Ji, C., Helmberger, D. V., Wald, D. J. & Ma, K. F. Slip history and dynamic implications of the 1999 Chi‐Chi, Taiwan, earthquake. J. Geophys. Res.: Solid Earth 108, 2412 (2003).

Ji, C., Wald, D. J. & Helmberger, D. V. Source description of the 1999 Hector Mine, California, earthquake, Part I: Wavelet domain inversion theory and resolution analysis. Bull. Seism. Soc. Am. 92, 1192–1207 (2002). The specific geometry of the inferred slow thrust faulting, with along-trench compression in the upper plate, is surprising, and if this model is correct, it comprises an unexpected tsunami hazard in the region. The presence of weak sediments near the shelf break may have influenced slow-slip rupture with 15 m of slip over ~300 s, as found for this successful model, which has fault dimensions of 20 km × 20 km. Such large slip over localized area has been observed in shallow megathrusts environments, typically involving a tsunami earthquake 23 or aseismic transient slip 24. Transpressional environments have been observed to have large slow thrust faulting along with dominant strike-slip faulting as well 25. Models with a larger fault area (30 km × 30 km; 40 km × 40 km) and lower slip (7 m, 4 m) that have similar total moment may be viable, but it is challenging to fit all of the tsunami data as well as in Fig. 8 (e.g., Supplementary Figs. 16, 17). While lower slip is appealing, larger fault dimensions imply more observable faulting in the wedge, for which available bathymetry and reflection profiling now provide independent evidence. The non-unique modeling suggests slow slip of from 4 to 15 m on the westward-dipping upper plate thrust fault. Ye, L. et al. Rupture model for the 29 July 2021 M W 8.2 Chignik, Alaska earthquake constrained by seismic, geodetic, and tsunami observations. J. Geophys. Res.: Solid Earth 127, e2021JB023676 (2022). To match the observed tsunami waveforms, an additional stronger source of tsunami excitation is required, but the two-fault fast-slip model alone already adequately accounts for the full set of seismic and geodetic data. This holds even for 256 s period Rayleigh and Love waves from global stations, for which the two-fault model predicts the four-lobed radiation patterns well (Supplementary Fig. 6). From the DART waveform comparisons, the additional source must have a 4–5 min delay relative to the initial compound faulting to account for the larger second peak, yet the nearby geodetic ground motions show no deformation after the first 60 s. The earlier deformation is well accounted for by the two-fault fast-slip model (Fig. 3c). Because the tsunami wave period is inversely proportional to the square root of the source water depth, the excitation most likely includes uplift of the sea surface over the continental slope to account for the impulsive peak along with some drawdown near the shelf break to match the wide trough that follows immediately. Li, S. & Freymueller, J. T. Spatial variation of slip behavior beneath the Alaska Peninsula along Alaska-Aleutian subduction zone. Geophys. Res. Lett. 45, 3453–3460 (2018).Tanioka, Y. & Satake, K. Tsunami generation by horizontal displacement of ocean bottom. Geophys. Res. Lett. 23, 861–864 (1996). Okal, E. A. & Hébert, H. Far-field simulation of the 1946 Aleutian tsunami. Geophys. J. Inter. 169, 1229–1238 (2007). Bai, Y., Ye, L., Yamazaki, Y., Lay, T. & Cheung, K. F. The 4 May 2018 M W 6.9 Hawaii Island earthquake and implications for tsunami hazards. Geophys. Res. Lett. 45, 11,040–11,049 (2018). Yamazaki, Y., Kowalik, Z. & Cheung, K. F. Depth-integrated, non-hydrostatic model for wave breaking and run-up. Int. J. Num. Meth. Fluids 61, 473–497 (2009).

