A reappraisal of active tectonics along the Fethiye-Burdur trend, southwestern Turkey

We investigate active tectonics in southwestern Turkey along the trend between Fethiye, near the eastern end of the Hellenic subduction zone, and Burdur, on the Anatolian plateau. Previously, regional GPS velocity data have been used to propose either (1) a NE-trending zone of strike-slip faulting coined the Fethiye-Burdur Fault Zone, or (2) a mix of uniaxial and radial extension accommodated by normal faults with diverse orientations. We test these models against the available earthquake data, updated in light of recent earthquakes at Acıpayam (20 March 2019, Mw 5.6) and Bozkurt (8 August 2019, Mw 5.8) — the largest in this region in the last two decades — and at Arıcılar (24 November 2017, Mw 5.3). Using Sentinel-1 InSAR and seismic waveforms and arrival times, we show that the Acıpayam, Bozkurt and Arıcılar earthquakes were buried ruptures on pure normal faults with subtle or indistinct topographic expressions. By exploiting ray paths shared with these well-recorded modern events, we relocate earlier instrumental seismicity throughout southwestern Turkey. We find that the 1971 Mw 6.0 Burdur earthquake likely ruptured a NW-dipping normal fault in an area of indistinct geomorphology near Salda Lake, contradicting earlier studies that place it on well-expressed faults bounding the Burdur basin. Overall, the northern Fethiye-Burdur trend is characterized by orthogonal normal faulting, consistent with radial extension and likely responsible for the distinct physiography of Turkey's 'Lake District'. The southern Fethiye-Burdur trend is dominated by ESE-WNW trending normal faulting, even though most faults evident in the topography strike NE-SW. This hints at a recent change in regional strain, perhaps related to eastward propagation of the Gökova graben into the area or to rapid subsidence of the Rhodes basin. Overall, our results support GPS-derived tectonic models that depict a mix of uniaxial and radial extension throughout southwestern Turkey, with no evidence for major, active strike-slip faults anywhere along the Fethiye-Burdur trend. Normal faulting orientations are consistent with a stress field driven primarily by contrasts in gravitational potential energy between the elevated Anatolian plateau and the low-lying Rhodes and Antalya basins.


INTRODUCTION
Southwestern Turkey is characterized by active crustal faulting and abundant seismicity, but the kinematics and dynamics of this deformation are both controversial. The region sits astride two arcuate, northward-dipping subduction zones -the Hellenic and Cyprus arcs -in which Nubian oceanic lithosphere is consumed beneath continental Anatolia (Figure 1). Kinematically, it is debatable whether or not the two subduction zones are linked by transform faulting across a triangular structural trend known as the 'Isparta Angle' (e.g. Glover & Robertson 1998). Dynamically, it is unclear whether crustal deformation in this region is driven mostly by plate boundary forces (Jiménez-Munt & Sabadini 2002;Reilinger et al. 2006), by contrasts in gravitational potential energy between thickened continental crust of the Anatolian plateau and low-lying oceanic lithosphere of the Mediterranean basin (England et al. 2016), or by a mixture of the two (Özeren & Holt 2010). The purpose of this paper is to reexamine the western limb of the Isparta Angle, between the cities of Fethiye, on the Mediterranean coastline, and Burdur, on the Anatolian plateau.
Our focus is on the kinematics of active faulting, though there are obvious implications for what is driving this deformation, which we discuss briefly toward the end of the paper.
The easternmost Hellenic subduction zone is characterized by parallel bathymetric troughs termed the Pliny and Strabo trenches, which are highly oblique to Nubia-Anatolia plate convergence and may involve some component of sinistral strike-slip faulting (McKenzie 1972;Hall et al. 2009;Shaw & Jackson 2010;Özbakır et al. 2013). It has been proposed that these faults continue across the Rhodes Basin and into Anatolia, forming a NE-trending zone of discontinuous, sinistral or sinistral-transtensional faults known as the "Fethiye-Burdur Fault Zone" (FBFZ) (e.g.

