Based on the important synergies of multiple variables in ENSO dynamic mechanisms, a deep learning model is designed in this paper, in which the graph convolution module is used to capture the interaction between factors. Recently, climate prediction using artificial intelligence methods is a fast-evolving area, a vast majority of the research efforts isolate machine learning from conventional knowledge-based modelling, which limits the potential of these two modelling paradigms. However, the author incorporates ENSO knowledge with the deep learning model, and the use of the graphic convolution module to describe the coupling between predictors in ENSO events is very novel. In particularly, the model works well both in prediction of Nino3.4 Index and simulation of the ENSO events. I have enjoyed reading this important article, and recommend publication. My minor comments are as follows:

• 260: There's an in this formula but was not explained ahead. This is supposed to be an identity matrix, right? Please introduce it clearly.

• Figure17-Figure20: The data seems to be standardized and should be mentioned on the figure.

• A suggestion: In this paper, it is found that the best effect is to set the input sequence length as 3. This may be due to selecting the predictors with short memory (vapor, cloud). If predictors with long memory (such as heat content) are added, it may be more effective to set the length longer. Although Table 3 shows the prediction effect of the model with increased heat content data, the input sequence length is the same. This may be taken into consideration in a future study using global data.

ENSO is a main driver of worldwide weather and climate anomalies. Its prediction has been subject to a relatively lower skill since 2000. In this new study, it is an interesting and innovative attempt for ENSO prediction with deep learning. The encoder-decoder structure can capture more nonlinear characteristics of ENSO evolutional process with the complex coupled multivariate interactions in ENSO events. The authors propose using the transfer learning to combine data of different resolutions, which is a good way to figure out the potential data restriction problem and to reduce the epochs on model training. By comparing with multiple traditional neural networks, the ENSO-ASC model clearly shows the improved prediction skills.

Minor comments:

(1) On the ablation experiment, “The calculation of this variable contains the region of SST, so the effect of the extra introduction of upper ocean heat content will be weakened” is at L533. I have a suggestion: if use upper ocean heat content to take place the SST in the model, how will the ENSO-ASC can perform?

(2) The initial letter of the sentence should be uppercase and some mistakes are found at line 103 and line 222.

(3) L374, “N40°-S40°, E160°-W90°”, should be expressed as the region of 40°N-40°S, 160°E-90°W.

(4) In Figure 6, the text looks too small.