Introduction

Tsunamis are secondary geohazards after earthquakes, landslides, volcanic eruptions, asteroid impacts, and meteorological events1. Among these, powerful underwater earthquakes are the most common cause of tsunamis that strike coastal areas near the epicenter within minutes of the seismic event. Annual global tsunami reports indicate that about 60,000 people and (US$) 4 billion in property are at risk of tsunamis2. Scientific investigations have consistently demonstrated the pivotal role of an accurate and robust tsunami early warning system (TEWSs) in mitigating the destructive effects of these natural disasters. Therefore, the urgent need for a robust and accurate TEWS is essential for countries that may be exposed to tsunami disasters. Although tsunami occurances are infrequent along Egypt’s coasts, the country remains at high risk due to high population density, vulnerable environmental systems, and low flat land of the Nile Delta. Egypt’s northern shore along the southern Mediterranean Sea is a dynamic system. Active tectonic movements surrounding Egypt have been documented extensively throughout history, specifically in the Mediterranean Sea and connected seas. These regions are among the most seismically active areas, presenting a significant threat toward Egypt3,4. Divergent and convergent motions of the African, Eurasian, and Arabian plates are responsible for seismic activity in the Mediterranean Sea. Previous studies have highlighted a relationship between shallow seismic activity and plate boundaries5,6, particularly within the Hellenic subduction zone, where the African plate subducts beneath the Eurasian plate, and the Cyprian Arc, where the African and Anatolian plates collide7,8,9. Historical documents from past civilizations have recorded violent earthquakes causing tsunamis that inundated Egyptian coasts since the 4th century10,11,12,13,14,15,16. Paleotsunami deposits along the Egyptian coast provide compelling evidence of tsunami layers deposited on the Egyptian coasts17. Thrust faults contribute significantly to large-scale tsunami events worldwide18. Large-scale events in this region have been predominantly associated with thrust faulting along the Hellenic and Cyprian Arcs19. Figure 1 shows a map of the major historical earthquakes in the Mediterranean Sea region plotted using topography and bathymetry data20,21. Table 1 indicates the major historical earthquakes responsible for generating tsunamis that inundated the Egyptian coasts, along with their fault parameters estimated by previous studies9,22,23,24,25,26,27,28. According to the worst modeling scenarios, considering the fault parameters of these historical earthquakes, the estimated wave heights showed a 9 m run-up through Alexandria9.

Figure 1
figure 1

Historical earthquakes with focal mechanisms that caused tsunamis; two nodal planes are shown for each focal mechanism. The fault parameters are shown in Table 1. The map is plotted using PyGMT python library version v0.13.029.

Full size image
Table 1 Historical earthquakes and fault parameters in the Mediterranean Sea: the region, year, focal depths, moment magnitudes, the run-up heights at Alexandria and the fault parameters are gathered from previous studies9,22,23,24,25,26,27,28, L is the fault length, W is the fault width, and D is the displacement.
Full size table

In order to reduce the impact of future tsunamis in Egypt, this study aims to rapidly and precisely determine the magnitude of earthquakes in the southeast Mediterranean Sea as a first step toward developing a local TEWS for Egypt. Seismic data were automatically analyzed by SeisComP3 software, and various magnitudes were computed at different post-event times (6 minutes, 11 minutes, and 20 minutes). Magnitudes were compared with USGS (M\(_{ww}\)) magnitude values to determine the most reliable type and timing for TEWS implementation. This study highlights the potential of integrating infrasound technology to enhance earthquake detection and localization. Unlike seismic and tsunami waves, which propagate through the Earth and ocean, respectively, infrasound travels through the atmosphere along a distinct path. Therefore, infrasound waves can arrive at remote sensors more rapidly than tsunami waves and with less distortion than seismic waves, providing a valuable complementary data source for improving the reliability of the TEWS. Data from two infrasound stations that are part of the International Monitoring System (IMS) were used for the analysis. The F–K method was used to analyze the data acquired from two IMS infrasound stations. The study also showed the contribution of infrasound wave propagation through the atmosphere, which highly affects infrasound detection.

