Research question: Can offshore optical fiber strainmeters detect repeating tectonic signals and candidate events that are absent from existing regional catalogs?

Methods: I process 8-Hz SOFS strain records using instrument correction, Butterworth filtering, wavelet denoising, and template-free matrix-profile analysis in an HPC environment. Motifs identify repeating strain signatures, while discords flag unusual or previously uncatalogued transients. I compare candidate detections with earthquake and tremor catalogs, including PNSN products, and evaluate consistency between the EW and NS strain components.

Why it matters: Most seismic and geodetic stations in Cascadia are located onshore, leaving offshore deformation near the plate boundary poorly resolved. Direct seafloor strain observations could reveal weak or previously uncatalogued tremor, LFE/VLFE, and repeating transient signals, improving our understanding of where and when strain accumulates and is released offshore.

Research question: How much can offshore SOFS measurements improve the location, timing, and uncertainty of Cascadia slow-slip estimates compared with inversions based primarily on onshore observations?

Methods: I am developing a time-dependent joint inversion that combines offshore SOFS strain with onshore GNSS and borehole strain observations, together with prospective InSAR constraints where appropriate. The framework uses elastic Green’s functions, sequential estimation, Kalman filtering, noise-covariance weighting, and backward smoothing to estimate the evolution of slip through time. I will compare models with and without offshore observations and evaluate their relationship with PNSN tremor activity.

Why it matters: Incorporating seafloor strain measurements could better constrain offshore slow-slip patches, reduce model uncertainty, and clarify the relationship between slip evolution and tremor.

Research question: Can physics-informed machine learning estimate the coupled evolution of fault slip, crustal deformation, and pore-fluid pressure while remaining consistent with the governing mechanics?

Methods: I am developing a physics-informed neural-network framework that uses spatial and temporal coordinates as inputs and estimates displacement, fault slip, pore pressure, porosity, and permeability. The loss function incorporates elastic deformation, pore-pressure diffusion, fault friction, dilatancy, and porosity-permeability relationships. Geodetic observations provide data constraints, while the governing equations reduce physically unrealistic solutions.

Why it matters: Conventional simulations of coupled deformation and fluid flow can be computationally expensive and difficult to calibrate. A physics-informed framework could accelerate model exploration, integrate observations directly with physical laws, and help evaluate how evolving fluid pressure influences slow-slip behavior.