Volume 46, Issue 15 p. 8721-8730
Research Letter
Free Access

Source Model for the Tsunami Inside Palu Bay Following the 2018 Palu Earthquake, Indonesia

Aditya Riadi Gusman

Corresponding Author

Aditya Riadi Gusman

GNS Science, Lower Hutt, New Zealand

Correspondence to: A. R. Gusman,

[email protected]

Search for more papers by this author
Pepen Supendi

Pepen Supendi

Geophysical Engineering Study Program, Faculty of Mining and Petroleum Engineering, Institut Teknologi Bandung, Bandung, Indonesia

Agency for Meteorology, Climatology, and Geophysics (BMKG), Jakarta, Indonesia

Search for more papers by this author
Andri Dian Nugraha

Andri Dian Nugraha

Global Geophysics Research Group, Faculty of Mining and Petroleum Engineering, Institut Teknologi Bandung, Bandung, Indonesia

Search for more papers by this author
William Power

William Power

GNS Science, Lower Hutt, New Zealand

Search for more papers by this author
Hamzah Latief

Hamzah Latief

Center for Marine and Coastal Development, Institut Teknologi Bandung, Bandung, Indonesia

Oceanography Study Program, Faculty of Earth Sciences and Technology, Institut Teknologi Bandung, Bandung, Indonesia

Search for more papers by this author
Haris Sunendar

Haris Sunendar

Center for Marine and Coastal Development, Institut Teknologi Bandung, Bandung, Indonesia

Search for more papers by this author
Sri Widiyantoro

Sri Widiyantoro

Global Geophysics Research Group, Faculty of Mining and Petroleum Engineering, Institut Teknologi Bandung, Bandung, Indonesia

Search for more papers by this author
Daryono

Daryono

Agency for Meteorology, Climatology, and Geophysics (BMKG), Jakarta, Indonesia

Search for more papers by this author
Samsul Hadi Wiyono

Samsul Hadi Wiyono

Agency for Meteorology, Climatology, and Geophysics (BMKG), Jakarta, Indonesia

Search for more papers by this author
Aradea Hakim

Aradea Hakim

Botram Ocean Technology Research and Management

Search for more papers by this author
Abdul Muhari

Abdul Muhari

Ministry of Marine Affairs and Fisheries, Jakarta, Indonesia

Search for more papers by this author
Xiaoming Wang

Xiaoming Wang

GNS Science, Lower Hutt, New Zealand

Search for more papers by this author
David Burbidge

David Burbidge

GNS Science, Lower Hutt, New Zealand

Search for more papers by this author
Kadek Palgunadi

Kadek Palgunadi

King Abdullah University of Science and Technology, Thuwal, Saudi Arabia

Search for more papers by this author
Ian Hamling

Ian Hamling

GNS Science, Lower Hutt, New Zealand

Search for more papers by this author
Mudrik Rahmawan Daryono

Mudrik Rahmawan Daryono

Research Center for Geotechnology, Indonesian Institute of Science (LIPI), Jakarta, Indonesia

Search for more papers by this author
First published: 29 July 2019
Citations: 56

Abstract

On 28 September 2018, a strike-slip earthquake occurred in Palu, Indonesia, and was followed by a series of tsunami waves that devastated the coast of Palu Bay. The tsunami was recorded at the Pantoloan tide gauge station with a peak amplitude of ~2 m above the water level and struck at high tide. We use the Pantoloan tsunami waveform and synthetic aperture rada displacement data in a joint inversion to estimate the vertical displacement around the narrow bay. Our inversion result suggests that the middle of the bay was uplifted up to 0.8 m, while the other parts of the bay subsided by up to 1 m. However, this seafloor displacement model alone cannot fully explain the observed tsunami inundation. The observed tsunami inundation heights and extents could be reproduced by a tsunami inundation simulation with a source model that combined the estimated vertical displacement with multiple subaerial-submarine landslides.

