Tsunami hazard in Lombok & Bali, Indonesia, due to the Flores backarc thrust

. The tsunami hazard posed by the Flores backarc thrust, which runs along the northern coast of the islands of Bali and Lombok, Indonesia, is poorly studied compared to the Sunda megathrust, situated ~250 km to the south of the islands. However, the 2018 Lombok earthquake sequence demonstrated the seismic potential of the western Flores Thrust when a fault ramp beneath the island of Lombok ruptured in two Mw 6.9 earthquakes. 15 Although the uplift in these events mostly occurred below land, the sequence still generated 1-2.5 m-high local tsunamis along the northern coast of Lombok (Wibowo et al., 2021). Historical records show that the Flores fault system in the Lombok and Bali region has generated at least six ≥Ms 6.5 tsunamigenic earthquakes since 1800 CE. Hence, it is important to assess the possible tsunami hazard represented by this fault system. Here, we focus on the submarine fault segment located between the islands of Lombok and Bali (below the Lombok Strait). We 20 assess modeled tsunami patterns generated by fault slip in six earthquake scenarios (slip of 1-5 m, representing Mw 7.2-7.9+), with a focus on impacts on the capital cities of Mataram, Lombok and Denpasar, Bali, which lie on the coasts facing the strait. We use a geologically constrained earthquake model informed by the Lombok earthquake sequence (Lythgoe et al., 2021), together with a high-resolution bathymetry dataset developed by combining direct measurements from GEBCO with sounding measurements from the official nautical charts for 25 Indonesia. Our results show that fault rupture in this region could trigger a tsunami reaching Mataram in <8 minutes and Denpasar in ~10-15 minutes, with multiple waves. For an earthquake with 3-5 m of coseismic slip, Mataram and Denpasar experience maximum wave heights of ~1.3-3.3 m and ~0.7 to 1.5 m, respectively. Furthermore, our earthquake models indicate that both cities would experience coseismic subsidence of 20-40 cm, exacerbating their exposure to both the tsunami and other coastal hazards. Overall, Mataram city is more exposed 30 than Denpasar to high tsunami waves arriving quickly from the fault source. To understand how a tsunami would affect Mataram, we model the associated inundation using the 5m slip model and show that Mataram is inundated ~55-140 m inland along the northern coast and ~230 m along the southern coast, with maximum flow depths of ~2-3 m. Our study highlights that the early tsunami arrival in Mataram, Lombok gives little time for residents to evacuate. Raising their awareness about the potential for locally generated tsunamis and the need for evacuation 35 plans is important to help them respond immediately after experiencing strong ground shaking. tsunami hazard for all of Indonesia. They used a bathymetry dataset that combined GEBCO data with measurements from Navy charts and multibeam surveys. For the western Flores 180 thrust, they set 1 m of slip on a range of 25-27°-dipping, 10x20 km sub faults (equivalent to a Mw6.4 earthquake) within the 3-30 km seismogenic depth of the Lombok Strait. Their results do not describe the regional hazard (e.g. wave heights, timing, inundation), but rather assess how much of the local hazard is contributed by this fault system rather than the megathrust. They showed that given an arbitrary 500-year return period on the fault, 10-30% of the tsunami hazard in Mataram is due to the shallow part of the Flores Thrust. horizontal of a m-high wave that arrives at the of Mataram ~18-20 on the posed by a Mw 7.4 earthquake on the Flores thrust to the northern coasts of Lombok Bali. set 2.7 of slip on a 27°-dipping fault plane with of 75 km x 27 km at 27 km depth. The fault parameters are on mean of the Lombok, low resolution near the coasts accurately model tsunami propagation and wave heights We improve the resolution of our bathymetry by digitizing sounding data the official nautical charts of Indonesia, which are densest in the coastal regions near the cities of Denpasar (Bali) and Mataram (Lombok) and therefore critical for modeling near-shore wave heights in these regions We also 300 trace the coastline using the National Digital Elevation Model (DEMNAS, and cross check it using satellite images from Esri World Imagery (https://www.arcgis.com/). depth measurements from both and the coastlines into single dataset, and then interpolate the data to Raster’ tool ArcGIS. This tool is on the and generates a continuous digital elevation model based on point data that takes into account the hydrological correctness of the resulting raster. While this method was developed on the basis of subaerial water flow, it has also been used to effectively generate bathymetries for tsunami studies in other regions (Fraser et al., 2014; Darmawan et al., 2020; Wilson and Power, 2020). We note that the shallow shelf regions of the Lombok Strait were likely incised subaerially during the late Holocene sea-level drop 310 (Boekschoten et al., 2000), and their morphologies therefore likely reflect subaerial water flow processes. the DSM is unavailable along the coast, due to difficulties in data processing associated with tides. We fill these areas with 1.5-m resampled elevation data from DEMNAS, the national elevation model for Indonesia, which has a coarser original horizontal resolution of 8 m. The vertical datum of the merged data is referenced to EGM2008. depths, to incorporate the effect of bottom friction. A Manning’s roughness coefficient of 0.013 is used for the water region, and 0.03 on land (Wang & Power, 2011). We run the tsunami simulation from the time of the earthquake for one hour; this is sufficient to capture both the 370 first wave and a series of smaller, later waves, since the coastal regions we are interested in are close to the source (<100 km). To observe the tsunami arrival pattern along the coasts of Mataram and Denpasar within the hour, we select virtual tide gauge locations along the 10-m bathymetric contour, facing the coastal areas where dense man-made structures are identified from satellite images. The results of the tsunami modeling are illustrated using maps of the initial sea surface deformation, maximum wave height, coseismic land subsidence in Bali and Lombok, 375 time series of wave arrivals at the virtual tide gauges, and maps of inundation depth in Mataram.


