Imaging the shear deformation zone along the Kefalonia transform fault zone

1 Introduction

Active fault zones are volumetric structures rather than single, infinitely thin surfaces. A fault zone is characterized by three main components: a cataclastic core, a dilatant damage zone, and a broader shear deformation zone. The cataclastic core contains the comminuted detritus produced by wear along the slipping surfaces, and its width is positively and approximately linearly correlated with fault displacement (Scholz, 1987). Beyond the core lies the dilatant damage zone, a region of tensile fracturing (Scholz, 1998) whose width is correlated with displacement in the same manner as the core but is generally considered to level out at several hundred meters for displacements exceeding a few hundred meters (Savage and Brodsky, 2011). The shear deformation zone encloses the dilatant damage zone and extends beyond it, defining the region within which enhanced seismicity is observed (Powers and Jordan, 2010). Within this zone, the density of near-fault seismicity falls off rapidly, following a power law, beyond an inner corner at distance W_S1 up to a full half-width W_S2, beyond which the active zone gives way to the largely undeformed host rock (Figure 1).

Characterizing the seismically active volumes of the dilatant damage zone (dominated by volumetric strain) and the shear deformation zone (dominated by shear strain) is important because their widths reflect the maturity of the fault. Because wear processes control the growth of these zones, quantifying the degree of wear provides a measure of fault maturity, while earthquake source parameters such as stress drop, radiated energy, and slip distribution may in turn be influenced this measure (e.g., Cao and Aki, 1986; Malagnini et al., 2010; Perrin et al., 2016; Qiao et al., 2022; Guo et al., 2023). In strike-slip settings in particular, mature fault zones may act as anisotropic permeability pathways in the brittle crust, making these segments more likely to host slow earthquakes or aseismic slip than immature, less-developed fault zones (Thakur et al., 2020).

High-resolution earthquake locations offer a means of imaging not only the surface trace of an active fault, but also the width and internal organization of the surrounding shear deformation zone, often at resolutions comparable to those of direct field observations (e.g., Hauksson, 2010; Powers and Jordan, 2010; Valoroso et al., 2014; Godano et al., 2025). Perrin et al. (2021) exploited this idea to analyze the aftershock distributions of seven large continental strike-slip earthquakes spanning a wide range of net displacement and showed that the width of the shear deformation zone (characterized by W_S1, the distance from the cataclastic core beyond which seismicity is greatly reduced and W_S2 the distance marking the edge of the shear deformation zone) decreases systematically with increasing cumulative fault displacement, initiation age, and long-term slip rate, so that mature fault sections exhibit narrower deformation zones than immature ones. In their framework, local fault geometry is estimated by moving-window (PCA) hypocentral distances are measured orthogonally to the local best-fitting plane, and the stacked fault-normal seismicity profiles are used to define W_S1, the inner distance at which event density begins to fall off rapidly, and W_S2, the outer limit of the seismically active shear zone.

The KTFZ is well suited to this type of analysis, because it is among the most seismically productive structures in the Mediterranean, with an average of about one Mw > 6.5 earthquake per decade over the last five centuries (Papazachos and Papazachou, 2003) and rapid dextral slip accommodated across a narrow plate-boundary zone (Kahle et al., 1995; Chousianitis et al., 2015). It is recognized as a prominent strike slip zone hosting frequent strong (M > 6.0) earthquakes, with the most recent being recorded by dense regional and temporary networks (University of Athens, 2008), and high-quality relocated catalogs are available for several sequences (e.g., Karakostas et al., 2015; Papadimitriou et al., 2017; Bonatis et al., 2024). At the same time, the zone is structurally segmented, comprising two principal dextral branches linked by a step-over transfer zone and accompanied by oblique secondary structures (Karakostas et al., 2015; Papadimitriou et al., 2017), so that different sequences are expected to sample different structural elements of the same plate-boundary system. This combination of high-quality data and along-strike structural variability makes the KTFZ a natural laboratory targeted at a quantitative, seismicity-based characterization of active shear-zone width.

In this study we apply the methodology of Perrin et al. (2021) to two relocated aftershock sequences of large crustal strike-slip earthquakes along the KTFZ: the 2014 Kefalonia doublet (M_w = 6.1 and M_w = 6.0) and the 2015 Lefkada earthquake (M_w = 6.5). For each sequence we estimate the local three-dimensional fault geometry, measure the fault-normal distribution of relocated hypocenters, and determine the inner and outer bounds (W_S1 and W_S2) of the seismically active shear deformation zone. Our aims are to (i) obtain reproducible, seismicity-based estimates of the active shear-zone width for the principal ruptured segments of the KTFZ; (ii) compare these widths and the associated PCA-derived fault geometries between the two sequences; and (iii) place the results in the context of the segmented, partly transtensional and transpressional structure of the transform zone.

