## Introduction

Rough subducting topography is invoked to explain shallow megathrust seismogenesis1,2,3,4. Particularly, in subduction systems with sediment-starved trenches5, the subducting topography plays an important role in determining interplate coupling and defining upper-plate tectonism6. In turn, upper-plate tectonics and the distribution of fracturing with depth determine the depth distribution of upper-plate elastic rock properties, key in controlling, to a large extent, the dynamic behaviour of megathrust earthquakes7. Yet, the linkage between upper-plate elastic structure and subducting topography, and its role in shallow tsunamigenic ruptures have not been quantified. Here we focus on the Mexican Pacific coast, offshore Jalisco and Colima, which is defined by the subduction of the Rivera oceanic plate beneath the North American (NA) plate with a convergence rate that increases from 1.5–2 cm yr−1 near Puerto Vallarta in the north to 3.5–4.4 cm yr−1 in the south8 (Fig. 1a). From north to south, the upper plate of this subduction system exhibits two different tectonic regimes. From the northernmost region of the Middle America Trench (MAT) to 19.2°N latitude, the continental margin is controlled by a subduction-accretion regime as evidenced by a well-developed accretionary prism observed from seismic images9,10. South of 19.2°N, the margin is highly eroded and non-accretionary, and it is characterised by several slope failure scars visible from bathymetry11 (Fig. 2c). Upper-plate submersible observations along this southern region provide evidence of episodic tectonic subsidence during the Neogene supporting that subduction-erosion controls upper-plate tectonics in this section of the12.

Tectonic regime variations along the margin correlate with along-strike topographic variations of the incoming Rivera plate (Fig. 1). Subduction accretion along the margin occurs in a region facing a comparatively smoother incoming oceanic Rivera plate, characterised by localised NW-SE trending seamount chains seemingly formed at the Pacific-Rivera Rise, and NNW-SSE trending normal faulting related to the bending of the incoming plate13 (Fig. 1b). In its southern region, the incoming Rivera plate features a rougher topography (Fig. 1) represented by sharp NW-SE trending ridges extending through the diffuse plate boundary between the Rivera and Cocos oceanic plates (Fig. 1b). According to plate reconstructions and magnetic lineaments, these ridges were not generated at the Pacific-Rivera Rise (Fig. 1), but formed during the northward propagation of the East Pacific Rise (EPR) at 1.7 Ma, segmented from the main EPR by the Rivera Fracture Zone at 0.95 Ma, and finally incorporated in the Rivera plate motion14,15,16.

Interplate seismicity in the southern subduction system is characterised by large (Mw ~8) shallow and tsunamigenic megathrust earthquakes, such as the two events in 1932 (Ms 7.8 and 8.0)17, and the Mw 8, 1995 Colima-Jalisco earthquake18,19. These events generated devastating tsunami waves indicating that the rupture reached the shallow portion of the plate interface20 (Fig. 1). In some cases, even buried megathrust aftershocks seem to have triggered localised tsunamis that are either explained by submarine landslides21 or hypothetical splay faulting22. In particular, the 1995 event released one-third of the total seismic moment near the trench18, producing the largest fault displacement at 10–15 km from the trench (Fig. 1b), and unusual near-trench aftershock activity (Fig. 1b)19. The shallow seismogenic character of this region was tentatively attributed to the lack of sediments with high pore pressure19, but the absence of geophysical observations yields large unknowns regarding the main contributing factors controlling the generation of large and shallow tsunamigenic coseismic slip. Here, we use spatially coincident 2D multichannel seismic (MCS) and wide-angle seismic (WAS) data acquired during the TSUJAL experiment in 201423 to investigate the shallow tectonic and elastic structure of the margin, and its role in the generation of shallow tsunamigenic megathrust ruptures (Fig. 1).

## Results

### Velocity structure and interplate topography

We apply seismic imaging techniques to obtain a 2D P-wave velocity (Vp) model of the upper plate and the geometry of the interplate (Fig. 2), and a spatially coincident pre-stack depth migrated (PSDM) seismic section of the first 20 km of the margin (Fig. 3) (Methods; Supplementary Figs. 16). From top to bottom, the velocity structure of the upper plate is characterised by <1-km-thick sedimentary layer with Vp < 2.5 km s−1 ± 0.1 km s−1 (Fig. 2a, b). Beneath, the sediment-basement boundary is defined by a sharp (1–2 s−1) velocity contrast (Fig. 2a). Overall, basement velocity increases downdip, as observed in subduction zones worldwide7, reaching maximum values of 5.0–5.5 km s−1 consistent with the presence of plutonic rocks as evidenced by seabed rock sampling12 (Fig. 2a). The toe of the upper plate (<20 km from the trench) is characterised by Vp < 4.5 km s−1 (Fig. 2a) and a prominent low-velocity anomaly (Vp < 4.0 ± 0.15-0.1 km s−1) at 15 km from the trench (Fig. 2a). The low-velocity body coincides with a highly fractured media observed in the PSDM image (Fig. 3) and spatially correlates with a prominent interplate high that defines a 2-km-high ~5 km-wide body (Fig. 2a). The interplate reflector is resolved as a ~8.5° SE-dipping irregular interface, and its topographic variations are within the range of our tomographic resolution (Methods; Supplementary Figs. 5, 7). Overall, the structure of the shallow subduction zone inferred here shows a more heterogeneous upper-plate velocity structure than previous forward models derived from land stations installed >60 km away from the trench11.