Data and methods

This study analyzed 32 earthquakes with W-phase moment magnitudes (M\(_{ww}\)) of 5.5 or greater, occurring between 2012 and 2023 within the region spanning 30\(^{\circ }\)N-38\(^{\circ }\)N latitude and 21\(^{\circ }\)E-36\(^{\circ }\)E longitude. During the study period, the infrasound data for the two IMS stations are available. Earthquake parameters were obtained from the United States Geological Survey (USGS) catalog and are listed in Table S1 in the Supplementary Data. Seismic data were automatically processed using the SeisComP3 (SC3) system to calculate magnitude values. Figure 2a shows the global seismic stations configured within SeisComP3 for automatic source parameters determination. In the figure, the study area is shaded in red, seismic stations are indicated by triangles, and yellow stars indicate the infrasound stations in this study. Figure 2b further illustrates the spatial distribution of earthquakes and station locations across the Mediterranean region. Typically, earthquakes with magnitudes below 7.0 do not generate strong tsunamis. However, several tsunamis in the historical tsunami dataset were generated by earthquakes with magnitudes below this threshold30. Although the earthquakes included in this dataset may not have caused strong tsunami events, accurately validating magnitude values and their computational times is beneficial as a first step to prepare for the possible future catastrophic events. Additionally, tsunami alerts are categorized to convey different threat levels, including information messages for earthquakes that are unlikely to generate a tsunami. Rapid earthquake magnitude determination has been investigated to test the TEWSs and validate the configured seismic stations, where the magnitude computation time depends primarily on the stations distribution.

Seismic magnitude estimation

SeisComP3 (SC3) developed by the Helmholtz Centre Potsdam, German Research Centre for Geosciences, in collaboration with gempa GmbH, is an automated tool for the detection of earthquake hypocenters and magnitudes31. The magnitudes analyzed automatically by SC3 include the body-wave magnitude (mB)32,33, short-period body-wave magnitude (mb)34, and broadband moment magnitude (Mwp)35,36. SC3 can rapidly estimate Mwp based on first-arrival P waves. M\(_w\)(Mwp) is an estimate of the moment magnitude based on Mwp, through a default conversion equation (1) implemented in SC337. Similarly, M\(_w\)(mB) is an estimate of the moment magnitude based on mB through a default conversion equation (2) implemented in SC337.

Figure 2
figure 2

(a) Distribution of seismic stations and infrasound stations used in this study. (b) The study area with the mechanism of significant earthquakes and available stations in the Mediterranean Sea region, the earthquakes, and the source parameters are computed from the USGS38. The (a) and (b) maps are plotted using PyGMT python library version v0.13.029.

Full size image
$$\begin{aligned} M_{w}(Mwp)= & 1.31 Mwp - 1.91 \end{aligned}$$
(1)
$$\begin{aligned} M_{w}(mB)= & 1.30 mB - 2.18 \end{aligned}$$
(2)

In this study, SC3 was used to estimate different magnitude types and determine their computational times and accuracy in detecting the most suitable magnitude type for our configuration. The efficiency of an early warning system depends on both time and accuracy. The stations distribution and significant configuration are essential for source parameter computation. SC3 playback was used to examine station distribution and source parameter determination for each event. Finally, the determined source parameters were subjected to the decision matrix (DM), as outlined in Table 2 to determine the first alert. The travel time in case of a tsunami due to these earthquakes was calculated using the Tsunami Travel Time (TTT) software package to validate the computational time of magnitude. The TTT software was produced by the International Tsunami Information Center, based on bathymetry data downloaded from GEBCO\(\_\)2023 grid39.

