Simplified rectangular fault models were adapted for the megathrust with parameters length (km), width (km), strike (°), dip angle (°), depth from the seafloor (km), rake (°), and slip amount (m). Insights from previous seismic tomography profiles (Fan et al., 2017; Fan and Zhao, 2019; Sevilla, 2011) on the geometry of the subducting slab reveal tapering of high P-wave velocity in the northern Negros Trench segment west of Panay Island, which provides limiting constraints on the length of the northern Negros Trench segment. Steepening of the subducting slab was also observed in the seismic tomography profiles around 100 km depth, although this is well beyond global seismogenic zones. Best-fit estimates on the slab dip of the shallow portion of the subducting slab derived from the tomography profiles within 50 km depth range from 9 to 30°. Focal mechanism solutions (ISC data from 1977 to 2023) of reverse-oblique rake values (10 to 170°) (Bondár et al., 2015; Di Giacomo et al., 2015, 2018; Storchak et al., 2013, 2015) also show a wide distribution of dip angles ranging from 9 to 90° (Fig. 2). To account for this uncertainty of the slab dip, we used three dip angles for all source parameters at 10, 20, and 30°.

We utilized a structural-based segmentation in which the megathrust fault length is derived from the orientation variations of the frontal wedge (i.e., Nawanao and Ramos, 2023), and the width and bottom depth are limited by the seismogenic zone. These led to four main segments of the Negros–Sulu subduction zone, namely, NT1, NT2, ST1, and ST2. The NT1 and NT2 are marked by north-northeast and north-northwest orientations, respectively, while the ST1 and ST2 by north–northeast and northeast, respectively (Fig. 3a). Due to the uncertainty of the northern extent of NT1 where P-wave velocity profiles of Fan et al. (2017) and Fan and Zhao (2019) indicate tapering of the slab northward, we included longer (NT1-1) and shorter (NT1-2) rupture lengths for NT1. For the Sulu Trench, an additional longer segment (ST) that combines ST1 and ST2 is also modeled due to their similar northeast trend. This segmentation results in six subfaults, i.e., NT1-1, NT1-2, NT2, ST, ST1, ST2.

A shallow blind rupture is assumed for all scenarios with a top depth of 5 km and a rake of 90° (purely reverse slip). Based on the interquartile range (1.5 IQR) of 207 reverse-oblique earthquakes (Fig. 2), a threshold depth of 50 km was set as the bottom depth of the seismogenic zone. The width of the megathrust is calculated using sine function from the top and bottom depth, with the given the dip angles, and we adopted a saturation limit of 196 km as observed in global subduction zones (Allen and Hayes, 2017; Skarlatoudis et al., 2016; Tajima et al., 2013). The dip angles 10, 20, and 30° correspond to 196 (maximum width limit), 132, and 90 km, respectively (Fig. 3).

To account for the variability of length and width, the area (S) of the fault plane is used as input in the linear rupture-scaling equations (Eqs. 1–2) of Allen and Hayes (2017) to estimate the maximum Mw and the average (Dave) slip (m). The linear area–Mw was preferred over the bilinear area–Mw scaling equation as the former produced slightly higher Mw, adopting a more conservative approach. A total of 18 sets of source parameters were constrained for elastic dislocation and tsunami modeling, with the ST (379 km) and NT1-1 (201 km) having the longest lengths (Table 1). Estimates of the moment magnitude from scaling equations range from Mw 8.0 to 8.9 (±0.27), where the ST segment with the longest fault length has the largest magnitude. These earthquake magnitudes correspond to an average slip of 1.68–6.23 m (±0.21). 1log10S=-3.63+0.96×Mw2log10Dave=-5.05+0.66×Mw

We utilized the 15 arcsec-resolution 2024 General Bathymetric Chart of the Oceans (GEBCO Compilation Group, 2024) bathymetry data as input for the tsunami simulation. To have a higher elevation accuracy of the topography, we resampled the interferometric synthetic aperture radar (IFSAR)-derived 5 m-resolution digital terrain model (DTM) (±1 m vertical accuracy) from the National Mapping and Resource Information Authority (NAMRIA) to 15 arcsec and merged it with the GEBCO bathymetry data. The resulting 15 arcsec IFSAR DTM–bathymetry raster served as the input depth in computing the initial seafloor and seawater displacement for tsunami modeling. As the entire coastal areas surrounding the Sulu Sea were considered for tsunami simulations, high-resolution subgrids for inundation models were not performed due to the absence of high-resolution nearshore bathymetry as well as restrictions in computational capacity and processing time.