InSAR observations and modelling
We used European Space Agency Sentinel-1 synthetic aperture radar interferograms and elastic dislocation modelling to characterize faulting in the 2019 Acıpayam, 2019 Bozkurt and 2017 Arıcılar earthquakes. For each event we used GAMMA software to construct short (6 or 12 day) coseismic interferograms on ascending track 58A and descending track 138D, choosing in each case the earliest available post-event scene in order to minimize the contribution from postseismic deformation. For the Arıcılar earthquake, we added a third interferogram from ascending track 131A; no Sentinel-1 scenes were captured between the two earthquakes, and so each interferogram captures the coseismic deformation of both events. Radar incidence angles are between 36 • and 38 • at both the Acıpayam and Bozkurt epicenters, and between 31 • and 43 • at the Arıcılar epicenters.
To model the interferograms we followed the routine procedures of Wright et al. (2003), which have been deployed on several other modern earthquakes across Turkey (Taymaz et al. 2007;Elliott et al. 2013;Karasözen et al. 2016Karasözen et al. , 2018Pousse-Beltran et al. 2020). We first downsampled the unwrapped interferograms using a Quadtree algorithm (Jónsson et al. 2002) and then solved for the fault plane parameters that minimize differences between these datapoints and synthetic displacements calculated for a rectangular fault plane embedded within an elastic half-space (Okada 1985). For the half-space, we chose Lamé parameters µ = 3.2 × 10 10 Pa and Poisson ratio 0.25, consistent with the velocity structure obtained and applied elsewhere in this study. We inverted for fault strike, dip, rake, uniform slip, center point, length, and top and bottom depths, as well as linear N-S and E-W orbital ramps and the zero displacement level. A global minimum misfit is achieved using Powell's algorithm with multiple Monte Carlo restarts (Press et al. 1992; Clarke

Regional waveform modelling
We estimated moment tensors for several additional earthquakes in the 2019 Acıpayam and Bozkurt and 2017 Arıcılar sequences by modelling regional waveforms. Having larger signal-to-noise than teleseismic waveforms, regional waveforms permitted assessment of far smaller earthquakes, down to M w 3.5 in this study. We investigated around fifty earthquakes, of which 36 yielded robust mechanisms that meet strict quality and variance reduction criteria (Zahradník & Sokos 2018) and which are presented here.
For each event, we gathered waveform data recorded over the distance range 50-300 km by stations belonging to several regional networks listed in the Acknowledgements. The preferred frequency band for the inversion was selected after a careful analysis of the signal-to-noise ratio and station epicentral distances, and Green's functions were estimated for the local velocity model (Section 2.4) using the discrete wavenumber method of Bouchon (1981) and Coutant (1989). We then used the iterative deconvolution inversion method of Kikuchi & Kanamori (1991), implemented in the ISOLA software package (Sokos & Zahradník 2008, to solve for the best point source representation of each earthquake. All but one of the events have majority doublecouple components (with half of them >90%) and we present here the best double-couple solutions.
Previous regional waveform modelling studies indicate that minimum misfit centroid depths can vary according to the station configurations, velocity models, and frequency bands used in the inversion (e.g. Zahradnik et al. 2008;Haddad et al. 2020). Accordingly, for a few of the critical, larger events analyzed, we repeated the inversion using perturbations to these parameters -including alternative, published regional velocity models -from which we estimated centroid depth uncertainties of ∼1-2 km. However, the smaller events studied here are likely to have greater uncertainties, perhaps up to around 5 km (Herman et al. 2014).
Active tectonics along the Fethiye-Burdur trend 9

Calibrated hypocenter relocations
We used local, regional and teleseismic arrival times to determine calibrated hypocenters for the 2019 Acıpayam and Bozkurt and 2017 Arıcılar sequences, as well as background seismicity across southwestern Turkey. We collated phase arrival times from regional archives listed in the Acknowledgements and from the global International Seismological Centre (ISC) bulletin.
The selected earthquakes were then separated into five distinct geographic clusters -one each for the Acıpayam and Bozkurt sequences, and three others centered on Ç ameli in the southwestern Fethiye-Burdur trend, Burdur in the northeastern Fethiye-Burdur trend, and Beyşehir in the eastern Isparta Angle (Figure 3). Finally, each cluster was relocated using the Hypocentroidal Decomposition (HD) method (Jordan & Sverdrup 1981) as implemented in the mloc program (Bergman & Solomon 1990;Walker et al. 2011;Bergman et al. submitted).