### The interplay between subducting ridges, upper and lower plate tectonics and margin erosion

The combination of the tomographic model with the PSDM image (Fig. 3) shows that interplate topographic variations relate to the relief of the incoming oceanic plate. A comparison between the inverted interplate reflector with the incoming bathymetry of the Rivera Plate (Fig. 2d) shows that the subducting topographic high is morphologically (height and width) comparable to the incoming oceanic ridges of the Rivera plate. Based on these results, and the lack of seamount chains in the southern portion of the Rivera plate (Fig. 1a), we interpret the subducted topography as a trench-parallel ridge subducting beneath the NA plate. This interpretation is consistent with the hypothesis that the ancient EPR fabric, observed in the outer rise (Fig. 1b), is subducting16.

Our geophysical results show that upper-plate tectonics is highly dependent on the subducted oceanic relief. This way, the upper plate features pervasive normal faulting and localised top of the basement depression at the wake of the subducted ridge indicating the collapse of upper-plate material (CDP4000-4500, Fig. 3a), while it is locally uplifted and highly fractured atop the ridge. Collapsing upper-plate material may seal upper-plate fractures previously generated by the subducting ridge and enhance compaction. Such a hypothesis would explain the higher upper-plate basement velocity in the wake of the ridge than above it (Fig. 3b). The complex network of fractures atop the ridge may prevent the amount of uplifting to be equivalent to the height of the subducted topography. Yet, margin uplifting is enough to increase the dip of the slope towards the trench from 5° to 14° (Fig. 3). Slope oversteepening is proposed to be the precursor for slope failure through submarine landslides and control margin erosion elsewhere along the MAT24. Consistent with this, at the same distance from the trench that we observe slope oversteepening, bathymetric data show several NW-SE trending margin scars towards the north and south of our seismic line (Fig. 2c), indicating the occurrence of past slope failure events.

The two-dimensional upper-plate deformation pattern observed here has been reported in Central and South America subduction zones and is mostly attributed to seamount subduction6,25,26. However, modelling results demonstrate that the three-dimensional structural pattern depends on the shape and size of the subducting body27,28. The erosive pattern associated with isolated subducting seamount causes localised basal erosion of the upper plate and localised embayment of the margin6,27,28, a structural framework somewhat different from that observed here. In the study area, margin erosion occurs over hundreds of kilometres along the trench axis, indicating that upper-plate basal erosion occurs similarly along strike, supporting the existence of subducting trench-parallel elongated ridges.

The subduction of rough-spreading fabric has contributed to the erosion of margins worldwide29,30. In some regions, the roughness of the lower plate is enhanced by the presence of bending-related faulting13. In northern Chile, the topographic roughness of the lower plate is enhanced by bending-related faulting oblique to the spreading fabric, as it induces up to 800 m of vertical seafloor displacement29. In our study area, the rough-spreading fabric and the bending-related faulting appear to have the same NW-SE trend in the southern region of the Rivera plate (Fig. 1b). Thus, it is likely that such faulting modifies the topography of the spreading fabric, increasing the roughness of the incoming plate, particularly near the trench as observed in Alaska30, where the subduction of trench-parallel rough subducting fabric enhances interplate seismicity.

## Discussion

The presence of subducting ridges beneath the continental margin carries important implications for interplate coupling, shallow seismogenesis and earthquake dynamic properties.

