy to local seismometer networks, but do not require expensive local infrastructure.

We report on our efforts to study global seismicity with Sentinel-1 TOPS data since April 2015. We searched the USGS/NEIC earthquake catalogue to identify earthquakes with epicenters located on land masses with deformation that was potentially detectable using InSAR. (As event detectability for a given hypocentral depth depends on event size, we select events in the following size and depth ranges: 6.0>Mw≥5.5, depth<10 km; 7.0>Mw≥6.0, depth<20 km; Mw≥7.0, depth<25 km.) Neglecting aftershocks or foreshocks that occurred within the same interferograms as their corresponding mainshocks, we identified 38 earthquakes that fit these criteria. Using the ISCE processing software, we then systematically processed interferograms for these events, prioritising image pairs with the shortest possible temporal baselines.

We find that we can identify deformation signals attributable to earthquakes in half of the events tested (19 out of 38 events). A further 13% of events (5 out of 38) have deformation patterns that may be due to earthquakes, but require additional processing and/or modelling for verification. 37% of events (14 out of 38) could not be identified from their interferograms. The majority of failed detections were due to interferogram decorrelation. This was particularly apparent in heavily vegetated tropical areas, such as Central America or northern South America, where 24 day (and in some cases even 12 day) repeat coverage is insufficient to achieve good coherence with Sentinel’s C-band radars. Heavy decorrelation was also observed over ocean islands, where image acquisitions have been infrequent and/or SLC data can be missing from the archive, allowing only long timespan interferograms to be processed.

At present, the largest earthquake that we have not detected is a Mw7.0 event whose epicenter was located on Melampa Island, Vanuatu (28 April 2016); we also did not detect a Mw6.2 event from Japan (21 October 2016) that was detected by the ALOS-2 satellite by other authors (GSI, Japan). We propose that these could be considered upper and lower estimates of the ‘magnitude of completeness’ (i.e. the magnitude above which all events are detected) for global earthquakes studied using Sentinel-1 data. In order to lower this magnitude to Mw6.0 or below, we suggest that more frequent acquisitions will likely be necessary over tropical continental areas and ocean islands near plate boundaries in future.

Estimates of global strain rates are important for both seismic hazard, and tectonic and geodynamic studies. Global InSAR displacement measurements can provide important constraints on global strain rates, but sensitivity in the north-south direction is limited. A side effect of the variable squint of the TOPS mode is much better sensitivity to displacements along track than the regular acquisition modes of all other SAR missions. Of particular note is the ability to isolate this along-track motion in areas of “burst overlap”, where the same area of ground is imaged twice or even three times during the same satellite pass, with different squints. This is achieved by differencing interferograms formed from the different bursts. The ability to image along-track motion with this technique has already been demonstrated for several large earthquakes

Although the precision is two orders of magnitude worse than regular InSAR, the limiting factor for InSAR accuracy is usually the tropospheric delay. An added advantage of the along-track InSAR technique is that the tropospheric influence cancels. In principle, this approach can therefore be extended to measure subtle long term along-track motions of mm/year, by applying time series analysis techniques and averaging over many pixels. However, although the tropospheric influence cancels, signals from different bursts sample different regions of the ionosphere, which leads to another spatially correlated signal that can be significant. Our time series analyses of several regions demonstrate that this ionospheric influence prevents an accuracy of mm/year being achieved even after many years of observation. The ionospheric contribution can, however, be reduced, using a combination of spectral diversity in range and azimuth. Deformation influences both of these spectral diversity measurements, but in different ways, allowing the ionospheric and deformation signals to be teased apart. We demonstrate the applicability of this approach in postseismic and interseismic cases and show that along-track motions can be estimated with similar accuracy to those in the line of sight, with the added advantage that they are tied to an absolute reference frame.