2 Seismotectonic setting

The area of central Ionian Islands (western Greece) provides an appropriate natural laboratory for imaging the seismically active shear deformation zone of a mature strike-slip system, because it accommodates one of the most active crustal seismic zones in the Aegean and the broader eastern Mediterranean (Papazachos and Papazachou, 2003; Ambraseys, 2009). Lefkada and Kefalonia fault segments belong to the Kefalonia Transform Fault Zone (KTFZ), an approximately 100 km long dextral strike-slip fault zone with a subordinate compressional component (Scordilis et al., 1985; Kiratzi and Langston, 1991; Karakostas et al., 2015; Özbakır et al., 2020). The KTFZ occupies a key position beyond the termination of the western Hellenic Arc, where the seismotectonic regime changes from subduction of the oceanic lithosphere of the eastern Mediterranean beneath the Aegean microplate to the south (Papazachos and Comninakis, 1971; Halpaap et al., 2018; Ganas et al., 2020) to continental collision between the Adriatic (Apulian) microplate and the Eurasian plate to the north (McKenzie, 1978). KTFZ comprises two principal branches with slightly different orientations: the Lefkada fault branch to the north, which strikes approximately NNE-SSW along the western coastline of Lefkada, and the Kefalonia fault branch to the south, which follows the western part of Kefalonia and the Paliki peninsula with a more NE-SW to N-S orientation (Louvari et al., 1999; Kokinou et al., 2006). Representative fault-plane solutions summarizing the two fault branches give strike, dip, and rake of about 11°, 60°, and 165° for the Lefkada branch and about 33°, 56°, and 163° for the Kefalonia branch, confirming their dominantly right-lateral character (Louvari et al., 1999; Karakostas and Papadimitriou, 2010).

The change in strike between the two fault branches is not merely geometric, it is associated with transtensional and transpressional motion in the Lefkada and Kefalonia domains, respectively (Papadimitriou et al., 2017). The two branches are linked by a step-over transfer zone of ENE-WSW striking secondary fault segments (Karakostas et al., 2015), consistent with numerous observations of earthquake clusters that align along an approximately E-W trend in this area (e.g., Sokos et al., 2015; Haddad et al., 2020; Anagnostou et al., 2025). Because strike-slip faults typically appear as echeloned, parallel fault strands separated by step-overs and jogs (Scholz, 2002), this transitional zone deflects the main fault azimuth and forms bends or duplexes that impede rupture of the large (M ≥ 6.0) main shocks, mitigating the regional seismic hazard, and commonly produce rupture heterogeneities and terminations. The transfer zone therefore acts as a barrier where rupture tends to terminate, in agreement with instrumental and historical seismicity (Papazachos and Papazachou, 2003) of strong earthquakes being associated with either the Lefkada or the Kefalonia fault branches, separately, and with synchronized activation of the two branches being attributed to stress transfer rather than unimpeded rupture propagation (Papadimitriou, 2002).

Geodetic data confirm that the KTFZ is a strongly coupled, high-slip-rate plate-boundary fault along which deformation is distributed across multiple structures. KTFZ is characterized by the highest crustal deformation rates within the Aegean, with slip rates reaching ∼20 mm/year, gradually increasing from the Lefkada segment where they are measured closer to 10 mm/year (Kahle et al., 1995; Briole et al., 2021), along with crustal shortening on the island itself (Ganas et al., 2013) to the Kefalonia segment where a combined InSAR and GNSS analysis yields slip rates between 17.3 mm/yr (Tsironi et al., 2024) and 19.5 mm/yr (Briole et al., 2015) values consistent with the larger magnitude and more in number earthquakes hosted by this fault branch.

The historical and instrumental earthquake record (Figure 2) confirms the elevated seismic hazard of the study area. Complete historical information for the last five centuries reveals an average of about one M_w > 6.5 earthquake per decade (Papadimitriou and Papazachos, 1985), with maximum reported magnitudes of about M_w7.2–7.4 for the Kefalonia branch and M_w6.6–6.7 for the Lefkada branch (Papazachos and Papazachou, 2003) with the magnitudes being estimated as equivalent moment magnitudes, after Papazachos et al. (1997). Strong (M ≥ 6.0) earthquakes have repeatedly concentrated along the western coastlines of both islands (Karakostas and Papadimitriou, 2010; Papadimitriou et al., 2017). The 1953 Kefalonia sequence, comprising four strong main shocks (9 August, M6.4; 11 August, M6.8; 12 August, M7.2; 21 October, M6.3), in an unprecedented multiplet, remains a benchmark seismic sequence because of its catastrophic consequences, whereas the more recent succession of M ≥ 6.0 main shocks in 2003, 2014, and 2015 (magenta stars in Figure 2) demonstrates continued activity on adjacent KTFZ segments over short time intervals (Karakostas et al., 2004; Karakostas and Papadimitriou, 2010; Karakostas et al., 2015; Papadimitriou et al., 2017; Kourouklas et al., 2023; 2024).

These recent sequences demonstrate that deformation along the KTFZ cannot be represented adequately by a single, throughgoing planar fault. The 2003 M_w = 6.2 Lefkada earthquake ruptured a dextral strike-slip segment almost 16 km long along the northwestern coastline of Lefkada, striking NNE-SSW and dipping steeply eastward (Karakostas et al., 2004). Dense local observations showed that the aftershock sequence delineated not only the main rupture but also along-strike adjacent activation and secondary nearby structures. Karakostas and Papadimitriou (2010) argued that such structures should be incorporated in seismic hazard assessment because they are capable of producing damaging moderate (∼M5.0) earthquakes. The 2014 Kefalonia doublet (M_w = 6.1 on 26 January and M_w = 6.0 on 3 February) occurred in the Paliki peninsula and further highlighted the segmented nature of the southern KTFZ (Sakkas and Lagios, 2015). Karakostas et al. (2015) interpreted the relocated sequence as associated with two adjacent, almost N–S striking, east–dipping dextral fault segments, along with ENE-WSW striking dextral lineaments forming the step–over transfer zone that links the Kefalonia and Lefkada fault branches. Using waveform modeling, Sokos et al. (2015) emphasized that the doublet did not simply rupture the main offshore Kefalonia Transform Fault but instead, the diffuse aftershock pattern implied the activation of an onshore fault network on Paliki, with strike-slip motion locally combined with thrust components. The 2015 M_w6.5 Lefkada earthquake ruptured the southern Lefkada fault branch, again with right-lateral strike-slip kinematics, on a fault running approximately parallel to the western coastline, striking at about N20 E and dipping steeply eastward (Ganas et al., 2016; Avallone et al., 2017). Papadimitriou et al. (2017) relocated the aftershocks and identified a main rupture of about 17 km in length, accompanied by energetic clusters beyond both its tips. These off-fault clusters were oblique to the main fault and were interpreted as en-echelon or secondary fault planes, indicating that strain was released on a complex fault network rather than on a single plane.

