The landslide source of the eastern Mediterranean tsunami on 6 February 2023 following the M_w 7.8 Kahramanmaraş (Türkiye) inland earthquake

Abstract

This paper presents the first example of how to systematically identify the submarine landslide source of a tsunami using an innovative hybrid approach. This ground-breaking method is developed to resolve the puzzle around the source mechanism of the mysterious tsunami observed on 6th February 2023 in the Eastern Mediterranean Sea. The tsunami followed the two inland M_w 7.8 and M_w 7.5 Türkiye–Syria earthquakes, which occurred consequently with a 9 h interval on this day. The first earthquake (M_w 7.8) had an epicentral distance of 90 km from the nearest coast, which is closer than the second one (M_w 7.5) to the coast and yet its crustal deformation was almost entirely limited to inland. Therefore, the co-seismic surface displacement generated by the earthquake was ruled out as the source of the tsunami, confirmed by numerical modelling. Here, we hypothesized that the tsunami was most likely generated by a submarine landslide triggered by the earthquake. Analysis of tide gauge observations revealed that the waves arrived from 27 min to 48 min after the first earthquake (M_w 7.8) at different coastal locations, implying that the potential submarine landslide was triggered by the first earthquake (M_w 7.8). Backward tsunami travel time mapping using tide gauge observations guided us to constrain the area of the potential landslide. We approximated the dimensions of the landslide using spectral analysis of the tsunami observations. Consequently, an iterative trial-and-error approach was employed to confirm the landslide source of the tsunami by defining various informed alternative landslide scenarios and applying numerical modeling. Modelling showed that a submarine landslide can reproduce the tsunami observations reasonably well. It is located on a steep slope of the seafloor approximately 50 km from Arsuz. The submarine landslide is estimated to have caused a seafloor deformation measuring approximately 16 km in length and 4.0 km in width.

Introduction

The Türkiye–Syria border was struck by two large earthquakes (M_w 7.8 and M_w 7.5; Fig. 1) on 6th February 2023 within nine hours leaving a death toll of over 50,000 (International Medical Corps 2023). According to the United States Geological Survey (USGS), both events were the results of strike-slip faulting and occurred at shallow depths of 17.9 km and 10.0 km for the M_w 7.8 and M_w 7.5 events, respectively. The respective origin times were 01:17:35 UTC (for M_w 7.8) and 10:24:49 UTC (for M_w 7.5). These devastating earthquakes remind the tragedy generated by the 17th August 1999 Izmit earthquake (M_w 7.4) in northwestern Türkiye (Fig. 1) which killed over 17,000 people (Barka 1999; Yalciner et al. 1999; Ozalaybey et al. 2002).

Fig. 1

Epicentral area of the two Türkiye–Syria earthquakes of 6th February 2023 (1st EQ and 2nd EQ). The tide gauges used in this study are shown as blue triangles, and the contours representing tsunami travel time (TTT) are shown as black dashed lines at 0.25 h (15 min) intervals. For TTT calculations, it is assumed that the earthquake source was the nearest water point to the epicenter of the 1st EQ. Brown dashed lines are approximations of adjacent plate boundaries. All epicenters and focal mechanisms are based on the USGS except for the event of 1999 which is from the solution made by the University of Tokyo (Japan)

The epicenters of the two recent 2023 earthquakes are located within a triple junction where three tectonic plates (the Anatolian, Arabian and African plates) meet (Fig. 1). Although theoretically triple junctions are thought to develop high levels of tectonic strains and stresses, the region had not experienced any earthquake larger than M 6.7 in the past 100 years. Possibly this is a reason that the recent M_w 7.8 and M_w 7.5 events came as surprises to the authorities and public and caused unprecedented damage and deaths.

