“…Another kind of widely used methods is meta-modeling to build cheap-to-compute surrogates or emulators of computationally expensive models so that performing a large number of model executions is computationally affordable (O"Hagan, 2004). The methods of developing surrogates include Taylor series approximation (Hakami et al, 2003), response surface approximation (Helton and Davis, 2003), Fourier series (Saltelli, et al, 1999) nonparametric regression (Helton, 1993;Storlie et al, 2009), Kriging (Borgonovo et al, 2012;Lamoureux et al, 2014), Gauss process (Rasmussen and Williams, 2006), polynomial chaos expansion (Garcia-Cabrejo and Valocchi, 2014;Oladyshkin et al, 2012;Sudret, 2007), and sparse-grid collocation (Buzzard, 2012;Buzzard and Xiu, 2012). However, the meta-modeling methods may still need a relatively large number of model executions to develop accurate surrogates, and the surrogate development is not always straightforward due to model nonlinearity (Razavi et al, 2012;.…”