Key Points

  • The series of tsunami waves that created the disaster in Palu was caused by a combination of seafloor uplift and multiple landslides
  • The seafloor vertical displacement was estimated using tsunami waveform and SAR data and is evaluated to be the source the largest tsunami
  • Ground subsidence of up to 1 m and a tide level of 1 m during the event enhanced the tsunami impact in Palu city

Plain Language Summary

The tsunami that devastated Palu and other coastal towns inside Palu Bay on 28 September 2018 was recorded at a sea level monitoring station in Pantoloan. The recorded tsunami wave data are used in an inversion method to estimate the source of the tsunami in the form of an initial seafloor displacement. We found that the seafloor displacement was the main cause of the large tsunami. Satellite images and field survey data suggest that landslides around multiple river deltas also generated local tsunami waves. Our numerical simulations of the tsunami inundation show that the disaster was caused by a combination of the sudden ground and seafloor changes from the earthquake, landslides, and the high tide at the time of the event.

1 Introduction

The Mw 7.5 Palu earthquake struck during sunset on 28 September 2018 at 18:02:44 local time (10:02:44 UTC), with epicenter located at 0.18°S and 119.85°E (Figure 1a), ~80 km north of Palu city in Central Sulawesi, Indonesia (Agency for Meteorology, Climatology, and Geophysics—BMKG). The left-lateral (sinistral) strike-slip faulting mechanism with north-south orientation (strike = 348°, dip = 40°, and rake = −9°) obtained from the global centroid moment tensor (gCMT) solution is consistent with the relative motion of the Palu-Koro fault. The aftershocks (28 September to 22 November 2018) are mostly located to the southeast of the mainshock (Figure 1a) thus showing a northwest-southeast trend with a total length of approximately 200 km and consistent with the eastward dipping nodal plane of the gCMT solution (Figure 1a). Strike-slip faulting earthquakes usually generate small tsunamis due to the lack of large amounts of vertical deformation (Gusman et al., 2017; Lay et al., 2018). However, the Palu earthquake generated large tsunamis along the coast of Palu Bay, a narrow bay with length of ~30 km and width of ~7 km (Figure 1b). The National Disaster Management Authority (BNPB) of Indonesia has reported that 4,340 people were killed and 10,679 were injured by either or combinations of the tsunami, landslide, liquefaction, and collapsing building following the earthquake, and 667 people have been declared missing.

Details are in the caption following the image
(a) Map showing the mainshock (yellow star), aftershocks (green circles), and global centroid moment tensor solution (red beachball diagram) of the 2018 Palu earthquake. Blue traces represent the Palu-Koro fault and red traces represent other major crustal faults in the region. (b) Topographic and bathymetric grid used in the tsunami numerical simulation. Inside the blue box, high resolution topographic data (light detection and ranging and interferometric synthetic aperture radar) and bathymetric contour were used to make the modeling grid, while BATNAS (BIG) data were used for the region outside the blue box. (c) Observed tsunami flow depths (above the ground) in Palu measured by Paulik et al. (2019).

Coastal villages within Palu Bay have been hit by significant tsunamis following earthquakes in the past. A 2-m tsunami was reported in Donggala, Sulawesi, following an earthquake on 30 January 1920 (Soloviev & Go, 1974). On 1 December 1927, a strong earthquake felt within Palu Bay, the subsequent tsunami completely washed away a pier in Talise and caused property damages in coastal villages; about 50 people sustained injuries and 50 died (Soetardjo et al., 1985; Soloviev & Go, 1974). The 20 May 1938 earthquake (M 7.6) located within the Tomini Gulf generated a tsunami along the coast between Toribulu and Parigi, but not inside Palu Bay (Soetardjo et al., 1985; Soloviev & Go, 1974). The 14 August 1968 earthquake (M 7.8) located at 0.157°N, 119.802°E generated an 8- to 10-m tsunami along the coast between Donggala (Tanjung), Manimbaya, and Sabang (Soetardjo et al., 1985; Soloviev & Go, 1974).

The Palu-Koro fault is a major transform zone in Southeast Asia, cutting across Sulawesi Island from NW to SE (Figure 1a), marking the convergence of the Eurasian, Australian, and Philippine plates with the Sula block located in the middle of the triple junction (Bellier et al., 2006). The collision of the Sula block (Rangin, 1989) into the Eurasian plate is accommodated at its western and southwestern limit by the Palu-Koro and the Matano faults, respectively, and accommodated at its northern limit by the North Sulawesi trench (Walpersdorf et al., 1998). The Palu-Koro fault strikes south-southeast and stretches from Palu Bay SSE for 300 km on land and has a 400-km long submarine extension in the Makasar Strait (Tjia & Zakaria, 1974). Plate convergence measurements by GPS from 1992 to 1999 showed that the Palu-Koro fault has a left-lateral slip rate of 34–38 mm/year (Vigny et al., 2002; Walpersdorf et al., 1998).