A moderate-size tsunami was generated in the eastern Mediterranean Sea following the 6th February 2023 earthquakes, which was reported by local people, and was recorded on tide gauges (Fig. 2). As the earthquakes’ epicenters were at least 90 km away from the coastline, it is not clear how the tsunami was generated. Normally, there are two possibilities: (1) Either the co-seismic crustal deformation was extended to the sea and generated the tsunami, or (2) the tsunami was generated by a coseismic submarine landslide or landslides triggered by the earthquakes. For the case of the 17th August 1999 Izmit Bay tsunami in Türkiye due to an inland M_w 7.4 earthquake (Fig. 1), it was shown by Tinti et al. (2006) that a submarine landslide contributed to the tsunami. Another example of a tsunami generated by an inland earthquake due to a coseismic submarine landslide is the tsunami of 24th September 2013 in the northwestern Indian Ocean following an inland M_w 7.7 earthquake in Pakistan (Heidarzadeh and Satake 2014). The recent February 2023 tsunami is the third tsunami to hit Türkiye in the past six years following the July 2017 and October 2020 tsunamis (Fig. 1) (Dogan et al. 2019, 2021; Heidarzadeh et al. 2017a, 2021). The region experienced historical tsunamis such as those in 749 AD and 1759 AD (Papadopoulos et al. 2014).

Fig. 2

The tide gauge records for six stations close to the epicenters of the 6th February 2023 Türkiye–Syria earthquakes (1st EQ and 2nd EQ). The blue arrows indicate tsunami arrivals in each station. Three stations (Bozyazi, Antalya, and Marmaris) do not show clear tsunami signals. TTT indicates tsunami travel time to each station. The red dashed vertical lines represent the origin times of the two earthquakes

For regional tsunami hazard assessment, it is critically important to identify which of the above two candidates (the earthquake, or a submarine landslide) was responsible for the February 2023 tsunami in the eastern Mediterranean Sea. Therefore, we developed this study with the objective of identifying the source of the tsunami. Here, we examine both possibilities (either co-seismic crustal deformation or co-seismic submarine landslides) through investigating tide gauge records of the tsunami and performing numerical simulations.

Data and methods

We collected and analyzed six tide gauge records of the tsunami (Table 1, Fig. 2). All records come with a sampling interval of 30 s and are provided by the Intergovernmental Oceanographic Commission’s sea level station monitoring facility (http://www.ioc-sealevelmonitoring.org/map.php). The data underwent quality control to remove spikes. Tidal analysis was performed using the TIDALFIT package (Grinsted 2023) to calculate tide signals which were then removed from the original tide gauge data to obtain de-tided tsunami waveforms (Fig. 2). TIDALFIT is a Matlab package (MathWorks 2023) that employs ordinary least squares regression methods to fit tidal components to sea level data (e.g., Grinsted 2023; Heidarzadeh et al. 2017b, 2017c). It is noted that, according to Fig. 2, the tsunami was initiated by the first earthquake (M_w 7.8). It is challenging to know whether the second earthquake has generated any additional tsunami or not. Throughout this research, we exclusively focus on the tsunami generated by the first earthquake.

Table 1 Information of tide gauge stations used in this research to study the 6th February 2023 Türkiye tsunami

Spectral analyses in this study were conducted through Fourier and Wavelet analyses. For Fourier analysis, we applied an updated version of the Welch’s (1967) power spectral density estimate presented by the function pwelch(x) in Matlab (MathWorks 2023) considering Hanning windows and 50% overlaps (Mulia et al. 2022; Wang et al. 2022; Heidarzadeh et al. 2017b; Heidarzadeh and Gusman 2021; Cheng et al. 2023a,b; Hu et al. 2023). Wavelet analysis was performed by applying the wavelet package of Torrence and Compo (1998) and considering the Morlet mother function.