Introduction
Most tsunami studies focus on earthquakes sourced by subduction zones, as they have high potential to generate destructive tsunamis (e.g., Mw 9.1 2004 Sumatra and Mw 9.0 2011 Tohoku earthquakes). Fewer studies focus on 40 tsunamis generated by back-arc thrust faults within the upper plate that accommodate a component of plate convergence (Silver and Reed, 1988). The Mw 7.7 1991 Limon, Costa Rica (Suárez et al., 1995), Mw 7.9 1992 Flores Island, Indonesia, and Mw 7.5 1999 Ambrym Island of Vanuatu (Regnier et al., 2003) earthquakes demonstrated that back-arc thrusts can generate earthquakes and tsunamis resulting in fatalities and severe damage and destruction to structures. Hence, it is important to study the tsunami hazard associated with back-arc thrusting. 45 Here, we assess the tsunami hazard associated with the westernmost segment of the Flores Thrust, a back-arc thrust that extends for >1,500 km, accommodating a portion of the convergence between the Indo-Australian and Sunda Plates (Fig. 1a). Unlike its eastern segment, where the 1992 Mw 7.9 Flores Island earthquake occurred, the western part of the fault has not hosted devastating tsunamis in recent years, although historical records and 50 previous studies show that it has generated at least eight tsunamigenic earthquakes (Fig.1b, NOAA database, Hamzah et al., 2000;Rastogi and Jaiswal, 2006;Musson, 2012;Nguyen et al., 2015, Tsimopoulou et al., 2020 The recent 2018 Lombok earthquake-triggered tsunamis were relatively minor because the earthquakes mostly occurred beneath the island itself and not offshore; nevertheless, the occurrence of the 2018 Lombok earthquakes gives new insights into the activity and geometry of this fault segment, and highlights the risk of earthquakes and 55 associated tsunamis along strike.
Our study focuses on the tsunami hazard caused by slip on the Flores Thrust in the Lombok Strait, a 20-60 kmwide body of water between the islands of Lombok and Bali that connects the Java Sea to the Indian Ocean.
Because of its geometry, slip on the thrust in the Lombok Strait could generate tsunamis that would efficiently 60 propagate southwards and hit the west coast of Lombok and the east coast of Bali, where their capital cities (Mataram and Denpasar) are located.

Regional setting
Bali and Lombok islands, east of Java, are part of the Lesser Sunda Islands (Fig. 1a). They are located along the 65 volcanic arc of the Java subduction zone, where the NNE-moving Indo-Australian Plate subducts beneath the Sunda Plate (Dewey and Bird, 1970;Hamilton, 1979;Bowin et al., 1980;Silver et al., 1983Silver et al., , 1986Hall and Spakman, 2015;Koulali et al., 2016). The Java trench lies ~250 km to the south. The Flores back-arc thrust belt, on the other hand, follows the northern edge of the islands. Here, the kinematics of fault slip and folding are consistent with the sense of movement of the Indo-Australian Plate and associated shortening, indicating that the 70 Flores back-arc thrust also formed to accommodate stress associated with the plate collision (Silver et al., 1983(Silver et al., , 1986).
The Flores back-arc thrust is an east-west-trending, south-dipping fault zone that extends for >1,500 km along strike. It is composed of two main segments: the Wetar thrust zone to the east and the Flores Thrust to the west 75 (Silver et al., 1983(Silver et al., , 1986; Fig. 1a). From east to west, the Flores Thrust traverses just north of central Flores, Sumbawa, Lombok and Bali (Fig. 1a). From central Flores to east of Lombok, the thrust zone reaches to the https://doi.org/10.5194/nhess-2021-343 Preprint. Discussion started: 20 November 2021 c Author(s) 2021. CC BY 4.0 License.
seafloor (Silver et al., 1983(Silver et al., , 1986Yang et al., 2020). As the deformation becomes blind from central Lombok to the west, the thrust zone has been mapped based on folds visible in seismic reflection data, and also manifests as a band of steeper north-facing slope on the seafloor (Silver et al., 1983;McCaffrey and Nabelek, 1987;Yang et 80 al., 2020). West of Bali, folds are fewer and have no to little seafloor expression (Silver et al., 1983;Fig. 1d), suggesting that the Flores Thrust terminates at Bali (Yang et al., 2020). However, GPS measurements show that the north-south convergence rate in Bali (5 ± 0.4 mm/yr) is similar to that onshore Java (6 ± 1 mm/yr), therefore back-arc shortening may continue across a segment boundary along the Kendeng thrust in Java (Koulali et al., 2016). 85