Decision matrix for the TEWS

The TEWS was developed under the umbrella of the United Nations Educational, Scientific, and Cultural Organization (UNESCO). The North-East Atlantic Mediterranean and Connected Seas Tsunami Warning System (NEAMTWS) is an intergovernmental coordination group (ICG) that covers the Mediterranean region. The Kandilli Observatory of the Institute of Earthquake Research Institute (KOERI) in Turkey, Centre d’Alerte aux Tsunamis (CENALT) in France, the National Observatory of Athens (NOA) in Greece, the Instituto Nazionale di Geofisica e Volcanologia (INGV) in Italy, and the Instituto Portugues do Mar e da Atmosfera (IPMA) in Portugal are the operational tsunami warning centers for NEAMTWS40,41. These five national monitoring centers act as tsunami service providers (TSPs), providing tsunami alerts for the Mediterranean Sea region to the forecast points. However, other Mediterranean countries can only receive tsunami alerts and warning information by subscription. Egypt is among those nations that subscribe to receive tsunami messages from the TSPs42. This study aims to start the first steps toward constructing a local system for Egypt. DM criteria generally depend on the region and share the same main underlying structure, with minor variations. Table 2 shows the decision matrix for the first alert, which mainly depends on the seismic source parameters, as seismic waves can propagate more than 40 times faster than tsunami waves, which propagate with a velocity (\(\sim\) 200 m/s in the Mediterranean Sea) of \(\sqrt{gh}\) depending on the water column thickness h in meters and g is the gravitational acceleration (\(\sim\) 9.81 \(m/s^2\))43. This first alert message is created after automatic magnitude determination and contains three types of warning messages, outlined in Table 3. The message types include the information alert level (no tsunami will reach the coast), tsunami advisory level (slight increase in water level on the coast and minor damage), and watch level, which can signal that the public must evacuate (a tsunami will hit the coast violently, causing coastal inundation)41,44.

Table 2 The Kandilli Observatory of the Institute of Earthquake Research Institut (KOERI) Decision Matrix that applied to this study41.
Full size table
Table 3 Tsunami alert levels and their effects on the nearby forecast points41.
Full size table

Infrasound coupling mechanism from earthquakes and tsunamis

The infrasound wave is defined as an acoustic wave below the range of human hearing that propagates in the atmospheric layers with a frequency range of less than 20 Hz45. A variety of natural and anthropogenic sources generate infrasound waves46,47. Infrasound from chemical and nuclear explosions has been studied as an artificial source48,49,50. Additionally, supersonic aircraft, rocket launches51, and wind turbines52 generate waves recorded by infrasound sensors. Natural sources such as volcanic eruptions produce strong infrasound waves that can be detected thousands of kilometers away53,54,55. Interactions between wind and ocean waves generate infrasound known as microbaroms56. Meteors, fireballs, and artificial reentries generate infrasound upon hypersonic entry into the Earth’s atmosphere57,58. Infrasound detection contributes to the detection of flash flood hazards59.

Lamb waves (atmospheric boundary waves) are generated from tsunami events due to the coupling mechanism between the ocean surface and the atmosphere60,61. Lamb waves, which are recorded with infrasound sensors, were excited by the tsunami generation related to large earthquakes, as in the case studies of the 2011 Great Tohoku-Oki earthquake62 and the Sumatra tsunami63.

Figure 3
figure 3

Diagram shows the different mechanisms of infrasound wave generation from earthquakes and tsunamis.

Full size image

Many researchers have also discussed the significance of infrasound technology in recording large earthquakes64,65,66,67,68. Earthquakes are impulsive sources that also generate acoustic waves in the atmosphere through different mechanisms63. Figure 3 shows a diagram of the mechanisms underlying infrasound wave generation from earthquakes and tsunamis, which can be classified as follows:

  1. I:

    Sudden vertical displacement near the infrasound station causing pressure fluctuations (local infrasound).

  2. II:

    Seismic waves and transferred energy coupling at the interfaces between solids and air, producing acoustic waves (primary infrasound).

  3. III:

    Seismic surface waves passing through mountainous areas generate acoustic waves (secondary infrasound).

In this study, the coupling of acoustic waves due to earthquakes as a primary source was analyzed. Infrasound waves propagate faster than tsunami waves over long distances without significant attenuation69. Therefore, infrasound technology can play a vital role as an additional data source integrated with the seismic tool. Understanding infrasound wave propagation, which is the most effective parameter for detecting infrasound waves, contributes to the incorporation of infrasound into TEWS, representing a critical advancement in the recording systems for early warning of the seismic source and tsunami.

