Research On Model Uncertainty Representation For Convection-Allowing Ensemble Prediction Based On CNOP-P

Convection-Allowing Ensemble Prediction System(CAEPS)has obvious advantages in predicting the convective events due to fruitful probabilistic forecast information.Comparing with global ensemble prediction system,representation of model uncertainty in CAEPS is lack of systematic research and theoretical basis,and becomes an important issue worthwhile further research.This thesis proposes a new idea to represent the model uncertainty in CAEPS based on Conditional Nonlinear Optimal Perturbation related to Parameters(CNOP-P).CNOP-P is used to detect and describe the maximum precipitation errors caused by the model parameter uncertainties.Meanwhile,the ensemble approach is used to compute CNOP-P,which can effectively reduce the computational cost of original adjoint method.The main results are summarized as follows:(1)GRAPES-Meso with a 3 km resolution set-up is utilized in our study.We select five typical rainstorm cases occurred in South China in 2017 as initial conditions for the CNOP-P computation.Using 15 parameters in boundary layer and cloud microphysics parameterization scheme,CNOP-P is solved by an ensemble-based algorithm.Here,CNOP could lead to the maximum variations in precipitation during the first 12 hours of model integration.According to the perturbation magnitude corresponding to each parameter in CNOP-P,the order of relative sensitivity of parameters is obtained.The first eight parameters are detected as sensitive parameters.These parameters play an important role in the calculation of the turbulent diffusion coefficient of the boundary layer,the boundary layer height,the vertical distribution of hydrometeor,and the automatic conversion in clouds.(2)A new model perturbation scheme(shortly “CNOP-P-based scheme”)is constructed by stochastically perturbing the sensitive physical parameters selected above.Verifications of this CNOP-P-based method are carried out in an experimental ensemble system with only considering the model uncertainty,Stochastic Perturbed Parameterization Tendencies(SPPT)scheme is used as a reference.The results from a typical case and 22 successive cases in 2017 show that: comparing with SPPT,the use of CNOP-P-based scheme give an improvement in the ensemble reliability and bring higher the probability forecast skills for the near-surface variables.For wind,temperature and other atmospheric variables,it can also increase ensemble spread,remarkably in the lower atmosphere.