Projections of future sea-level changes are usually based on global climate models (GCMs). However, the changes in shallow coastal regions, like the China marginal seas, cannot be fully resolved in GCMs. To improve regional sea-level projection, a high-resolution (~8 km) regional ocean model is set up for the China marginal seas for both the historical (recent 2-decades) and future periods (2081-2100) under representative concentration pathways (RCPs) 8.5. The historical ocean simulations are first evaluated against observations at different spatiotemporal scales, and the model is then integrated for the future period. The downscaled ocean changes derived by comparing historical and future experiments reveal greater spatial details than those from corresponding GCMs. The results of mean sea level (MSL) downscaling indicate that forcing of the MSL change and increased cyclonic circulation in the SCS are dominated by the climate change signals from the Pacific, while the MSL change in the East China marginal seas is caused by both local atmosphere forcing and signals from the Pacific. For the extreme sea level, we find the MSL changes explain most of the extreme high sea level (EHSL) changes in the future, while the probability density function (PDF) shape change (e.g., skewness and kurtosis) is the key factor for extreme low sea level (ELSL) changes. Perturbation experiments with different surface forcing indicate that the high-frequency signal (shorter than three months; F_H) change is the major contributor to the PDF shape change. Specifically, the stronger F_H southerly wind and weaker F_H northerly wind lead to strengthened EHSL and weakened ELSL in the Bohai Sea and Yellow Sea, resulting in significant skewness and kurtosis changes in this region. The method of dynamical downscaling developed in this study is a useful modelling protocol for adaptation and mitigation planning for future oceanic climate changes.