The Mw 7.5 Tadine (Maré, Loyalty Is.) earthquake and related tsunami of December 5, 2018: implications for tsunami hazard assessment in New Caledonia

On the 5 of December 2018, a magnitude Mw 7.5 earthquake occurred southeast of Maré, an island of 15 the Loyalty Archipelago, New Caledonia. This earthquake is located at the junction between the plunging Loyalty ridge and the southernmost Vanuatu arc, in a tectonically very active area regularly subjected to strong seismic crises and events higher than magnitude 7 and up to 8. Widely felt in New Caledonia it has been immediately followed by a tsunami warning, confirmed shortly after by a first wave arrival at the Loyalty Islands tide gauges (Maré and Lifou), then along the east coast of Grande Terre of New Caledonia and in several islands 20 of the Vanuatu Archipelago. Seafloor initial deformation linked to tsunami generation has been modeled with MOST numerical code using earthquake parameters available from seismic observatories. Then the wave propagation has been modeled using SCHISM, another modelling code solving the shallow water equations on an unstructured grid based on a new regional DEM of ~180 m resolution and allowing refinement in many critical areas. Finally, the results have been compared to tide gauge records, field observations and testimonials 25 from 2018. The arrival times, wave amplitude and polarities present good similarities, especially in far-field locations (Hienghène, Port-Vila and Poindimié). Maximum wave heights and energy maps for two different scenarios highlight the fact that the orientation of the source (strike of the rupture) played an important role, focusing the maximum energy path of the tsunami south of Grande-Terre and the Isle of Pines. However, both scenarios indicate similar propagation toward Aneityum, Vanuatu southernmost island, the bathymetry acting 30 like a waveguide. This study has a significant implication in tsunami hazard mitigation in New Caledonia as it helps to validate the modelling code and process used to prepare a scenarios database for warning and coastal evacuation. https://doi.org/10.5194/nhess-2021-58 Preprint. Discussion started: 18 March 2021 c © Author(s) 2021. CC BY 4.0 License.

It appears clearly that the successive seismic crises are quite similar and included both interplate thrust fault type earthquakes northeast of the trench and normal fault type events southwest of the trench in the plunging plate 135 ( Figure 2). The strong spatiotemporal pattern between these two types of events suggests that static stress interactions may account for triggering non-distant earthquake, normal faulting on the plunging plate triggering interplate thrust faulting or the reverse. Archipelago. Being strongly felt in New Caledonia (Loyalty Islands and the Grande Terre) as far as Nouméa, more than 300 km west from the source (Roger et al., 2019a(Roger et al., , 2019b(Roger et al., , 2019c, it has been also weakly felt in Port-

Tsunami
This earthquake is added to the two local earthquakes reported by the past in the south Vanuatu Subduction zone that triggered major tsunamis in the Loyalty Islands in March 28, 1875and September 20, 1920(Sahal et al., 2010 with estimated magnitude of 8.  Figure 3 locates the different tide gauges that were able to record the tsunami within the southwestern Pacific Region and illustrates the recorded maximum wave height (ITIC communication from Stuart Weinstein, 2018).

Eye-witnesses' observations
In the aftermath of this event, two videos have been collected for two different locations: Yaté (Figure 4a

Tsunami modelling
Numerical models are commonly used to assess the tsunami hazard. In this section, a suite of models used to simulate bottom deformation, tsunami generation and propagation and their settings are presented, including details about the Digital Elevation Models (DEM) used in computational grid generation. Tsunami modelling 240 sensitivity to detail the rupture model is presented and finally tsunami simulation results are compared to observations.

Bathymetric grids
It is well known that tsunami's behavior is dependent upon the bathymetric features and the coastal geometries 245 (e.g., Matsuyama, 1999;Hentry et al., 2010;Yoon et al., 2014). When it approaches coastlines or seamounts, the wave shoaling leads to the rising-up of the amplitude and slows down the tsunami as the water depth reduces. It is even worse when the tsunami enters harbors, bays, lagoons or fjords able to produce resonance, a phenomenon This technique allows more flexibility in mesh design and can capture more coastline details than regular meshes at the same computational cost.
In this study, bathymetric grids have been built using: 1) Smith and Sandwell (1997)