The earthquake slip model is provided by the United States Geological Survey (USGS) (Fig. 3a). The corresponding coseismic crustal deformation is shown in Fig. 3b. The USGS fault model consists of three segments named S1, S2, and S3 discretized into subfaults with a size of approximately 8 km². Details on the fault slip modeling setup are available at the USGS website at https://earthquake.usgs.gov/earthquakes/eventpage/us6000jllz/finite-fault, with main fault parameters tabulated in Table 2. We calculated the coseismic crustal deformation from the slip model using Okada’s (1985) analytical formula, assuming an instantaneous crustal deformation. The effect of horizontal displacements on the tsunami genesis (Tanioka and Satake, 1996) was incorporated, which is necessary particularly for a strike-slip event. A maximum slip of 11 m occurred on S2, the largest segment, resulting in an uplift of up to 0.5 m and a subsidence of approximately 0.9 m inland (Fig. 3). The likely segment responsible for the tsunami was S3, being the closest to the sea. However, the average slip on this segment (i.e., S3) was only 1.0 m; hence it produced insignificant vertical displacements of less than 0.05 m on the seafloor.

Table 2 Fault parameters for the source model developed by the USGS for the 6th February 2023 M_w 7.8 Türkiye–Syria earthquake

Fig. 3

The earthquake source model for the 6th February 2023 Türkiye–Syria M_w 7.8 earthquake developed by the USGS. The red stars indicate the epicenter. The model comprises three fault segments marked by S1, S2, and S3

The modelling package COMCOT (Cornell Multi-Grid Coupled Tsunami model) was used to simulate the tsunami from various source models evaluated in this study. COMCOT was originally developed at Cornell University (USA) in the 1990s (Liu et al. 1998; Wang 2008), and it has been under development at GNS Science, New Zealand since 2009 (Wang and Power 2011). In this study, a single bathymetric grid layer with a grid spacing of 15 arc-sec was used to simulate tsunami propagation and coastal amplifications. The Digital Elevation Model (DEM) data for the grid layer were extracted from the GEBCO (2022) (General Bathymetric Chart of the Oceans) bathymetric data which has an original resolution of 15 arc-second. We solved linear shallow water equations considering a time step of 0.1 s. Total simulation time was 4 h.

For landslide tsunami modelling, a Gaussian non-symmetric dipole initial source was considered as the input initial wave and thus a static tsunami initiation was applied following previous studies such as Synolakis et al. (2002), Satake and Tanioka (2003), Okal and Synolakis (2004), and Tappin et al. (2008). Although this method has been established as one of the reliable methods for modelling landslide tsunamis, its details yet to be developed. Among such details is the relationship between the dimension of the initial surface water displacement ($L$) and the length of the submarine landslide ($b$). This motivated us to conduct a few limited laboratory experiments, which were delivered through moving a submarine mass (a 20 cm long concrete block—see Fig. 4a) down a slope, with a slope angle of 47°, in a wave flume of the size 10 m (length) × 0.45 m (height) × 0.3 m (width). The water depth was 35 cm, the submergence depth was 5 cm (Fig. 4a), and the sliding mass was released at rest and moved down the slope under gravity. Results indicated the following relationship:

$$L \sim 2 b$$

(1)

where $L$ is the dimension of the initial surface water displacement, and $b$ is the length of the submarine landslide. We acknowledge that Relationship (1) depends on the water depth, slope angle and the type of the material (either solid-block or granular), which are not examined in our limited physical experiments. Also, the slope angle of the physical experiment may be higher than the actual slope angle on the seafloor. Therefore, Relationship (1) is a preliminary result, and can be used only to give initial insights. It is challenging to know the duration of landslide occurrence due to limited data. However, we estimate it to be short (around 1–2 min) and speculate that landslides should have occurred at a high speed (approximately 20–40 m/s) because slow-moving landslides are not efficient towards tsunami generation. For example, by assuming an average velocity of 30 m/s and considering a travel distance of 2 km for the landslide, the duration of landslide occurrence will be 1.1 min.

