The Indian Ocean Dipole (IOD), which is one of the dominant interannual variability modes of SST (sea surface temperature) in the tropical Indian Ocean, has striking impacts on regional and global climate. Thus, finding ways to achieve accurate and short-term climate predictions of IOD is an important subject of research. Recently, a near-term climate prediction system called IAP-DecPreS was constructed by the Institute of Atmospheric Physics (IAP), Chinese Academy of Sciences. It is based on a newly developed initialization scheme (EnOI-IAU), which can assimilate the observed ocean temperature profile records. In this paper, the authors compare the differences in skills of the IAP-DecPreS for the IOD in fall (September-November) between the following two distinct initialization approaches: anomaly and full-field initializations. The results indicate that, for predictions starting from August, the hindcast runs based on the full-field initialization are more skilled at both deterministic and probabilistic predictions compared with those based on the anomaly initialization. For predictions starting from May, the predictive skill of the hindcasts based on the two initialization approaches are similar. Compared with the anomaly initialization, the full-field initialization is superior because it improves the predictive skill for the IOD events occurring together with ENSO. The wind-evaporation-SST positive feedback over the tropical eastern Indian Ocean, which is excited by the ENSO remote forcing, is key for the development and maintenance of the IOD. The hindcasts based on the full-field initialization can reproduce the spatial distributions of precipitation and wind anomalies associated with the ENSO during the IOD development stage. In contrast, for the hindcasts based on anomaly initialization, the biases of precipitation and wind anomalies are much larger. Full-field initialization can reduce the initial errors in the climatological precipitation over the tropical Indian Ocean, thus improving the accuracies in simulating the response of precipitation and wind anomalies over the tropical Indian Ocean to the ENSO remote forcing. In comparison, the anomaly initialization nearly does not change the model inherent climatology. Thus, the ENSO-related precipitation and wind anomalies over the tropical Indian Ocean simulated by the hindcasts based on anomaly initialization show biases similar to those of the model’s historical runs.