Earthquake parameters
Most of tsunami modelling codes are using Okada (1985)  where Mn is Manning's roughness coefficient set spatially uniform with a value of 0.025 s.m -1/3 . All tsunami simulations were performed assuming that prevailing tide was static (no flow) and equal to high water (+1.6m).
To limit undesirable wave reflection, a Flather radiation condition (Flather, 1987)  To detect changes due to fault parameters, total wave energy (E, unit j.m-2) is added in SCHISM outputs, as the 345 sum of two components, kinetic energy (first term) and gravitational potential energy (second term): It is again important to underline that the sea-level has been set to a high tide value of 1.6 m, which corresponds to the situation when the tsunami reached New Caledonia on December 5, 2018. Figure 6 presents the maximum wave energy map obtained after 3 hours of tsunami propagation over the TIN DEM. It highlights the important role played by the strike angle of the fault plane. This parameter should absolutely be chosen accurately in good agreement with the geology. A 298° (USGS) and a 312° (GCMT) strike will lead to a different behavior of the tsunami, focusing its main energy path generally perpendicularly to the strike of the fault plane with respect to the slip angle (=rake) (Okal, 1988). But if the waves encounter submarine 355 features like seamounts or ridges, the trajectory of the tsunami could be dramatically modified as these features act as wave guides, focusing the wave train in another direction due to the fact that the tsunami speed relies only on the bathymetric depth in the open ocean (Satake, 1988;Titov et al., 2005;Swapna and Srivastava, 2014).

395
At Tadine, Maré, the modelling is not able to reproduce correctly the tide gauge record in terms of arrival time and wave amplitude (Figure 8a). It shows a delay of ~5 min, the modelling being faster than the reality. Also, it does not reproduce the oscillation of period ~4-5 min with amplitudes more than three times those that are modeled.
At Wé (Lifou), the simulated signal exhibits some strong similarities with the real one recorded in terms of 400 polarity, wave amplitude and periodicity, but there is a delay of more than 5 minutes, the modelling being faster than the reality (Figure 8b).
At Thio, the modelling is able to reproduce the real record for what concerns the polarity, the amplitude or the periodicity but not exactly the arrival time, being still early of a couple of minutes (Figure 8c). At Ouinné, the modelling is not able to reproduce the recorded signal, except for the first wave polarity, showing 405 a strong delay of nearly 5 min, the modelling being the fastest (Figure 8d). An oscillation with a period of ~6-8 min seems to occur after the first arrival.
At Poindimié -Passe de la Fourmi, there is a good agreement between the modelling and the reality: the arrival time only exhibits a small delay of 1-2 min, the modelled signal being the fastest (Figure 8e). The wave amplitude and polarity are quite good, and the periodicity shows only a few differences that will be discussed 410 further.
At Hienghène, there are differences in arrival time (~2-3 min) between the modelled and the real tide gauge records, the modelled one being the fastest (Figure 8f). The wave polarity and periodicity are well reproduced but the amplitude is slightly overestimated by the modelling.

Discussion
The comparison of the maximum energy path of the tsunami as a function of strike on the energy maps shown on Figure 6 highlights the fact that a 312° angle has a slightly bigger impact on the Isle of Pines matching much better with the observations than a 298° angle. The maximum wave height map calculated over a high-resolution 430 TIN grid (Figure 7) , 2008;Rabinovich, 2009;Aranguiz, 2015). The fact that the high-resolution coastal zones surrounding the location of the tide gauges have been built from sparse bathymetric data coming from low resolution nautical charts and aerial pictures interpretation could explain that the modelling is not able to reproduce the resonance as the shape of the water bodies, and thus their natural 450 oscillation modes are not exactly the same. According to previous studies, it is a safe bet that either a source refinement (complex source showing slip heterogeneity for example) or high-resolution bathymetric data coming from multibeam or LIDAR surveys would be able to reproduce such phenomenon in these small and complicated places (e.g. Vela et al., 2014).
Considering both maximum amplitude maps compared to the testimonials (locations and amplitudes) and the 455 tide gauges simulation results comparison to the real recorded data, the simple fault plane rupture scenario chosen for this study provides quite good results.
It is interesting to notice that, nearly two years after the tsunami occurred, hidden observations are still transmitted by witnesses. Tsunami modelling showing that the west coast of the Isle of Pines would have also been impacted by the tsunami, we questioned the diving center and the Kodjeu Hotel located within the Ouaméo 460 bay: the final testimony is that the diving club boat, supposed to be load at high tide, was laying on the sand instead at the exact arrival time of the tsunami (P.-E. Faivre, pers. comm., 2020). Then the water came back and the sea rose above its natural maximum (according to a local fisherman, 2019).