On the Segmentation of the Cephalonia–Lefkada Transform Fault Zone (Greece) from an InSAR Multi-Mode Dataset of the Lefkada 2015 Sequence

Abstract

We use Interferometric Synthetic Aperture Radar (InSAR) to study the Cephalonia–Lefkada Transform Fault Zone (CTF) in the Ionian Sea. The CTF separates continental subduction to the north from oceanic subduction to the south, along the Hellenic Subduction Zone. We exploit a rich multi-modal radar dataset of the most recent major earthquake in the region, the 17 November 2015 Mw 6.4 event, and present new surface displacement results that offer additional constraints on the fault segmentation of the area. Based on this dataset, and by exploiting available information of earthquake relocation, we propose a new rupture process for the 2015 sequence, complementary to those published already. Our modelling includes an additional southern fault segment, oblique to the segment related with the mainshock, which indicates that the CTF structure is more complex than previously believed.

1. Introduction

The Cephalonia–Lefkada Transform Fault Zone (CTF) in western Greece (Figure 1a), connects two different subducting regimes: subduction of continental lithosphere to the north, with subduction of oceanic lithosphere to the south [1]. The CTF is known for its frequent seismic activity with a well-known historical record [2] and is related to dextral-strike slip motions, which sometimes are connected with thrust components [3,4,5,6]. The CTF consists of two active segments: the Cephalonia segment and the Lefkada segment, which run parallel and very close to the western coastline of these islands [4]. The exact mapping of the fault trace is still not known, and its position is inferred, mainly, from the bathymetry.

On 17 November 2015, an earthquake (Mw 6.4) (Figure 1b) struck the western coast of Lefkada island, which stimulated a wealth of publications [7,8,9,10,11,12,13,14,15,16,17,18]. Sokos et al. [19] were the first to identify that the mainshock broke a strong asperity, left unbroken, in-between two large subevents of the previous 2003 doublet earthquake of Mw 6.2. The same authors, using seismic waveforms, modelled the 2015 event as a multiple point source event, with two well-resolved sources close to Lefkada and a third southern one, less reliable, located at the latitude of the tip of Cephalonia island, offshore. It is noteworthy that the recent (in 2003, 2014, and 2015) events that ruptured parts of the CTF, were all modelled as multiple source events, which indicates faulting complexity [13,19,20]. Papadimitriou et al. [21], using high precision aftershocks of the 2015 sequence, proposed that the area has a network of fault segments, which in some cases are not clearly identified or easily distinguished.

This complexity is also reflected in the modelling of geodetic data. For example, Ganas et al. [10], based on InSAR and Global Positioning System (GPS ) data, presented two uniform slip fault models; one with a unique fault and another one with two faults. Their preferred solution is the two-fault uniform slip model, which comprises of the main fault alongside a smaller fault, parallel to the main one, located partly onshore the island of Lefkada, close to the latitude of the mainshock. Bie et al. [8] used Sentinel-1 imagery as input and performed a series of tests to identify the slip distribution. They show three slip patches in their preferred slip distribution. The southern patch is related with the displacement detected at the tip of Cephalonia island, and according to the authors this southern slip patch was constrained only by their Sentinel descending data. Moreover, this slip patch was defined to be at depths >10 km, and [8] claim that it was not well resolved by their dataset. It must be noted, though, that monitoring tectonic phenomena in the Ionian Sea is a challenging task, and to elucidate the fault segmentation is difficult due to the sparse distribution of the seismographic and geodetic data, as previously mentioned also by [8,9,12,19].

The advent of remote sensing and, more specifically, the radar satellite data have provided new means of research to the scientific community for earthquake studies [22,23,24,25,26,27]. The aim of this study is to examine the fault segmentation of the CTF, taking advantage of new remote sensing radar data. The new radar datasets used here (Radarsat-2 and ALOS-2) have not been previously exploited, and they consist of imagery from different viewing geometries. These are utilized together with Sentinel-1 data and provide a clearer picture regarding the displacement that occurred on both the Lefkada and the Cephalonia islands.

