The research, led by Guoxing Chen of Nanjing Tech University together with Zhijie Jiang, Yanzhen Wang, Kai Zhao, Weiyun Chen, and Charng-Hsein Juang of Clemson University, presents a two-dimensional nonlinear seismic response analysis of a full cross-strait transect, modeled on the Qiongzhou Strait between China’s mainland and Hainan Island. Rather than treating the seabed as a flat, uniform layer of soil, the team built a numerical model that captures nearly everything that makes marine sites different from land sites: the geometry of the strait basin, small-scale seabed microtopography, spatially varying soil properties including both shear (S) and compressional (P) wave velocity structures, a nonuniform finite element mesh refined where wave resolution demands it, and artificial boundary conditions that absorb outgoing waves instead of reflecting them back into the domain.

At the heart of the study is the coupling between seawater and the seabed. Water is roughly 1,500 times less dense than soil and cannot sustain shear stresses, but it is not inert. During an earthquake, the seafloor moves, and the water column above it responds with its own inertia and compressibility, exerting fluctuating pressures back on the seabed. The authors simulated this two-way interaction using a weak coupling algorithm for fluid–solid interaction, alternating between the soil domain and the water domain until their interface conditions converged at each time step. On the soil side, they represented the material’s nonlinear hysteretic behavior, the well-known tendency of soft sediments to soften and dissipate energy as shear strain grows during strong shaking.

The team first ran simulations without seawater to establish a baseline, and those dry-seabed results already contain valuable physics. Bedrock motion components near the seabed’s fundamental frequency were amplified as they propagated upward through the soft deposits, the classic resonance effect familiar from sedimentary basins on land. But two distinctly marine findings emerged. First, the seabed’s microtopography, ridges, channels, and paleochannels carved into the seafloor, influenced vertical seafloor motions far more strongly than horizontal ones, a consequence of how irregular topography scatters and converts P and SV waves at the seafloor interface. Second, the simulations revealed a resonance-like peak near 2 hertz in both horizontal and vertical components of seafloor motion, produced by the complex interplay of seismic wave reflection, refraction, and interference within the heterogeneous sediment fill of the strait basin.

The picture changed dramatically once the water was switched on. With seawater–seabed coupling included, the models showed significant suppression of seafloor peak accelerations, and the effect was strongest in deep-water regions. The physical explanation lies in the mass of the water column: when the seabed accelerates, it must drag the overlying water along, and the resulting inertial loading resists the motion, damping peak accelerations at the mudline. In effect, the ocean acts as an enormous, distributed damper sitting on the seafloor.

Frequency-domain analysis sharpened the finding. The coupled model showed pronounced suppression of higher-frequency content near 4 to 5 hertz and, conversely, amplification of very low frequencies below about 0.5 hertz. The resonance-like response near 2.0 hertz, so conspicuous in the dry-seabed runs, was diminished in both horizontal and vertical components. Vertical motions proved especially sensitive: seawater suppressed them within specific, narrow frequency bands more aggressively than it suppressed horizontal motions, consistent with the water column’s ability to carry compressional pressure waves but not shear waves.

Perhaps the most consequential discovery concerns intensity dependence. The degree to which the resonance-like responses near 2 hertz were suppressed or amplified correlated positively with the intensity of the bedrock motion. In other words, the ocean’s moderating effect is not a fixed correction factor that can be applied uniformly. Weak earthquakes and strong earthquakes interact with the water column differently, and the soil’s nonlinear hysteretic response compounds that dependence. A design spectrum calibrated from moderate events could therefore misrepresent the hazard from a rare, large-magnitude offshore earthquake.

The stakes are high because the observational record points in the same direction. Seafloor instruments in Japan, including the DONET and S-net networks, have documented anomalously large amplifications at some ocean-bottom stations and clear nonlinear soil response during magnitude 7-class events, while other studies have noted systematic differences between seafloor and land ground motions at comparable distances. Numerical work on tsunami generation has similarly shown that the compressibility and inertia of the ocean layer matter for offshore wave propagation. What has been missing, the authors argue, is an integrated treatment that brings the basin geometry, realistic marine soil dynamics, and seawater coupling into a single two-dimensional framework spanning an entire strait.

Building such a model is far from trivial. Seismic waves must be resolved with enough elements per wavelength to avoid numerical dispersion, a constraint the team addressed with a nonuniform mesh layout across the transect. Because a truncated computational domain artificially reflects outgoing waves, the model employs consistent artificial boundary conditions that let energy radiate away as it would in the real Earth, allowing seismic input to be imposed at the bedrock through equivalent nodal forces. The soil’s spatial variability, drawn from marine geotechnical investigations of the Qiongzhou Strait including measured dynamic shear modulus and damping relationships for its deep marine silty clays, means that amplification patterns vary laterally across the basin rather than behaving as a single layered column. The result is a model in which wave fields are genuinely two-dimensional, with energy scattering between locations, and the amplification patterns that emerge are intricate and strongly coupled.

For offshore engineering practice, the implications are direct. Tunnels bored beneath straits, the piers of sea-crossing bridges, submarine pipelines and cables, and jacket foundations of offshore wind turbines are all designed against seafloor ground motions. If standard onshore-style site response analysis is applied at such sites, it may overpredict peak accelerations in deep water, potentially driving unnecessary cost, while simultaneously underpredicting the long-period content that water coupling can amplify, which governs the response of long, flexible structures such as tunnels running along the basin. The finding that vertical motions are selectively filtered in narrow frequency bands is especially relevant for structures sensitive to vertical excitation, and the intensity-dependent suppression means that probabilistic seismic hazard analysis for offshore sites cannot simply scale land-based transfer functions.

The work also suggests a research agenda. Extending the coupled framework to three dimensions, incorporating excess pore pressure generation and potential liquefaction of saturated marine sands, and validating against dense seafloor accelerometer networks are natural next steps. The authors’ own prior work on undersea shield tunnels and on nuclear-island buildings founded on soft soil demonstrates the engineering payoff of rigorous nonlinear site response analysis; the present study extends that rigor to the harshest site condition of all, a deep, water-covered basin.

As offshore infrastructure proliferates, from intercontinental power cables to planned strait-crossing megaprojects, the ground motion that such structures feel will increasingly be a seafloor motion, not a land motion. This study makes a compelling case that the ocean itself must be treated as part of the site. The sea, it turns out, does not merely sit above the earthquake hazard; it actively shapes it, filtering some frequencies away and handing others an amplification that no land-based model would predict.