An exceptionally strong and persistent snowstorm event (PSSE) occurred in southern China from 10 January to 3 February 2008, which caused considerable economic losses and many casualties. To what extent that dynamical model can predict such extreme event is crucial for disaster prevention and mitigation. Here, we found the three S2S models (ECMWF, CMA1.0 and CMA2.0) can predict the distribution and intensity of precipitation and surface air temperature (SAT) associated with the PSSE at 10-day lead and 10-15-day lead, respectively. The success is attributed to the models’ capability in forecasting the evolution of two important low-frequency systems in the tropics and mid-latitudes (the persistent Siberian High and the suppressed phase of the Madden–Julian Oscillation), especially in the ECMWF model. However, beyond the 15-day lead, the three models show almost no skill in forecasting this PSSE.
    The bias in capturing the two critical circulation systems is responsible for the low skill in forecasting the 2008 PSSE beyond the 15-day lead. On one hand, the models cannot maintain the Siberian High, which results in the underestimation of negative SAT anomalies over southern China. On the other hand, the models cannot accurately capture the suppressed convection of the MJO, leading to weak anomalous southerly and moisture transport, and therefore the underestimation of precipitation over southern China.
    The SVD analyses between the critical circulation systems and SAT/precipitation over southern China shows a robust historical relation, indicating the fidelity of the predictability sources for both regular events and extreme events(e.g., the 2008 PSSE).
 

S2S prediction model,subseasonal prediction skill,MJO,Siberian High