Seismicity of the Flores Thrust
Focal mechanisms show that from February 1976 to February 2021, the Flores Thrust generated 29 Mw 5.5 to 7.8 earthquakes within the upper 40 km of the crust (GCMT; Fig. 1a). Earthquakes in this region can be caused by either tectonically driven fault slip or volcanic activity. In this back-arc region, most of the focal mechanisms are 90 characterized by east-west striking nodal planes with a fault plane dipping 26±8°S; we infer that these are associated with the Flores Thrust.
This fault system has also produced uplift on its hanging wall. From eastern Sumbawa to central Flores, uplift is recorded by elevated terraces on the northern sides of the islands (Van Bemmelen, 1949). We suggest that the 95 Quaternary reef terraces in northwest Bali (Boekschoten et al., 2000) are also related to tectonic uplift above the Flores thrust system, suggesting that the fault extends all the way to the western coast of the island (Fig. 2).
Although the earthquakes in this region are largely consistent with tectonic shortening, the active volcanoes not only generate their own seismicity, but also play a role in the horizontal and vertical distribution of fault-generated 100 earthquakes (Lythgoe et al., 2021). A relationship between faulting and volcanic activity was observed for the 2018 Lombok earthquake sequence, which generated four >Mw 6 events between 28 th July to 19 th August. These earthquakes did not occur offshore on the northern frontal thrust of the Flores Thrust, but instead involved slip along the deeper part of the fault and associated imbricate thrusts beneath Lombok, to the north of the active Rinjani volcano (Salman et al., 2020;Yang et al., 2020;Lythgoe et al., 2021). While these earthquakes were not 105 directly caused by volcanic activity, the presence of the volcano constrained the earthquake distribution by elevating the downdip limit of the seismogenic zone in the crust (Lythgoe et al., 2021). Based on relocated earthquakes and seismic reflection data analysis, the earthquakes occurred on the Flores fault ramp, a blind thrust dipping 25°S that flattens updip onto the Flores Thrust décollement at ~6 km depth (Lythgoe et al., 2021;Fig. 1c).

Tsunamigenic earthquakes of the Flores Thrust
Historical records (NOAA database, www.ngdc.noaa.gov) and tsunami studies (Hamzah et al., 2000;Rastogi and Jaiswal, 2006;Musson, 2012;Nguyen et al., 2015) document at least four tsunamigenic earthquakes on the Flores 115 Thrust, in addition to the two earthquakes in 2018, which produced local inundation (Fig. 1b) . Three of these events occurred in the western part of the thrust zone, north of Bali. The oldest event on record is the 1815 Ms 7 earthquake, which triggered a landslide and tsunami; together, these events killed >1,200 people. NOAA categorizes this as a probable tsunamigenic event, as it is unclear whether the tsunami was caused only by the coastal landslide, or by the earthquake and landslide together. The 1857 Ms 7 and 1917 Ms 6.5 events are 120 described by NOAA as definite and probable tsunamigenic earthquakes, respectively. The 1857 event generated four consecutive tsunami waves, at least 3 m high, northwest of Flores Island (NOAA, 2021). In addition, in the Lombok Strait, a 1979 Ms 6.6 tsunamigenic earthquake left 200 injured and killed 27 people, although the tsunami is poorly documented and may have played a minor role in the destruction (Hamzah et al., 2000).      The best-documented tsunamigenic earthquake on the Flores Thrust occurred in its far eastern part (Yeh et al., 1993;Imamura and Kikuchi, 1994;Tsuji et al., 1995;Pranantyo et al., 2021). The 1992 Mw 7.9 Flores Island earthquake injured 2,144 people and killed 2,080 (Yeh et al., 1993;Tsuji et al., 1995;Fig. 1a). This earthquake occurred at ~16 km depth (Beckers and Lay, 1995), and generated a tsunami that propagated to the northern coast of Flores Island within five minutes (Yeh et al., 1993). Field mapping shows that the tsunami 130 inundated the land as far as 600 m, with an average run-up height of ~2 to 5 m (elevation reached above sea level).
Anomalously high run-up heights of 20-26 m to the northeast may be associated with submarine landslides (Yeh et al., 1993).
The recent 2018 Lombok earthquake sequence occurred primarily below land, but nevertheless small-scale 135 tsunamis were reported by the residents of northern Lombok (Tsimopoulou et al., 2020). When the Mw 6.4 July event occurred, the northern coast of Lombok subsided by ≤0.1 m , and the northeastern coast was hit by a tsunami at the towns of Labuhan Pandan and Tanjung, which were inundated 10-70 m with run-up heights of ~1-2.5 m. For the Mw 6.9 August 5 event, although the northern coast was uplifted by ≤0.5 m , the residents of the northwest towns, Tanjung and Kayangan, reported a tsunami that 140 inundated 7-40 m inland with a run-up height of ~1.7-2 m (Fig. 1b).
Together, these records show that the Flores Thrust is capable of generating significant thrust earthquakes with associated land uplift/subsidence as well as local tsunamis. The full tsunamigenic potential of this fault system is not known, as the observational window is short compared to typical earthquake recurrence intervals. Here, we 145 explore what could happen when coseismic slip occurs on the Flores thrust ramp within the Lombok Strait, and how the generated tsunami and coseismic land deformation would together affect the coastal cities of Mataram, Lombok and Denpasar, Bali.

