Reducing Evapotranspiration Simulation and Forecast Uncertainties Due To Initial and Model Errors Over the Tibetan Plateau

Abstract Simulations and predictions of evapotranspiration (ET) over the Tibetan Plateau (TP) are typically affected by both model uncertainties and the accuracy of the initial conditions. In this study, a minimization technique and a data assimilation method are first applied to reduce the impact of initial errors on the uncertainties in the ET simulation over the TP. Second, the conditional nonlinear optimal parameter perturbation ensemble prediction (CNOP‐PEP) method is employed to evaluate the simulation and forecast performances with imperfect model physical parameters based on the reduction of impact of initial errors. The main features are the combined contribution for ensemble prediction of ET over the TP through the perturbed parameter and the optimization and assimilation of the initial conditions. The numerical results show that the minimization technique and the data assimilation method are useful for reducing uncertainties in ET over the TP due to errors of initial values. Moreover, the ensemble prediction of ET over the TP with assimilated initial conditions is better than that without assimilated initial conditions using the CNOP‐PEP method. Moreover, the ensemble prediction of ET over the TP using the CNOP‐PEP method (17.3%) is better than that using traditional methods, such as the one‐at‐a‐time (OAT) method (2.0%) and the stochastically perturbed parametrization (SPP) method (10.6%) with and without assimilated initial conditions. And, the absolute errors using anomaly‐based metrics by removing the climatological mean are 71.80, 101.06, and 105.52 mm/year using the CNOP‐PEP, OAT and SPP methods.