CMA Global Ensemble Prediction Using Singular Vectors from Background Field

Abstract

China Meteorological Administration Global Ensemble Prediction System (CMA-GEPS) adopts singular vector method to generate initial perturbations. CMA-GEPS currently uses the initial analysis field from CMA Global Forecast System (CMA-GFS) data assimilation to calculate singular vector (ANSV). With this dependency, in the operational running procedures of CMA numerical weather prediction systems, the singular vector calculation starts when CMA-GFS analysis job is finished. With the improvement of model resolution, especially the horizontal resolution, the computation time of data assimilation analysis and the ensemble forecasts would be lengthened. Given relatively limited high-performance computational resources, it would bring great challenge for delivering the ensemble forecast products on time. ECMWF uses the data assimilation background field to calculate singular vector (FCSV), which could be implemented earlier than the computation of ANSV in the operational flow and then optimize the computation time for ensemble prediction system (EPS), and it shows that the performance of FCSV ensemble is comparable to ANSV ensemble.Based on CMA-GFS background field and SV calculation module, the feasibility of applying FCSV in CMA-GEPS is investigated. First, the spatial structures of ANSV and FCSV and their similarity index are analyzed. And then, two ensembles based on ANSV and FCSV are conducted for 10 cases in summer and autumn. The forecasts from ANSV ensemble and FCSV ensemble are comprehensively evaluated in terms of the ensemble prediction skill of barometric surface variables, the probability prediction of 24 hours accumulated precipitation in China, tropical cyclone track ensemble prediction skill, and the forecast skill of the minimum sea level pressure at tropical cyclone center. The results show that for the dominant extra-tropical singular vector in CMA-GEPS, the ANSV and FCSV have similar horizontal and vertical structures, their general similarity index is 0.6-0.8, and two ensembles have the comparable forecast skill over extratropics. For tropical singular vector which are only calculated when tropical cyclones are observed, their similarity index between ANSV and FCSV is relatively lower than that in extratropics, and FCSV ensemble shows slightly smaller ensemble spread but comparable error for tropical cyclone tracks. For the precipitation forecast, two ensembles have similar forecast skills for moderate to heavy rain. For mean sea level pressure forecast of strong tropical cyclone case, two ensemble have members showing the skill in terms of structures and magnitude. Therefore, it is feasible to apply FCSV in CMA-GEPS, and it could be an option to construct singular vector-based initial perturbations for future high-resolution operational CMA-GEPS.

Keywords

• singular vectors
• ensemble forecasts