This study presents a spatially varying parameter optimization for a process-based model to better capture the spatial variation in observation-driven GPP, aboveground biomass and mortality rates over Amazon basin. The authors first identified the two most sensitive parameters (Coefficient of the self-thinning relationship, α, and Nitrogen use efficiency of Vcmax, η) affecting the simulated GPP, AGB and mortality rate. They then optimized the two parameters for each grid cell by using a set of parameter samples generated from the Latin hypercube sampling (LHS) method to drive the model, interpolating the simulated the results, and obtaining the optimized parameters with minimized the quadratic losses. The results showed a large improvement in model simulation of GPP, AGB and mortality rate using spatially optimized parameters against observations. This study further identified the potential driving factors of the optimized spatial pattern of the parameters using random forest regression and SHAP values, and discussed the underline mechanisms. The study is well designed and provides useful implication in future parameter optimization for similar model developments. The manuscript is well written with justified results and details supporting the conclusion. I have a few minor comments as follows:

• Section 2.5, L259-260: it is not clear how the spatially constant parameters were derived. Was it done using regional total GPP, AGB and mean mortality rate from simulations and observation to derive the two parameters? Specification is needed.
• Figure 2: the caption seems to be different from the text in the figure. E.g., a-c are observations.
• Section 3.3, L337-338: the purpose for running the using parameters predicted by random forest is not clear. As the influencing factors can capture only 45% and 48% of the variation in α and η, it is expected the reduced accuracy.
• Section 4.1, L351: it seems that “our model”means the simulating using optimized spatially varying parameters, but to be specified to avoid confusion.
• Section 4.4: though an improved simulated spatial patterns of the multi-year mean GPP, AGB and mortality rate are the purpose for the optimization, how such parameterization might affect the interannual variation and long-term trends of the variables could be briefly discussed here.

The authors presented spatially explicit parameter tuning to improve a DGVM simulation of biomass, GPP and mortality gradient in intact Amazon forests. The work is very comprehensive and the results make sense for me. I have only minor comments below for the authors to consider:

The Introduction would benefit from an enhanced discussion on how the spatial gradient of morality rates was driven by different types of mortality, e.g., drought, wind and self-thinning as optimized in this study.

Line 117: How wood density influences carbon cycle/mortality processes in ORCHIDEE? I noticed section 1.2 in Supplement but still it does not explain the physiological impacts of wood density.

Line 164-170: what is this ‘background mortality’? I would think this is just self-thinning before it is represented in the model. Now that we already have self-thinning explicitly represented, why we still need this ‘background mortality’? How its magnitude is determined?

Line 204: the spinup process was already influenced by the parameters governing the self-thinning mortality and hence there is the issue of circularity here. How is this considered/addressed?

Line 205-206: “An equilibration of the soil biogeochemistry was not needed because of absence of any effect on vegetation processes. ”: I found this statement too much absolute. In principle the soil must influence vegetation processes through nutrient supply and the regulation on soil water process (transport, soil water holding capacity) that are influenced by soil biogeochemistry. Pls rephrase.

Fig. 2 has a wrong caption when indicating the subpanels. Correct.