Probabilistic corrosion prediction of marine steel under future climate change scenarios using physics-informed Gaussian process regression

This study proposes a framework of corrosion prediction by utilizing physics-informed Gaussian process regression (PI-GPR) and future climate change simulation. Long-term corrosion prediction of marine steel structures is routinely based on stationary environmental assumptions; however, future sea surface temperature (SST) changes should be taken into account as it may affect corrosion degradation over structural service life. The present work develops a probabilistic corrosion prediction based on the PI-GPR model, which is coupled with CMIP6-based SST projections. The proposed PI-GPR model incorporates a physics-based corrosion mean function into a probabilistic machine learning framework, enabling physically consistent long-term extrapolation with 95% credible intervals. Annual SST projections from 31 CMIP6 models under SSP126 and SSP585 were processed using a multi-model ensemble approach and applied to the seas surrounding the Korean Peninsula. The results demonstrate that scenario-based corrosion loss exceeds fixed-temperature estimates, with the deviation increasing over multi-decadal exposure. Especially, the West Sea exhibited the largest climate-induced corrosion deviation under SSP585, indicating strong regional sensitivity. These findings suggest that future SST trajectories should be considered as essential inputs for corrosion allowance, inspection planning, and reliability assessment of marine steel structures under climate change.