A q-Likelihood Particle Filter in PDAF: From Lorenz-96 to Offline-Coupled MITgcm Assimilation in the Northern South China Sea

High-dimensional ocean data assimilation (DA) remains challenging when dynamics are strongly nonlinear, observations are sparse or irregular, or error statistics deviate from Gaussianity. Ensemble Kalman filters (EnKFs) are efficient and widely used, but they rely on approximate Gaussianity and can degrade when strongly non-Gaussian posterior distributions arise. Particle filters (PFs) offer a fully Bayesian alternative, yet they are hampered by weight degeneracy in high-dimensional geophysical applications. Here we introduce and implement a q-likelihood PF (q-PF) in the Parallel Data Assimilation Framework (PDAF), where the Gaussian likelihood is replaced by a q-Gaussian/Tsallis likelihood to temper the influence of outliers and mitigate weight degeneracy. The PDAF-MITgcm system uses PDAF offline coupling in a file-based forecast-analysis-restart workflow with cycled ensemble forecast priors. We evaluate q-PF against a localized ensemble transform Kalman filter (LETKF) and a baseline PF in (i) Lorenz-96 experiments spanning weak to strong nonlinearity and (ii) a 33-day winter regional case over the northern South China Sea and adjacent western Pacific assimilating satellite sea surface height (SSH) and sea surface temperature (SST). For ensemble size N = 40, q-PF yields lower period-mean continuous ranked probability score (CRPS) for both SSH and SST and reduces coherent SSH departure patches in eddy-active corridors. Eddy census and shelf-break diagnostics further show that the DA-cycled SSH output supports organized eddy genesis and propagation patterns consistent with energetic variability near the Luzon Strait and the continental slope.