### Implications for interplate coupling

Rough subducting seafloor influences interplate coupling28,31,32. Conceptual models of seamount subduction propose that the uplift of the upper plate increases normal stress, enhancing the coupling between plates33,34. Yet, these models did not take into account the complex pattern of fractures that are generated atop the subducted feature. Alternative models predict that interplate coupling is reduced by this network of upper-plate fractures32, promoting stable sliding and preventing the accumulation of elastic energy. Yet, numerical simulations reveal that overpressure builds up landward of the subducting roughness28,35, favouring hydrofracturing and pore fluid drainage26, which in turn, increases effective stress and interplate coupling36, promoting the occurrence of confined ruptures landward of the subducted seamount. Supporting this scenario, the PSDM section reveals reverse polarity reflections within the subducting sediments trenchward of the subducted ridge, indicating the presence of overpressured pore fluids (CDP4000-5000; Fig. 3). Yet, reverse polarity reflectors are not observed landward of the ridge in subducting sediments, indicating that fluid drainage may have occurred as a result of hydrofracturing. We hypothesise that fluids may have escaped upwards through the complex network of fractures atop the ridge. The presence of fluid-bearing fractures would contribute to reducing upper-plate seismic velocities atop the ridge, and thus explain our tomographic results.

The presence of comparatively higher coupling landward of the rough subducting oceanic fabric is also supported by the rupture pattern of past earthquakes in the area. Analogue modelling shows that the rupture generated by rough subducting topography is controlled by its spatial extent and shape31. Considering that oceanic ridges converge trench parallel beneath the margin in the southern subduction system8, stress accumulation would occur along-strike rather than being localised as in the case of isolated seamounts28, resulting in a set of NW-SE-elongated asperities. Finite-fault inversion results of the 1995 earthquake show that coseismic slip released near the trench (Fig. 1) followed a trench-parallel patchy distribution18,37 (Fig. 1b), consistent with the presence of trench-parallel asperities along the interplate. Similarly, the distribution of aftershocks of the 1995 Colima-Jalisco Earthquake shows several events in the leading flank of the subducting ridge19 (Fig. 2c), supporting the presence of overpressured areas. Based on these aspects, we propose that the rough subducting topography of the southern Rivera Plate contributes to expel pore fluids from subducing sediments landward of subducted ridges, enhancing interplate coupling, and promoting the formation of a pattern of trench-parallel shallow asperities.

### Subducting ridges determine upper-plate elastic properties

Megathrust earthquakes offshore Colima-Jalisco exhibited a depth-dependent dynamic behaviour22,37. In particular, the 1995 Colima-Jalisco event displayed the largest slip and slowest propagation in the shallow region, with rupture velocity ranging between 1.0–1.5 km s−1 in the first 10–12 km from the trench37. In addition, the event was characterised by a discrepancy between the surface wave magnitude (Ms 7.4) and the seismic moment magnitude (Mw 8)18,19, implying a depletion in high frequencies. Such dynamic behaviour is characteristic of shallow megathrust earthquakes38, including tsunami earthquakes39, and largely explained by the trenchward decrease in rigidity of rocks overlying the fault7,40,41. Thus, to understand the dynamic behaviour of shallow events in the area we calculate the distribution of upper-plate rigidity along our seismic line using the Vp tomographic model (Methods). Upper-plate tectonics and the distribution of elastic properties along our line are intimately related to the subducting oceanic topography (Figs. 3, 4). Our results show that intense fracturing atop the ridge causes comparatively lower rigidity areas in the upper plate (Fig. 4b). In addition to this local variation, the overall depth distribution of rigidity of rocks overlying the interplate decreases trenchward. Rigidity drops from ~30 GPa at 20 km depth to 2–5 GPa at the trench, in agreement with the global depth-distribution of rigidity derived from upper-plate tomographic models7 (Fig. 4a), and thus, favouring the enhancement of coseismic slip trenchward. Further, assuming that rupture velocity is within 70 and 90% of the S-wave velocity of rocks overlying the fault7,40,42, the low Vs estimated through the first 12 km of the fault implies that earthquakes propagate at velocities ~1.0 km s−1 (Fig. 4c), consistent with estimates of rupture velocity of the 1995 earthquake in the shallow megathrust37. In summary, the elastic structure of the margin promotes shallow ruptures with large slip, long duration, and consequently a depletion in high frequencies. A pattern that explains the dynamic behaviour of past shallow events in the study area.

## Conclusions

Our results show that rough subducting topography beneath non-accretionary margins not only controls upper-plate tectonics and margin erosion but also upper-plate elastic structure, setting ideal conditions for large near-trench slip and slow ruptures. In the study area, the shallow distribution of slip of the 1995 Colima-Jalisco earthquake suggests that, rather than acting as barriers32,43, subducting ridges may act as unstable asperities, assisting deeper ruptures in reaching shallower regions of the fault, and propagating laterally along the shallow non-accretionary portion of the margin. Such seismogenic behaviour likely prevails along the entire portion of the Colima-Jalisco margin affected by the subduction of rough ancient EPR fabric, exposing more than a hundred kilometres of continental margin to shallow seismogenesis and associated tsunamigenesis. Overall, our study shows that quantifying the impacts of subducting topography on upper-plate rigidity through joint tomography of MCS and WAS data may help to explain the large tsunamigenic potential of other non-accretionary margins globally.

