Tsunamigenic potential of crustal faults and subduction zones in the Mediterranean

We compiled a database and systematically evaluated tsunamigenic potential of all up-to-date known crustal fault systems and subduction zones in the entire Mediterranean region that has experienced several catastrophic tsunamis in historical times. The task is accomplished by means of numerical modeling of tsunami generation and propagation. We have systematically simulated all representative ruptures populating known crustal faults and subduction interfaces with magnitudes ranging from 6.1 up to expected Mwmax. Maximum tsunami heights calculated everywhere along the coasts allowed us to classify the sources in terms of their tsunamigenic potential and to estimate their minimum tsunamigenic magnitude. Almost every source in the Mediterranean, starting from Mw = 6.5, is capable to produce local tsunami at the advisory level (wave height >20 cm and ≤50 cm). In respect to the watch level (wave height >50 cm) larger magnitudes are needed (Mw ≥ 6.9). Faults behave more heterogeneously in the context of far field early warning. De-aggregation of the database at any selected coastal location can reveal relevant sources of tsunami hazard for this location. Our compilation blueprints methodology that, if completed with source recurrence rates and site-specific amplification factors, can be considered as a backbone for development of optimal early warning strategies by Mediterranean tsunami warning providers.


Tsunamigenic potential: a coastal perspective.
We present here the tsunamigenic potential of the source database by simple counting the number of scenarios (individual sources) producing different threshold levels at POIs. Note that while presenting the absolute numbers (counts), we still effectively provide the relative information since the total number of our scenarios (about 20000) depends on the adopted discretization of CSS and SUBD interfaces. Moreover, these counts have no direct relation to the tsunami impact probability because we do not assign likelihoods to our sources.
Nevertheless, plotting all scenarios together gives an idea about general tsunami hazard distribution in the whole region. Figure S1 presents such compilation for CSS sources only. From this compilation we can see that almost the whole Mediterranean coast is subjected to potential tsunamis at the 'advisory' (i.e., wave height exceeding 20 cm and under 50 cm; blue bars) and 'watch' (i.e., wave height exceeding 50 cm; red bars) alert levels. The largest number of occurrences is close to the Hellenic arc. The total number of events is high along north-African and Italian coasts.

Figure S2 -Number of total advisory (blue) and watch (red) tsunami occurrences due to the whole set of SUBD. Deaggregation of events counting produced by each subduction interface is also given in panels for A) Calabrian arc, B) Hellenic West and C) East arcs, D) Cyprus West and E) East arcs.
Noteworthy that gaps and peaks in bar heights characterize the East-Tunisia-Libya and South Turkey coasts, respectively. A relatively low number of 'watch' events was predicted for Spain and France. Among subduction zones ( Figure S2), events produced at the Calabrian Arc can be considered as more local ( Figure S2a). Relatively high number of 'advisory' or never exceeds 50 occurrences. Few occurrences affect the eastern Greek and Libyan coasts at the longitude of Benghazi.
The Hellenic West and East Arcs are the most effective tsunamigenic sources in the region due to their greater size and higher maximal assigned magnitude (Mw max 8.4). Considering results for the West Hellenic Arc ( Figure S2b), peaks of occurrences (> 100) are attributed to the nearby coasts of Peloponnese and Crete and, in a more regional perspective, to the Benghazi promontory in Libya. A relative large number of occurrences is also recorded in the Aegean Sea as well as along Sicily and Calabria in South Italy. Few events reach the easternmost Mediterranean coasts of Syria and Lebanon due to the barrier represented by Cyprus Island and, south of it, the Eratosthenes Seamount (see Figure 1 of the main text).
With the exception of some coasts suffering more than 75-100 events in the very near islands, the East Hellenic Arc affects the Eastern Mediterranean more evenly and with less occurrences ( Figure S2c). Only few green spots (i.e., over 75 events) can be recognized at the

Comparison with results by Necmioglu and Ozel (2015) 11 for the East Mediterranean.
Recently, Necmioglu and Ozel 11 published an extensive databank of synthetic tsunami scenarios in the East Mediterranean with earthquake sources distributed by location, depth ('shallow' with upper edge at 5 km and 'deep' with that at 40 km), magnitude and focal mechanism in accord with regional tectonics and seismicity. To compare our results with this dataset, we have post-processed their data and extracted minimum tsunamigenic magnitudes according to our definition (see main text). Figure S3 presents minimum magnitudes for 'advisory' (upper-left panel) and 'regional watch' (upper-right panel) tsunami alerts.
Minimum tsunamigenic magnitudes demonstrate the same tendency to increase from 'advisory' to 'regional-watch' alert levels. However, in comparison to our simulation results, these simulations generally show lower minimum tsunamigenic magnitudes, especially for main text. Apparently, this difference is attributed to the methodological issues of the both studies. Necmioglu and Ozel 11 scenario database is aimed for the tsunami early warning. At each gridded geographical location, the database contains several scenarios varying by magnitude and depth; the latter can be either 5, or 40 km (top edge of a fault). Another methodological difference lies in the post-processing. As described above in the main text. In order to 'label' a source as 'tsunamigenic', we counted at least 10 POIs over the corresponding threshold. This was done to remove possible outliers since our POIs (>20 000) were distributed automatically and not checked for occasional mal-positioning. Experiments with less minimal counts always resulted in lower Mw min (sometimes. maybe due to outliers).
By post-processing Necmioglu and Ozel (2015) scenario database, we did not apply this 'minimum-count' criterion: the source was marked as 'tsunamigenic' if the wave height overridden an alert level threshold at any single POI.