The HD algorithm divides the relocation procedure into two distinct inverse problems that each utilize customized phase arrival time data (e.g. Karasözen et al. 2016Karasözen et al. , 2018. The first step uses arrival times of all phases recorded at all distances to determine 'cluster vectors' that relate the locations and origin times of each individual event with respect to the geometrical mean of all events, the 'hypocentroid'. The second step uses direct P g and Sg phases at epicentral distances <2 • -at which biases from unknown velocity structure are minimal -to establish the absolute location and origin time of the hypocentroid. The cluster vectors, added to the absolute hypocentroid, yield the 'calibrated' coordinates of all events: latitude, longitude, focal depth, origin time, and their uncertainties. The HD method can solve for focal depth as a free parameter if all events in the cluster have near-distance readings; around one third of the nearly 700 relocated earthquakes were determined in this way. For most of the remainder, we set the depths manually by minimizing the residuals at close-in stations. For around 100 events, focal depths were fixed to a default value of 10 km for the Ç ameli cluster and 15 km for the other clusters. By analyzing fits to P g and P n at the closest stations and P n and Sn at regional distances, we settled upon a two-layered crustal velocity model with V P 5.7 km/s and V S 3.25 km/s for the upper 20 km and V P 6.2 km/s and V S 3.6 km/s from 20 km to the Moho at 40 km. Below the Moho, we used velocities from the ak135 1-D Earth model (Kennett et al. 1995). The relocation procedure eliminates systematic biases of up to ∼0.5 sec and ∼1.5 sec in P g and Sg residual travel times, respectively, and reduces their root mean square errors from starting values of ∼1-2 sec down to ∼0.3-0.6 sec. We have posted detailed information on each cluster -such as arrival time compilations, station coordinates and calibration raypaths, velocity models, travel time residual plots, focal depth histograms, and epicentral uncertainties -to the Global Catalog of Calibrated Earthquake Locations database (Bergman et al. (submitted); see Data Availability).
Resulting, calibrated hypocenters have typical uncertainties of ∼1-2 km in latitude and longitude. Focal depth accuracy depends strongly on the availability of close-in stations, meaning those at epicentral distances less than ∼1-2 times the focal depth (e.g. Gomberg et al. 1990). In two previous studies of ours in neighbouring regions of western Turkey, we estimated these uncertainties at ∼2 km where close-in stations are available and ∼5 km where they are not (Karasözen et al. 2016(Karasözen et al. , 2018. This marks a significant improvement on the relocated ISC-EHB catalogue, whose focal depth uncertainties have been estimated at ∼10-15 km (Engdahl et al. 2006). However, a comparison between our calibrated focal depths and centroid depths from regional waveform modelling reveals the former to be on average several kilometers deeper, with respective means of ∼8 km and ∼14 km (Supplementary Figure S1). This discrepancy holds for individual seismic sequences and is consistent across three orders of magnitude (M w 3-6). It also mimics patterns observed elsewhere in western Turkey (Karasözen et al. 2016(Karasözen et al. , 2018Mutlu 2020) and in similarly well-instrumented regions of Alaska (Gaudreau et al. 2019) andIsrael (Haddad et al. 2020). Our interpretation is that for most of the events analyzed, calibrated relocations provide an upper bound on focal depth while regional waveform modelling is better at resolving the shallowest earthquake depths.