Fig. 4

a, b Snapshot of a submarine landslide physical experiment at rest (i.e., t = 0) and 0.5 s after landslide onset showing generation of an initial negative (depression) phase shoreward and an initial positive (elevation) phase seaward of the landslide. c Sketch showing the shape of the initial wave generated by a submarine landslide. “$L$” is the initial surface water displacement, and “$b$” is the length of the landslide as shown in panel a

Two pieces of information were gained from our physical experiments: (1) The initial wave generated by a submarine mass movement at the end of the wave generation has a dipole shape made of a negative (depression) phase shoreward and a positive (elevation) phase seaward (Fig. 4). And (2) The length of the initial surface water displacement is estimated at approximately twice of the length of the sliding mass (Relationship 1). Previously, other authors also considered dipole waves for modelling submarine landslide tsunamis (Synolakis et al. 2002; Satake and Tanioka 2003; Okal and Synolakis 2004; Watts et al. 2005; Grilli and Watts 2005; Tappin et al. 2008; Yalciner et al. 2014; Heidarzadeh et al. 2019). It is noted that the limited physical experiments conducted here were only used as a guide for developing alternative landslide source scenarios. Therefore, the limitations of the physical experiments do not have any impact on the outcomes of our research because the final source model are carefully tested through detailed numerical modeling on the actual bathymetry of the region and were validated through comparison with real tide gauge observations.

Characterization of the waves and estimating source lengths

Spectral and wavelet analyses are performed to characterize the waves (Fig. 5). For these analyses, we only considered two stations that registered most clear tsunami signals (Arsuz and Erdemli). Comparison of the tsunami spectra (blue-colored spectra) with the background ones (black-colored spectra) reveals peak tsunami periods. According to Fig. 5a, the peak tsunami periods are 5.4 min, 7 min, 12 min and 17 min in Arsuz, and they are 7 min, 13 min, and 21 min in Erdemli.

Fig. 5

Spectra a and wavelet b plots for the tide gauge records of the 6th February 2023 eastern Mediterranean tsunami. Here, 1st EQ and 2nd EQ refer to the two earthquakes on 6th February 2023 at 01:17:35 UTC (M_w 7.8) and 10:24:49 UTC (M_w 7.5), respectively (see Fig. 1 for their locations)

The tsunami source periods are normally those that occur in several stations and show large spectral energy differences with the background spectra or the peak periods that appear in spectral ratio plots (e.g., Rabinovich 1997; Heidarzadeh and Satake 2017; Zaytsev et al. 2021; Heidarzadeh et al. 2022a; Wang et al. 2023). Based on the peak periods in Arsuz and Erdemli, we could establish three dominant tsunami period bands of 5.4–7 min, 12–13 min, and 17–21 min (Fig. 5a). Wavelet analyses reveal that the period bands of 12–13 min, and 17–21 min are more persistent and stronger in Arsuz than in Erdemli (Fig. 5b). On the other hand, the period band of 5.4–7 min is stronger in Erdemli than in Arsuz. Wavelet plots reveal distinct tsunami oscillations patterns between Arsuz and Erdemli: while most of the tsunami energy is channeled to the period band of 12–13 min and 17–21 min in Arsuz, two evident channels of 7 min and 21 min are seen in the wavelet plot of Erdemli. This may indicate that the tsunami wavefield was more complicated in Erdemli than in Arsuz.

The peak tsunami periods (${T}_{p}$) can be used to estimate the characteristic length of the initial sea surface displacement of a tsunami ($L$), for example, using the following equation (Heidarzadeh and Satake 2015):

$$L=\frac{{T}_{p}}{2}\sqrt{gd}$$

(2)

where, $L$ is the characteristic length (or width) of the initial surface water displacement generated by a submarine landslide, ${T}_{p}$ is peak tsunami period, $g$ is gravitational acceleration, and $d$ is water depth around the source region. Considering the water depths of $d=$ 100–200 m (150 m as average) for the source region (Fig. 1), and the peak periods (${T}_{p}$) of 5.4–7 min (6.2 min as average), 12–1  min (12.5 min as average), and 17–21 min (19 min as average), we estimate the characteristic length of the initial surface water displacement to be approximately in the range of 5 – 28 km for these period bands.

