Cascadia Plate Interface Structure Between 40-45°N From Amphibious Regional-Earthquake-Tomography Vp and Vs Models

Cascadia Region Earthquake Science Center 2023-24 Seed Grant Program Final Report Cascadia Plate Interface Structure Between 40-45°N From Amphibious RegionalEarthquake-Tomography Vp and Vs Models Liam P. Moser and J. Pablo Canales Woods Hole Oceanographic Institution, Woods Hole, MA https://doi.org/10.5281/zenodo.17253338 1 Project Report for CRESCENT Seed Grant Number CR2024-05 Cascadia Plate Interface Structure Between 40-45 ° N From Amphibious RegionalEarthquake-Tomography Vp and Vs Models Liam P. Moser and J. Pablo Canales Dept. of Geology & Geophysics, Woods Hole Oceanographic Institution, Woods Hole MA The goal of this project was to conduct a three-dimensional (3-D) regional earthquake tomography study of southern Cascadia using Pand S-wave first arrival times from a recent catalog (Morton et al., 2023) derived from Cascadia Initiative data. The study region extended from the south of the Mendocino transform at 39.9 ° N up to northern Oregon at 45.3 ° N, a region characterized by strong north-south gradients in extent and intensity of deformation of the incoming plates (e.g., Gulick et al., 2001). At the southern end of our study region, Vp/Vs ratios of ~1.85, elevated relative to background, were determined along the down-going Gorda slab in a previous local earthquake tomography study (Guo et al., 2021). The presence of elevated Vp/Vs along the top of the slab correlates with geodetically inferred plate locking. Further to the north plate locking and episodic tremor and slip rates decrease, but the Vp/Vs structure along the plate interface is unconstrained. The main scientific question of this project is to evaluate the continuity of this high Vp/Vs feature along the Oregon margin and evaluate if its extent or magnitude changes with the inferred decrease in plate locking. The main methodological advancement proposed is the use of a new Poisson Voronoi cell-based inversion (Fang et al., 2019, White et al., 2020). This technique solves smaller over-determined problems that are combined into a final model, therefore removing the need to define inversion regularization parameters. With support from CRESCENT, we have achieved three main outcomes: (1) PoissonVoronoi cell parameter testing, (2) Poisson-Voronoi cell checkerboard testing, and (3) a preliminary Vp and Vs model of the southern Cascadia subduction zone. 1. Parameter testing. The primary parameter tested was the number of Voronoi realizations (NREAL) used in each inversion iteration in order to converge into a stable solution. In Fig. 1 we show the data misfit vs. iteration number for several values of NREF: 20, 50, and 100 Voronoi realizations per iteration. Each Voronoi realization consists of a single inversion using N Voronoi cells of variable size dependent on ray coverage (i.e., areas with denser data coverage are automatically discretized with smaller cells). We kept N=400 constant for all tests of NREAL. The solution for each iteration is the average of the N realizations. We found that the number of Voronoi realizations does not significantly affect the final misfit, but larger number of realizations result in a smoother model with less artifacts from the Voronoi cell averaging. 2. Checkerboard tests. We conducted checkerboard tests as a means to assess spatial resolution given the data and method. In Fig. 2 we show checkerboard resolution tests for the Swave model at 100 by 100 km scales. Seaward from the deformation front, resolution of the incoming plate is acceptable at depths of 20 and 30 km (Fig. 2b,c). Beneath the accretionary prism in the region encompassed by the deformation front and the coast, the tests show the highest model resolution with checkerboard patterns recoverable from shallow (10 km) up to 40 km depth. Beneath onshore portion of the model, the S-wave model is well resolved between 10 and 30 km depth (Fig. 2a-c) but degrades deeper (Fig. 2d). P-wave resolution at a 100 km length scale is similar (not shown), but with reduced coverage between 41.5-43 ° N. 3. Preliminary model. In Fig. 3 we show the preliminary Vs model of the study region. At 10 km depth velocities decrease from the incoming to the down-going plate across the deformation 2 front (Fig. 3a). At 20 and 30 km depth Vs is <~4 km/s between the deformation front and onshore at ~123 ° W, which may correspond to the high Vp/Vs feature reported by Guo et al. (2021). At 30 km depth this Vs feature extends to 41.5 ° N, slightly north