Significant earthquakes along the Palu-Koro fault occurred in 1907, 1909, 1937, and 2012. Among these events, the 1909 was recorded as the most violent one with 7 km of surface ruptures (Abendanon, 1917; Daryono, 2016). Geological trenching at Omu village north of Lindu lake showed that the 1909 earthquake had sinistral movement of 1.5 m and vertical displacement of 1.5 m (Daryono, 2016). The paleoseismology study also found geological evidence of two large Palu-Koro earthquakes in 1338 and 1468.

Field surveys following the recent Palu earthquake were conducted by Indonesian and international teams to measure tsunami heights, impacts, and collect eyewitness accounts (BMKG, 2018; Arikawa et al., 2018; Cipta et al., 2018; Muhari et al., 2018; Omira et al., 2019; Paulik et al., 2019). The tsunami was also recorded by a tide gauge station at Pantoloan harbor located inside Palu Bay. Satellite images taken before and after tsunami (https://www.digitalglobe.com/ecosystem/open-data) provide a way to identify the damage extent and the limit of tsunami inundation. The Geospatial Information Agency of Indonesia (BIG) has provided locations of destroyed and damaged buildings around Palu from a rapid assessment using the satellite images. The earthquake and tsunami were recorded by smartphone and closed-circuit television cameras at different locations around Palu Bay (Table S1 in the supporting information). The video footage provides several kinds of information such as tsunami arrival time, inundation height, period, incoming direction, and potential locations of subaerial-submarine landslides (Carvajal et al., 2019).

Coseismic fault slip models have been estimated in previous studies using seismic data (United States Geological Survey, 2018) and interferogram synthetic aperture radar (InSAR) data sets (Socquet et al., 2019). Tsunami simulation results from the fault slip distributions (Carvajal et al., 2019) did not fully reproduce the observed tsunami waveform at Pantoloan station or the inundation heights along the coast of Palu Bay. Therefore, in this study, we are using another approach to estimate the tsunami source model for the 2018 Palu earthquake.

Here, synthetic aperture radar (SAR) offsets from Sentinel-1 satellite images are used to map the ground deformation due to the earthquake. We use the tsunami waveform and SAR vertical displacement data in a joint inversion method to estimate the initial seafloor displacement of the tsunami source. The tsunami source model is validated by comparing the simulation result with the observed tsunami inundation heights and extents in Palu city from the field survey and satellite images. The available data allow us to estimate a source model for the tsunami that occurred inside Palu Bay but unfortunately not for the area outside the bay. River deltas collapsed during the earthquake (Carvajal et al., 2019; Cipta et al., 2018) and might have made some contributions to the tsunami generation, which we evaluate by tsunami numerical simulations.

2 Data and Methods

2.1 Synthetic Aperture Radar Offsets

Using SAR data acquired by the Copernicus Sentinel-1A/Sentinel-1B satellites, we use ISCE (InSAR Scientific Computing Environment software, Rosen et al., 2012) to generate range and azimuth offsets for both ascending (13 March 2017 to 4 October 18) and descending (7 June 2018 to 5 October 2018) tracks 112 and 134, respectively. Due to the dense vegetation of the region and the short wavelength of the Sentinel satellites, coherence is almost completely lost preventing the use of traditional interferometry. After their creation, the offsets are geocoded and median filtered with a width of ~500 m. Pixels with a signal to noise ratio of less than 5 are removed and, based on the average look vector covering the epicentral region, we invert for the three-dimensional displacement field (Figure S1; e.g., Hamling et al., 2017; Michel et al., 1999). The inversion is weighted by the far field variance for each data set and any pixel in the final displacement map with an error greater than 1 m is removed. While noisy, the resolved ground displacements match well with ground observations (Figure S2; Irsyam et al., 2018; Supartoyo et al., 2018). N-S offsets along the main strand of the fault of 3–4 m (Figure S1) with subsidence of up to ~1 m observed within the Palu basin and at the coast (Figure 2a).