Active tectonics along the Fethiye-Burdur trend 11 2.5 Regional compilation of well-located earthquake focal mechanisms Lastly, we compiled a regional catalogue of well-located earthquake focal mechanisms by combining our own results with source parameters from the literature. We found a total of 299 earthquake focal mechanisms across the region shown in Figures 2 and 3; the full catalogue, with references, is given in Supplementary Table S3. Of the larger events (greater than M w ∼5) between 1955 and 2019, fifteen mechanisms were estimated using first motion polarities, thirty-six using teleseismic long-period body waveform modelling, and sixty-five were determined by the Global Centroid Moment Tensor (GCMT) project. In addition, 183 smaller events (M w 3-5) were calculated using regional waveform modelling or first motions (mostly the former), but these go back only as far as 2001, around the time that station coverage across Turkey started to improve markedly. Of the 299 focal mechanism events, 241 have hypocenters determined from calibrated relocations, either in this study or by Karasözen et al. (2016Karasözen et al. ( , 2018. Most of the remainder are offshore earthquakes characterized by large azimuthal gaps at regional distances, making their precise relocation difficult. For these earthquakes, we choose the best available hypocenter from the ISC where possible: in most cases, we took the parameters listed in the relocated ISC-EHB catalogue (Engdahl et al. 1998;Weston et al. 2018). InSAR data reveal a NW-SE-oriented elliptical fringe pattern with line-of-sight displacements of up to ∼5 cm away from the satellite (Figure 5a, left column). Since the pattern is similar in ascending and descending interferograms, these displacements must be dominated by vertical rather than horizontal motions. Our elastic dislocation modelling best reproduced the observed ground deformation with normal slip on a buried, moderately (54 • ) NE-dipping model fault that projects to the surface within the flat, central Acıpayam basin (Figure 5a, center and right columns; Figure 6a; and Table 1). Our relocated hypocenter lies just down-dip of the southeastern extent of model slip patch, suggesting that the mainshock rupture propagated upwards and unilaterally towards the NW ( Figure 6a). An alternative, SW-dipping model fault reproduced the data nearly as well, but we consider this geometry unlikely on the basis that the relocated hypocenter would be located updip of the main slip area (Supplementary Figure S1). On our preferred, NE-dipping model fault, slip is restricted to a depth range of ∼4-9 km with peak slip of ∼0.3 m at ∼6 km depth (Figure 6b), matching the minimum misfit centroid depth from teleseismic body waveform modelling ( Figure 7a) and only slightly shallower than the ∼7 km centroid depth estimated using regional waveforms. The InSAR model moment lies in the middle of the range of seismological estimates, suggesting negligible contribution to modelled slip from early aftershocks or afterslip. Finally, we note that our preferred source parameters are in good agreement with alternative InSAR-derived slip models by Yang et al. (2020) and Elliott et al. (2020), with discrepancies of 10 • or less in strike, dip and rake, and near-identical slip depth ranges.

THE RECENT ACıPAYAM, BOZKURT AND ARıCıLAR EARTHQUAKE
The mainshock was preceded ∼5 hours earlier by a moderate (M w 3.7) foreshock, located ∼1 km to the SE and with a similar normal mechanism (Figure 6a and Supplementary Table S2). An abundant aftershock sequence includes 193 earthquakes with sufficient station picks for precise relocation, of which twenty-three were sufficiently large (M w 3.5-5.1) that we could obtain robust focal mechanisms and centroid depths. The aftershocks form a diffuse distribution, with several colocated with the mainshock slip region but others lying well away from it. Centroid depths range from 3-15 km, with the greatest concentration at 4-5 km, but likely uncertainties of up to a few kilometers make it difficult to ascertain whether the colocated events lie on, or off (below or Active tectonics along the Fethiye-Burdur trend 13 above), the mainshock fault plane. Southern aftershocks -including a cluster around the southern end of the mainshock slip region -tend to have normal mechanisms similarly oriented to that of the mainshock and so might plausibly lie on the same fault plane. Northern aftershocks, on the other hand, involve normal faulting with a greater diversity of orientations including a few orthogonal to the main fault plane. The northern aftershocks also include a few oblique slip events and a single strike-slip earthquake.
The mainshock fault is highly oblique to the sinistral-normal Acıpayam fault in the southern Acıpayam basin (Kürçer et al. 2016;Emre et al. 2018) and somewhat oblique to a number of un- with the aid of high-resolution topographic imagery (Elliott et al. 2020). This suggests either that shallow extension is accommodated elsewhere -perhaps by distributed deformation -or that the fault is structurally immature, by which we mean that it has yet to accommodate appreciable cumulative slip. The inference of structural immaturity is consistent with our observation of diffuse aftershock seismicity, much of it presumably on structures subsidiary to the mainshock fault where peak line-of-sight displacements are ∼4 cm away from the satellite. We replicated the observed deformation most closely with normal slip on a buried, ∼Nor ∼S-dipping model fault, though we found fault strike to be poorly resolved due to the circular deformation pattern. We favour the N-dipping model since its parameters are in much closer agreement with our teleseismic body waveform focal mechanism than those of the S-dipping model fault ( Figure 6c, center and right columns; Figure 7b; Table 1). Our relocated hypocenter lies at the western edge of the modelled fault slip, suggesting unilateral, eastward rupture. Model fault slip occurs at depths of ∼6-10 km with peak slip of ∼0.6 m at ∼8.5 km ( Figure 6d). Our teleseismic waveform model centroid depth is somewhat deeper at ∼12 km, though we find similar waveform misfits across the centroid depth range 9-14 km. Our minimum misfit centroid depth from regional waveform modelling lies near the shallow end of this range, at ∼10 km. The InSAR model moment and moment magnitude (M w 6.0) are larger than any of the available seismological solutions, hinting that the modelled fault slip incorporates some postseismic afterslip.