## Methods

### Seismic data acquisition and processing

Wide-angle seismic (WAS) and multichannel seismic (MCS) data used in this study were collected during the TSUJAL survey (2014) along the upper-plate segment of transect TS0223 (Supplementary Figs. 1, 2). WAS data were recorded by 4 ocean-bottom seismometers (OBS) (Supplementary Fig. 1), and data processing involved instrument relocation on the seafloor, band pass filtering (3–7; 20–25 Hz), deconvolution and a wiener shaping filter that increased the signal-to-noise ratio at far offsets (>80 km offset). The WAS source was generated with 14 air guns towed at 15 m depth that released a total volume of 6800 cubic inches every 120 s along the line. MCS data were recorded with a 6-km-long hydrophone streamer deployed at 10 m depth that included 468 active channels (5850 m length) separated 12.5 m apart. CMP distance is 6.25 m and allows a CMP nominal fold of 58–59 traces. The MCS sources consisted of an array of 12 guns divided into four subarrays of three guns that released a total volume of 3540 cubic inches. The source was towed at 8 m depth and the shooting interval was every 50 m. MCS processing has been adapted to image the deep geometry of the interplate megathrust, whose depths range between 5–9 km. A preliminary post-stack time migration has been obtained with a model derived from velocity analyses to identify the most striking structures. A pre-stack depth migration was subsequently carried out using the interval velocity model obtained in the tomographic OBS model derived from a coincident wide-angle profile (Supplementary Fig. 2). This step requires a special pre-processing flow to improve the noise suppression and the multiple removals to reveal the deeper structures. Noise suppression includes removing burst and attenuating noise without affecting the spectrum of the remaining record (TFDN), deghosting and debubbling the signal, and removing coherent noise to increase the resolution ratio via predictive deconvolution (44 ms prediction gap, and 360 ms operator length). Removing multiple includes the use of prediction of multiples following SRME strategies (RadexPro software). We use the tomographic velocity model obtained from OBS and MCS data as starting model to perform a Pre-stack depth migration with pre-computed travel-times calculated by a Kirchhoff eikonal equation approach in the common-offset domain, well suited for moderately complex geology settings handling steep dips imaging, prior to testing the aperture, the dip limits and the anti-alias filter (PRoMAX software).

### Travel-time picking

We have picked travel-times of P-waves refracted through the upper-plate, or PgC seismic phases, from OBS records and MCS shot gathers (Supplementary Figs. 1, 3). We have applied the downward continuation method44 to shot gathers located between 15 and 40 km of profile distance from the trench to retrieve PgC first arrivals masked by reflected phases in the original gathers (Supplementary Fig. 4). We picked one gather every 200 m, and within each gather one trace every 200 m. Testing denser picking interval increase the computational cost without significant tomographic improvement. This way, a total of 7239 PgC travel-times were manually picked. We have also included in the inversion MCS travel-times of P-waves reflected at the interplate, or PiP phases, identified in CDP gathers (Supplementary Fig. 3). We used the stack section in Supplementary Fig. 3a to guide the manual picking of CDP gathers. We resorted the picking geometry from CDP to shot gathers and selected one shot every 400 m, yielding a total of 1053 PiP travel-times. Such decimation was selected on the basis of several tests and it provides the best balance between tomographic resolution and computational cost. Travel-time uncertainty was set on the basis of the S/N ratio of the trace 250 ms before and after the selected travel pick as proposed by ref. 45. This way, pick uncertainty of PgC travel-times from WAS data ranges between 20 and 90 ms, while PiP and PgC travel-times from MCS data is 40 ms.

