Abstract: In order to resolve the dynamically inharmonic problem between assimil ated observational data and dynamic model in the original initialization scheme of the NCCo model, and improve its prediction skill on ENSO, a new initializatio n scheme called dynamically harmonic initialization scheme (DHI hereafter) base d on a data reconstructed method was proposed. The main function of this data re constructed method is to separate the model compatible part of observational inf ormation from the original observational data through inversing the information of model climate attractor contained in the longterm coupled simulations. To v erify the capability of DHI, the prediction skill of DHI and its relative mecha nism were also comprehensively analyzed. The results show that most of the model related ENSO scale noises in the intialfield caused by disharmony between the assimilated data and the model from the original initialization scheme have been squeezed out, while the model related ENSO scale signals in the intialfield hav e been well preserved. The corresponding fast prediction errors arisen from prop agation of none ENSO scale noises are effectively halted. Thus, the prediction s kill is improved greatly, and the harmony between the NCCo model and assimilated data is easily reached when it is initialized.