The elastic deformation from the finite rectangular fault source is analytically solved through Green's function, assuming a homogeneous isotropic elastic half-space (Okada, 1985) and utilizing the average slip from the scaling equation. The tsunami wave propagation is modeled using two-dimensional nonlinear shallow water equations through the finite difference method (FDM) in a staggered grid deployed in the JAGURS numerical code (Baba et al., 2015, 2017). In JAGURS, both the initial vertical and horizontal seafloor displacement and Boussinesq dispersion effects are accounted for in the tsunami propagation. The modeling assumes a uniform surface roughness (Manning's coefficient = 0.025 s m^−1∕3), and we set a time step (Δt) of 0.5 s to maintain stability conditions based on the maximum depth (d), bathymetry resolution (Δx), and acceleration due to gravity (g) in Eq. (3) (Satake, 2002): 3Δt≤Δx2gd

The tsunami wave propagation time was set to 4 h, and 34 synthetic tide gauge stations (TGS) (Fig. 4 and Table S1 in the Supplement) were positioned across the coastal areas surrounding the Sulu Sea. The coastal areas surrounding the Negros–Sulu Trench are subdivided into 16 regions: I, Visayan Sea; II, Mindoro; III, Coron–Cuyo; IV, West Panay; V, Panay Gulf; VI, Sipalay–Bayawan; VII, Tañon–Cebu Strait; VIII, Bohol Sea; IX, Leyte Gulf; X, Palawan; XI, Cagayancillo; XII, Sindangan–Labason; XIII, Siocon–Sibuco; XIV, Basilan–Sulu; XV, Moro Gulf; and XVI, northeast Borneo. To represent the minimum arrival time, maximum wave height, and flow velocity across the coastlines, 6879 sampling points were set at 10 km intervals, each with a radius of 5 km.

Figures 5 and S1–S5 (in the Supplement) show the representative wave propagation snapshots from rupture scenarios of six segments (NT1-1, NT1-2, NT2, ST, ST1, ST2) based on the 20° slab dip. The initial wave propagation from all ruptures is characterized by strong east–west and northwest–southeast wave crests in the Negros and Sulu Trench, respectively. Short-period wave dispersion and interferences are also prevalent as waves propagate toward shallower depths. In I, III, VII, VIII, and IX, the combined effects of dispersion, diffraction, and refraction caused significant deformation of wave crests as they propagated toward the coasts.

Coastal islands in direct proximity have <2 min arrival time, including V and VI for Negros Trench segments and XII, XIII, and XIV for Sulu Trench segments. Arrival times from Negros Trench segments in XI, XII, XIII, and northeast of X were about 10, 15, 30, and 45 min, respectively. As the initial wave approaches shallow depths, wave speed significantly drops as indicated by the compression of arrival time contours (Fig. 6). For instance, the first wave arrives in the southwest of X about 1 h after the initial displacement, while about 2 h in XVI. The effects of wave refraction, diffraction, and interference are also apparent in the deformed arrival time contours in the majority of coastal regions.

For ST and ST1 segments, the initial wave arrives in VI, XI, IV, and III by 10, 20, 25, and 45 min, respectively. For ST2 segments, arrival times are around 30 min in VI and XI, while around 35 min, 45 min, and 1 h in XVI, V, and III, respectively. Tsunami waves from Sulu Trench segments can also pass through the XIV and propagate across the Celebes Sea, arriving in XV at about 10 min (Fig. 6).

Waveforms from 34 synthetic tide gauge stations (Figs. 7–8 and S6–S7) also indicate the arrival as the first crest or trough. Relatively long-period waves are apparent in TGS 1 Coron, 2 Pandarochan Bay, 13 Himamaylan, 14 Bacolod, and 18 Cebu, while the majority are dominated by short-period waves due to nonlinear and dispersion effects. Tsunami waves from all segments did not propagate in TGS 22 Surigao City within the 4 h simulation. In TGS 33 Jolo, dissipation also led to no tsunami waveforms from NT and ST1, while the ST and ST2 ruptures caused coastal uplift.