Previous tsunami modelling studies
Tsunami modelling studies in this region commonly focus on the segment of the Sunda Megathrust along the Java trench (Okal and Borrero, 2011;Kurniawan and Laili, 2019;Suardana et al., 2019;Kardoso and Dewi, 155 2021) (Fig. 1a), with a few studies evaluating the western segment of the Flores Thrust (Løvholt et al., 2012;Rusli Modified from Yang et al., 2020 Esri  Afif and Cipta, 2015), and four considering an earthquake sourced within the Lombok Strait (Rakowsky et al., 2013;Horspool et al., 2014;Pradjoko et al., 2018;Fig. 1b). All four studies show tsunami results in Mataram, Lombok; however, each study focuses on different aspects of tsunami modelling, and three predate the 2018 Lombok earthquake sequence, which illuminated important aspects of the 160 fault geometry. The only study after the 2018 earthquakes  did not update their fault model to reflect new information about the geometry of the Flores Thrust derived from studies of the 2018 Lombok earthquake sequence. Overall, these prior results do not address the potential earthquake scenarios that we consider plausible: Rakowsky et al. (2013) study the sensitivity of inundation to land friction, Horspool et al. (2014) describe the probabilistic tsunami hazard, Pradjoko et al.(2018) considers a fault that is much too steep and uses 165 bathymetry that is too coarse to produce reliable results, and  did not consider the post-2018 earthquake studies of the fault geometry of the Flores Thrust. Rakowsky et al. (2013) studied the sensitivity of inundation models in the region to the topography and friction parameters of the land surface. Their tsunami modeling was done using the ~900-m-resolution GEBCO dataset 170 interpolated with measurements from ships and nautical charts; the interpolation method is not described in detail.
They considered an unrealistically large Mw 8.5 earthquake, and produced a maximum flow depth (vertical distance between the land and inundating water surface) of 10 m, with an inundation extent ranging from ~1-1.6 km in Mataram. They found that inundation distance depended on the topographic parameters: lower bottom friction or a bare earth digital terrain model produced higher inundation compared to higher friction or a digital 175 surface model (with structures, e.g., houses). Their results highlight the importance of using an accurate surface model when assessing potential inundation. Horspool et al. (2014) focused on probabilistic tsunami hazard for all of Indonesia. They used a bathymetry dataset that combined GEBCO data with measurements from Navy charts and multibeam surveys. For the western Flores 180 thrust, they set 1 m of slip on a range of 25-27°-dipping, 10x20 km sub faults (equivalent to a Mw6.4 earthquake) within the 3-30 km seismogenic depth of the Lombok Strait. Their results do not describe the regional hazard (e.g. wave heights, timing, inundation), but rather assess how much of the local hazard is contributed by this fault system rather than the megathrust. They showed that given an arbitrary 500-year return period on the fault, 10-30% of the tsunami hazard in Mataram is due to the shallow part of the Flores Thrust. (BATNAS) dataset as input bathymetry in the numerical simulations. Their focus was on the impact along the northern coasts, but they note that the tsunami arrives in Mataram and Denpasar in 9 and 25 minutes, respectively.
They also find that the maximum wave height is 1.5 m in Mataram and 1 m in Denpasar.
Following the 2018 Lombok earthquake sequence, we now have a more accurate understanding of the location 205 and subsurface geometry of the Flores Thrust in this region. Hence, the earthquake models we use in our study are geologically well-constrained. In addition, since tsunami propagation in shallow water depends strongly on the bathymetry, we develop and incorporate a new bathymetric model by combining the GEBCO dataset with sounding measurements from the official nautical chart for Indonesia. This is particularly important along the shallow coast, where seafloor roughness has a strong control on wave propagation. In our study, we show the 210 tsunami results from six different earthquake scenarios within the Lombok Strait, highlighting impacts on the populated capital cities of Mataram, Lombok and Denpasar, Bali, as both cities face the Strait. We also calculate the coseismic uplift and subsidence for varying slip amounts, and report this together with the tsunami time history and pattern and the maximum wave height. An inundation scenario is also included for the city of Mataram.