The deformation induced by the 2015 earthquake sequence, detected by this new multi-sensor InSAR dataset, together with the exploitation of external data (relocated aftershocks), lead to a more complex fault setting, and to new insights in terms of the fault segmentation and the tectonic processes currently acting at the CTF.

2. Data and Methods

SAR interferometry was used here to calculate displacement of the earth’s surface caused by the Lefkada earthquake, based on a diverse dataset consisting of multi-band (C-, X-, and L-) multi-resolution SAR imagery. Initially we considered 4 Sentinel-1 C-band (European Space Agency) images, 2 from ascending and 2 from descending orbits, 4 Radarsat C-band (Canadian Space Agency) images from ascending orbits, 2 ALOS L-band (Japanese Aerospace Exploration Agency,) wide-swath images from descending orbits, and 8 ascending COSMO-SkyMed X-band images (Italian Space Agency). After calculating several differential interferograms, a number of them was discarded due to the excessive impact of low interferometric coherence (e.g., Figure S1). The analysis was therefore carried out with the datasets presented in

Table 1. Radar data used.

Satellite	Master Date (m/d/y)	Slave Date (m/d/y)	Temporal Baseline (days)	Normal Baseline (m)	Pass	Linear Inversion RMS (m)
Radarsat-2	2/21/2014	12/1/2015	648	165.4	Ascending	0.016
Sentinel-1	11/5/2015	11/17/2015	12	22.5	Ascending	0.018
ALOS-2	10/12/2015	11/23/2015	42	77.35	Descending	0.025
Sentinel-1	11/11/2015	11/23/2015	12	66.2	Descending	0.012

.

InSAR processing was carried out with the “repeat-pass” approach [28] using the SARscape software. After coregistering the pre-event and the post-event images, the phase difference was calculated and the topographic phase contribution removal was achieved using the shuttle radar topography mission (SRTM) digital elevation model [29]. Spatially correlated noise was filtered with the approach of Goldstein et al. [30], and the wrapped interferometric phase was then unwrapped with a minimum cost flow algorithm [31]. We then proceeded to the phase (radians) to displacement (meter) conversion and geocoding, and finally we obtained deformation maps in the satellite line-of-sight (LoS) (Figure 2).

To avoid excessive and unnecessary computational load, all displacement maps were downsampled at regular intervals of 200 m in the epicentral area and of 500 m outside it. We identified the seismic sources by means of equations describing the surface displacement induced by a dislocation in an elastic, homogeneous, and isotropic half-space [32]. The source retrieval was carried out in two steps: initially, the fault geometry and mechanism were modelled through a nonlinear optimization scheme based on the Levenberg–Marquardt algorithm [33,34]. The next stage of the modelling followed the retrieval of the fault plane, and included modelling the dislocation distribution, adopting a linear inversion algorithm [35]. In this inversion approach, we extended the size of the retrieved uniform slip fault planes in order to let the slip vanish at the edges, we then subdivided the faults in 1 km × 1 km sized patches. Imposing constraint of a damped and non-negative solution [36], we solved the system:

$$\left[\begin{array}{c}d\\ 0\end{array}\right]=\left[\begin{array}{c}\mathrm{G}\\ \mathsf{\epsilon }{\nabla }^2\end{array}\right]·\mathrm{m}$$

(1)

where $d$ is the InSAR data, $\mathrm{G}$ is the Green’s functions matrix, ${\nabla }^2$ is the Laplacian operator, and $\mathsf{\epsilon }$ is the damping factor obtained by trial-and-error [36]. To adopt a value for the latter, we selected the best compromise between best data fit and fault reliability according to the resulting slip distribution and peak slip values. Finally $\mathrm{m}$ is the vector of parameters.