Assigning grid nodes to test-sites
When computing tsunami threads for the selected test-sites via de-aggregation (see main text, Figure 4), we identified each test site through a cloud of grid nodes rather than assigning a single node to it. As noted in the main text, it was done to suppress possible outliers and to obtain an averaged, smoothed and, thus, more representative, tsunami impact over the community coastline. Effective wave height for a test-site was calculated as an average wave height over the cloud of grid nodes assigned to the test-site. Since the average value is always smaller than the maximum over population, Mw min based on community-averaged wave height should be a conservative estimate for the community. Figure S4 demonstrates assignment of computational grid nodes to test-sites: selection of points generally follows geographic location and extent of corresponding communities.

Sensitivity to propagation model: placement of the reflecting wall and POIs
We simulate tsunami propagation in a linear long-wave approximation with full-reflection at the water-land boundary; see Methods section in the main text for more detailed description. Coastal points-of-interest (POIs) where we do measure wave heights are distributed along the reflecting coastal wall. Here we test sensitivity of our metricminimum tsunamigenic magnitude Mw minto the positioning of the wall. In the first case ( Figure S5, left), we set the wall along the zero-meter isobaths, i.e., directly along the coast as given by the topobathymetric grid (in our case: Gebco_08). In the second case ( Figure S5, right), for which we actually present all our results in this paper, the wall was placed along the 20-meter isobaths.
Moving the reflecting wall into deeper water is a usual approach when using linear tsunami simulations: it restricts computations to the validity domain of linear approximation and also allows avoiding small-scale coastal features like shallow bays which would require much higher spatial resolution. Comparison of the two models ( Figure S5) demonstrates no visible difference between the two propagation models.

Sensitivity to source parameters
Composite seismic sources (CSS) are described by their geographical shape and representative focal parameters including strike, dip and rake angles. For the most CSS's, these parameters are not unique but allow some variations. To check influence of the possible variations on our results, we conducted a sensitivity test based on one selected CSS structure at the North-Algerian margin (DZSC001).
This CSS hosts 10 individual source locations: 5 at a shallow level and 5 at a deeper level (see Figure S6). For each location and magnitude (6.3 -8.0), we have systematically repeated our Mw min analysis by disturbing reference rupture parameters. In particular: -Each individual source location was shifted 30 km NW and 30 km SE (see Fig. S6); -Strike, dip, and rake angles were varied by +/-10 degrees. Tables ST1 and ST2 summarize the variability of the resulting Mw min metric in respect to the given variations of source parameters. Table ST1 presents variation range of Mw min at the 'advisory' alert level, while Table ST2 refers to the 'regional watch' alert level. Minimum tsunamigenic magnitude does not vary markedly due to the variations of focal parameters: typical variation is 0.1 with maximum value of 0.2 (fifth column of the Tables). Additional incorporation of the geographical shift for 30 km has more pronounced effect: see the last column of the both Tables. Mw min variations are significantly larger for individual sources positioned close to the shoreline-compare, e.g., position 6 against 5. The reason is obvious: if source is located close to the shoreline, volume of the uplifted water drastically depends on the exact source position. This is clearly represented in Figure S7 where the relative error is shown. Regardless of the considered warning level, the variation due to shift in the source position is higher but always less than 10% moving the source to the north, i.e. toward the open Sea. Moving the source on-shore (to the south) produces variations in the obtained Mw min less than 5%.  Variations by strike, dip and rake … plus position shift 1 3.41259 36.8704 6.5 6.5 ÷ 6.5 6.5 ÷ 7.0 2 3.65318 37.0147 6.6 6.5 ÷ 6.6 6.5 ÷ 6.7 3 3.93598 37.1317 6.6 6.6 ÷ 6.6 6.5 ÷ 6.7 4 4.24246 37.2237 6.6 6.6 ÷ 6.6 6.5 ÷ 6.7 5 4.54894 37.2799 6.7 6.6 ÷ 6.7 6.6 ÷ 6.7 6 3.49432 36.7018 7 6.9 ÷ 7.1 6.6 ÷ 7.6 7 3.73491 36.8462 6.9 6.8 ÷ 6.9 6.7 ÷ 7.4 8 4.01771 36.9632 6.9 6.8 ÷ 6.9 6.8 ÷ 7.1 9 4.32419 37.0551 6.8 6.8 ÷ 6.8 6.8 ÷ 6.9 10 4.63066 37.1113 6.8 6.8 ÷ 6.8 6.8 ÷ 6.8

Sensitivity to the POI 'count threshold'
As explained in the main text, we employed a group POI evaluation to declare a particular individual source as tsunamigenic or not. A source was declared as tsunamigenic if it was able to trigger tsunami over given wave height (wave-height threshold) at more than given number of POIs (POI count threshold). Such group evaluation plays a role of filtering of possible outliers, but, on another hand, tends to increase the Mw min estimate (that is why we consider our resulting minimum tsunamigenic magnitudes as rather conservative estimates).
In order to evaluate the sensitivity of Mw min to the count threshold, we have made a detailed systematic analysis taking the above mentioned CSS DZCS001 as an example. Table ST3