In recent years, the damage caused by extreme weather events has frequently been reported in many parts of the world. Therefore the improvement of prediction accuracy for extreme events is an urgent issue. This study conducted to evaluate the sub-seasonal forecast skill with land initializations in an Atmospheric General Circulation Model (AGCM) targeting a severe drought event which occurred in the North American continental in boreal summer of 1988. The experiment using land initialization with influence of human activities, sub-seasonal forecast skill for near surface temperature improved its skill comparing with another forecast experiment without realistic land initializations. We obtained the result of forecast that soil moisture states could come to have a long dry state by considering human activities at the drought event.