Details are in the caption following the image
(a) Coseismic vertical displacement estimated from tsunami waveform recorded at Pantoloan tide gauge station and synthetic aperture radar data. Black dots represent the location of the unit sources and green triangles represent tide gauge stations. (b) Tide gauge record at Pantoloan station. (c) Time series plot comparing the observed (black line) and simulated (red line) tsunami waveforms and showing data used in the inversion (blue line).

2.2 Tsunami Source Inversion

We use the tsunami waveform recorded at Pantoloan station (0.7117°S, 119.8572°E) and vertical displacement data estimated from SAR offsets with a joint inversion method (Gusman et al., 2010; Gusman et al., 2018) to estimate the initial seafloor displacement only within Palu Bay. The available data is not enough to estimate the tsunami source outside the bay. The Pantoloan tide gauge record (Figure 2b) has a sampling interval of 1 minute. The ocean tides at the station were removed by a high-pass filter with cutoff period of 100 min to obtain the tsunami waveform. The tsunami arrived at Pantoloan with small positive waves (0.4 m) followed by a negative wave down to −1.9 m and then a positive wave up to 1.9 m (Figure 2c). The tsunami was also recorded at Mamuju tide gauge station (inset in Figure 2a). A previous study by Heidarzadeh et al. (2018) shows that simulated tsunami waveform at Mamuju tide gauge from the United States Geological Survey earthquake source model has an arrival time that is approximately 45 min earlier than the first tsunami signal. The study suggests that the source for the initial tsunami signal at this gauge might be located outside Palu Bay. For whatever the reason, the tsunami waveform in Mamuju is therefore not useful in constraining the tsunami source model inside the bay.

For the tsunami Green's functions, we distribute 12 × 4 B-spline unit sources (Koketsu & Higashi, 1992) with spatial interval of 3 km to cover Palu Bay. Here we assume that sea surface displacement is the same as seafloor displacement. The synthetic tsunami waveforms at the station are calculated for unit sources by solving the linear shallow water equations (Gusman et al., 2010). For the tsunami numerical modeling, we use a combination of bathymetric contour data and Batimetri Nasional (BATNAS) gridded bathymetric data with grid size of 6 arc sec, which are both available from BIG (Figure 1b). The middle of Palu bay is located approximately 60 km from the epicenter. Fault movement around the bay occurred 15–25 s after the start of the earthquake as indicated by the source locations of significant beam power and by the earthquake's average rupture velocity (4 km/s) as evaluated in the backprojection study of Bao et al. (2019). We assume a time delay of 25 s for the tsunami inversion as the tide gauge is located inside the bay. The result from this inversion is not significantly different from that of an equivalent inversion that assumes a time delay of 15 s, because a tsunami travels with relatively slow speed inside the bay (60 m/s at 360 m of water depth). The time window used for the inversion is 10 min, which includes the first main tsunami wave cycle.

The main purpose of using the SAR-derived ground displacements in tsunami source inversion is to better estimate the seafloor vertical displacement around the coast (e.g., Lorito et al., 2011; Williamson et al., 2017). To ensure comparable sensitivity between the data sets, we apply a weighting factor of 15 for the tsunami data because the number of tsunami data points is much smaller than the SAR data points. The weighting factor was found after experimenting with different values, to find a weighting that reproduced the important features of both the tsunami waveform and the SAR displacement image in the inversion. For the joint inversion, we use the least squares method of Lawson and Hanson (1995), in which the unknown parameters are the amplitude scaling of the unit sources. A spatial smoothness constraint through a Laplacian operator with an assumed weight of 0.12 is used.

We assess the reliability of the inversion methods by checkerboard sensitivity tests. The synthetic tsunami waveform and SAR vertical offset include an added noise (random number from a Gaussian distribution) with a noise factor equaling 10% of the signal's peak amplitude or offset. The checkerboard test result shows that the tsunami waveform inversion can only roughly reproduce the offshore target pattern within a radius of ~10 km from the tide gauge (Figure S3). A joint inversion is preferable as the target displacement pattern can be much better reproduced by a joint inversion than by a tsunami waveform inversion (Figure S3).