The maximum wave height and flow velocity from six segments are shown in Fig. 9 (Figs. S8–S9) and Fig. 10 (Figs. S10–S11), respectively. The highest wave heights (>2 m) are observed in coastal areas parallel to the rupture segments, as also indicated in six tide gauge stations (13, 16, 28, 29, 30, 32). For NT1, tsunami waves reached up to 4–5 m in VI. For the NT2 segment, tsunami waves of up to 4 m are localized in VI. For ST and ST1 segments, the highest wave heights are located in XII (8 m) and XIII (5–7 m). Tsunami waves from ST segments propagate across the Celebes Sea, reaching 3–7 m in XV.

Coastal areas on the rear side and parallel to the trench segment also have relatively high tsunami waves. Negros Trench segments produced up to 2–3 m of tsunami wave heights in III and 1–2 m in XI and northeast of X. Sulu Trench segments also produced up to 3–4 m in X and up to 1–2 m in XI and III. The majority of the central Philippines (I, VII, VIII, IX), as well as in XVI, have wave heights of <1 m, except in some embayments and narrow channels where amplifications can reach 1–2 m.

The maximum tsunami wave height and flow velocity are coupled, i.e., coastal areas with the highest tsunami waves also have the highest flow velocity. For Negros Trench segments, the coastal areas with high tsunami waves in VI correspond to flow velocities of 3–7 m s⁻¹, while reaching up to 3 m s⁻¹ in III. For Sulu Trench segments, coastal areas in XII, XIII, XIV, and XV with the highest tsunami waves are exposed to a flow velocity of 3–6 m s⁻¹. In coastal areas of northeast Palawan, the flow velocity reached 3–4 m s⁻¹. The majority of coastal areas with <1 m tsunami wave height correspond to a maximum velocity of <1 m s⁻¹.

Table 2 summarizes the maximum, minimum, and average differences of the maximum tsunami wave height and flow velocity from megathrust segments at 10° dip compared to 20 and 30°. A shallow-dipping megathrust tends to produce the largest earthquakes (Bletery et al., 2016; Schellart and Rawlinson, 2013), hence the use of the 10°-dip rupture parameter as a reference. This tendency is highlighted by the positive average and maximum differences among parameters. However, in certain regions, steeper initial seawater displacement from rupture parameters with 20 and 30° dip angles produced waveforms that preferentially led to higher maximum wave height and flow velocity as represented by the negative minimum differences. The spatial distribution of the maximum wave height and wave velocity differences among parameters is indicated in Figs. 11 and S12–S14.

For Negros Trench segments, the range of maximum difference is 0.934–2.507 m and 0.968–1.811 m s⁻¹ for maximum wave height and flow velocity, respectively, while the average differences are 0.008–0.104 m and 0.002–0.071 m s⁻¹. The minimum differences range from -0.529 to -1.141 m (maximum wave height) and -1.099 to -2.140 m s⁻¹ (maximum flow velocity). For Sulu Trench segments, the differences are higher than those from Negros Trench segments, with a range of 2.289–5.622 m (maximum wave height) and 2.369–4.509 m s⁻¹ (maximum flow velocity) for the maximum, 0.014–0.116 m and 0.003–0.017 m s⁻¹ for the average, and -0.481 to -1.890 m and -1.020 to -2.164 m s⁻¹ for the minimum.

Wave directivity mainly controls the distribution of the highest tsunami wave heights, which are coastal regions directly facing (i.e., V, VI, XII, XIII, XIV, XV) and on the rear side (III, X) of the segments. As tsunami waves propagate in the interior central Philippines, wave interactions, refraction, diffraction, reflection, and nonlinear effects result in significant dissipation, with the majority of coastal areas in I, VII, XIII, and IX having <1 m wave height. However, the morphology of coastlines, especially along embayments, can amplify tsunamis due to constructive resonance, which can even generate up to 2 m in interior islands (Fig. 12b–f). Previous studies have also acknowledged the effects of morphology on tsunami wave resonance (e.g., Aranguiz et al., 2019; Roeber et al., 2010; Vela et al., 2014; Yamazaki and Cheung, 2011).

The foremost islands can also function as a natural barrier for the rear islands as they diffract and attenuate the tsunami waves. This case is apparent for the higher tsunami waves in III compared to its rear islands (Fig. 12g). Another factor playing a role in the resulting tsunami waves is the nearshore width along the coasts, as nonlinear shallow-water waves dominate. Arbitrarily setting the distance between the -50 and 0 m depth contour as a reference for nearshore width in 6879 sampling points, the nearshore width interquartile range of the coastal areas with ≥2 m maximum wave height is 2–4.5 km with a maximum bound of 8 km (Fig. 13). On the other hand, coastal areas with <2 m maximum wave height have a nearshore width interquartile range of 2.8–10.6 km with a maximum bound of 22 km. All coastal areas with nearshore width ranging from 30 to 77 km also have <2 m maximum wave height. This indicates that as the nearshore width widens (e.g., >20 km), wave dissipation increases, reducing the tsunami wave as it reaches the shoreline. This is exemplified in the difference in distribution of maximum tsunami wave heights between the northern and southern coastline of XVI (Fig. 12h).

