The study uses a new version of the ORCHIDEE model to study elevated CO2 impact on forest growth and mortality in the Amazon in the past decades. The model was previously calibrated at several Amazon sites and was applied at regional scale with and without historical CO2 increase. The simulations with elevated CO2 can better reproduce the temporal trend of C gain and C loss estimated from long-term field plots. Comparison between the simulations with and without CO2 effects show that elevated CO2 increased both growth and mortality while the latter is caused by increased competition and elevated CO2 reduced drought-induced mortality. Further spatial analysis reveals that the CO2 effect is stronger in drier regions.

Overall, it is neat to use a model to separate the processes (CIM and DIM) over the Amazon. The manuscript is clear and well written. I feel the mortality response makes sense but I am not sure how much we should trust simulated growth responses to eCO2 as outlined below.

I am concerned that models overestimated average carbon gain and carbon loss by ~30% or more (3.0-3.5 Mg/ha/yr vs observed 2-2.5 Mg/ha/yr, Fig.3) in simulation A2 but not in A1. To me, this means simulations with elevated CO2 greatly overestimated baseline growth (and thus mortality), suggesting the CO2 fertilization effect might be overestimated. It would be important to explain this difference in baseline values.

In addition, the positive trend of carbon gains in observation is mainly due to increase from 1980s to early 1990s. I believe the trend is much weaker after 1990s and in the same Hubau et al. study, there was not growth trend in Africa, suggesting CO2 fertilization effect on growth is quite uncertain. For instance, van der Sleen et al. 2014 reported no growth simulation by CO2 from tropical tree rings. More recently, Jiang et al. 2020 reported eCO2 increased GPP but not woody NPP in an Eucalyptus woodland. Such allocation changes are briefly mentioned in Discussion (line 415 - 425) while I think it should be highlighted as one of the major limitation/uncertainty of the study. For example, how would your conclusion change if the CO2 effect on growth is overestimated by 50% - 100%?

Finally, since AmazonFACE is mentioned, it would be interesting to provide results from some short-term (e.g. 5-10 years, single site) simulation results using similar magnitude of CO2 increase. This can serve as a priori estimate of AmazonFACE results (not necessarily correct).

In this study, Yao et al. used a well-established ecosystem model equipped with plant physiology, demography, and hydraulic processes to simulate the carbon sink response to CO2 fertilization in the Amazon rainforest. The results in the figure and texts are well presented, and the experiment simulations are reasonable. While I enjoy reading this work, I found that the paper needs to extract more clear messages especially in the Abstract and Conclusion. For example, what do we learn from this advanced improvement of the model process related to mortality and hydraulic resistance to droughts, and what does this imply for the carbon cycling in Amazonia? The message is not totally clear to me though detailed results have been reported. 

My other minor comments are mainly about clarification issues. In Lines 145-150, since the carbon gain and loss time series are from Brienen et al. (2015), why do you say in the first paragraph of the results that the model simulates these two? How do you get carbon gain and carbon loss from the model output? What are the output variables? 

The definitions of drought resistance and resilience are not entirely clear to me. The equations are clear, as in Equations (5) and (6). But what do these metrics imply for drought resistance and resilience? More explanations are needed.

Overall, I think this work is very novel and represents our newest process understanding of the Amazonian carbon sink from CO2 forcing from the perspective of models. But the messages need to be clearer.