This talk will review the recent advances of GFDL’s high-resolution seasonal climate prediction system in terms of modeling, initialization, seasonal prediction, and predictability sources. Beyond the primary roles of ocean state (e.g., ENSO) in seasonal climate prediction, we find that the atmosphere initial state plays significant roles in improving the short-term seasonal climate prediction. The physical processes translating the observed atmosphere initialization into predictive skill, such as stratosphere-troposphere interaction and air-sea coupling, will be discussed. As a case study, we will discuss the roles of atmospheric initial state in predicting the winter precipitation anomalies over the western United States during the major 2015/16 El Niño event.