2.3 Tsunami Inundation Simulation and Validation

Field surveys were conducted to measure tsunami inundation heights and collect eye witness accounts (BMKG, 2018; Cipta et al., 2018; Paulik et al., 2019). The observed tsunami heights (Figure 1c; Paulik et al., 2019) are compared with the simulation result (Figure 3a) to verify our high-resolution tsunami inundation simulation in Palu city. The coastal city of Palu is located at the southern end of Palu Bay with flat topography. The coastal area is densely populated with residential, commercial, and hotel buildings located very close (<100 m) to the shoreline, and cafes built on the beach as shown by maps from BIG, OSM (Open Street Map) and satellite images. Many buildings within 200 m from the shoreline were destroyed or damaged by the tsunami and ground shaking as mapped by BIG. We use the satellite images taken before and after the earthquake provided by DigitalGlobe to trace the limit of tsunami debris (Figure 3).

Details are in the caption following the image
Comparison of observed and simulated tsunami inundation. (a) Observed and simulated tsunami inundation heights from the model that combines the vertical displacement (VD) and multiple landslides (LS) models. (b) Simulated tsunami inundation from VD model. (c) Simulated tsunami inundation from LS model. (d) Simulated tsunami inundation from VD + LS model. Red lines represent the limit of tsunami inundation as interpreted from tsunami debris visible on satellite images. Black rectangles represent the landslide block models. To obtain the tsunami heights (above the mean sea level) from the measured flow depths (above the ground), the topographic and synthetic aperture radar vertical displacement data were used.

To verify the estimated sea surface displacement, we simulate tsunami inundation in Palu city and then compare the simulation result with measured tsunami inundation heights and the traced limit of tsunami debris. For precise tsunami simulation inundation along the coast of Palu Bay, we use a grid size of 0.48 arc sec (~14 m). The modeling grids in Palu city are based on topographic data from Light Detection and Ranging (LiDAR) measurements with an original spatial resolution of 15 cm obtained in 2011. For other places within the blue box in Figure 1b that are not covered by the LiDAR data, we use interferometric SAR data (5 m resolution). The vertical datum of the bathymetric and topographic data is mean sea level. The bathymetric data used here is the same as the one that we use in calculating the tsunami Green's functions explained above. We simulate the tsunami inundation using a well-validated tsunami code, Cornell Multi-grid Coupled Tsunami (COMCOT), which can solve the nonlinear shallow water equations (Liu et al., 1995; Wang & Power, 2011; Wang & Liu, 2006).

2.4 Landslide Tsunami Simulation

Video footage and satellite images suggest that multiple landslides contributed to the source of the tsunami (Carvajal et al., 2019; https://blogs.agu.org/landslideblog/2018/10/19/landslide-tsunamis-sulawesi-earthquake/). We estimate model parameters for eight subaerial-submarine landslides at the mouth of several rivers in Palu Bay (Table S2 and Figure S4), which are assumed to occur simultaneously. The source model parameters are not strongly constrained except for the landslide width from the satellite images. The length and thickness of the landslide model are assumed to be 50% and 3% of the width, respectively. The direction of landslide movement in the models is taken to be roughly perpendicular to the coastline and is stopped at the nearest flat bathymetry along the sliding trajectory. According to eyewitness accounts and video footage, the collapse of river deltas in Talise (LS3) and Buluri (LS6; Figure S4) occurred during the earthquake ground shaking, the times of the river delta collapses at the other locations are not known but are probably almost the same as those at these two locations.

We also use COMCOT to simulate landslide tsunami scenarios in Palu. The equation of solid block motion presented in Enet and Grilli (2007) is used to describe the landslide failure process. To simulate the resulting tsunami, the landslide motion model is coupled with a tsunami model through a dynamic boundary interface (Liu et al., 2018; Mountjoy et al., 2018; Wang et al., 2016). We will discuss the contribution of the collapse of river deltas in generating the tsunami in the discussion section.

3 Results

3.1 Seafloor Vertical Displacement

The tsunami source model estimated from the Pantoloan tsunami waveform only, without SAR data, shows an uplift area in the middle of the bay (Figure S5). However, the model also indicates inland uplift (Figure S5a) that contradicts the SAR vertical offset data (Figure S1). To overcome this inconsistency, we estimate the seafloor vertical displacement using tsunami waveform and SAR offset data in a joint inversion. The SAR data are used to help constrain the estimated seafloor vertical displacement near the coast. The by-product of estimated inland vertical displacement from the joint inversion is removed because the unit source distribution was not designed to estimate inland displacements. The estimated seafloor vertical displacement from the joint inversion is then combined with the SAR vertical displacement data to obtain a coseismic vertical displacement model around Palu Bay (Figure 2a). The tsunami recorded at Pantoloan station could be generated by a combination of faulting and landslide. Therefore, the seafloor displacement model estimated from the joint inversion of tsunami waveform and SAR data should be regarded as a static representation of the tsunami source from both the faulting and landslides.