The area north of Lefkada marks the transition from transform faulting to continental collision. The February-March 2019 offshore north Lefkada swarm occurred close to this junction and activated structures that rotated in strike from the NNE-SSW orientation of the Lefkada fault branch toward NE-SW, while maintaining a dominantly dextral sense of motion (Kostoglou et al., 2020). This activity was interpreted as evidence for a local fault system that accommodates the transition from right-lateral transform motion along the KTFZ to thrust-dominated deformation along the collision front. More recent microseismicity work similarly describes the Lefkada branch as bounded by the step-over zone to the south and the Adriatic-Eurasian collision to the north, with a transition zone where fault strikes rotate and thrust components become more important (Kostoglou et al., 2026).

The crustal seismogenic layer along the KTFZ exhibits a thickness of about 12 km and focal depths commonly between about 3 and 15 km. Papadimitriou et al. (2017) relocated the 2015 Lefkada aftershocks within a seismogenic layer extending from about 3 to 16 km depth, within which multiple well-defined fault planes were activated. This depth range is important for the present analysis, because the PCA-derived planes and fault-normal distances are intended to represent the active brittle shear zone sampled by relocated hypocenters rather than only near-surface geological fault traces.

For the present study, the key implication of this setting is that the KTFZ contains both mature, rapidly slipping dextral branches and local zones of structural complexity, including step-overs, transfer faults, oblique secondary structures, and transpressional segments. A seismicity-defined shear deformation zone may therefore differ from one sequence to another, depending on whether the analyzed catalog mainly samples a principal rupture, a transfer zone, or a compact secondary structure. The two datasets examined here (for the 2014 Kefalonia doublet and the 2015 Lefkada sequence) are consequently treated not as isolated earthquake sequences, but as different structural expressions of the same plate-boundary system.

3 Data

We analyze two relocated aftershock datasets, associated with the most recent aftershock sequences, namely, of the 2014 Kefalonia doublet and the 2015 Lefkada main shock. The first dataset samples the aftershock distribution of M_w6.1 on 26 January and M_w 6.0 on 3 February main shocks on the Paliki peninsula, associated with fault segments belonging to the southern branch of the transform zone (Karakostas et al., 2015). The second dataset corresponds to the 2015 Lefkada earthquake (M_w6.5, 17 November) and samples the main rupture along the southwestern coast of Lefkada on the northern branch, along with the associated activity extending toward the offshore Lefkada-Kefalonia area (Papadimitriou et al., 2017). The two datasets are treated independently, but an identical analysis workflow is applied to each one of them, for their fault-normal profiles and W_S1-W_S2 values being directly comparable. Key catalog characteristics are summarized in Table 1.

TABLE 1. Summary of the two analyzed datasets.

Both catalogs were compiled by relocating the recorded seismicity with the double-difference technique (HYPODD; Waldhauser and Ellsworth, 2000), using arrival times from the permanent regional network, adopting the one-dimensional velocity model of Haslinger et al. (1999) with a P- to S-wave velocity ratio of V_p/V_s = 1.80 (Karakostas et al., 2015; Papadimitriou et al., 2017). The 2014 Kefalonia catalog contains 1,154 aftershocks with magnitudes 1.4<M_w < 6.1 for the period of 26 January to 17 February 2014, and the 2015 Lefkada catalog contains 1790 aftershocks with 0.2<M_w < 6.5 for the period 17 November 2015 to 31 March 2016. Reported focal depths lie mainly within the seismogenic layer of the KTFZ (about 3–16 km; Papadimitriou et al., 2017), with median depths of 8.5 km (Kefalonia) and 7.9 km (Lefkada), consistent with a mid-crustal brittle zone.

For each dataset, the hypocenters were projected into a local Cartesian coordinate system with horizontal and vertical distances expressed in kilometers. Depth was retained as the vertical coordinate, and the epicentral distribution was inspected to define an approximate mean strike and the spatial limits of the moving-window analysis. The analysis uses the full three-dimensional hypocentral coordinates rather than map-view distances alone, because the objective is to estimate local fault planes and to measure hypocentral distances orthogonal to them in three dimensions. Depth coloring in the map views (i.e., Figure 3) is used only for visual interpretation and does not influence the plane fitting.