3. Deformation and Seismic Source Modelling Results

The 2015 sequence affected the surface of both Lefkada and Cephalonia islands (Figure 2). The differences in patterns of the displacement results is expected, due to the different viewing angles of the satellites. When looking at all the unwrapped results, the differences between the ascending and the descending acquisitions indicate a horizontal movement (Figure 2). The Sentinel and Radarsat ascending interferograms indicate that both islands (Lefkada and Cephalonia) approached the satellite sensors. As also stated by Bie et al. [8], the Sentinel descending displacement result is more complicated: looking from north to south, on Lefkada at first there is a small signal of negative displacement and then a lobe of positive displacement. Then at the tip of Cephalonia there is also a displacement signal. The ALOS descending result confirms the pattern of the Sentinel descending and offers additional input on the descending viewing geometry. The two lobes on Lefkada, as seen in the descending unwrapped pairs, are part of the classic strike-slip deformation pattern (the rest is covered by the sea body). Previous studies have noted the existence of displacement at the tip of Cephalonia [7,8,10]. The geodetic InSAR results, presented here (Figure 2), provide additional evidence on the displacement at the northern tip of Cephalonia. This, in turn, offers additional constraints regarding the fault segmentation, as shown in the next section. As can be seen in Figure 3, the northern edge of Cephalonia has a positive displacement pattern, which is evident in all the InSAR results. The wrapped interferograms, the consideration of potential atmospheric artefacts, alongside the InSAR uncertainties, can be found in the supplement.

3.1. Single Fault Model

Initially, we attempted to jointly model our InSAR results with GPS data [9,10,12]. However, the joint geodetic inversion led to large residuals in both datasets. The projection of the three components of GPS measurements into the LoS of every InSAR dataset confirmed a disagreement between the GPS projected values and the InSAR results. In fact, the GPS and InSAR inconsistency is not surprising and it was already observed [11]. While we cannot exclude potential sources of error, this divergence could be attributed, as also stated by Melgar et al. [11], to early afterslip signal contained in InSAR maps, taking into account that the InSAR data include few days of possible slow deformations after the mainshock [7,8,11]. It is thus possible that our dataset contains postseismic signal; the InSAR input is, of course, still essential since our goal is to define the fault segmentation of the area. Given the unavailability of published GPS measurements capturing the displacement at the northern tip of Cephalonia, we proceeded using only the InSAR results in the inversion. Moreover, InSAR offers spatially well-constrained slip patterns, especially for the shallow earthquake events, and is able to identify complex fault segmentations [37].

We firstly performed the nonlinear inversion to identify the faulting geometry and mechanism, assuming that rupture occurred on a single fault. For setting an initial range of allowed values for the fault parameters, we took into consideration the already published ones for the Lefkada 2015 event. In this inversion, every parameter was left free, except the dip value that could not be constrained, thus we adopted a dip of 70° in accordance with [7]. However, a single fault solution was able to model only the displacement of the Lefkada island. A successful reproduction of the displacement at the northern tip of Cephalonia could not be attained (Figure S5). Therefore, the single fault solution was considered insufficient for modelling our dataset.

3.2. Two-Fault Model

Subsequently, we attempted to model the detected displacement using a two-fault model, by adding also a southern fault segment (F2), and we repeated the previous procedure. The deformation signal demanded to lower the number of free parameters. Thus, we kept fixed the parameters of one of the faults, adopting the values previously found with the one-fault solution, which had successfully modelled the displacement of Lefkada. The location and the strike of the second fault segment (F2, Figure 1b) was based on the relocated seismicity of Papadimitriou et al. [21]. Moreover, topographic profiles along part of the F2 segment show that the fault passes by a bathymetric feature that could be related with tectonic activity. More specifically, the fault is located at the peaks of the bathymetric curvature (Figure 4). The dip angle was grid searched in terms of the lowest root-mean-square (RMS) values, which were obtained for dips in the range 80° to 90°, with the lowest values indicating a vertical fault. This is in accordance with Papadimitriou et al. [21], which state that the aftershock locations possibly indicate a vertical structure at that specific area. Results of the nonlinear inversion are summarized in Table S1 and uncertainties are shown in Figures S6 and S7. The RMS values are included in Table S2.

