Abstract: Convection-allowing ensemble prediction (CAEP) is an important approach to improve the capability of strong convective weather prediction, and how to construct reasonable initial disturbance is one of the key issues of CAEP. In this paper, the experiments of the per⁃ turbed-observation (PO) method in the CAEP system in the Inner Mongolia region were carried out and evaluated by comparing them with the downscaling (DOWN) method. The performance of the PO method in the CAEP system was then analyzed, which will provide a technical ref⁃ erence for the construction of the CAEP system in Inner Mongolia. The results are as follows. (1) The initial perturbation constructed by the PO method can effectively include the observations in the Inner Mongolia region, which can reduce the uncertainty of the background field and the perturbation has sufficient growth capacity. (2) Compared with the DOWN method, the PO method can significantly reduce the short-term forecast error of CAEP. The root mean square error (RMSE) of upper-level elements is reduced by 4% ~43%, and the RMSE of ground surface elements is reduced by 3% ~9%, suggesting a slightly decreased ensemble spread. The continuous ranked probability score (CRPS) of upper-level elements can be reduced by up to 53% and the CRPS of ground surface elements is reduced by an average of 6%, which generally indicates an improvement in the quality of convective scale ensemble forecasts. (3) The PO method can also improve the capability of short-term precipitation forecasts. The TS score for precipitation levels of 0.1 mm, 4 mm, and 13 mm increased by 0.015, 0.003, and 0.0015, re⁃ spectively. Furthermore, the case study shows that the PO method is more accurate in predicting the precipitation areas and intensity levels.