A M w 4.1 foreshock and six M w 3.6-4.0 aftershocks were sufficiently well-recorded for regional waveform modelling, and seven smaller aftershocks could also be precisely relocated (Figure 6c).
The larger events involved predominantly normal faulting mechanisms -mostly oriented ∼E-W except for one which was oriented ∼N-S -at centroid depths of 5-11 km. Several of the aftershocks are located close to the up-dip edge of the InSAR-derived model slip distribution, though the limited depth resolution precludes any firm association or interpretation.
The surface projection of our model fault aligns closely with a mapped, N-facing scarp in the southern part of Acıgöl basin, ∼3 km north of the main, rangefront-forming Acıgöl fault (Figure 5b). Topographic profiling indicates that the scarp is around 5-10 m high. Its involvement in All of the available InSAR imagery captures both the foreshock and mainshock. Ascending and descending coseismic interferograms each exhibit an E-W-oriented, elliptical fringe pattern with peak line-of-sight displacements of ∼11-14 cm (Figure 9, left column). Observed displacements were best reproduced by normal slip on a S-dipping model fault that extends from the surface to ∼4 km depth (Figure 9, center and right columns; Figure 10; Table 1). The foreshock and mainshock are both colocated with the model slip area and their combined seismological moments approximate the InSAR model moment, suggesting that both events contributed to the observed surface deformation. Model slip peaks at ∼30-40 cm at ∼2 km depth, and few centimeters of model slip reaches the surface over a distance of 4 km, suggesting that a small surface rupture may have occurred (Figure 10b). Very shallow coseismic slip is further supported by our regional moment tensor centroid depths of ∼1-2 km, which additional depth resolution tests (Section 2.3) confirmed as being robust. Such shallow rupture is unusual in continental crust of the eastern Mediterranean and Middle East, but we note that it is not unprecedented (Savidge et al. 2019;Elias et al. 2021).
The causative fault is not evident in the topography and was not known prior to the earthquake.
However, it is only a few kilometers along strike from -and only ∼20 • oblique to -the easternmost mapped extent of the SSW-dipping Mugla normal fault, which has a similar geological slip vector to that of our InSAR model (Howell et al. 2017). We therefore consider that the 2017 earthquakes ruptured an eastern continuation of the Mugla fault zone.

DISCUSSION
Next, we discuss the broader patterns of seismicity along the Fethiye-Burdur trend revealed by our new compilation of focal mechanisms and relocated epicenters. Where possible, we compare focal mechanisms with geological or geomorphic indicators of fault kinematics, and we also test their consistency against the GPS-based tectonic models shown in Figure 2. Our analysis starts in the northern study area where we also add a detailed reassessment of the 12 May 1971 M w 6.0 Burdur earthquake, one of the largest and most destructive instrumental events in western Turkey.
Our focus then switches to the southern study area, between Fethiye and Ç ameli. The section concludes with a brief discussion of the forces likely to be driving the observed deformation.