After defining a geometry assuming a uniform slip, we estimated the slip distribution on the two fault planes. It can be seen in Figure 5a,b that the northern fault segment (F1), which projects underneath the Lefkada island, hosted three separated slip patches. The middle patch, which was the one located at the locus of the mainshock (green asterisk in Figure 5a), had a length of about 8 km and a width of 6 km; the peak slip is 2.9 m and this was the highest slip value overall. The southernmost slip patch of F1 expressed a peak slip of 1.7 m. The northern part of F1 hosted a smaller-sized slip patch with peak slip of 0.6 m that did not reach the surface. The locus of the major slip patch and its peak values at shallow depths can explain the damage pattern (rockfalls, landslide phenomena, building failures) observed in-situ [10,14,15,16,17,39]. The main amount of slip on F1 is expressed above 7 km and there is no slip below 10 km. The southern fault segment (F2), hosted a single slip patch with a peak slip of 1.6 m that did reach the seafloor surface. This is the slip patch that predicts the deformation signal of northern Cephalonia (Figure 3). Results indicate that on F2 there was no on-fault dislocation below 6 km.

The root-mean-square (RMS) values of linear inversion for each SAR dataset are shown in Table 1, the source results of this study are in

Table 2. Source parameters of the two-fault segments.

Segment F1
Strike°	Dip°	Rake°	Length (km)	Width (km)	Max Slip (m)
13	70	158	32	10	2.9
Segment F2
Strike°	Dip°	Rake°	Length (km)	Width (km)	Max Slip (m)
43	90	108	25	10	1.6

, while the fits to the data are included in the central and right panels of Figure 6. To evaluate the resolution power of the data, we also performed the source modelling using the variable patch size “full resolution” method introduced by Atzori and Antoniolli [40] (Supporting Information). The variable patch size slip pattern is in accordance with the slip distribution presented here, validating our preferred solution.

Since we present here a new fault segmentation (2 segments, striking obliquely to each other), a straight comparison of our slip distribution with previous studies is not feasible. Our F1 segment is the only one that could be compared to other slip models, at a first approximation, as published slip distributions are only provided for a single fault segment. Our estimated peak slip of 2.9 m is comparable to the 2.3 m calculated by Chousianitis et al. [9]. Additionally, our model predicts a small slip pattern to the northeast of the Lefkada island. This patch, with slightly different characteristics, also appears in the models of Avallone et al. [7] and Chousianitis et al. [9].

4. Further Interpretation of the Two-Fault Model

Stress Transfer Loading

We used the obtained source mechanisms (Table 2) to investigate a potential stress interaction between F1 and F2 (with F2 as receiver fault). A similar scenario was previously examined [21], with a different fault configuration for the receiver fault, than the one we adopt. Here, based on our slip distributions, the static Coulomb stress changes ΔCFF, taking into account the pore pressure contribution, were estimated as [41]:

$$\mathsf{\Delta }\mathrm{CFF}=\mathsf{\Delta }\mathsf{\tau }+{\mathsf{\mu }}^{\prime }\mathsf{\Delta }{\mathsf{\sigma }}_{\mathrm{n}}$$

(2)

where $\mathsf{\Delta }\mathsf{\tau }$ is the shear stress, $\mathsf{\Delta }{\mathsf{\sigma }}_{\mathrm{n}}$ is the normal stress change (unclamping is positive), and μ′ is the apparent friction coefficient where μ′ = μ (1 − B) and B is the Skempton coefficient. Here, μ′ was adopted to be 0.4 [42]. Positive ΔCFF indicates that failure is promoted. In the opposite case (negative value), failure is suppressed.