The tsunami source model has a main uplift area (maximum 0.8 m) in the middle of the bay and two subsidence areas near the end and mouth of the bay. The northern and southern parts of the bay are both estimated to have subsided by up to around 1 m, which is consistent with the subsidence pattern determined from SAR data (Figure 2a). The synthetic tsunami waveforms from the tsunami source model can explain well the observed first tsunami cycle (Figure 2c). The simulated later phases do not perfectly match the observed small waves, but the simulated peak amplitudes resemble the observation. The discrepancy in the later phases may indicate that some complexities in the tsunami generation are not captured by the source model. It may also be because the bathymetry and coastal structures around the tide gauge are not represented well enough by the bathymetric grid.

3.2 Validation for the Vertical Displacement Model

The topographic data that were collected before the event was corrected by the vertical displacement model that is based on the SAR data. This corrected topographic data are then used in the tsunami inundation modeling. The ground subsidence along the coast of Palu city of up to 1 m as indicated in the SAR data increases the tsunami potential impact in the region because it lowers the topography. Moreover, the tidal range as recorded at Pantoloan station is ~2 m and the tsunami occurred during high tide at approximately 1 m above mean sea level (Figure 2b); this also increases the tsunami potential impact. The tidal offset of +1 m is included in the simulation. The simulation with the tidal offset can reproduce the observed limit of tsunami inundation (200–400 m) in Silae and Lere areas quite well (Figure 3b), and the simulated tsunami heights in these areas are generally close to the observation. While in Besusu Barat and Talise, the model underestimates the observed tsunami heights and limit of tsunami inundation (Figures S6a and 3b). A tsunami simulation with this tsunami source model but without the tidal offset reduces the simulated limit of tsunami inundation by approximately 20%.

A ferry crew at Wani harbor (Video 3 in Table S1), located 2.5 km northwest the Pantoloan tide gauge station, described that the land was inundated soon after the earthquake by initial waves and then the sea receded significantly. The ferry was dragged by the receding water but remained near the harbor because it was secured to a mooring. The receding water was also recorded by Pantoloan tide gauge. After that, a large tsunami swept the ferry ~50-m inland; this description is consistent with the tide gauge record at the nearby Pantoloan harbor and our simulation result (Figure 2c). At the end of Palu Bay, in Palu city, eyewitness accounts and video footages (e.g., Videos 4 and 5 in Table S1) also show that small tsunami waves hit the coast shortly after the earthquake and that small waves were followed by a large tsunami. Our vertical displacement model can only explain the large tsunami that arrived 5 min after the earthquake (Figure S7) but not smaller waves that arrived earlier. This suggests that additional tsunami sources are required and subaerial-submarine landslides are likely candidates.

4 Discussion

4.1 Landslide Tsunami Contribution

The simulation result using the model for multiple landslides in eight river deltas in Palu Bay cannot reproduce either the amplitude or periods of the largest signals observed within the first 20 min at Pantoloan (Figures S8 and S9). The model explains the observed initial tsunami waves and gives locally large simulated tsunami near the sources but largely underestimate the observed tsunami inundation heights in Palu city (Figures 3c and S6b).

We simulate the tsunami inundation in Palu city using a combination of the vertical displacement model and the multiple landslides model. The simulation shows that in Palu city, the initial tsunami waves can be explained by the three subaerial-submarine landslide models near Palu city. These waves arrive in less than 2 min (Figures S10 and S11), while the large tsunami was generated in the middle of the bay and arrived in around 5 min after the earthquake (Figures S7 and S11). The combined model can explain well the observed limit of tsunami inundation (Figure 3d) and tsunami inundation heights in Palu (Figure 3a). It is important to emphasize that landslide tsunami is not yet commonly considered in tsunami hazard assessments or included in forecasting systems.