Fault model setup
The 2018 Lombok earthquake sequence illuminated the geometry of the Flores Thrust beneath Lombok (Fig. 1c).
Together, relocated aftershocks, earthquake slip distributions, and seismic reflection imaging indicate a blind fault ramp dipping 25°S that flattens updip to a décollement at ~6 km depth and continues north below the Bali Sea. 220 The part of the thrust ramp that ruptured in the 2018 sequence extends 45 km downdip and 116 km lengthwise (Lythgoe et al., 2021;Fig. 1c & 3).
We use these fault parameters to set up our fault model, choosing a fault with an east-west strike, similar to the general trend of the Flores Thrust, positioned across the Lombok Strait. The eastern boundary of the fault is defined by the western limit of the 2018 earthquake sequence. We extend the western edge of the model to below 225 the eastern edge of Bali, in order to span the width of the Strait; the fault likely continues further west (as evidenced by uplifted terraces and seismicity), but rupture to the west would occur below land and would not contribute to a tsunami. We trace the upper blind tip of the fault ramp following the southern edge of a north-facing seafloor slope. This surface morphology coincides with folding interpreted from seismic reflection surveys (Silver et al., 1983;Yang et al., 2020), and we interpret that the folding formed due to slip across a bend at the upper tip of the 230 blind fault ramp (Fig. 1b). We extend the fault ramp to a depth of 25 km below the seafloor, which represents the maximum seismogenic depth in this region based on historical seismic records and the maximum depth of seismicity observed in the 2018 sequence (Lythgoe et al., 2021).

Slip model
For both Models A and B, we consider three scenarios with uniform slip of 1, 3, and 5 m (six scenarios total). In order to focus on the impact of tsunami generation, we include only slip on the fault ramp (no slip transferred onto the northern décollement). This updip termination of slip was observed in the Lombok sequence (Lythgoe et al., 2021) and is therefore realistic in our region to the west as well. Although we consider uniform slip, earthquake 245 slip is known to be spatially variable, and in particular to taper around the edges of the slip patch. We evaluate the impact of this taper on the initial seafloor deformation using the Green's function for rectangular dislocations (Okada, 1992) in the code Unicycle (Moore et al., 2019); we find that tapering the slip slightly modifies the uplift profile by broadening it and shifting it to the south (downdip direction) but does not significantly change the model (Fig. 4).  To better translate the models into equivalent earthquakes, we calculate the equivalent Moment Magnitude (Mw) for each modeled event, using a standard rigidity of 30 GPa. Since Model A has a wider fault surface, for the same amount of slip, it produces larger magnitudes compared to Model B (Table 1). In each model, we consider only 260 the part of the fault that lies below the Lombok Strait, since this is the part of the fault that is submarine and therefore capable of generating tsunamis. We note that an earthquake rupturing this fault segment could involve slip further along strike, either to the west (below Bali) or to the east (below Lombok, although this part of the fault recently ruptured in multiple earthquakes and is relatively less likely to slip again). Indeed, reaching 5 m of slip within the Lombok Strait alone would likely require a longer rupture, and therefore a larger magnitude than 265 the values reported in Table 1, given known scaling relationships between fault area and coseismic slip (Wells and Coppersmith, 1994;Hanks, 2002;Biasi and Weldon, 2006;Hanks and Bakun, 2008).

Bathymetry
Accurate modeling of tsunami wave propagation requires a high-resolution bathymetric map, especially in shallow 275 water. By using detailed bathymetry together with a fine grid size, modelled simulations of tsunami wave heights have been shown to effectively match real near-coast waveforms (Satake, 1995). However, in many parts of the world, high resolution bathymetric data are unavailable. In general, regional tsunami studies use only one bathymetric dataset (e.g., Satake, 1988), commonly either ETOPO (https://www.ngdc.noaa.gov/mgg/global/) or GEBCO (https://www.gebco.net/), because they are publicly available and have wide coverage. However, these 280 datasets have an artificially smooth seafloor (Marks and Smith, 2006), especially at shallow depths, because of the low density of interpolated points (e.g., Fig. 5). In local tsunami studies, the detailed seafloor morphology in shallow water is critical, since seafloor roughness in these regions has nonlinear effects on wave propagation (Wang and Power, 2011). Kulikov et al. (2016) demonstrated that tsunami propagation modeled using the GEBCO dataset results in substantial errors in the estimation of wave propagation. 285 We generate a high-resolution bathymetric model of the region of interest by combining water depth measurements from GEBCO with sounding measurements from the official nautical charts of Indonesia (http://hdc.pushidrosal.id/). The publicly available GEBCO dataset is provided as an interpolated raster, but also includes the original data points used for interpolation. These data points (water depths) are derived from a variety 290 of sources, both direct (echo soundings, seismic reflection, isolated soundings, electronic navigation chart soundings) and indirect (e.g. satellite altimetry, flight-derived gravity data). Using the Type-Identifier Grid file from GEBCO, which includes the source of the depth data, we identify and extract only the water depths acquired by direct measurement (Fig. 5).

