Abstract: The initial condition and lower boundary forcing are the main sources of the predictability in extended range, and they play different roles in different time scales and different regions. Using the latest version of BCC_AGCM, four experiments with different combinations of atmospheric initial condition and sea surface temperature (SST) boundary forcing are designed in this study to investigate the roles of atmospheric initial condition and SST boundary forcing in the extended prediction over different regions around globe. Results show that the skills in extended prediction are strongly dependent on initial condition at lead times less than three weeks, and different initial conditions with the same boundary conditions lead to significant differences in the prediction skills at sub-monthly time scales. However, the initial condition can still provide some useful information for the prediction exceeding monthly time scales. From a global perspective, SST boundary forcing affects the prediction skill at lead times of more than one week. It will take roughly four to five pentads for boundary conditions to have the same effect as initial conditions have on the predictability of geopotential height at 850 hPa in the tropics, and this lag time will be longer in other regions. Similar results can also be found in the geopotential height prediction at 500 hPa except that the lag time is 5 pentads in the northern hemisphere and 6 pentads in other regions, which means the above effects in the upper atmosphere lag behind that in the lower atmosphere by one to two pentads. In the region of East Asia, the contribution of boundary forcing to the prediction skill appears on the second pentad accompanied with significant improvements in forecast at this time. As for the extended forecast, impacts of the lower boundary forcing are concentrated in low-latitude regions, and are more stably shown in the forecast of geopotential height at 500 hPa. Significant improvements caused by the boundary forcing are shown in the forecast after the 5th pentad. The results indicate that both initial conditions and boundary conditions are very important in extended forecast, and understanding the relative roles of initial and boundary conditions in the extended prediction is the basis to improve forecast skills.