Accounting for all 18 megathrust rupture scenarios, Figs. 14, S15, and S16 present the maximum wave height (m), flow velocity (m s⁻¹), and minimum arrival times (min), respectively, in coastal areas surrounding the Sulu Sea. Of the 6879 sampling points, 8 % have a maximum wave height of ≥2 m. Sindangan–Labason (XII) and Siocon–Sibuco (XIII), Sipalay–Bayawan (VI), Moro Gulf (XV), and Basilan–Sulu (XIV) have the highest exposure of 5–8 m wave height and 5–7 m s⁻¹ flow velocity. Certain coastlines in Palawan (X), Panay Gulf (V), Coron–Cuyo (III), West Panay (IV), northeast Borneo (XVI), and Bohol Sea (III) are potentially exposed to 2–4 m wave height and 2–4 m s⁻¹ flow velocity. The majority of Tañon–Cebu Strait (VII), Mindoro (II), Leyte Gulf (IX), and Visayan Sea (I) coastal regions have a maximum wave height and flow velocity of <1 m and <1 m s⁻¹, respectively, with minor amplification of up to 2 m and 2 m s⁻¹ in a few areas. Comparisons of wave heights across 16 coastal regions are also shown in Fig. 15 with an average ranging from 0.1 to 2.1 m and standard deviations of ±0.06–0.77 m.

The majority of coastal regions with the highest tsunami waves (XII, XIII, VI, XV, XIV) also have the shortest arrival times of <2 min due to their proximity to the rupture segments. This short arrival requires the need for quick evacuation strategies. In certain areas where established evacuation sites are far, vertical evacuation, i.e., reaching the closest elevated and structurally strong location, may be the most viable option. While the variability of coastal surface roughness is not accounted for in these models, pre-existing natural and man-made barriers may delay or inhibit inundation, providing additional time for evacuation.

We compared the maximum wave heights from the models to the eyewitness accounts of historical tsunamis in the Sulu and Celebes seas from Bautista et al. (2012). Although the source mechanisms of the majority of earthquakes were unknown, we can attribute them to megathrust, upper-plate, outer-rise, or intraplate ruptures, depending on the hypocenter when available and the position of the estimated epicenter relative to the trench, as labeled in Fig. 15. The estimated epicenter of the 1897 Ms 7.5 event was along the Sulu Trench with a 2 m maximum wave height in the majority of the Sulu Sea basin (III, IV, VI, X, XIII, XV, XVI) and up to 6 m in coastal region XIV. In III, IV, VI, X, XIII, XV, and XVI, the highest modeled tsunami wave in this study is slightly greater than the eyewitness accounts. However, in XIV where a 6 m wave height was estimated, the highest modeled tsunami wave is slightly lower at 5.4 m. Eyewitness accounts of the more localized tsunamis by the 2012 Ms 6.9, 1925 Ms 6.8, and 1948 Ms 8.2, reported up to 2 m wave height, which is lower than the highest modeled tsunami. In Moro Gulf (XV), six historical tsunamigenic earthquakes (2002 Ms 7.2, 1976 Ms 8.1, 1928 Ms 7.5, 1923 Ms 7.2, 1918 Ms 8, 1917 Ms 6.4) were generated along the Cotabato Trench (CT) with eyewitness accounts reporting tsunami wave heights ranging from 1 to 9 m. In this region, the highest modeled tsunami from the Sulu Trench reached 6 m. This indicates that this coastal region is not only susceptible to the Cotabato subduction zone rupture but also to that of the Sulu subduction zone. In addition, prehistoric extreme wave deposits that may be recovered from this coastal region (e.g., Claro et al., 2021) may also be associated with the Sulu megathrust rupture and tsunami.