Many researchers have discussed the effects of earthquake source parameters, such as the magnitude and source mechanism, on recorded infrasound waves66,70,71,72. Earthquakes with magnitudes greater than 5.5 can generate atmospheric infrasound waves near the epicenter and in surrounding regions67. Although infrasound wave propagation is among the most significant parameters influencing infrasound wave recording, propagation models for case studies of earthquakes in the Mediterranean Sea contribute to our understanding of infrasound observations under different propagation conditions.

F–K array analysis

The International Monitoring System (IMS) is a global network for the Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO). It contains 53 established infrasound stations among 60 proposed stations73,74, which are designed to detect low-frequency infrasound waves from events such as nuclear tests. In this study, acoustic waves generated from large earthquakes were examined by two infrasound stations in an array surrounding the Mediterranean region: I48TN, located in Tunisia, and I26DE, in Germany (yellow stars, Figure 2). Numerous seismic signal processing analyses have utilized the Frequency–Wavenumber (F–K) method75. Due to the similarities between seismic and infrasound arrays, this approach has become a highly effective method for analyzing waveforms and detecting signal coherence, signal slowness, and azimuth from digitally recorded arrays76,77.

The F–K method supposes that the infrasound wave can be modeled as a plane wave propagating toward an infrasound station in an array with N elements, whose locations are represented by vectors r\(_{i}\) as (i = 1, 2, ..., N). The Fourier transform in Equation 3 is applied to the original signal to obtain the frequency wavenumber spectrum.

$$\begin{aligned} F(\omega )= & \int _{-\infty }^{\infty }f(t)\,e^{-i \omega t} dt \end{aligned}$$
(3)

According to the sampling theorem, the distance between elements (d) in the array must be less than half the wavelength of the signal. Depending on the relative positions of the sensors, the wave arrives at sensors i and j with a time delay (\(\Delta\)t\(_{ij}\)), which can be represented by Equation 4. The F–K spectrum is constructed by beam-forming the array data for different slowness and azimuth values with different possible wavenumbers77,78.

$$\begin{aligned} \Delta t_{ij}= & S \, \{(x_i -x_j)\, \sin {\theta } \, + (y_i -y_j)\, \cos {\theta }\} \end{aligned}$$
(4)

Slowness (S) can be estimated from the relationship between the wavenumber (k), the apparent velocity (v), and the angular frequency (\(\omega\)). Both S and the back-azimuth (\(\theta\)) can be calculated using Equations 5 and 6.

$$\begin{aligned} |K|= & (K_x^2 + K_y^2)^{1/2} = \frac{\omega }{v} \end{aligned}$$
(5)
$$\begin{aligned} |S|= & \frac{1}{v} \end{aligned}$$
(6)

Slowness is the reciprocal of the apparent velocity for the wavefront across the array, and the back-azimuth angle (\(\theta\)) in Equation 7 represents the angle between the direction of the epicenter and north when the wavefront arrives at the array.

$$\begin{aligned} \theta= & \tan ^{-1}({K_x /K_y}) \end{aligned}$$
(7)