The table below contains all postcodes on a two day service. Please note all deliveries to Northern Ireland are also on a 3-5 days service. As seismic and geodetic data can provide complementary constraints on the rupture process, we used both data types to invert the rupture process of the 19 October 2020 event assuming first one and then two fault segments. We performed non-linear finite fault inversions 29, 30, involving the joint analysis of coseismic static offsets, hr-GNSS time series, and seismic waveforms. A simulated annealing algorithm was used to solve for the slip magnitude and direction, rise time, and average rupture velocity for subfaults on the two segments. For each parameter, we set specific search bounds and intervals. The subfault size is chosen as 5 km × 5 km, and the rake angles on the two fault segments are constrained to be right-lateral purely strike-slip and purely dip-slip, respectively. We allowed both the rise and fall intervals of the asymmetric slip rate function for each subfault to vary from 0.6 to 6.0 s; thus, the corresponding slip duration for each subfault is limited between 1.2 and 12 s. We let the slip vary from 0.0 to 8.0 m, and the average rupture velocity is allowed to vary from 0.5 to 3.0 km/s. Green’s functions for static displacements and seismic waveforms are computed using a 1-D layered velocity model 31. Equal weighting among the data functionals for GNSS statics and seismic waveforms was used in this study. Tsunami modeling Hayes, G. P. et al. Slab2, a comprehensive subduction zone geometry model. Science 362, 58–61 (2018).Cheung, K. F., Bai, Y. & Yamazaki, Y. Surges around the Hawaiian Islands from the 2011 Tohoku tsunami. J. Geophys. Res.: Oceans 118, 5703–5719 (2013). Herman, M. W. & Furlong, K. P. Triggering an unexpected earthquake in an uncoupled subduction zone. Sci. Adv. 7, eabf7590 (2021). Crowell, B. W. & Melgar, D. Slipping the Shumagin gap: A kinematic coseismic and early afterslip model of the M W 7.8 Simeonof Island, Alaska, earthquake. Geophys. Res. Lett. 47, e2020GL090308 (2020). Dziewonski, A. M. & Anderson, D. L. Preliminary reference Earth model. Phys. Earth Planet. Inter. 25, 297–356 (1981).

An upper plate splay-fault model for the additional source of tsunami waves involves a compact 20 km × 30 km slip patch with an upper edge 3 km deep, and strike 250°, dip 35°, and rake 90°, with 12 m of pure thrust slip. The slow-fault ruptures at the same time as the initiation of the earthquake and lasts for 5 min. Assuming a rigidity of 30 GPa, appropriate for the shallow megathrust environment, the seismic moment is 2.16 × 10 20 Nm ( M W 7.49). The computed seafloor deformations for the two-fault coseismic rupture and the slow thrust slip on the splay patch are shown in Supplementary Fig. 10, separately and combined. The thrust splay patch is located near the shelf break and similar to the dipole fitting has about 20 km absolute uncertainty, but cannot locate significantly out onto the continental slope, as the tsunami excitation changes rapidly along the slope and incompatible waveforms are produced at the DART stations. The resulting seafloor deformation again resembles a scaled-up version of the 2-fault model with uplift and subsidence straddled across the shelf break. Comparisons of the observed and computed tsunami signals for the three-fault model are shown in Supplementary Fig. 11, with clear uniform improvement relative to the two-fault solution in Fig. 4. The fits are slightly improved in comparison to those for the optimal megathrust slow-slip model in Supplementary Fig. 9. The large second arrival and the following trough in the DART waveforms are matched well by the slow-slip event. The computed tsunami waves from the two sources are out-of-phase in Hawaii waters