Unlike deterministic approaches, probabilistic seasonal precipitation forecasts can effectively quantify climate uncertainty. Within this framework, skillful prediction of extreme anomalies is especially critical for agricultural planning, water resource allocation, and flood mitigation. Single-model deterministic dynamical systems are inherently limited in probabilistic forecasts of summer extreme precipitation. They often yield spurious certainty and negative predictive skill. To address this issue, this study systematically evaluates the probabilistic prediction skill of China Multi-Model Ensemble (CMME) prediction system for summer extreme precipitation and explores the potential sources of its advantages. We evaluate extreme summer precipitation forecasts from 1993 to 2016 using the Brier Skill Score. A 25-member mini multi-model ensemble (mini-MME) is compared with a single-model ensemble and the full multi-model ensemble (MME). This comparison isolates the distinct impacts of model diversity and ensemble size on forecast reliability and resolution. Results indicate that, the multi-model ensemble successfully reverses the negative skill of the single model by leveraging the diversity of model physical processes. In tropical and South Asian regions dominated by large-scale forcing, the skill improvement mainly comes from better resolution of extreme signals. For mid-to-high latitude regions such as East Asia, where internal variability plays a larger role, the ensemble system substantially improves forecast reliability by increasing  ensemble members and the spread of the probability distribution. This helps correct the overconfidence inherent in the single model. Furthermore, the optimal probability threshold is identified using the Heidke Skill Score (HSS) across the entire probability range. Based on this optimal threshold, the multi-model ensemble demonstrates stable and positive forecast skill both globally and across key regions. It effectively overcomes the limited applicability of the Extreme Forecast Index (EFI) in mid-to-high latitudes. This study confirms the advantages and exploits the sources of multi-model ensembles in probabilistic forecasts of summer extreme precipitation, providing a robust scientific basis for the further objective extraction and utilization of probabilistic information in extreme prediction.

CMME、Probabilistic seasonal prediction Extreme precipitation Ensemble Forecast