Stabilizing REGARD with real-time MCMC uncertainty quantification and precise point positioning

This study introduces structural upgrades to REGARD, Japan’s real-time GNSS-based coseismic analysis system, addressing two long-standing limitations in real-time geodetic source modeling: the absence of explicit uncertainty quantification in fault estimates and positioning instability due to reference‑station dependence. These limitations reduce the reliability and physical interpretability of automated products used for rapid earthquake and tsunami assessments. To quantify uncertainty under strict real-time constraints, we implement a Bayesian inversion framework based on Markov chain Monte Carlo (MCMC), termed RUNE, for estimating a single rectangular fault model. RUNE outputs posterior probability density functions for moment magnitude (Mw) and fault parameters. Stable and efficient operation is achieved through physically informed priors and likelihood regularization through variance lower bounds designed to ensure statistical consistency, thereby suppressing noise-driven overfitting while retaining sensitivity to genuine crustal deformation. To improve positioning robustness, we introduce a real-time precise point positioning (PPP) cluster that eliminates reference-station dependence and mitigates network-wide translational artifacts inherent in real-time kinematic (RTK) positioning. Additional procedures—automatic station screening based on per-epoch position uncertainty and removal of common-mode trends estimated from distant stations—further stabilize displacement fields and improve the validity of the statistical assumptions used in Bayesian inversion. Applications to recent Japanese earthquakes, including the 2024 Noto Peninsula and Hyuga-nada events, together with 5 years of operational statistics, demonstrate that the combined MCMC–PPP upgrades markedly reduce false alarms while remaining consistent with post-processed solutions. Overall, the introduction of explicit uncertainty quantification and stabilized real-time positioning enhances the reliability and interpretability of automated GNSS-based coseismic source modeling for rapid hazard assessment.