Abstract: Using the 500 hPa geopotential height (H₅₀₀) forecast data of the CMA global ensemble prediction operational system (CMA-GEPS) for a 1-year period from 1 June 2020 to 31 May 2021, the scale-dependent characteristics of error growth and forecast performance of the CMA-GEPS in the Northern Hemisphere are evaluated. The H₅₀₀ field is decomposed into different scales (including the planetary scale, the synoptic scale and the sub-synoptic scale) by using the spectral filtering method. The relationship between the Root Mean Square Error of the ensemble mean (RMSE) and ensemble spread (SPD) indicates that the CMA-GEPS is over-dispersive (RMSE is smaller than SPD) in the early forecast stage (before 108 h), which is mainly attributed to the excessive dispersion on the synoptic scale. In the subsequent forecast period (beyond 108 h), the CMA-GEPS is under-dispersive (RMSE is greater than SPD), which is caused by insufficient spread on both the planetary scale and the synoptic scale. The error growth model modified by Dalcher et al. in 1987 is applied to diagnose the characteristics of the H₅₀₀ forecast error growth. It is found that the error growth processes of CMA-GEPS are reasonable, and the initial error grows fastest on the sub-synoptic scale and slowest on the planetary scale. In terms of absolute (relative) errors, impacts of model errors on forecast errors increase (decrease) with increasing spatial scale. In addition, taking the climatological distribution derived from the daily dataset of the ERA-Interim reanalysis for 30 years from 1989 to 2018 as the reference forecast, the Continuously Ranked Probability Skill Score (CRPSS) is computed to verify the probabilistic forecast skills of H₅₀₀ within CMA-GEPS together with its components on different scales. Results reveal that the forecast skills for the planetary scale are the highest, and those for the sub-synoptic scale are the lowest. Moreover, the probabilistic skills of the unfiltered H₅₀₀ lie between skills of the planetary and synoptic scales. Above diagnostic results can provide an objective basis for further improvement of the CMA-GEPS.