295
The GEBCO data in this region are concentrated along the heavily-travelled ship tracks between the islands of Bali and Lombok, and are too low resolution near the coasts to accurately model tsunami propagation and wave heights (Fig. 5a). We improve the resolution of our bathymetry by digitizing sounding data from the official nautical charts of Indonesia, which are densest in the coastal regions near the cities of Denpasar (Bali) and Mataram (Lombok) and therefore critical for modeling near-shore wave heights in these regions (Fig 5b). We also 300 trace the coastline using the National Digital Elevation Model (DEMNAS, http://tides.big.go.id/DEMNAS/), and cross check it using satellite images from Esri World Imagery (https://www.arcgis.com/). We combine the water depth measurements from both sources and the coastlines into a single dataset, and then interpolate the data using the 'Topo to Raster' tool in ArcGIS. This tool is based on the ANUDEM program 305 developed by Hutchinson (1989), and generates a continuous digital elevation model based on point data that takes into account the hydrological correctness of the resulting raster. While this method was developed on the basis of subaerial water flow, it has also been used to effectively generate bathymetries for tsunami studies in other regions (Fraser et al., 2014;Darmawan et al., 2020;Wilson and Power, 2020). We note that the shallow shelf regions of the Lombok Strait were likely incised subaerially during the late Holocene sea-level drop 310 (Boekschoten et al., 2000), and their morphologies therefore likely reflect subaerial water flow processes. https://doi.org/10.5194/nhess-2021-343 Preprint. Discussion started: 20 November 2021 c Author(s) 2021. CC BY 4.0 License.
We set the resolution of our interpolated raster to 30 m, as this is similar to the mean distance between the data points along the coasts of Mataram (~27) and Denpasar (~36 m). Our final bathymetry represents a reasonable balance between achievable accuracy at shallow depths and computational efficiency. We validate the interpolated 315 bathymetry by comparing its values with the source data; the mean difference in the shallow regions offshore Mataram and Denpasar is <0.4 m.

Topography in Mataram, Lombok
Based on our tsunami model runs, the highest wave heights are observed along the coast of Mataram, Lombok.
In order to further explore the tsunami hazard in this populated area (Fig. 6), we model the inundation of the onshore region. The inundation distance and run-up height of a tsunami can vary significantly depending on factors such as the average slope of the coast and the land cover roughness (Kaiser et al., 2011;Griffin et al., 2015); an accurate forecast requires a high-resolution Digital Surface Model (DSM) that maps the buildings and 330 trees.
We use a Digital Surface Model generated by Apollo Mapping based on Pleiades satellite imagery. The DSM has a horizontal resolution of 1.5 m and a vertical error of ±3 m. This vertical error is the lowest possible for digital elevation models without ground control points, which we do not have access to. We use a DSM rather than a 335 DTM (Digital Terrain Model) to better represent the man-made structures (e.g., houses, infrastructure) present in Mataram city. There are a few areas where the DSM is unavailable along the coast, due to difficulties in data processing associated with tides. We fill these areas with 1.5-m resampled elevation data from DEMNAS, the national elevation model for Indonesia, which has a coarser original horizontal resolution of 8 m. The vertical datum of the merged data is referenced to EGM2008. 340 In order to run the inundation modelling, the topographic data must be merged with the bathymetry so that the incoming wave can be smoothly modeled across the sea-land interface. To match the resolution of the DEMNAS- 'Topo to Raster' interpolation method as used previously for the bathymetry. We match the coastlines of the two 345 datasets to generate the final combined model.

Tsunami modelling using COMCOT
We model the tsunami generation, propagation, run-up and inundation using the Cornell Multi-grid Coupled Tsunami (COMCOT) model developed by . This modeling system solves linear and nonlinear 350 shallow water equations using a modified leap-frog finite difference approach (Wang & Power, 2011). It uses a nested-grid layer algorithm to increase its computational efficiency. The Okada (1985) model is used to calculate surface deformation due to fault slip. We use this model in our study as it has been extensively adopted and