Figure 7 summarizes the results, which indicate that the slip distribution of F2 occurred at an area of positive stress loading (max 0.35 MPa) caused by F1.

InSAR cannot offer information on the deformation at the areas that are covered by water bodies. The same limitation exists for the few on-land GPS stations that are installed at the area. Based on all the previous results of this study, we present in the supplementary material the reconstruction of the three dimensional displacement that took place during the 2015 Lefkada sequence and how it evolved.

5. Discussion and Conclusions

We exploited new InSAR multi-modal data from different satellites and orbit modes to revisit the displacement pattern of the Lefkada and Cephalonia islands (Ionian Sea) caused by the occurrence of the 2015 Lefkada earthquake sequence. The displacement fields retrieved here support the operation of a more complex set of structures for the 2015 case study. The InSAR input indicates that a rupture of a single fault, parallel to the western coastline of Lefkada island, is not able to account for the entire observed displacement pattern. It is the observed displacement at the northern tip of the Cephalonia island, which requires a second fault segment, to be sufficiently modelled.

We agree with Ganas et al. [10] about the general concept of a second fault segment, and we also agree with Bie et al. [8] about the fact that the signal at the tip of Cephalonia is an important feature to be considered in the modelling procedure. The importance of the signal at the tip of Cephalonia was furtherly highlighted in our study, and additional information on it was obtained by the specific InSAR dataset used. Our radar signal provided more insights on the on-land deformation pattern of both islands, and as a consequence, it offered better constraints regarding the fault segmentation. We further investigated the multi-segment hypothesis, and building upon the study of Papadimitriou et al. [21] we were able to present a new fault segmentation that predicts well all the InSAR displacement data (both ascending and descending), which is at the same time justified by the relocated seismicity, the seafloor bathymetry, and the stress transfer computation.

Our preferred model identifies two distinct fault segments, which are both required to successfully interpret the geodetic data: a northern fault, bordering the western coastline of Lefkada, associated with the mainshock, encompassing three discrete slip patches, and a secondary southern fault segment, oblique to the northern segment, hosting a single slip patch, at a shallow depth. For this southern segment, no slip is evident below 6 km. The detected dip-slip of the movement of F2 is unfavourably oriented, being a pseudo-vertical fault, not encouraging a dip-slip mechanism; we propose here a possible explanation involving block rotation bounded with pre-existing structure, shown in Figure 8. Regardless of the real block length, by assuming L equal to 1 km, a dislocation of 1.6 m at the surface can be obtained with a negligible rotation α of 0.09°. This movement could have been further facilitated by the existence of evaporites at the broader area [43,44,45,46,47] that can promote frictional instabilities under specific conditions [48].

To summarize, the Cephalonia–Lefkada Transform Fault (CTF) zone is better viewed as a wide zone where the deformation is taken up by multiple strike-slip fault segments (Figure 9), whereas the thrust components are not negligible at all, possibly related to the Ionian Thrust zone. Here, we proposed the existence of an oblique fault segment (F2) in the intersection of the Cephalonia branch with the Lefkada branch, and it is probable that other similar structures co-exist in the broader region, as Papadimitriou et al. [21] proposed. This oblique fault to the main CTF zone highlights the importance of pre-existing structures in the rupture processes occurring on the CTF. These could partially explain why the Cephalonia Fault Zone tends to rupture in segments with characteristic doublets. At the same time, we believe that the definition of the fault segmentation in the Ionian Sea is still an open issue and further studies are needed to have a clearer picture on the local tectonics.

As the area of the Ionian islands is very active seismically, in the future, new seismic sequences might reveal more information on the processes that take place. However, monitoring should not be focused only on the cases of the major seismic events such as the 2015 one. A continuous monitoring of microseismicity could provide insights about the potential pre-existing structures. These appear to play an important role in the evolution of the seismic sequences. The study of the specific fault network is crucial since it is a threat to the Ionian Islands, which are densely populated, especially during the summer months.