Seismicity along the northern Fethiye-Burdur trend
Within the northern study area, earthquake focal mechanisms indicate a predominance of shallow normal faulting with a wide diversity of orientations (Figure 4). There are only a very few scattered strike-slip events -all with small to moderate magnitudes -and there is certainly no evidence for a through-going strike-slip fault zone as depicted in early GPS models (e.g. Figure 2c). Nodal A m b 5.3 earthquake on 9 September 1971, relocated to the Korkuteli basin in the SE of Figure 4, was previously assigned a pure strike-slip mechanism from teleseismic P waveform modelling (Yılmaztürk & Burton 1999). However, only ten waveforms were used in this study and the authors recognize that there are large residuals at some stations leading to large uncertainties in the mechanism. Moreover, Yılmaztürk & Burton's centroid depth of 34 km is inconsistent with our focal depth of 15 km and with other regional focal depths. This earthquake has been used elsewhere to argue for a left-lateral FBFZ (Hall et al. 2009), but we consider its published source parameters to be questionable and do not include it in our focal mechanism database. Further south and east, the Bey Dagları mountains and Aksu basin are characterized by mostly N-S-trending normal faulting mechanisms (Figure 3), consistent with regional E-W extensional strain (Figure 2d). This style of faulting also seems to predominate further east still, in the western Taurus mountains. Our hypocentral relocations place the Burdur mainshock and largest two aftershocks close to Lake Salda, ∼30 km WSW of Lake Burdur (Figure 4). Smaller relocated aftershocks form a broader distribution between Lake Salda in the WSW and the southern end of Lake Burdur in the ENE.
The orientation of the aftershock cloud matches the strike of the mainshock nodal planes but its length of 30-40 km likely exceeds that of the M w 6.0 mainshock fault plane based on scaling relations (Wells & Coppersmith 1994). The easternmost aftershocks are therefore likely to be situated some distance along strike from the mainshock rupture. Collectively, this suggests that the Burdur mainshock propagated unilaterally towards the ENE from its epicenter near Lake Salda, but that it Active tectonics along the Fethiye-Burdur trend 19 terminated well short of the Hacılar and Suludere faults that were attributed to this earthquake by Taymaz & Price (1992). The heavy damage to villages at the southern end of Lake Burdur likely reflects this rupture directivity, while the cracks observed along the SE margin of the lake might reflect secondary deformation related to liquefaction or landsliding which were also observed in this area.
The Burdur mainshock faulting is therefore confined to the area between Salda and Yarışlı Lakes, which exhibits indistinct surface geomorphology and lacks mapped surface faulting. The tight clustering of the mainshock and two largest aftershocks coupled with their diversity of nodal plane dip angles suggests high structural complexity within the source region. These observations hint that the M w 6.0 Burdur earthquake ruptured an immature fault with low cumulative slip, much like the 2019 M w 5.7 Acıpayam and M w 5.9 Bozkurt earthquakes analyzed in Section 3.1-3.2. Focal mechanisms of the Burdur sequence, dominated by NE-SW-trending normal faulting mechanisms, are also consistent with our inference of radial extension within the northern Fethiye-Burdur trend in Section 4.1.

Seismicity along the southern Fethiye-Burdur trend
In the southern part of the study area, the fifteen moderate magnitude earthquakes (up to M w 5.4) with assigned focal mechanisms almost exclusively involve ESE-WNW-oriented normal faulting and there is no evidence of any strike-slip activity (Figure 8) . This is clearly insufficient to account fully for the roughly ∼60 • difference in strike between the instrumental earthquake nodal planes and the largest faults. A second possibility is that there has been a recent change in the regional strain field, from NW-SE-directed extension to NNE-SSW extension. Fault kinematic and tectonostratigraphic data from the Ç ameli basin support such a change and constrain its timing to the late Quaternary (Alçiçek et al. 2006). We speculate that the switch might be related to eastward propagation of the Gökava graben into the area (Tur et al. 2015) and/or to lateral gradients in gravitational potential energy introduced by rapid subsidence of the Rhodes basin (Hall et al. 2009) (Figure 3).

Dynamics of the deformation
Data from multiple sources indicate radial horizontal extension at the northern end of the Fethiye-Burdur trend. We now discuss why this radial extension might occur. Processes that are thought to drive deformation in the Aegean and Anatolia include: (1) slab rollback in the Hellenic and (2) a NW-SE gradient between Burdur and the Antalya basin. Both west and east of Burdur, the stress field predicts more uniaxial horizontal extension associated with each GPE gradient; radial horizontal extension is only expected in the region equidistant from the Rhodes and Antalya basins. Lateral variations in GPE are therefore sufficient to explain the large-scale pattern of surface deformation in SW Turkey, although it is hard to rule out contributions from other dynamic processes.