The method assumes that the catalogs are sufficiently accurately relocated for a sub-kilometer geometric analysis, and this assumption is central to the interpretation of W_S1, which is of the order of 0.5–0.8 km. Perrin et al. (2021) restricted their analysis to sequences with relative horizontal location errors typically at or below about 0.03 km, so that the width of the near-fault seismicity cloud reflects the structure of the fault zone rather than location scatter. The relocated catalogs used here were obtained with the same double-difference, waveform cross-correlation approach. The reported bootstrap location uncertainties are of the order of 0.250–0.975 km in both the horizontal and the vertical direction (Papadimitriou et al., 2017). These uncertainties are comparable to W_S1 itself, and they are the relevant measure of positional accuracy here. The travel-time residuals quoted above (mean RMS of 0.24 s for Kefalonia and 0.13 s for Lefkada) describe the misfit of the location procedure and are not a measure of hypocentral accuracy; we therefore do not use them to support the geometric resolution of W_S1. To quantify how strongly scatter at this level can affect the results, independent Gaussian noise was added to all three hypocentral coordinates and the full workflow was repeated (Sections 4.5 and 5.3). Noise with a standard deviation of 0.5 km per coordinate, at the upper end of the reported bootstrap uncertainties, increases W_S1 from 0.83 to 0.93 km for Kefalonia and from 0.50 to 0.67 km for Lefkada, whereas W_S2 changes by less than 0.2 km in both cases. Location error therefore inflates W_S1 rather than creating it: the near-fault plateau survives a level of added scatter comparable to the reported uncertainties, so the W_S1 values reported here are best read as upper bounds on the true inner half-width. We therefore interpret W_S1 and W_S2 as first-order estimates of the inner and outer bounds of the active shear zone, while recognizing that residual location uncertainty, and especially depth uncertainty, contributes to the apparent near-fault width and is considered in the sensitivity discussion.

4 Methods

4.1 Overview of the PCA and fault–normal workflow

The analysis follows the approach of Perrin et al. (2019), Perrin et al. (2021), adapted to the KTFZ seismotectonic setting and to the available relocated catalogs. First, each catalog is transformed into a local Cartesian coordinate system. Second, rectangular boxes are superimposed along the inferred strike of the target structure and stepped along strike at fixed distance increments. Third, within each box, a local best-fitting plane is estimated by principal component analysis (PCA) of the three-dimensional hypocentral coordinates. Fourth, the orthogonal distance of each hypocenter from the local plane is computed and used to build stacked, normalized fault-normal distance profiles, from which the shear-zone width parameters, W_S1 and W_S2, are derived.

The PCA plane is defined by the two principal axes that account for the greatest variance of the hypocentral cloud within a box, the third principal axis, corresponding to the smallest variance is taken as the local normal to the seismicity plane. Following Perrin et al. (2019), a constant dip (extracted from the previous works done in the area) is assumed for the plane within each box. This formulation does not assume that the entire sequence lies on a single planar fault: the best-fitting plane orientation is allowed to vary between adjacent boxes along strike, which is essential for a segmented system such as the KTFZ.

4.2 Moving–window geometry

For each dataset, a mean strike was adopted from the observed seismicity alignment and from the published fault orientation, and the hypocenters were transformed into along-strike and across-strike coordinates. Rectangular moving windows (boxes) were then stepped along strike. The boxes are elongated parallel to the strike direction and broadened in the across strike direction, so that each box captures the full fault-normal distribution of seismicity while still resolving along-strike changes in fault geometry. Box dimensions of the order of several kilometers along strike and a few kilometers across strike, stepped at approximately 1 km increments, were used. For the KTFZ sequences we adopt boxes elongated along strike (10 km long and 3 km wide) so that each local best-fitting plane is constrained by the along-strike trend of seismicity rather than by its across-strike spread. For the KTFZ datasets we chose the reference box size of 10 × 3 km, because it is the shortest along-strike length for which every box along both branches retain at least 20 events, which is the minimum required here for a stable three-dimensional PCA plane. The 3 km across-strike width, instead, is wide enough to contain the full fault-normal distribution of both sequences (more than 95% of events lie within 1.5 km of the local plane) without reaching into the adjacent branch across the step-over zone. Boxes containing fewer than 20 events were excluded from the PCA. Alternative box dimensions and minimum event counts are explored in the sensitivity analysis (Section 4.5). Each retained box yields one local best-fitting fault plane and one fault-normal event-distance distribution.

The choice of box size controls the trade-off between local resolution and stability. Short boxes can follow rapid along-strike changes in orientation but are more sensitive to event clustering and location errors whereas long boxes produce more stable planes but may average over adjacent structures, particularly in the step-over region between the Kefalonia and Lefkada branches. For this reason, W_S1 and W_S2 are recomputed for several reasonable box-size choices as part of the sensitivity analysis (Section 4.5).

4.3 Orthogonal distances and stacked distance profiles

For every retained box, the orthogonal distance between each hypocenter and the local PCA plane was computed. The absolute value of this distance was used to construct a one-sided fault-normal distribution. Aftershocks were counted as a function of distance from the plane in bins of 0.03 km, following Perrin et al. (2021), and each box histogram was normalized by the event count of the corresponding box. Normalization prevents boxes with very high aftershock density from dominating the stacked profile.

The normalized profiles were then stacked by computing the mean normalized number of earthquakes in each distance bin. In the figures, the gray curves represent the individual normalized profiles and the black curve denotes their average. A systematic decrease of earthquake density away from the local plane indicates that the local PCA planes capture a coherent central structure around which the seismicity is organized.