Results

Seismic magnitude determination

TEWSs are mainly based on seismic analysis to issue the first alert within a few minutes from the origin time of an earthquake. Therefore, a robust computation of the source parameters is a challenge. The magnitude was automatically calculated using the configured network in SC3. The magnitude values showed the most change specifically after 6 and 11 minutes for the studied earthquakes. Additionally, 20 minutes were used as a final estimate for the magnitudes values. Magnitude accuracy was verified at different computation times, at 6, 11, and 20 minutes after the origin time, to determine which magnitude and time were appropriate for a TEWS for Egypt. The decision matrix was then applied to the magnitude values obtained automatically by the SC3 equations 1 and 2 to determine the alert messages (Information, Advisory, and Watch) in 6, 11, and 20 minutes. This study mainly concentrates on the magnitude computation with our system, considering that the USGS magnitude value is correct. However, the depth and location values are obtained from the USGS. As the maximum magnitude of the earthquakes in the dataset used in this study was insufficient to produce a tsunami that crossed the Egyptian coastlines, information and advisory messages were the results of these earthquakes. Furthermore, for the 32 earthquakes included in the analysis, a watch message was not issued toward the Egyptian coasts, reflecting the absence of a tsunami threat from these earthquakes. The USGS magnitude value used as a reference for the accurate alert. False alerts include both overestimates and underestimates alerts. Overestimated alerts are for cases in which the system produces a higher alert level than the USGS magnitude alert. Underestimated for cases in which the system alert is lower than the USGS magnitude alert. Figure 4a shows the accurate and false alerts for \(M_w\)(Mwp) obtained automatically by SC3 by Equation 1. Figure 4b shows the accurate and false alerts for \(M_w\)(mB) obtained automatically by SC3 by Equation 2.

Figure 4 shows that M\(_w\)(Mwp) produced a more accurate alert than M\(_w\)(mB). Using the M\(_w\)(Mwp) values, five false alerts were expected within 6 minutes after the earthquake origin time and four false alerts within 11 minutes, compared to M\(_w\)(mB), which expected 10 false alerts within 6 minutes and the same within 11 minutes. So, the M\(_w\)(Mwp) is suggested for establishing the TEWS in Egypt. Comparison between magnitude values within 6 and 11 minutes after origin time is essential to assess the time required to determine the magnitude of the warning for Egypt. The results show a slight difference in magnitude accuracy between 6 and 11 minutes, as the five false alerts in 6 minutes were reduced to four false alerts within 11 minutes (Figure 4a). Therefore, M\(_w\)(Mwp) magnitude type estimates obtained at 11 minutes were more effective for magnitude accuracy.

Figure 4
figure 4

Panels (a) and (b) show the numbers of accurate and false alerts within 6, 11, and 20 min for M\(_w\)(Mwp) and M\(_w\)(mB), which are calculated by Equations 1, 2, respectively.

Full size image

Figure S1 (a) and (b) in the supplementary data show the least-squares fit between M\(_{ww}\), as calculated by the USGS, with Mwp and mB calculated automatically by SC3, respectively. The two equations S1 and S2 shown in the supplementary data were obtained and considered as correction equations for the study area, which are suggested for inclusion in future SC3 to reduce the false alerts of magnitude. These correction relations were suggested instead of Equations 1 and 2, which were empirically configured by the default SC3. Figure S1 (c) shows the accurate and false alerts for \(M_w\)(Mwp) obtained using correction Equation in S1. Figure S1 (d) shows the accurate and false alerts for \(M_w\)(mB) obtained after using correction Equation S2. The correction equations obtained in this study reduced false alerts for M\(_w\)(mB). However, the accuracy of the M\(_w\)(Mwp) alert obtained using SeieComp3 was the same as that obtained using the correction equation developed in this study.

The travel times of the tsunami waves due to these earthquakes were calculated using the TTT software package to validate the magnitude accuracy through time of the earthquakes. Figure S2 in the supplementary data was calculated to show the travel time for the sea waves due to the 32 earthquakes in the case of a tsunami. The tsunami waves took more than 65 minutes to reach the supposed forecast point (Alexandria). Although the TTT for a tsunami occurrence caused by this earthquake might have been shorter if a finite fault size were considered, the findings suggest M\(_w\)(Mwp) at 11 minutes offers the best tradeoff between timelines and accuracy to detect magnitude and warn the public of the impending tsunami disaster.

