The deterioration of long-term structural performance of marine reinforced concrete (RC) structures involves a large number of spatiotemporal variables related to material properties and environmental actions. The independent consideration of spatiotemporal variables in time and space may result in the loss of valuable probabilistic information, lacking the necessary physical mechanisms and theoretical support for reliable probabilistic analysis of structural performance. To address the challenges, this study introduces a spatiotemporal variability modeling and evaluation framework based on the space–time-dependent Bayesian Network (STBN). In the STBN, some parameters of material properties and environmental actions are constructed and updated at each space-time-slice by integrating diverse models and analysis methods, with the pinch-point variables serving as bridges between the different sub-Bayesian Networks. The example of Hangzhou bay cross-sea bridge is used to validate the proposed framework for assessing long-term structural performance. The results show that this framework provides an effective solution to handle the complexities of information updating and probabilistic inference for the evaluation of long-term performance of engineering structures.