4.4 Estimation of W_S1 and W_S2

The mean normalized profile was fitted with a regularized power-law function, following Perrin et al. (2021) and Powers and Jordan (2010). W_S1 is defined as the distance from the local fault plane at which the fitted curve exhibits its maximum curvature (the maximum of the second derivative), marking the transition from the near-fault high-density region to a more rapid fall-off in hypocentral density. W_S2 is defined as the outer half-width of the active shear deformation zone, the distance beyond W_S1 at which the observed profile departs from the fitted power law, either flattening or dropping toward zero. We identify this transition with the smaller of two distances: (i) the distance beyond which fewer than 20% of the boxes used to count aftershocks still contain events (the coverage criterion of Perrin et al., 2021), and (ii) the distance at which the observed profile first departs from a factor-of-two envelope around the fitted curve, that is, where the observed to fitted density ratio falls below 0.5 or rises above 2.0 (the departure criterion). Taking the minimum of the two places W_S2 where the profile first visibly departs from the fit, rather than only where box coverage becomes sparse. For the Kefalonia sequence the two criteria coincide, whereas for the Lefkada sequence the departure criterion is reached first and controls W_S2. The factor-of-two envelope is introduced in the present analysis rather than adopted from previous work. Perrin et al. (2019), Perrin et al. (2021) defined W_S2 through the coverage criterion alone. Because this criterion controls W_S2 for the Lefkada sequence, envelope factors (f) between 1.3 and 4.0 were tested (Section 5.3). W_S2 for Kefalonia is unchanged for every f ≥ 2.0, and for Lefkada it is 1.48 km throughout the interval f = 1.5–2.5, relaxing to the coverage-controlled value of about 2.2 km only for f ≥ 3.0, whereas f < 1.5 causes the criterion to fire inside the near-fault plateau, where bin-to-bin counting scatter alone exceeds the envelope. The adopted value f = 2.0 therefore lies inside a stable interval rather than at a threshold that controls the result. W_S2 marks the outer limit of the volume within which aftershock density still follows the fault-zone decay law, that is the outer edge of the seismically active shear-deformation zone.

4.5 Uncertainty and sensitivity tests

W_S1 and W_S2 are treated as estimates of the inner and outer bounds of the active shear zone rather than as exact physical boundaries. Four main sources of uncertainty are considered. Location uncertainty broadens the hypocentral cloud and can inflate the apparent near-fault width, particularly through depth error. Box geometry matters because different box lengths and widths can either isolate or merge adjacent structures. Structural complexity is also influential, because an oblique secondary fault falling within the same box can rotate the PCA plane away from the principal rupture surface. Fitting uncertainty matters because W_S1 depends on the curvature of the fitted function and W_S2 depends on the behavior of the outer tail of the distribution.

To assess robustness, the workflow is repeated for several combinations of box length, box width, minimum event count, and depth cutoff. The two remaining analysis parameters, the 0.03 km distance bin width and the factor-of-two departure tolerance, are varied separately, as are the reference strike and the hypocentral coordinates themselves, the latter by adding synthetic location noise. The stability of the stacked profile with respect to individual boxes is assessed by bootstrap resampling and by leaving each box out in turn (Section 5.3). The main text reports the preferred values and the range spanned by these tests. The objective is to establish whether the estimated widths, and any difference between the two sequences, remain broadly consistent across reasonable analysis choices rather than being artifacts of a single parameter set.

5 Results

The moving-window PCA and fault-normal stacking workflow were applied independently to the two relocated catalogs, restricted to a 2 month window for the 2015 Lefkada earthquake. In both cases the reference strike used to define the along-strike and across-strike coordinates was obtained from a two-dimensional principal component analysis of the epicentres of the whole sequence, which gives 15° for Kefalonia and 18° for Lefkada. The mean azimuth of the local three-dimensional PCA planes agrees with these values to within 3°. This strike of approximately 15–18° is in agreement with the north-northeast orientation of the KTFZ branches (Karakostas et al., 2015; Papadimitriou et al., 2017). The local best-fitting planes are steep, with mean dips of about 88° for the 2014 Kefalonia dataset and about 80° for the 2015 Lefkada dataset, consistent with the near-vertical to steeply east-dipping strike-slip geometry reported for these ruptures. The along-strike sequence of local planes (Figure 3) defines a coherent principal fault trace in each case rather than a scattered set of orientations, confirming that the relocated seismicity is organized around a well-defined central structure and that the PCA plane fitting is meaningful for these datasets. It is worth mentioning that the fitted planes bend in the northern parts of the Kefalonia segment (Figure 3A) and the southern parts of the Lefkada segment (Figure 3B), where the edges of the transfer zone between the two KTFZ fault segments are located. A counterclockwise rotation of the traces can also be observed in the northern parts of the Lefkada segment, where smaller structures perpendicular to the main KTFZ strike have been observed before (e.g., Kostoglou et al., 2020).

5.1 The 2014 Kefalonia doublet

The 2014 Kefalonia dataset (1,154 relocated events, 1.4 ≤ M ≤ 6.1, from 26 January to 17 February 2014) yields 38 valid along-strike boxes covering the Paliki peninsula and its offshore continuation. The stacked, normalized fault-normal profile (Figure 4a) shows a near-fault region of high, nearly constant event density followed by a power-law decay. Fitting the regularized plateau-and-power-law function gives an inner half-width W_S1 = 0.83 km, marking the transition from the near-fault high-density core to the decaying tail, and an outer half-width W_S2 = 1.88 km, the distance at which the coverage and departure criteria coincide. The corresponding full width of the active shear deformation zone is 2·W_S2 = 3.75 km. The near fault density contrast for this sequence is relatively low, consistent with the doublet nature of the 2014 activity, in which two adjacent principal patches ruptured within a few days (Karakostas et al., 2015), broadening the near-fault seismicity cloud.