of the Guo et al. (2021) study region (Fig. 3c), but shallower at 20 km depth the feature is present up to 43°N where Vs increases beneath the accretionary prism but remains low beneath the onshore portion of the model (Fig. 3b). This transition at 43°N occurs where the Blanco shear zone enters the subduction system, and may constitute a primary structural control on the seismic properties of the shallow slab. The next steps in this project are: (1) replace the current one-dimensional starting model with a 3-D starting model using the results of the CASIE-21 seismic experiment (Miller et al., 2024), (2) refine checkerboard and parameter tests as needed with the new starting model, and (3) obtain final models and interpret along slab structure. Care must be taken with selecting the starting 3-D model as it inherently includes a decision about plate interface geometry. The selected plate interface geometry will be a combination of the CASIE-21 model and deeper receiver function constraints (Bloch et al., 2023, Carbotte et al., 2024). The results of this grant will be part of the Ph.D. Thesis of L. Moser. They have not yet been presented or published, but we expect to disseminate results during 2026. References Bloch, W., Bostock, M.G. and Audet, P., 2023. A Cascadia Slab Model From Receiver Functions, Geochem. Geophys. Geosyst., 24, e2023GC011088, doi:10.1029/2023GC011088. Carbotte, S.M., Boston, B., Han, S., Shuck, B., Beeson, J., Canales, J.P., Tobin, H., Miller, N., Nedimovic, M., Tréhu, A., Lee, M., Lucas, M., Jian, H., Jiang, D., Moser, L., Anderson, C., Judd, D., Fernandez, J., Campbell, C., Goswami, A. and Gahlawat, R., 2024. Subducting plate structure and megathrust morphology from deep seismic imaging linked to earthquake rupture segmentation at Cascadia, Sci. Adv., 10, eadl3198, doi:10.1126/sciadv.adl3198. Fang, H., van der Hilst, R.D., de Hoop, M.V., Kothari, K., Gupta, S. and Dokmanić, I., 2019. Parsimonious Seismic Tomography with Poisson Voronoi Projections: Methodology and Validation, Seismol. Res. Lett., 91, 343-355, doi:10.1785/0220190141. Gulick, S.P.S., Meltzer, A.S., Henstock, T.J. and Levander, A., 2001. Internal deformation of the southern Gorda plate: Fragmentation of a weak plate near the Mendocino triple junction, Geology, 29, 691-694, doi:10.1130/0091-7613(2001)029<0691:IDOTSG>2.0.CO;2. Guo, H., McGuire, J.J. and Zhang, H., 2021. Correlation of porosity variations and rheological transitions on the southern Cascadia megathrust, Nat. Geosci., 14, 341-348, doi:10.1038/s41561-021-00740-1. Miller, N.C., Canales, J.P., Carbotte, S.M., Han, S., Boston, B., Jian, H., Moser, L.P., Mann, M.E. and Shuck, B., 2024. A New 3D Reference Velocity Model for Offshore Cascadia Based on CASIE21 Data, Abstract T53A-3189 presented at 2024 Fall Meeting, AGU, Washington, D.C., 9-13 Dec. Morton, E.A., Bilek, S.L. and Rowe, C.A., 2023. Cascadia Subduction Zone Fault Heterogeneities From Newly Detected Small Magnitude Earthquakes, J. Geophys. Res., 128, e2023JB026607, doi:10.1029/2023JB026607. White, M.C.A., Fang, H., Nakata, N. and Ben-Zion, Y., 2020. PyKonal: A Python Package for Solving the Eikonal Equation in Spherical and Cartesian Coordinates Using the Fast Marching Method, Seismol. Res. Lett., 91, 2378-2389, doi:10.1785/0220190318. 3 Figures Figure 1. Travel time misfit descent as a function of iteration and number of Voronoi realizations (NREAL). Inversions use 400 Voronoi cells. Iteration #8 is selected as trade-off between run time and misfit. Misfit is given as a root-mean-square. 4 Figure 2. S-wave checkerboard recovery test. Each panel is a horizontal slice at (a) 20, (b) 40, (c) 30, and (d) 40 km depth. Solid grid lines define the input checkerboard pattern consisting of 100 by 100 km2 synthetic anomalies with ±5% velocity perturbation against a 1-D starting velocity model. No noise is added. The inversion is run for 3 iterations with 300 Voronoi cells and 20 Voronoi realizations per iteration. Increased number of cells and/or realizations will likely improve recovery. White lines mark California to Oregon state border, the coastlines, the deformation front, and the plate boundaries. 5 Figure 3. Preliminary S-wave model with eight iterations, 400 Voronoi cells, and 100 Voronoi realizations per iteration. Each panel is a horizontal slice at (a) 20, (b) 40, (c) 30, and (d) 40 km depth. Black dots correspond to earthquake locations from the data catalog used in this study. Lines mark California to Oregon state border, the coastlines, the deformation front, and the plate boundaries.