Impacts of different stochastic physics perturbation schemes on the GRAPES Global Ensemble Prediction System

In order to better understand the impacts of different stochastic physics perturbation schemes on global medium ensemble forecasts, this research conducts a comparative analysis of the features of perturbations yielded by the Stochastically Perturbed Parameterization Tendencies (SPPT) scheme, the Stochastic Kinetic Energy Backscatter (SKEB) scheme, and the combination of the SPPT and SKEB schemes as well as the impacts of these three model perturbation methods on ensemble forecasts based on the GRAPES Global Ensemble Prediction System (GRAPES-GEPS). To avoid the impacts from initial perturbations, initial perturbations are disabled when the features of perturbations produced by the above stochastic physics schemes are explored. Via the perturbation versus error correlation analysis (PECA), it is found that perturbations yielded by different stochastic physics perturbation schemes have the capability to capture forecast errors. Furthermore, when the combination of the SPPT and SKEB schemes is applied, the produced perturbations best simulate forecast errors. For all stochastic perturbation schemes, the total energy of perturbations is initially concentrated in the middle and upper troposphere and the lower stratosphere of the tropics. In addition, the total energy of perturbations increases with the forecast lead time, for which the maxima keep propagating towards the extratropical regions. From the spectra of total energy of perturbations, it is observed that the perturbation energy evolves upscale. In the GRAPES-GEPS built on the initial perturbations derived from singular vectors, the applications of stochastic physics perturbation schemes increase the ensemble spreads for fields at different isobaric surfaces in different regions and improve the root-mean-square errors of the ensemble means to some extent. Due to the increased ensemble spreads, outliers are significantly decreased. The continuous rank probability scores are also reduced, which is more pronounced in the tropics. Furthermore, the probabilistic forecast skills of rainfall in China for light rain, moderate rain, heavy rain, and rainstorm are also improved to some extent. The above-mentioned improvements are the largest when the combination of the SPPT and SKEB schemes is employed. This is consistent with the results from the analyses on total energy of perturbations and PECA.