Coseismic deformation and maximum wave height
When slip occurs on the Flores Thrust ramp during an earthquake, the elastic response of the crust will lead to broad changes in the elevation of the ground surface. In the north, above the fault ramp, the seafloor will rise (uplifting any ocean column above), whereas the southern region will subside (Fig. 7a-c). Associated with this process, the islands of Bali and Lombok will tilt towards the south (Fig. 7a-c, 8a-c). As the initial sea surface 385 deformation will have the same magnitude as the land deformation, the initial wave will be unnoticeable relative to the coast, which experiences the same vertical motion (Fig. 7d-f, 8d-f). The initial waves in our models correspond to tsunami energies of 1, 13, and 36 TJ for Model A and 1, 7, and 20 TJ for Model B for 1, 3, and 5 m of slip, respectively (Felix et al., 2021)  Along the southern coasts, on the other hand, coseismic subsidence acts to increase the relative tsunami heights. 400 The subsidence in southern Lombok and Bali can reach as high as ~0.3-0.4 m for 5 m of fault slip, ~0.1-0.25 m for 3 m slip, and <0.1 m for 1 m slip. We find that overall, the west coast of Lombok experiences higher tsunamis than the southeast coast of Bali, because it is closer to the tsunami source and the coastline is perpendicular to the source, making it more exposed to the propagating waves. The maximum tsunami height on the west coast of Lombok is ~1.6-3.7 m for 3-5 m of coseismic slip. On the other hand, the more distant and better protected 405 southeastern coast of Bali has a maximum wave height of ~0.7-2.2 m given the same slip amount, with slightly higher waves within the semi-enclosed bays (Figs. 7d-e).
When only the upper half of the fault ramp slips (model B), the uplift patch is narrower and the subsidence region is broader, covering about three quarters of the coasts of Lombok and Bali. Unlike in model A, the headlands at 410 8.38°S are now within the area of subsidence. This results in an increase in the relative maximum wave height at the headlands, with ~1.5-4 m high tsunamis for 3-5 m of coseismic slip (Fig. 8d-e). Similarly, the west coast of Lombok is hit by ~1.5-3.4 m high tsunamis, while southeastern Bali experiences 0.5 to 2 m high tsunamis for 3-5 m of coseismic slip.

415
The two fault models generate similar maximum wave heights along the west coast of Lombok (Fig. 9), while the tsunamis generated by model A are slightly higher than model B along the southeastern coast of Bali (Fig. 10). In both models, however, we consistently observe higher tsunami waves in Lombok compared to Bali. This difference is best observed using the virtual tide gauge records situated near the cities of Mataram and Denpasar. 420

Tsunami time series in Mataram, Lombok and Denpasar, Bali
The tide gauge records show that the tsunami arrival times in Mataram and Denpasar are insensitive to the fault model geometries that we consider. The first and highest wave in Mataram arrives <8 minutes after the earthquake and it reaches its peak at ~11 minutes, followed by a drawdown at ~15 minutes. Three more waves reach the coast at ~20, ~35 and 45 minutes (Fig. 11). The first wave in Mataram is ~2.

430
In Denpasar, the waves are smaller and take longer to arrive. The first wave arrives at ~10-15 minutes and reaches its peak at ~30 minutes. It is followed by a drawdown at ~40 minutes and a second wave at ~45 minutes (Fig. 11).
As Denpasar is further from the tsunami source and has a complex coastline, its wave records are not as uniform as those along the linear coast of Mataram. For both fault models A and B, relatively higher tsunami waves are generated within the semi-enclosed bay in the northeast of Denpasar, while lower waves reach southwestwards 435 along the concave coastline (Fig. 11). Although they have a similar trend, the wave heights generated by model

Inundation in Mataram, Lombok
Tsunami waves of a given height at the coastline can have variable impact depending on the topography and infrastructure on land. Because inundation modeling requires a detailed Digital Surface Model for accurate results 465 and significant computational time, we limit the inundation modeling to the city of Mataram, Lombok, because this region is densely populated (Fig. 6) and is exposed to the highest waves in our tsunami models. We run the modeling for the highest value of fault slip (5 m) for fault model A (full rupture) to represent our worst-case scenario.

470
Based on our results, 5 m of fault slip generates two >2 m high waves followed by two lower waves that hit the coast at Mataram city (Fig. 11)  . When the beach is wider and the structures are further from the coast, the inundation extends further inland . We note that in our model, clustered vegetation on the beach is 485 represented in the DSM as a solid barrier, and thus is able to entirely block the flow (upper part of Fig. 12c-d). In reality, clustered vegetation can slow but not completely obstruct the flow; the inundation extent at this site is therefore likely underestimated. Our results may be more realistic in regions where vegetation is absent, as in the lower part of Fig. 13a-

