The work built a hybrid coupled model based on ROMS and a statistical atmosphere. The paper specifically present the model formulation and its performance evaluations about ENSO. Overall, the work is interesting. The developed model will be a useful for future ENSO studies.

Table 1: I can’t understand how your “complexity” is defined, as well as your degree of freedom? The definition of “variables” for dynamical models is different from that for AI. Your rating/table is misleading, and gives one a feeling that the AI models are much more complex than the CGCMs.

Fig. 1/L121: Not sure how the first SVD modes are derived? SVD needs to be performed by a pair of fields. Please explain how the three fields are used?

Fig. 9/10: The simulated ENSO cycles are too regular. One suggestion for your future experiments is to add some state-depend noise to your statistical atmospheric model.

Fig. 14(e-h): Are you presenting streamlines in the figures? If so, are they derived based on mean currents?

L471: eastward advection anomalous => anomalous eastward advection;

Also, the statements are at least incomplete. The vertical advection of anomalous warm water by mean upwelling is important contributor to the onset of El Ninos. The subsurface warming is also related to the westerly wind anomalies through their triggered downwelling Kelvin waves.