and the matching with the tide gauge records through destructive interference is remarkable (Supplementary Fig. 11). Again, we reject this specific model despite its ability to match the tsunami data because it predicts larger dynamic displacements at GNSS stations AC12 and AC28 (Supplementary Fig. 10), which are not observed after the motions from the fast rupture. Coulomb failure stress Given the guidance provided by the simple dipole modeling, we considered physical fault dislocation models for plausible geometries that can match the salient features of seafloor deformation from the dipole model that leads to successful match of the tsunami waveforms. This includes simultaneous assessment of the seismic and geodetic motions produced by such models for the sensitive high-rate GNSS recordings at nearby stations AC12 and AC28. The latter constraint is very important; there is essentially no geodetic or seismic signature of the second (dominant) tsunami source, and models that violate this can be rejected with confidence. We considered appropriately placed models with delayed slow thrust slip on the shallow megathrust (Methods, Supplementary Figs. 8, 9) or slow thrust slip on an upper plate splay fault with a strike parallel to the trench (Methods, Supplementary Figs. 10, 11) and allowed sufficiently long source process times to obscure the seismic and geodetic expressions while giving strong tsunami excitation, finding models that match the tsunami signals by extensive searches over model parameters (fault dimensions, slip, absolute location, etc.). However, those models that do match the tsunami observations acceptably all badly violate the geodetic observations at AC12 and AC28 (Supplementary Figs. 8, 10). This eliminates the more obvious candidate model geometries. Successful slow-slip faulting geometry Zhao, B., Bürgmann, R., Wang, D., Zhang, J. & Yu, J. Aseismic slip and recent ruptures of persistent asperities along the Alaska-Aleutian subduction zone. Nat. Comm. 13, 3098 (2022). Li, L. & K. F. Cheung, K. F. Numerical dispersion in non-hydrostatic modeling of long-wave propagation. Ocean Modelling 138, 68–87 (2019). Drooff, C. & Freymueller, J. T. New constraints on slip deficit on the Aleutian megathrust and inflation at Mt. Veniaminof, Alaska from repeat GPS measurements. Geophys. Res. Lett. 48, e2020GL091787 (2021).Ye, L., Lay, T., Kanamori, H., Yamazaki, Y. & Cheung, K. F. The 22 July 2020 M W 7.8 Shumagin seismic gap earthquake: Partial rupture of a weakly coupled megathrust. Earth Planet. Sci. Lett. 562, 116879 (2021). For the intraslab fast-slip strike-slip fault, computations use seismic moment M 0 = 2.43 × 10 20 Nm, strike 350°, dip 50°, rake 173°, and depth 35.5 km. For the upper plate fast-slip oblique normal fault, computations use M 0 = 0.29 × 10 20 Nm, strike 260°, dip 35°, rake 225°, and depth 15 km. For the upper plate slow-slip thrust fault, computations use M 0 = 1.8 × 10 20 Nm, W = 20 km, L = 20 km, slip 15 m, strike 190°, dip 30°, rake 90°, and depth 8 km. The rigidity used for the strike-slip faulting was 5.4 GPa, and it was 3.2 GPa for the oblique faulting and 3.0 GPa for the thrust faulting. Slow megathrust rupture Xu, W. et al. Transpressional rupture cascade of the 2016 M W 7.8 Kaikoura earthquake, New Zealand. J. Geophys. Res.: Solid Earth 123, 2396–2409 (2018). Yamazaki, Y., Cheung, K. F. & Lay, T. Modeling of the 2011 Tohoku near-field tsunami from finite-fault inversion of seismic waves. Bull. Seism. Soc. Am. 103, 1444–1455 (2013). Bai, Y., Liu, C., Lay, T., Cheung, K. F. & Ye, L. Optimizing a model of coseismic rupture for the 22 July 2020 M W 7.8 Simeonof earthquake by exploiting acute sensitivity of tsunami excitation across the shelf break. J. Geophys. Res.: Solid Earth 127, e2022JB024484 (2022).