Infrasound observations

Infrasound could be recorded over long distances from the earthquake’s epicenters. Detecting an infrasound signal over this global range highlights the importance of infrasound as additional technology for hazard warning. The F–K analysis method is used as a data analysis method to detect acoustic signals from 12 earthquakes with a magnitude of \(\ge\) 6 in the Mediterranean. Although the two infrasound stations (I48TN in Tunisia and I26DE in Germany) are considered far from the epicenters (\(\ge\) 1500 km), most of the earthquakes were recorded by at least one of the two stations (8 out of the 12 earthquakes). Station I48TN detected three earthquakes, and three were detected by station I26DE; in addition, two further events were recorded simultaneously by both stations. The earthquakes detected by the I48TN occurred in June and July, as the stratospheric jet flowed toward the west in summer and spring79. Figure 5 presents the results of the data analysis for the acoustic waves recorded at both stations due to an earthquake of magnitude M\(_{ww}\) = 6.6 that occurred on July 20, 2017. Data filtered between 0.2 and 1.5 Hz showed high coherence for pixels at an azimuth of 80\(^{\circ }\) for the I48TN station. The coherence of pixels for the I26DE station at an azimuth of 134\(^{\circ }\). Figure 6 shows the back projection of the pixel of an infrasonic detection toward a possible source, assuming a stratospheric phase speed of \(\approx\) 305 m/s. The figure also shows the back-azimuth projections from each station to the event. The cross-bearing between the two stations shows the power of infrasound technology to monitor such hazards over a large distance. However, non-recording events should be discussed, especially when the largest earthquake was recorded at neither of these stations. A non-recorded earthquake was compared to a recorded earthquake to present the effect of infrasound propagation. Table 4 shows a comparison between the parameters for both recorded and non-recorded earthquakes. The table shows that the two cases (the recorded case and the non-recorded case) occurred approximately at the same epicenter. Both earthquakes were shallow, with a slight difference in depth. Moreover, the two earthquakes have the same focal mechanisms. Generally, the most common parameters are almost the same in both studied cases, except for the date and season of occurrence. The recorded case occurred in July, which was in the summer season, and the non-recorded case was in late October, which was in the autumn. Hence, the propagation models for the two earthquakes were studied to enhance the effect of propagation on infrasound recording.

Figure 5
figure 5

F–K analysis of the recorded infrasound signals from the earthquake at magnitude M\(_{ww}\) = 6.6 on 20 July 2017 with I48TN and I26DE, respectively.

Full size image
Figure 6
figure 6

Pixels back-projection and azimuth projection from the infrasound stations back to the event for the case of the 20 July 2017 earthquake (map is plotted using PyGMT v0.13.0)29.

Full size image
Table 4 Comparison between the recorded and non-recorded earthquake cases.
Full size table

Infrasound propagation modeling

Parabolic equation models were applied using the NCPAprop package80 to analyze the infrasound propagation probability of both recorded and non-recorded events. The models assumed that the sources were on the ground; a source frequency of 0.5 Hz, because the lowest mean frequency in detections from all stations was 0.596 Hz; and a range of propagation of 2000 km. The atmospheric parameters (temperature, zonal wind speed, meridional wind speed, and air pressure with respect to altitude up to 80 km) used for infrasound propagation models were extracted from the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5)81. The thermodynamic sound speed \(c_{th}\) was driven by the following Equation:

$$\begin{aligned} c_{th} = \sqrt{\gamma R T} \; \; , \end{aligned}$$
(8)

where \(\gamma\) is the ratio of specific heat (= 1.4), R is the specific gas constant for dry air (= 286.9 J \(kg^{-1}\) \(K^{-1}\)), and T is the temperature. The parabolic equation solution is dependent on the effective sound speed (\(c_{eff}\)), which can be estimated as follows:

$$\begin{aligned} c_{eff}(z) = c_{th}(z)+ u(z) \sin {\phi } + v(z)\cos {\phi } \;\;, \end{aligned}$$
(9)

where u is the zonal wind speed, v is the meridional wind speed, and \(\phi\) is the azimuth direction of propagation.

For the July 20, 2017, earthquake, the resulting transmission losses were lower than -60 dB from both azimuths (271\(^{\circ }\) and 324\(^{\circ }\)) toward I48TN and I26DE, respectively (Figure 7). The atmospheric duct occurred between 40 km and 55 km in altitude (within the stratosphere), confirming the arrival of the infrasound signal from the earthquake. The parabolic equation models were estimated in all directions with an azimuth step of 2\(^{\circ }\), between 0\(^{\circ }\) and 360\(^{\circ }\) from an earthquake that occurred on October 30, 2020 (Figure 8). Multi-propagation models were applied to explain a lack of signals recorded at I48TN and I26DE. Transmission losses in the directions of both stations exceeded -140 dB, reflecting the difficulty of remote signal detection at the time of the event at these stations.