5.2 The 2015 Lefkada earthquake

The 2015 Lefkada dataset, restricted to the 2 months following the 17 November 2015 mainshock (1,291 relocated events, 0.2 ≤ M ≤ 6.5), yields 58 valid boxes along the northern branch. Its stacked profile (Figure 4B) is more sharply peaked at the fault, with a near-fault density roughly twice the mid-distance level and a steeper power-law tail. The fit results to W_S1 = 0.50 km and W_S2 = 1.49 km. Here, the departure criterion is reached before the coverage criterion, so that the observed profile leaves the fitted power law at about 1.5 km, earlier than the distance (about 2.2 km) at which box coverage becomes sparse, and W_S2 is set by the departure distance. This corresponds to a full active shear zone width of 2·W_S2 = 2.97 km. The sharper near-fault peak, is consistent with a more localized principal rupture on the Lefkada branch, in agreement with the compact main-rupture geometry inferred from the relocated aftershocks (Papadimitriou et al., 2017). Additional tests support reading this departure as a property of the sequence rather than of the stack. Resampling the 58 boxes with replacement (300 realizations) gives a median W_S2 of 1.48 km with a fifth–95th percentile range of 1.30–1.57 km, and removing any single box in turn leaves W_S2 unchanged in 57 out of 58 cases (1.42–1.48 km overall). No individual box, therefore, controls the result, and the coverage-only value of about 2.2 km is not reached by any resampling of the data. Individual box profiles are too sparse at these distances for the departure to be identified box by box, which is precisely why the profiles are stacked, but the stacked departure is not driven by a small subset of them. We note that the event density does not vanish beyond 1.5 km: bins between about 1.6 and 2.1 km remain populated and locally scatter to either side of the fitted curve. W_S2 is not a distance beyond which aftershocks cease, but the distance beyond which their dense concentration stops following the fault-zone decay law; the tail of events outside W_S2 is expected and is consistent with off-fault aftershocks on secondary structures. We adopt the departure-bound value as the outer half-width because the departure criterion identifies where the aftershock density physically ceases to follow the fault-zone decay law, whereas the coverage criterion is a sampling measure that is controlled by where along-strike box occupancy thins rather than by where deformation ceases Because the along-strike extent of the Lefkada aftershock zone is long, boxes remain well populated out up to greater distances, and the coverage-only estimate (about 2.2 km; teal dotted line in Figure 4B) is pushed outward by data availability rather than by a genuine broadening of the damage zone. We therefore prefer the departure distance of about 1.5 km as the outer limit of the shear-deformation zone and regard the coverage-only value as an upper bound, while acknowledging that the two values bracket the transition rather than pinpoint it.

5.3 Comparison of the two sequences and sensitivity tests

The two sequences give outer half-widths of W_S2 = 1.88 km (Kefalonia) and 1.49 km (Lefkada) and inner half-widths of W_S1 = 0.83 and 0.50 km, respectively. To assess whether these values are robust, W_S1 and W_S2 were recomputed for 27 combinations of box length (6–10 km), box across-strike width (2–4 km), and minimum event count (15–30), keeping the 0.03 km bin width and the departure tolerance fixed (Figure 5). For the 2014 Kefalonia sequence, the range of W_S1 is 0.51–1.01 km whereas that of W_S2 is 1.06–2.17 km. For the 2015 Lefkada sequence, W_S1 extends from 0.31–0.66 km, whereas that of W_S2 is 1.12–1.90 km. Across these configurations the coefficient of variation of W_S2 is nearly 25% for Kefalonia and 18% for Lefkada, and essentially all of the scatter is controlled by the across-strike box width, with the minimum event count threshold having no measurable effect: wider across-strike boxes average over a broader swath and systematically increase the estimated half-width. Four further tests address parameters held fixed in that grid. Varying the distance bin width between 0.02 and 0.15 km leaves W_S1 within 0.75–0.86 km (Kefalonia) and 0.48–0.55 km (Lefkada); W_S2 is stable at 1.73-1.88 km and 1.43–1.48 km, respectively, for bins of 0.02–0.05 km and drifts upward to about 2.2–2.3 km for bins of 0.08 km and coarser, where the bin becomes too broad to resolve the departure. By varying the departure envelope factor f from 1.3 to 4.0 changes W_S2 as described in Section 4.4. Rotation of the reference strike by + -5° changes W_S1 by at most 0.11 km and W_S2 by at most 0.64 km, with the larger excursions arising for Lefkada, where a rotated frame mixes the main branch with the obliquely oriented structures at its northern edge. Adding Gaussian location noise of 0.2–1.0 km per coordinate raises W_S1 monotonically (to 1.03 and 0.79 km at 1.0 km noise) while leaving W_S2 within 0.2 km of its adopted value, confirming that location scatter inflates the inner half-width but does not generate the outer one. Bootstrap resampling of the boxes gives W_S1 = 0.82 km [0.71–0.93] and W_S2 = 1.73 km [1.57–1.88] for Kefalonia and W_S1 = 0.49 km [0.44–0.55] and W_S2 = 1.48 km [1.30–1.57] for Lefkada (median and fifth–95th percentiles). The preferred values reported above lie near the medians of these ranges, and the difference between the two sequences is comparable to the scatter induced by the parameter variation, so that we do not consider the small difference in W_S1 to be physically significant. For the Lefkada branch this W_S2 is the departure bound value since the coverage criterion alone would place the outer boundary near 2.2 km (Figure 4B, teal line), which we regard as an upper bound for the reasons given in Section 5.2. All of these tests are summarized in Supplementary Figure S1 and listed in full in Supplementary Table S1 of the supplementary material.

