In response to the problem that the statistical characteristics of wave height no longer satisfy the stationarity assumption due to global climate change and anthropogenic factors, this study aims to more accurately estimate design wave heights for marine engineering. An improved compound extreme value distribution (ICEVD) model is developed based on the traditional compound extreme value distribution (CEVD) by integrating non-stationary modeling techniques with extreme wave height data from both typhoon and non-typhoon periods. The computational results of the traditional CEVD and the proposed ICEVD under non-stationary conditions are compared. Using four methods—the expected number of exceedances (ENE), design life level (DLL), equal reliability (ER), and average design life level (ADLL)—design wave heights are derived, and an uncertainty analysis is conducted. The results demonstrate the improved performance of the ICEVD. This model not only overcomes the limitation of the traditional CEVD, which ignores non-typhoon processes, but also more accurately reflects the non-stationary characteristics of extreme wave heights under climate change. It is particularly suitable for sea areas jointly influenced by typhoon and non-typhoon weather systems, providing a new technical approach for estimating design wave heights in marine engineering.