Figure 7
figure 7

A and C panels show the estimation of the effective sound speed in the direction of both stations I48TN and I26DE, respectively, as the pink areas in both panels indicate the possible duct altitudes from both directions; B and D panels show the 2D transmission losses being modeled for the 20 July 2017 event.

Full size image
Figure 8
figure 8

The left panel refers to the multi-propagation PE model for the 30 October 2020 event; the middle panel shows the horizontal wind speed profile at the time of the event; and the right panels reflect the 2D transmission losses toward I48TN and I26DE, respectively (plotted using the Matplotlib python library82).

Full size image

Discussion

This study analyzed tsunami alerts based on a USGS catalog containing 32 earthquakes with magnitude (M\(_{ww}\ge\) 5.5) in the Mediterranean Sea. However, significant tsunamis are typically triggered by stronger earthquakes, the suitable magnitude and its calculation time are important to be tested through the decision matrix. Tsunami alerts do not definitively indicate that a tsunami will impact the coastline, as the alert messages in this study are mostly information messages that indicate that there is no tsunami threat due to these earthquakes.

Different magnitude types calculated by SC3 were compared with the M\(_{ww}\) data calculated by USGS. Different computation times were tested to evaluate their accuracy through different computational times. The DM was applied to these magnitudes to assess the accuracy of the messages estimated in 6, 11, and 20 minutes. The \(M_w\)(Mwp) obtained automatically by SC3 by equation 1 shows higher accuracy than \(M_w\)(mB) obtained by equation 2. The magnitude accuracy for \(M_w\)(Mwp) is not very significant when comparing 6 minutes, 11 minutes, and 20 minutes. The false alerts were five for 6 minutes, four for 11 minutes, and three for 20 minutes. We concluded that 11 minutes was suitable for balancing TEWS magnitude accuracy and computation time.

The TTT was calculated for Mediterranean Sea earthquakes toward Egyptian forecast points to validate the 11-minute computational time efficiency. The results in Figure S2 show that the wave may take longer than 65 minutes for earthquakes in the Cyprian and Hellenic arcs to hit the forecast point (Alexandria). The TTT was calculated for a small event (M\(_{ww}\)=4.9) near Egypt (green star, figure S2 in supplementary data). Tsunami waves triggered by this earthquake would take 55 minutes to reach Alexandria. Furthermore, the TTT could be slightly shortened if the fault parameters are considered.

Time consumption and magnitude accuracy depend highly on the distribution of configured stations in the Mediterranean region. However, considering the challenges related to limited network and data availability in this region, issuing the first alert within 11 minutes represents an efficient approach for saving lives under the current configuration, regardless of earthquake magnitude. The issuance of the alert within approximately 11 minutes represents a significant advancement towards the objectives of the Ocean Decade Tsunami Programme (ODTP 2023–2030), which was initiated by the Intergovernmental Oceanographic Commission of UNESCO and aims to provide tsunami confirmation in 10 minutes by 203083. This study relied mainly on global seismic stations. Including local data could reduce computational time and increase the magnitude accuracy, which should be considered for future system upgrades.