It should be noted, however, that these eyewitness accounts pose uncertainties and are only available in certain coastal areas that have been inhabited before these events. For events before the 1960s, the estimated earthquake magnitudes are based on reported damages or anecdotes. The magnitude of the 1897 Ms 7.5, for example, may have been underestimated. This event may have also been a tsunami earthquake which generates large tsunami waves compared than its earthquake magnitude (e.g., Tanioka and Seno, 2001), or an earthquake-triggered submarine landslide also contributed to the basin-wide tsunamis as these submarine features are prevalent in this active margin (Nawanao and Ramos, 2023). As our models assumed blind megathrust rupture, these historical events may also have been triggered by other rupture mechanisms (e.g., outer rise, tsunami earthquake, intraplate, upper-plate earthquakes) with a different set of parameter estimations. With the scarcity of information, these historical accounts are still valuable to compare with tsunami models. What is consistent is that the coastal positions and configuration relative to the trench dictate where the highest exposures are identified, i.e., controlled by wave directivity.

A major limitation of this study is the uniform slip distribution along the megathrust fault plane, which is based on the average slip from the scaling equation. Earthquake slip distributions are typically heterogeneous, and they control whether the initial seafloor displacement can generate tsunamis. Megathrust ruptures with larger slip distributions downdip but lower on the updip can produce lower tsunami waves (e.g., Delouis et al., 2010; Fujii et al., 2020), which can lead to overestimation from these rupture scenarios. Conversely, large slip updip, especially for trench-breaching and splay faulting rupture types, can produce larger seafloor displacement (e.g., Baba et al., 2006; Fujiwara et al., 2011) that may underestimate the resulting tsunami wave heights from this study. Splay faulting and trench-breaching megathrust scenarios are also not accounted for in this study, as these modes of rupture are rarely reported, but they must be considered in future tsunami hazard assessments in this subduction zone. While the geometry of the Negros–Sulu subduction zone is currently not well constrained, the generally thick sediment cover overlying the Negros and Sulu subduction zone (Fig. 1) may decrease the slab dip and can widen the seismogenic zone (Brizzi et al., 2020). With the bathymetry data used in this study and the lack of variable and asperity-based slip distribution when considering megathrust ruptures, we recommend that the maximum wave heights are treated as minimum and that the actual highest tsunami waves are more likely underestimated.

Future studies along the Negros–Sulu subduction zone should aim to constrain both the geometry of the subduction slab and coupling ratios where slip deficits can be derived as applied in other subduction zones (e.g., Chlieh et al., 2008; Métois et al., 2013, 2016; Michel et al., 2019; Moreno et al., 2010; Ozawa et al., 2012). The role of submarine landslides in tsunami generation can also be explored in future studies as these features are prevalent in this active margin (Nawanao and Ramos, 2023). Submarine landslides have been shown to amplify tsunamis when the failure is coeval to megathrust earthquakes or can also be the main driver of localized but large tsunamis (e.g., Heinrich et al., 2001; Imamura et al., 1995; Nakata et al., 2020; Ramirez et al., 2022; Schambach et al., 2020; Takagi et al., 2019). Other mechanisms such as tsunami earthquakes and outer rise faulting may also be possible where focal mechanism solutions indicate normal faulting along these sections of the subduction zone (Fig. 2).

Updated seismic reflection and geophysical surveys across the Negros–Sulu Trench may also reveal underlying complexities that can influence the potential to generate megathrust rupture, including splay faults, fluids, and subducting seamounts. Gathering high-resolution bathymetry data is necessary to capture the local-scale wave amplification and deformation. The accuracy of nearshore bathymetry connecting inland and offshore must also be assessed, as the nearshore width shown in this study has significant effects on the resulting maximum tsunami waves along the coasts. Furthermore, mapping of pre-existing natural and man-made barriers for site-specific simulation and vulnerability assessment will also be important to assess their effectiveness in minimizing tsunami inundation.

The identified localities with the highest exposure to tsunami waves are potential sites (e.g., V, VI, XII, XIII) in searching for paleotsunami deposits that may serve as records for prehistoric earthquakes. Most importantly, the estimated moment magnitude, maximum tsunami wave heights, flow velocity, and minimum arrival times can provide invaluable information in creating strategies and policies for local government offices and coastal communities to mitigate hazard risk. These include information dissemination in coastal communities, delineation of evacuation routes and sites, and proper engineering designs such as coastal forests, green embankments, seawalls, offshore breakwaters, and dikes (e.g., Benazir et al., 2024; Dengler, 2005; Oetjen et al., 2022; Strusińska-Correia, 2017). In addition, as the majority of these coastal areas tend to have short arrival times, site-specific inundation modeling and geotechnical analysis are recommended for identifying and creating infrastructure for vertical evacuation.