“…Near or below the LUF [1], this assumption may be unsustainable. In that case, a characterization based on sampling circular, elliptic [17], or even more general [18] joint pdfs for E i and E j may be appropriate. A complete generalization involves additional parameters of the distribution, each requiring individual estimation (e.g., using Bayesian inference), and the characterization of the dependence structure between E i and E j , e.g., using copulas [18].…”