Conclusions
The Flores Thrust is an active south-dipping back-arc fault system traversing north of the Lesser Sunda Islands.
The 2018 Lombok earthquake sequence and prior historical events show that the western part of the fault zone is 505 capable of generating tsunamigenic earthquakes. In this work, we study the tsunami potential associated with coseismic slip on the blind fault ramp below Lombok Strait, located between the islands of Lombok and Bali. We focus on the tsunami patterns near the capital cities of Mataram, Lombok and Denpasar, Bali, which both lie on the coasts facing the strait. Our modeling is based on a geologically constrained model of the fault, informed by the 2018 earthquake sequence. Tsunami propagation is modeled using a high-resolution bathymetry dataset 510 generated by combining data points from the global GEBCO dataset with sounding data digitized from the official nautical charts of Indonesia, interpolated using the Topo to Raster tool in ArcGIS.
Our results show that fault rupture in this region with 1-5 m of coseismic slip could trigger a tsunami that would hit Mataram, Lombok in ~11 minutes and Denpasar, Bali in ~30 minutes with multiple waves. Furthermore, both 515 cities would experience coseismic subsidence of 20-40 cm, exacerbating their exposure to the tsunami hazard and leading to more long-lasting coastal vulnerability. The maximum wave heights in Mataram are 1.5 to 3.3 m for 3-5 m of coseismic slip, while Denpasar has maximum wave heights of 0.7 to 1.5 m. Overall, the coast along Mataram city is more prone than Denpasar to high tsunamis arriving quickly.

520
Because Mataram experiences higher wave heights, we also modelled the inundation in this region for our worstcase scenario (5 m slip) using a high-resolution DSM. We found that the inundation extends for ~55-140 m inland with a maximum flow depth of ~2-3 m, and, except in the region just south of the city, where the inundation reaches 230 m. This difference in inundation extent appears to be primarily influenced by the structures present near the coast, which are denser in the north. However, if structures are destroyed by flow, inundation could reach 525 further inland.
Because of the proximity of the Flores thrust ramp to the coasts of Lombok and Bali, associated tsunamis would hit within <15 minutes after the earthquake. This early tsunami arrival would mean little time for evacuation. In the case of the 2018 Lombok earthquake, the residents of northern Lombok started evacuation only after a 530 government announcement, and the evacuation took at least 20 minutes (Tsimopoulou et al., 2020). For a potential tsunami in Mataram caused by slip on the Flores thrust, there is insufficient time to wait for an announcement after the earthquake. Hence, raising community awareness about earthquake-generated tsunamis and evacuation plans is important, so that residents will know to respond immediately after experiencing strong ground shaking.
Furthermore, the initial polarity of the waves would be positive, and thus there would be no warning signal from 535 drawdown prior to inundation. In addition, a second high wave would hit Mataram coast at ~20 minutes, emphasizing the need for continued heightened alert following the first inundation.
We finally note that some of the structures built along the coast are industrial, with several petroleum tanks and a gas power plant. The impacts of natural disasters can be multiplied when natural events trigger industrial events 540 ('Natural Hazards Triggering Technological Disasters,' or Natech) (Cruz and Suarez-Paba, 2019). Tsunamis in particular have a history of causing Natech events (e.g. (Suppasri et al., 2021); for instance, the 2011 Mw9.1 Tohoku earthquake and tsunami led to not only meltdown at the Fukushima-Daichi nuclear power plant, but also fires, explosions, and hazardous materials release at industrial sites (Krausmann and Cruz, 2013). In Mataram, damage to the petroleum tanks, power plant, and other industrial equipment by groundshaking or inundation could 545 trigger Natech events, including fires, explosions, and pollution of the coastal water and associated ecological damage. Evaluating these sites to understand and strengthen their resilience to these hazards should be a priority.
While most tsunami modeling studies in Indonesia have focused on the hazard associated with large tsunamis triggered by megathrust ruptures, such as the devastating 2004 Indian Ocean earthquake and tsunami (e.g. Wang 550 and Liu, 2007), we highlight here the hazard associated with smaller, local events caused by slip on a back-arc thrust system. One of the challenges with local studies is the need for detailed and accurate fault models and bathymetry datasets. We show that geological information such as regional and nearby seismicity can be combined with bathymetry, topography, and seismic reflection data to model fault geometry, and that a high-resolution bathymetry dataset can be generated by combining globally available bathymetric data with sounding 555 measurements collected for navigation purposes. Specifically, for earthquake-triggered tsunamis in Indonesia, the https://doi.org/10.5194/nhess-2021-343 Preprint. Discussion started: 20 November 2021 c Author(s) 2021. CC BY 4.0 License. official nautical charts for Indonesia provide dense measurements offshore shallow coastal cities. Integrating these datasets can provide more accurate forecasts and hazard estimations for both tsunami wave height and arrival time, for local and regional studies, and could be replicated for other fault systems and areas.

DATA AVAILABILITY
The animation of the tsunami propagation for the 5 m coseismic slip on the full fault ramp can be accessed at https://researchdata.ntu.edu.sg/privateurl.xhtml?token=ed262ac9-0649-4d39-9c34-104c0e93f6f1 The