Extreme earthquake loss assessment using spliced marginal distributions and SJC Copula based joint modeling

Earthquake hazards, though occurring infrequently, can produce catastrophic losses with pronounced right-skewed and heavy-tailed characteristics in both economic damages and fatalities. These impacts often intensify jointly under extreme conditions, posing challenges for reliable regional catastrophe risk assessment. Using earthquake records from 1980 to 2024 in selected provinces (autonomous regions and municipalities) of China, this study develops variable weights spliced Gumbel–GPD and Weibull–GPD distributions to model the marginal behavior of extreme losses. The dependence between economic losses and casualties—particularly in the upper tail—is captured using the SJC Copula, allowing for asymmetric co-extremal behavior. Due to the limited number of high-loss historical events, we analyze the sensitivity of parameter estimates to sample size through numerical simulations. After confirming that the GAN-generated samples are statistically consistent with the original data, they are employed to strengthen the robustness of parameter estimation. Integrating the spliced marginal models with a copula-based dependence framework, this study evaluates extreme loss levels for different regions under historical seismic conditions. The resulting estimates offer quantitative support for identifying key areas requiring enhanced seismic protection and for informing regional disaster-risk management.