Modelling tsunamis from the earthquake source

We applied the earthquake source model of the USGS as the initial source of the tsunami and modelled its propagation (Fig. 6). Simulation results reveal that the modeled tsunami is negligible and the maximum simulated amplitude at the closest tide gauge to the epicenter (Arsuz) is around 2 cm, whereas observed tsunami has a maximum amplitude of 15.2 cm in Arsuz (Table 1). In addition, the simulated waves show longer wave periods compared to the observations. For the other two stations of Erdemli and Tasucu, it can be seen that the simulated tsunami waveforms are less than 1 cm and bear no resemblance to the observed waveforms. Therefore, there is no doubt that the crustal deformation from the earthquake cannot be considered as the source of the tsunami observed in these stations. In other words, it is likely that a secondary source, triggered by the earthquake, was responsible for the generation of the tsunami. The most plausible secondary source can be a submarine landslide.

Fig. 6

Results of tsunami simulations (red waveforms) by considering the earthquake (M_w 7.8) as the source of the tsunami. The “Eq. source” refers to the USGS source model (see Fig. 3). The black waveforms show tsunami observations. The red vertical dashed line represents the origin time of the first earthquake (M_w 7.8)

Normally submarine landslides generate short-period tsunamis as compared to earthquakes due to the relatively smaller sizes of submarine landslides (Synolakis 2003; Heidarzadeh et al. 2014; Takagi et al. 2019). The fact that the observed tsunami on 6th February 2023 shows shorter periods compared to the simulated waves from the earthquake source (Fig. 6) may strengthen our hypothesis that the February 2023 tsunami was produced by a submarine landslide or landslides.

Backward tsunami travel time mapping

Backward tsunami travel time mapping is based on considering hypothetical tsunami sources at the location of a tsunami observation station (e.g., a tide gauge) and allowing the waves to propagate backward from that station to the open sea for the amount of actual tsunami travel times (TTT) of that station. This will produce a reverse (or backward) tsunami travel time contour. The zone where all reverse travel time contours from all stations meet is considered as the source of the tsunami. Backward tsunami travel time mapping is a powerful tool for pinpointing the location of tsunami source, and thus has been used in the past for both tectonic and landslide tsunamis. For example, Hayashi et al. (2011) and Mulia and Asano (2016) applied this technique to constrain the extension of the earthquake source of the March 2011 Japan tsunami due to the M_w 9.0 off-Tohoku earthquake. Heidarzadeh and Satake (2014) employed this method to identify the location of the submarine landslide responsible for the September 2013 tsunami following M_w 7.7 Pakistan inland earthquake. Heidarzadeh et al. (2022b) applied this method on the 16 June 2021 Seram Island (Indonesia) tsunami.

For the case of the 6th February 2023 tsunami, the TTT for each tide gauge station is given in Table 1 and shown in Fig. 2. The outcome of backward tsunami travel time mapping is presented in Fig. 7. Various uncertainties are associated with this analysis including those arising from bathymetry data and TTT estimates. For example, considering that the area of the tsunami is shallow with water depth of 100–200 m, a 30 m error in bathymetry data results in an error of approximately 11 km for the location of the tsunami source. Also, an error of 4 min for estimating TTT would result in a dislocation of approximately 9 km for the tsunami source. In addition, a point source is considered for backward travel time mapping while actual tsunami source is not a point source. Therefore, it is natural that the analysis (Fig. 7) gives a zone with length of approximately 45 km as the potential area of the tsunami source. It is noted that the boundaries of the tsunami source region (dashed box in Fig. 7) are not definitive, and we consider an uncertainty zone of at least ± 11 km. Despite such uncertainties, it is believed that the potential landslide zone obtained by backward tsunami travel time mapping is a success, and is helpful towards identifying the source of the tsunami.