### Combined MCS and WAS joint refraction and reflection travel-time tomography

We have jointly inverted PgC and PiP travel-times using the modified joint refraction and reflection travel-time inversion code TOMO2D46,47. This modified version allows the source to be outside the irregular velocity mesh, which enables the inversion of travel-times recorded with streamer geometry48. The velocity grid is parametrised as a set of nodes hanging from the seafloor with a regular horizontal spacing of 100 m and a variable one in the vertical direction, with node spacing increasing from 25 m at the top of the grid to 0.5 km at the bottom of it. Regularisation parameters are defined by a set of correlation lengths (CL) that vary from top to bottom of the grid. In the horizontal direction CL increase from 1 km at the top to 5 km at the bottom of the grid, while vertical correlation lengths increase from 0.5 at the top to 1 km at the bottom of the grid. The horizontal correlation length for the reflector is 0.8 km. Damping and additional smoothing constraints are also set to stabilise the inversion. The inversion was performed following a Monte Carlo (MC) uncertainty analysis (Fig. 2a, b, and Supplementary Fig. 5). We created a set of 500 MC realisations each realisation consisting of the randomly generated 1D velocity model, an initial landward dipping interplate reflector, and a noisy travel-time dataset. Each velocity model was generated by randomly varying by 10 % the velocity of the reference 1D model in Supplementary Fig. 5a. Initial reflectors were randomly generated with landward dips between 5° and 15° (Supplementary Fig. 5a). Finally, the 500 noisy travel-time datasets were generated by adding random Gaussian noise to each manually selected travel-time. The range of noise is set on the basis of picking uncertainty. The initial uncertainty of Vp and interplate depth is shown in Supplementary Fig. 5b, while the random distribution is depicted by initial travel-time residuals in Supplementary Fig. 5d. After inverting the 500 MC realisations the overall root mean square (RMS) decreases from 350 to 65 ms after 15 iterations each (Supplementary Figs. 5d, 6). The final average velocity model reproduces first order variations of the upper-plate velocity structure and geometry of the interplate reflector as shown in Fig. 3. Small inconsistencies between the PSDM section and the tomographic result are due to the inherent limitation of travel-time tomography in resolving small-scale velocity variations. We use the average velocity model for the final interpretation (Fig. 2a), while we take the final standard deviation as a proxy of the model parameter uncertainty49 (Fig. 2b). Integration of MCS PiP and PgC travel-times increases the amount of inverted travel-time information by an order of magnitude respect WAS data. This yields a denser ray coverage of the upper plate, and thus, lowers velocity and reflector geometry uncertainties, particularly in the first 20 km from the trench (Supplementary Fig. 5c). This way, velocity and interplate location uncertainty are lowest (<0.2 km s−1) in the first 20 km from the trench (Fig. 2c). At a further distance from the trench the upper plate is mostly covered by PiP rays and few PgC rays from WAS data (Supplementary Fig. 5c). As a result, velocity and interplate geometry uncertainty increases (>0.2 km s−1) because of the inherent velocity-depth trade-off of near-vertical reflections (Fig. 2c). Additional larger velocity uncertainty (>0.4 km s−1) is associated with sharp velocity contrasts at the top of the basement, but only imply a ~10% variation of the average velocity (5.0–5.5 km s−1), yielding a similar outcome and interpretation.

### Resolution test

we have tested the sensitivity of our travel-times and tomographic method to retrieve the velocity and interplate topographic variations observed in the final model (Fig. 2). We have built a true model based on the final solution (Supplementary Fig. 7a). We calculated the set of synthetic travel-times and added random Gaussian noise on the basis of the picking uncertainty as in the MC approach. The set of synthetic noisy travel-times were inverted using the same 1D reference velocity model as in the MC approach and a straight 7° landward dipping interplate reflector. The retrieved tomographic model has an RMS of ~70 ms, showing a similar velocity structure and interplate geometry than the true model, particularly in the first 20 km from the trench (Supplementary Fig. 7b). Velocity differences are overall >0.25 km s−1 in the first 20 km from the trench, with some local differences of 0.5 km s−1 landward of the subducting ridge at 14–16 km from the trench (Supplementary Fig. 7c, d). Larger velocity differences (>0.5 km s−1) are also observed at further offsets than ~30 km from the trench and deeper than 4 km depth (Supplementary Fig. 7c, 7d), consistent with larger uncertainties. Regarding the reflector geometry, we are able to retrieve the true interplate geometry with minor variations up to ~25 km from the trench (Supplementary Fig. 7c, d). At further offsets, we are not able to rely on our results given the increasing velocity-depth trade-off of reflections.

### Elastic parameters calculation

To calculate rigidity (µ) at the base of the upper plate we first converted the final Vp model into density (ρ) and Vs using empirical relationships between Vp-Vs and Vp- ρ from ref. 50. In particular, we used the Nafe-Drake curve (Eq. 1 in ref. 50) to estimate ρ, while we used eq 6 in ref. 50 to derive Vs. Both relationships are valid for Vp between 1.5 and 8–8.5 km s−1, and valid for a suite of crustal rock types that includes plutonic rocks such as those sampled in the study area (i.e., granodiorites12). Rigidity was then calculated as:

$${{{{{\rm{\mu }}}}}}\,=\rho {V}_{s}^{2}.$$
(1)

Finally, to extract rigidity profiles in Fig. 4a, b, we averaged rigidity values within 0.5 km above the interplate reflector.