Our preferred estimates place the full active shear-deformation zone at about 3–4 km wide, with 2W_S2 approximately 3.0 km for the Lefkada branch and 3.8 km for the Kefalonia branch. These are preferred values rather than tightly determined ones: the sensitivity ranges of W_S2 for the two branches (1.06–2.17 km and 1.12–1.90 km) overlap almost completely, so the apparent difference between 3.8 and 3.0 km is comparable to the variability induced by the analysis choices and should be carried as such through the comparisons below. The 0.03 km bin width is the numerical resolution with which the stacked profiles are discretized and is not a statement of uncertainty, which is set instead by the methodological scatter quantified in Section 5.3 and by the hypocentral location uncertainties, both of which are more than an order of magnitude larger. These half-widths place the two KTFZ ruptures within the range documented by Perrin et al. (2021) for mature continental strike-slip faults, where W_S1 is of the order of a few tenths of a kilometer to about 1 km and W_S2 is of the order of one to a few kilometers. We deliberately refrain from converting these values into a maturity ranking of the two branches, because the two catalogs differ in completeness magnitude, and because W_S1 for the Kefalonia sequence partly reflects the compound geometry of a doublet rather than a single fault core. The comparison is presented as a first quantitative characterization of the width of the active shear deformation zone along the two principal branches of the KTFZ.

6 Discussion

The principal outcome of this study is a direct, catalog-based estimate of the width of the active shear-deformation zone along the two main branches of the Kefalonia Transform Fault Zone. Applying the PCA and fault-normal stacking method of Perrin et al. (2019), Perrin et al. (2021) to the relocated aftershock sequences of the 2014 Kefalonia doublet and the 2015 Lefkada mainshock, we find an inner half-width W_S1 of 0.5 km for the Lefkada and 0.83 km for the Kefalonia branches and an outer half-width W_S2 of 1.49 km and 1.88 km, respectively. The full active shear-deformation zone is therefore approximately 3.8 km wide along the Kefalonia branch and about 3.0 km wide along the Lefkada branch. To our knowledge this is the first time that the width of the seismically active fault zone of the KTFZ has been quantified in this way, complementing the extensive body of work on the fault geometry, kinematics, and stress interactions of these ruptures (Karakostas et al., 2004; Karakostas et al., 2025; Karakostas and Papadimitriou, 2010; Papadimitriou et al., 2017).

W_S1 marks the outer edge of the near-fault region within which the aftershock rate is saturated, and W_S2 marks the outer limit of the seismically active shear-deformation zone, beyond which the aftershock density no longer follows the fault-zone decay law. We emphasize that the reduced aftershock density beyond W_S2 constrains the extent of the seismically active volume only, it does not imply that the surrounding host rock is undeformed, since aseismic deformation and inherited structures may well extend further. The interpretation follows from the mechanics of slip on a non-planar fault: slip on a geometrically rough or segmented surface produces high stresses in a narrow region adjacent to the fault, within which aftershocks are generated, whereas a perfectly smooth fault would produce a near-fault stress shadow and few near-fault aftershocks (Kostrov and Das, 1984; Perrin et al., 2021). The steep, coherent principal fault planes recovered here (mean dips of approximately 80–88°), along with the well-defined near-fault density peak, indicate that both KTFZ ruptures are organized around a narrow central slip zone bordered by a distributed damage zone a few kilometers wide.

The W_S1 and W_S2 values obtained for the two branches are broadly comparable, with the between-branch differences of the same order as the parameter-induced scatter of the sensitivity analysis. This indicates that, at the resolution afforded by these catalogs, the two principal strands of the KTFZ have shear-deformation zones of similar width. This is consistent with the two branches being kinematically and structurally similar right-lateral strike-slip faults that together accommodate the relative motion between the Adriatic microplate and the Aegean (Karakostas et al., 2015; Papadimitriou et al., 2017), rather than structures at markedly different stages of development.

The W_S1 and W_S2 values determined for the KTFZ fall within the range reported by Perrin et al. (2021) for mature continental strike-slip faults, which yield W_S1 estimates of approximately 0.06–1 km and W_S2 of approximately 0.5–4.5 km. In their global compilation, W_S1 and W_S2 correlate with each other and decrease with increasing cumulative fault displacement and structural maturity, reflecting the progressive smoothing and localization of faults as they accumulate slip (Mayolle et al., 2021). The KTFZ estimates fall within the wider W_S1 portion of this distribution rather than at the narrow highly localized end, occupied by the most mature faults such as the Parkfield section of the San Andreas Fault (W_S1 approximately 0.06 km).

We interpret this cautiously. The comparatively large W_S1 could reflect a genuinely broad near-fault damage zone associated with a structurally young and geometrically segmented transform boundary; the KTFZ is characterized by step-overs and en echelon segmentation between the Lefkada and Kefalonia branches. However, part of the apparent width almost certainly reflects the location uncertainty of the catalogs used here, which might be larger than in the very well-controlled sequences that anchor the narrow end found by Perrin et al. (2021). Absolute location errors and, in particular, focal-depth uncertainty broaden the fault-normal distribution and increase the apparent W_S1. For this reason we refrain from assigning the KTFZ a position on a quantitative maturity scale relative to the global compilation. What the widths do support is the weaker statement that they are of the order expected for an established, long-lived transform boundary rather than for an immature or a highly localized fault; whether the KTFZ is more or less mature than any individual fault in that compilation cannot be resolved with catalogs of the present location accuracy.

