The numerical seasonal prediction system using the Scale Interaction Experiment–Frontier version 1 (SINTEX-F) ocean–atmosphere coupled model has so far demonstrated a good performance for prediction of the Indian Ocean dipole mode (IOD) despite the fact that the system adopts a relatively simple initialization scheme based on nudging only the sea surface temperature (SST). However, it is to be expected that the system is not sufficient to capture in detail the subsurface oceanic precondition. Therefore, the authors have introduced a new three-dimensional variational ocean data assimilation (3DVAR) method that takes three-dimensional observed ocean temperature and salinity into account. Since the new system has successfully improved IOD predictions, the present study is showing that the ocean observational efforts in the tropical Indian Ocean are decisive for improvement of the IOD predictions and may have a large impact on important socioeconomic activities, particularly in the Indian Ocean rim countries.