4.2 Faulting Mechanism Beneath Palu Bay

The uplift in the middle of Palu Bay indicated by the joint inversion could be a result from either contractional bend or stepover faulting mechanisms. We prefer these two possibilities to that of a concentrated slip with highly variable rake angle on a simple fault geometry like the one in the United States Geological Survey (2018) fault slip model, although it also predicts uplift in the bay. The SAR N-S displacement map can be used to trace the inland Palu-Koro fault (black traces in Figure S12), while the exact fault trace beneath the bay is unknown. Palu-Koro fault may have a contractional bend (purple trace in Figure S12a) in the middle of the bay. A rupture of a contractional bend fault typically has an oblique sense of movement which generates uplift.

Another possible fault mechanism that allows uplift in the middle of the bay is a step over jump with a push-up block between two separate faults. This could be broadly consistent with the multiple dislocations model of Socquet et al. (2006), which has two dislocations on both sides of the bay. The two dislocations could be two active branches which are surface splays of a strike-slip flower structure (Socquet et al., 2006). The inland fault traces from SAR offsets may also suggest a gap of less than 4 km between two separate faults (blue traces in Figure S12b). Considering the fault dimension of 200-km long and 15-km wide from the aftershock distribution, the average stress drop of ~11 MPa calculated (Kanamori & Anderson, 1975; Noda et al., 2013) for this earthquake allows a 4-km step over jump (Liu & Duan, 2015).

5 Conclusions

The estimated coseismic seafloor vertical displacement pattern is complex with up to 0.8-m uplift in the middle of the bay and two subsidence areas on the northern and southern parts of the bay. The tsunami source model reproduces the tsunami waveform at Pantoloan tide gauge station with tsunami wave amplitude of 2 m and a wave height of 4 m. The area of large inferred uplift in the middle of the bay is interpreted to be the main tsunami source region for this event. Such uplift could be either from contractional bend or step over in the middle of the bay. A contractional bend may have oblique thrust motion that produces uplift, while a step over jump mechanism may have a push-up area between the faults. The possibility of a very large submarine landslide in the middle of the bay, with an equivalent effect to the vertical deformation, cannot currently be excluded. To further investigate if the largest tsunami wave was generated by a huge submarine landslide in the middle of the bay, more data from subbottom profiler and multibeam surveys are required.

The coseismic vertical displacement model can reproduce the observed 2- to 4-m tsunami inundation heights above mean sea level and ~400-m limit of inundation in the western part of Palu city but underestimates the observation along the coast east of Palu River. Our simulation results suggest that eight subaerial-submarine landslides located at the river deltas along the coast generated only local tsunami inundations and they could not be the main cause for the devastating tsunami in Palu city. Nevertheless, the initial tsunami waves observed in Palu can be explained by the three subaerial-submarine landslide models near Palu city. We show that the observed tsunami inundation in the city can be explained by a model that combined uplift in Palu Bay and eight landslides at the river deltas. Our simulations also suggest that the high tide (~1 m) and the coseismic subsidence (up to ~1 m) in Palu city during the event are both significantly increased the tsunami impact.

Acknowledgments

We thank BMKG for providing the earthquake data catalog used in this study. Tide gauge data at Pantoloan tide gauge station was provided by the Geospatial Information Agency of Indonesia (BIG) http://tides.big.go.id/. The DEMNAS (BIG) topographic data with grid resolution of 0.27 arc sec can be downloaded from http://tides.big.go.id/DEMNAS/. The BATNAS (BIG) bathymetric data with grid resolution of 6 arc sec can be downloaded from http://tides.big.go.id/DEMNAS/. National tsunami and earthquake damage assessment and base map in Palu are available from BIG https://cloud.big.go.id/. Satellite image data for Palu Bay before and after the earthquake can be downloaded from http://www.digitalglobe.com/ecosystem/open-data. The satellite data are distributed under creative commons attribution noncommercial 4.0 data license, in Open Data Program Digital Globe a Maxar Company Indonesia earthquake and tsunami. LiDAR data for Palu city was provided by Australia-Indonesia Facility for Disaster Reduction (AIFDR). Sentinel-1 satellite data are available from Copernicus (https://scihub.copernicus.eu/). The aftershock data are from the Agency for Meteorology, Climatology, and Geophysics of Indonesia (BMKG) earthquake catalog. We thank the Editor and two anonymous reviewers for their comments and suggestions to improve this article. This work was supported by public research funding from the Government of New Zealand.