Abstract: Aiming at the event of medium-range forecast "drop-outs" or forecast busts of CMA-GFS in East Asia and using the forecast case from 12:00 UTC on 30 December 2021, backward error tracking and ensemble sensitivity are used to give a first guess for the source region and relaxation experiments are used to confirm that the region is east of Greenland. The results indicate that the 24th hour forecast bias in the east of Greenland and downstream propagation finally lead to the forecast drop-outs in East Asia. On this basis, by relaxing different model prognostic variables we can track and analyze the contribution of each variable to the total forecast errors and its downstream impacts. It is found that the forecast bias are caused by the prognostic variable θ in the key error region, and v component also contribute lesser with the time integration. The reduction of the forecast error in the east of Greenland region can improve the forecast performance of the CMA-GFS greatly in East Asia and Northern Hemisphere.