Although the arrival of infrasound waves is slower than that of seismic waves, it is still faster than that of tsunami waves. As ground positioning systems and sea-level stations play vital roles in early warning systems, integrating infrasound technology enriches data fusion, which is beneficial in locating the epicenter of earthquakes over large distances without considerable attenuation compared to seismic waves. Infrasound observations were analyzed for 12 earthquakes with magnitude (M\(_{ww}\ge\) 6). Most of these earthquakes (8 earthquakes) showed the arrival of an infrasound signal from a distance of more than 1500 km. Two cases had been studied by infrasound technology. For the two case studies (recorded and non-recorded earthquakes), the source parameters are slightly similar to each other except for the occurrence date and season. So, the infrasound wave propagation is the most effective parameter contributing to the detectability of these cases. The propagation model was produced for a former case study earthquake that occurred on 20 July 2017. Transmission losses were lower than -60 dB from both azimuths (271\(^{\circ }\) and 324\(^{\circ }\)) toward I48TN and I26DE, respectively. The atmospheric duct occurred between 40 and 55 km within the stratosphere, confirming infrasound wave arrival at both stations. The latter large earthquake (October 30, 2020; M\(_{ww}\)=7) was not recorded by both stations. Multi-propagation models were applied to determine the reasons for the lack of recording at these stations. Transmission losses in the direction of both stations exceeded -140 dB, which reflects the difficulty of detecting signals at the time of this earthquake.

In summary, TEWS alerts can be established with high accuracy depending on the M\(_w\)(Mwp) within 11 minutes, as this computational time was found to allow the public to be warned before the arrival of a tsunami on the Egyptian coasts. The study also highlighted the importance of infrasound technology observations as a data source for large earthquakes, which contribute to the location of epicenters over large distances. Infrasound detections are affected by atmospheric propagation. The 11-minute estimation obtained using M\(_w\)(Mwp) produced an acceptable accuracy level of 87.5%. These results represent an important initial step in TEWS establishment compared to the previous study, which evaluated the NEAMTWS Decision Matrix for Italian events, showing 45–55\(\%\) correct alerts84. Moreover, the W-phase technique is expected to provide significant improvement to the TEWS in the foreseeable future85,86. In the studied region, there are several challenges in implementing a TEWS due to the lack of seismic and infrasound stations. Therefore, we recommend the establishment of seismic and infrasound stations along the Egyptian coastlines. A single infrasound array in Egypt is susceptible to high ambient noise, which must be studied before establishing infrasound sensors on the coastlines50.

Conclusions

This study focused on 32 large earthquakes in the eastern Mediterranean Sea with a magnitude of M\(_{ww}\ge\) 5.5 during 2012-2023. Different magnitudes had been automatically calculated by SC3 software at 6, 11, and 20 minutes. The first tsunami alert messages were estimated after the rapid determination of the earthquake magnitude. The M\(_w\)(Mwp) was found to be the most suitable magnitude type for establishing the TEWS for the first step, as it produced highly accurate alert messages at different times. The M\(_w\)(Mwp) results after 6 minutes showed an 84.4% alert accuracy rate, compared with 11 minutes, which had an alert accuracy rate of 87.5%. While results after 20 minutes show an alert accuracy rate of more than 90%, M\(_w\)(Mwp) magnitude value within 11 minutes was selected as the most suitable time to balance between accuracy and computational time.

The incorporation of infrasound data was also found to be useful in detecting large earthquakes for the second step of TEWS. IMS infrasound station I48TN recorded three earthquakes, I26DE recorded three earthquakes, and another two earthquakes were recorded in both stations during the period. The largest earthquake did not show apparent detection of infrasound waves at both stations. Infrasound propagation models were estimated for a signal recorded on July 20, 2017, for which transmission losses were lower than -60 dB. A stratospheric duct occurred between 40 and 55 km, confirming infrasound wave arrival. Multi-propagation models were also applied to the non-recorded earthquake on October 30, 2020. Transmission losses in the directions of both stations exceeded -140 dB, which explained the absence of acoustic signals at both stations. As the two cases studied by infrasound have similar fault parameters, the atmospheric propagation played a vital role in affecting the arrival of infrasound waves.

In conclusion, the findings suggest M\(_w\)(Mwp) at 11 minutes offers the best tradeoff between timelines and precision to detect magnitude to warn the public of the imminent tsunami disaster, as the travel time for a tsunami from Hellenic and Cyprian arc earthquakes to reach the Egyptian coast exceeded 65 minutes. Our findings also suggested the importance of using propagation models to incorporate infrasound data in determining the detectability of infrasound waves, especially when establishing infrasound stations throughout the Egyptian coasts.