CONCLUSIONS
Our refined and updated earthquake catalog for southwestern Turkey reveals no evidence for NEtrending, active strike-slip faults along the Fethiye-Burdur trend, as has previously been posited.
Instead, the western limb of the Isparta Angle is characterized by shallow normal faulting earth-quakes, with a diversity of orientations in the north (across Turkey's Lake District), mostly N-S nodal planes in the east (in the Bey Dagları mountains), and ESE-WNW nodal planes in the south (near Fethiye and Ç ameli). In each case, fault orientations are consistent with the principal axes of the horizontal strain rate tensor calculated from regional GPS velocities (Howell et al. 2017).
These kinematics appear to be driven principally by lateral gradients in gravitational potential energy between the high Anatolian plateau and the deep Rhodes and Antalya basins.
Three earthquake sequences associated with clear InSAR signals provide additional information on how active faulting is manifest in the topography. The 2019 M w 5.6 Acıpayam earthquake involved buried slip on a previously unrecognized fault with no discernible geomorphic expression.
The 2019 M w 5.8 Bozkurt earthquake was also buried, but its fault plane aligns with subtle (5-10 m-high) surface scarps that had previously been mapped. The 2017 M w 5.3 Arıcılar earthquake slipped at shallower depths but also failed to break to the surface; its causative fault lies a few kilometers along strike of the mapped Mugla fault zone but also appears indistinct in the topography.
All three of these earthquakes can therefore be inferred to have ruptured structurally-immature (low cumulative slip) faults. Our relocation of the destructive 1971 Burdur sequence hints that this also ruptured a structurally-immature fault zone with an indistinct expression in the topography.
These observations raise the spectre that across southwestern Turkey damaging earthquakes (of up to at least M w 6) are possible on faults that would prove difficult to identify beforehand. Active tectonics along the Fethiye-Burdur trend 23

Data Availability
Interferograms were constructed using Copernicus Sentinel-1 data (2017, 2019) available from https://scihub.copernicus.eu/. Corresponding interferograms are also available to download from the COMET LiCS database (Wright et al. 2016), which we exploited during our initial reconnaissance of the Acıpayam, Bozkurt and Arıcılar earthquakes. Teleseismic waveforms were accessed through IRIS Data Services, and specifically the IRIS Data Management Center (https://ds.iris.edu/ds/nodes/dmc/). partner networks. Complete references for these earthquake parametric data sources are given in Supplementary Table 2. The ISOLA software can be downloaded from http://seismo.geology.upatras.gr/isola/ and Mloc source code from https://www.sciencebase.gov/catalog/item/ 59fb91fde4b0531197b16ac7. Other codes used in the paper will be shared upon reasonable request to the corresponding author. All figures in this paper were plotted using Generic Mapping Tools (Wessel et al. 2013). Table 1. Source parameters of the 2017 Arıcılar foreshock and mainshock and the 2019 Acıpayam and Bozkurt mainshocks from catalogues (GCMT = Global Centroid Moment Tensor project; USGS = United States Geological Survey ANSS Comprehensive Earthquake Catalog (ComCat); RMT = Regional Moment Tensor solution) and from our own modelling. The listed origin times are those yielded by calibrated earthquake relocations. Location refers to the latitude and longitude of the GCMT and GEOFON centroids, the USGS epicenter, the relocated epicenter for our seismological solutions, and the peak slip patch our InSAR solutions. Depth refers to the centroid depth for all of the seismological solutions, and the depth of peak slip for the InSAR solution.     Emre et al. (2018) and the topography is as in Figure 1. Earthquake focal mechanisms (beach balls) and epicenters (circles) are coloured by year of occurrence and plotted at their relocated epicenters (Karasözen et al. (2016(Karasözen et al. ( , 2018, and this study), except for a few offshore events which could not be reliably relocated and which are marked with shadows. We only plot earthquakes whose best available focal or centroid depths are <35 km; a few deeper events, in particular in the Antalya Bay region, are excluded. shown; the second has a similar mechanism but its relative location is unconstrained (Taymaz & Price 1992). SL = Salda Lake and YL = Yarışlı Lake.