Fig. 7

Backward tsunami travel time mapping for the 6th February 2023 eastern Mediterranean tsunami using the records of three tide gauges that registered tsunami signals (Arsuz, Erdemli, and Tasucu)

As the co-seismic crustal deformation was far from the coastal area (Fig. 3), it is most likely that the tsunami was produced by a submarine landslide that occurred in the area marked in Fig. 7. Another evidence that further supports this hypothesis is that the observed tsunami waves show shorter periods as compared to the simulated waves from the earthquake source (see Fig. 6). We test the hypothesis of a landslide being the source of the tsunami by performing full numerical modelling of tsunamis from potential submarine landslides.

Modelling tsunami from potential landslide sources

In this section, we apply an iterative trial-and-error approach by numerical modelling of landslide source candidates to identify the most credible landslide source. By considering the landslide zone identified by backward tsunami travel time mapping (Fig. 7) and estimates of the sizes of the initial water displacement, we considered seven candidate submarine landslide sources (LS) located in and around the zone (Fig. 8, Table 3). The source scenarios have maximum positive initial amplitudes of 0.5 m–2.0 m and maximum negative initial amplitudes of 0.6 m – 2.8 m (Table 3). The lengths and widths of the initial sea surface displacements for these LS scenarios range in the domains of 18.0 km–32.0 km and 8.0 km–8.5 km, respectively (Table 3, Fig. 8). The scenarios are located within or around the potential landslide zone detected by backward tsunami travel time mapping (Fig. 8). To consider the uncertainties associated with backward tsunami travel time mapping, some of the LS scenarios are placed around the borders of the zone (Fig. 8). The volume estimates of the landslide scenarios in Table 3 are based on the volumes of the total displaced water multiplied by two. The rational for applying a factor of two is that the displaced water by each landslide scenario is assumed to be half of the volume of the landslides.

Fig. 8

All submarine landslide (LS) alternative source scenarios considered in this study. The dashed box shows the potential landslide zone identified by backward tsunami travel time mapping (see Fig. 7)

Table 3 Seven candidate submarine landslide scenarios considered in this study to model the 6th February 2023 eastern Mediterranean tsunami

Figures 9, 10 present the results of numerical modelling of tsunamis generated by all seven LS scenarios by comparing the simulated waveforms with observations. LS1, LS2 and LS4 successfully reproduce the initial depression wave observed in Arsuz, but the simulated initial waves arrive earlier, and the amplitudes of the simulated waves are much larger than the observations. Another major problem is that all these three scenarios (LS1, LS2, LS4) produce short-period waves as compared to the observations. LS3 gives good results in Arsuz in terms of both amplitudes and arrival times, and the results are acceptable in Erdemli only for the first wave (Fig. 9). A good agreement between the period of the simulated and observed waves is seen for LS3, which is considered as an advantage of LS3 over the previous three scenarios. LS5 overestimates the waves both in Arsuz and Erdemli and the periods of the simulated waves are noticeably shorter than the observations (Fig. 10). The last two scenarios, LS6 and LS7, appear to underestimate the waves and produce short-period waves as compared to observations.

Fig. 9

Results of tsunami simulations (red waveforms) from hypothetical submarine landslide scenarios LS1–LS4. The black waveforms show tsunami observations at various tide gauges. The vertical dashed line represents the origin time of the first earthquake (M_w 7.8)

Fig. 10

Results of tsunami simulations (red waveforms) from hypothetical submarine landslide scenarios LS5–LS7. The black waveforms show tsunami observations at various tide gauges. The vertical dashed line represents the origin time of the first earthquake (M_w 7.8)

In summary, simulations reveal that LS3 is the best scenario among the examined ones and can reproduce the amplitudes and periods of the observations reasonably well.

The most credible landslide tsunami source and discussions

The final best-fit source model (model LS3; Fig. 11) can well reproduce the tsunami observations. In terms of waveform spectra, a relatively good agreement is seen between the spectra of the observations and simulations based on the final best-fit source model (Fig. 12). Applying Relationship (1), and by considering that the initial water surface displacement due to LS3 was 32 km × 8.5 km (Table 3), we estimate the size of the submarine landslide LS3 on the seafloor to be approximately 16 km (length) × 4 km (width).