The 2014 Kefalonia sequence is a doublet in which two adjacent principal patches ruptured within a few days (Karakostas et al., 2015). This is expressed in the fault-normal profile as a broad, low-contrast near-fault plateau, and it may contribute to the preferred W_S1 for Kefalonia (0.83 km) being somewhat larger than for the single-rupture Lefkada sequence (0.50 km). We have not isolated this effect: splitting the boxes of each sequence into a straight and a bent population according to the deviation of the local strike from the reference strike, changes W_S1 by 0.13 km for Kefalonia (0.81 against 0.68 km) and not at all for Lefkada (0.46 km in both populations), and the corresponding W_S2 values agree to within 0.15 km. Geometric complexity at the scale sampled by the boxes therefore does not by itself reproduce the between-branch difference in W_S1, and the doublet geometry should be regarded as one plausible contribution among several others, rather than as a demonstrated cause. Where two subparallel or slightly offset rupture patches fall within the same along-strike box, the local PCA plane is fitted to their combined cloud, which increases the apparent inner width. This is an inherent limitation of applying a single-plane PCA method to a segmented source, and it reinforces the decision not to interpret the modest W_S1 difference between the two branches as a difference in intrinsic fault-zone width. It also illustrates why the along-strike, box-by-box formulation of the method is important: it allows the fitted plane orientation to follow the segmentation of the KTFZ rather than forcing a single planar fit to the entire sequence.

Three limitations should be kept in mind. First, the analysis is only as good as the relative location accuracy of the input catalogs permit. Although the double-difference, cross-correlation relocations minimize relative errors, they do not reach the sub-0.03 km relative precision of the best sequences in Perrin et al. (2021), and depth uncertainty inflates the apparent near-fault width. Second, the two catalogs differ in completeness magnitude and network configuration, so the comparison between branches is qualitative rather than a controlled experiment. Third, the two catalogs cover post-mainshock intervals of unequal length: the Kefalonia catalog spans 22 days (26 January to 17 February 2014) while the Lefkada catalog was truncated at 2 months for consistency with Perrin et al. (2021). Two tests bound the effect. Truncating the Lefkada catalog to the same 22 days gives W_S1 = 0.51 km and W_S2 = 1.54 km, within 0.06 km of the 2-month values, and the unequal window lengths do not compromise the comparison. Extending the Lefkada catalog to its full 4.5-month duration (1790 events) results to W_S1 = 0.52 km and W_S2 = 1.90 km, and restricting it to the late aftershocks alone (months two–4.5) gives W_S1 = 0.29 km and W_S2 = 1.25 km. Late aftershocks, therefore, do not reveal a wider outer zone; if anything, they are more tightly clustered on the fault, as expected once the early, broadly distributed stress-transfer aftershocks have decayed. Applying a common magnitude threshold of M ≥ 1.4, the completeness level of the Kefalonia catalog, changes W_S1 to 0.67 and 0.45 km and W_S2 to 1.57 and 1.54 km for the two sequences, bringing them into even closer agreement and confirming that the differing completeness magnitudes do not drive the reported contrast.

7 Conclusion

The application of the moving-window PCA and fault-normal stacking method of Perrin et al. (2019), Perrin et al. (2021) to the relocated aftershock sequences of the 2014 Kefalonia doublet and the 2015 Lefkada earthquake resolves steep, coherent principal fault planes on both branches (mean dips of about 80–88°) and comparable shear-zone widths, with an inner half-width W_S1 of 0.5–0.8 km and an outer half-width W_S2 of about 1.5–1.9 km, giving a full active shear-deformation zone approximately 3–4 km wide. These estimates remain broadly consistent across the tested parameter configurations: a sensitivity analysis over box length, box width and minimum earthquake count yields a coefficient of variation of about 18%–25%, dominated by the choice of across-strike box width, while separate tests on the distance bin width, the departure tolerance, the reference strike, the length of the aftershock window, the completeness magnitude and the hypocentral coordinates themselves leave the outer half-width within about 0.2–0.4 km of the preferred values and inflate only the inner one. The variability is not negligible, and the differences between the two branches fall within the parameter–induced scatter, so the somewhat larger W_S1 for Kefalonia is not interpreted as a difference in intrinsic fault-zone width; the compound rupture geometry of the doublet might be considered as the causative factor. The KTFZ widths fall within the range documented globally for mature continental strike-slip faults but sit toward the wider–W_S1 part of that distribution, which we attribute to a combination of the genuine geometric complexity of this segmented transform boundary and the location uncertainty of the available catalogs; we therefore refrain from assigning the KTFZ a quantitative structural–maturity rank. To our knowledge these are the first estimates of the width of the seismically active fault zone along the two principal branches of the KTFZ, and they provide a quantitative baseline against which future, more precisely located sequences can be compared.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/feart.2026.1939552/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Mourad Bezzeghoud, Universidade de Évora, Portugal

Reviewed by: Andreas Nikolaos Karakonstantis, University of Patras, Greece

Leonardo Colavitti, University of Genoa, Italy