Mulia, I. E., Heidarzadeh, M. & Satake, K. Effects of depth of fault slip and continental shelf geometry on the generation of anomalously long-period tsunami by the July 2020 M W 7.8 Shumagin (Alaska) earthquake. Geophys. Res. Lett. 49, e2021GL094937 (2022). Four levels of telescopic grids are needed to model the tsunami from the sources with increasing resolution to the Kahului tide gauge. An additional level is needed to resolve the more complex waterways leading to Hilo, King Cove, and Sand Point. Supplementary Fig. 7 shows the layout of the computational grid systems. The level-1 grid extends across the North Pacific at 2-arcmin (~3700 m) resolution, which gives an adequate description of large-scale bathymetric features and optimal dispersion properties for modeling of trans-oceanic tsunami propagation with NEOWAVE 35. The level-2 grids resolve the insular shelves along the Hawaiian Islands at 24-arcsec (~740 m) and the continental shelf of the Alaska Peninsula at 30-arcsec (~925 m), while providing a transition to the level-3 grids for the respective islands or coastal regions at 6-arcsec (~185 m) resolution. The finest grids at levels 4 or 5 resolve the harbors where the tide gauges are located at 0.3-arcsec (9.25 m) or 0.4 arcsec (12.3 m). A Manning number of 0.025 accounts for the sub-grid roughness at the harbors. The digital elevation model includes GEBCO at 30-arcsec (~3700 m) resolution for the North Pacific, multibeam and LiDAR data at 50 m and ~3 m in the Hawaii region, and NCEI King Cove 8/15-arcsec dataset and Sand Point V2 1/3-arcsec dataset, which also covers the Shumagin Islands. Long-period spectral analysis Fukao, Y. et al. Detection of “Rapid” aseismic slip at the Izu-Bonin trench. J. Geophys. Res.: Solid Earth 126, e2021JB022132 (2021). A plate boundary thrust-fault model for the additional source of tsunamis involves a compact 20 km × 20 km slip patch with an upper edge 22 km deep, and strike 250°, dip 12°, and rake 90°, with 16 m of pure thrust slip. The slow-fault ruptures 30 s after the initiation of the earthquake and lasts for 5 min. The time-varying seafloor deformation of the slow-slip event is approximated by the Okada solution at each computational time step together with those from the fast-slip event, and the associated evolution of the tsunami is dynamically and internally resolved by NEOWAVE driven by the prescribed kinematic seafloor conditions to fit the DART records. Assuming a rigidity of 30 GPa, appropriate for the shallow megathrust environment, the seismic moment is 1.92 × 10 20 Nm ( M W 7.46). The computed seafloor deformations for the two-fault coseismic rupture and the delayed slow slip on the thrust patch are shown in Supplementary Fig. 8, separately and combined. The thrust slip patch is located near the shelf break and similar to the dipole fitting has about 20 km absolute uncertainty, but cannot locate significantly out onto the continental slope, as the tsunami excitation changes rapidly along the slope and incompatible waveforms are produced at the DART stations. The resulting seafloor deformation resembles a scaled-up version of the 2-fault model with uplift and subsidence straddled across the shelf break. Comparisons of the observed and computed tsunami signals for the three-fault model are shown in Supplementary Fig. 9, with clear evidence of uniform improvement relative to the two-fault solution in Fig. 4. The fits are slightly improved in comparison to those for the optimal dipole model in Fig. 6. The large second arrival and the following trough in the DART waveforms are matched well by the slow-slip event. The tide gauge records, which were not used in the source deduction, provide independent validation of the model results. In particular, the computed tsunami waves from the two sources are out-of-phase in Hawaii waters and the matching with the tide gauge records through destructive interference is remarkable (Supplementary Fig. 9). We reject this specific model despite its ability to match the tsunami data because it predicts larger dynamic displacements at GNSS stations AC12 and AC28 (Supplementary Fig. 8), which are not observed after the motions from the fast rupture. Slow splay fault rupture Scholz, C. H. The Mechanics of Earthquakes and Faulting. 439 (Cambridge Univ. Press, New York, 1990).Yamazaki, Y., Cheung, K. F. & Kowalik, Z. Depth-integrated, non-hydrostatic model with grid nesting for tsunami generation, propagation, and run-up. Int. J. Num. Meth. Fluids 67, 2081–2107 (2011). Xiao, Z. et al. The deep Shumagin gap filled: Kinematic rupture model and slip budget analysis of the 2020 M W 7.8 Simeonof earthquake constrained by GNSS, global seismic waveforms, and floating InSAR. Earth Planet. Sci. Lett. 576, 117241 (2021). Niazi, M. & Chun, K. Y. Crustal structure in the southern Bering Shelf and the Alaska Peninsula from inversion of surface-wave dispersion data. Bull. Seism. Soc. Amer. 79, 1883–1893 (1989). We select 62 P and 50 SH broadband recordings from the Incorporated Research Institutions for Seismology (IRIS) data management center with well-distributed azimuthal coverage at teleseismic epicentral distances between 30° and 90° (station distributions and data are shown in Supplementary Fig. 3). Instrument responses are removed to obtain ground velocities in the passband 1–300 s with waveform durations of 100 s. We precisely aligned P and SH wave initial motions manually.