Fig. 11

The final submarine landslide best-fit source model (LS3) for the 6 February 2023 eastern Mediterranean tsunami following the M_w 7.8 inland Türkiye–Syria earthquake. The blue dashed contours at the top panel show bathymetry contours in meters. The red vertical dashed lines at the bottom panels represent the origin time of the first earthquake

Fig. 12

Comparison of Fourier analyses of the observed (blue) and simulated (pink) waveforms for the final best-fit source model (LS3). The black spectra show background signals using the tsunami observation waveforms

As for limitations of this study, it is worth noting that our final best-fit source model (Fig. 11) is exclusively based on waveform analyses and numerical modelling, and it is not confirmed by geological mapping of the seafloor. The bathymetry data used in this study (i.e., GEBCO) is known as being uncertain in shallow water around the coastal areas and thus is not suitable for such geological studies. Despite this, it is helpful noting that our final best-fit source model is located on a relatively steep slope on a transition zone on the seafloor where water depth rapidly changes from 200 m to over 800 m.

As marine geological surveys are expensive and time-consuming, our results offer precious information regarding the location of submarine landslides, and thus can be used as guides in future marine geophysical surveys to save time and costs. The other limitation of this study concerns the quality of the bathymetric data, which is based on the GEBCO digital atlas and is not generally considered as high-resolution bathymetry data. It is recommended that potential future bathymetric surveys of the area produce high-resolution bathymetry data. There is no concrete and universal relationship to relate the size of the initial water surface displacement of a tsunami to the dimension of the seafloor deformation of a landslide. Our proposed relationship (1) should be considered as a rule of thumb.

It may be helpful noting that landslides may generate some seismic waves that could be recorded on seismic stations. The clarity of such signals among many other signals from the mainshock and aftershocks depends on the size and speed of the landslide and the distance of the seismic stations to the landslide location. In general, such seismic signals from landslides are hidden among other signals and our previous efforts for identifying them for the September 2013 offshore Pakistan event (Heidarzadeh and Satake 2014) was unsuccessful. Nevertheless, we recommend looking at such signals for future studies.

Conclusions

It has been a puzzle as how the 6th February 2023 East Mediterranean tsunami was generated following the M_w 7.8 Türkiye–Syria earthquake considering that the epicenter was approximately 90 km inland. We developed a novel hybrid methodology for the first time and applied it here to solve this puzzle and showed that the tsunami was generated by a submarine landslide triggered by the earthquake. Our hybrid methodology employs a combination of waveform, spectral, and backward tsunami travel time mapping analyses as well as numerical modeling. Main findings are:

• Three dominant tsunami period bands were identified through spectral analyses of the tide gauge waveforms, which are: 5.4–7 min, 12–13 min, and 17–21 min. Based on these dominant tsunami period bands, the dimension of the initial water surface displacement of the tsunami is approximated in the range of 5–28 km. This theoretical dimension guided us to decide the dimensions of the candidate tsunami sources.

• Tsunami simulations based on the earthquake source (M_w 7.8) revealed that the simulated tsunami was too small compared to the observations, and thus the co-seismic crustal deformation of the earthquake was ruled out as the tsunami source.

• The tsunami was most likely generated by a submarine landslide and the location of the landslide was constrained using backward tsunami travel time mapping. A zone was given by backward tsunami travel time mapping for the location of the potential submarine landslide.

• By considering seven candidate landslide sources and conducting numerical modelling, we were able to reproduce the observed tsunami using one of the submarine landslide scenarios reasonably well. It is located on a steep slope of the seafloor approximately 50 km from Arsuz. The submarine landslide is estimated to have caused a seafloor deformation measuring approximately 16 km in length and 4.0 km in width.