“…For least squares regression‐based calibration, correlated residuals can be handled by using a full error covariance matrix (i.e., with nonzero off‐diagonal entries) in the objective function [ Lu et al ., ]. In Bayesian calibration, customized likelihood functions are used to characterize model residuals that are non‐Gaussian, biased, skewed, heteroscedastic, and correlated [ Beven and Freer , ; Erdal et al ., ; Nearing et al ., ; Schoups and Vrugt , ; Shi et al ., ]. The above methods have been applied to various fields, including rainfall‐runoff, unsaturated flow and groundwater reactive transport modeling.…”