Reza et al. improved the fire module in the SEIB-DGVM model and used the model to project Siberia wildfire dynamics, and emissions under various climate scenarios. I appreciate the valuable model development and analysis, and it provide us important implications on how future climate change affects Siberia boreal forest. The paper is informative and has a lot of important results to discuss, but it is not well-written in terms of smoothness and readability. There are confusions on methodology and data usage. Most importantly, this manuscript over-emphasizes future RCP projections. However, historical validation is not clear and convincing. A more reasonable approach is to first thoroughly demonstrate the model performance during the historical period by comparing it with observations. Then discuss future projections and caveats. Below are my major comments.

• There is no quantitative evidence of the model's performance. During the historical period, how did the model simulate burned area, and fire emissions, compared with existing datasets (e.g., GFED) at gridcell level? Plot scatter plots of modeled versus observed variables (burned area, emissions, biomass) will be super helpful. Showing R2 and RMSE of gridcell level comparison in the abstract is highly encouraged. As it is stated in the abstract the motivation for integrating SPITFIRE into DGVM is to improve the accuracy of fire modeling. Therefore, it is also important to show how much improvement has been achieved by SPITFIRE compared with DGVM’s default fire model.

• Improvement from annual time step estimate to monthly is an important change. However, most of the fires last less than one month. How does the monthly time step fire module resolve a process that lasts much shorter than a month?

• Human ignition is considered in E term (equation 2), how about human suppression on fire spread?

• Methodology, between 2.2 model description and 2.3 model application, there seems missing a section about model calibration. How was the SPITFIRE module calibrated for burned areas or emissions? How was the DGVM model calibrated to capture observed AGB?

• For future projection, are population density changes and lightning flash changes considered under future scenarios?

• Section 2.4, the model is validated by GFED4 burned area and GFED4s burned fraction. However, GFED4s and GFED4 are different because GFED4S include smaller fires. How to reconcile GFED4s and GFED4 during validation?

• Section 3.1-3.4. Historical validation is an important component of this study. I would like to suggest before discussing RCP results thoroughly compare the simulated burned area, emission, and biomass with observations. Discuss model performance and biases at gridcell level, and the regional level.

• Section 3.5 Fire-off simulation is not mentioned in the methodology section. When was the fire turned off? Does the model run an extended spinup with fire-off? One would expect much larger aboveground biomass when the fire is turned off.
• Figure 7 showing spatial coverage is confusing because GFED4s provide burned fraction, which should be directly compared with SEIB-DGVM at gridcell level. Figure 8, a and b are duplicated, both showed long-term monthly average dry matter emission. Figure 9, why dry matter emission is perfectly simulated by SEIB-DGVM (shown in Figure 8), but the performance of CO2 emission is much worse. It seems inconsistent. What does it look like, if plot a spatial average comparison of DM emissions between SEIB-DGVM with GFED4s?

Please find my review attached.

This study projected future wildfire activities in Siberia by implementing a process-based fire module into a dynamic vegetation model. The main conclusion was that fire emissions continued to increase due to climate change. The authors also quantified the negative impacts of fire activity on tree mortality and NPP. While the topic of the study well fit the scope of BG, I found the presentation was very poor and many results were contradicting. I suggested a major revision but with strong doubts about the credibility of the model and the projections.

First, the model descriptions were unclear. The authors put most of details, including equations and descriptions, to the supplementary material. However, the key processes should be listed in the main text for clarity. Most important, the links between the fire module and vegetation model should be explicitly explained. For example, what parameters did the SEIB-DGVM provide for SPITFIRE, and how the fire activity affect the vegetation distribution and ecosystem productivity. Such information determines which variables to be validated against observations.

Second, the model validations were questionable. Normally, the validations come before the projections. However, this study put the validations of fire models in the section 3.7, far behind the future projections. Some of the validations were not necessary. For example, Figure S2 compared the input and output of lightning flash rate and population density, which were actually the input of fire models instead of the prediction. Some validations were not consistent between different presentations. For example, Figure S17a compared the simulated and observed dry matter. The model predictions showed poor performance by missing almost all the observed fire episodes. However, in Figure 8, the monthly validations of simulated dry matter showed perfect performance with almost the same magnitude (Figure 8a) and R2=1 (Figure 8b) against observations. These results were too good to be true and seemed not consistent with Figure S17. Furthermore, the perfect simulation of dry matter (Figure 8b) resulted in a poor prediction of CO2 emissions (Figure 9). Was such bias attributed only to the emission factors? Then why making such a great effort to develop the sophisticated fire module when the simplest parameters caused the largest biases? BTW, I do not believe the R2 >0.6 in Figure 9 given such a wide range of scattering points.

Third, the future projections were doubtful. This study used daily meteorology from four RCP scenarios output by MirocAR5. How to reduce the uncertainties from a single climate model projection? Normally, these scenarios showed very different tendencies of warming, indicating different fire probability. However, the projections of fire activities showed very similar results among these scenarios (e.g., Figure 4, Figure S5, Figure 12 …). Does it mean that the future wildfire activity in Siberia will increase at the similar rate no matter how warm the climate becomes? Furthermore, the updated fire module showed good correlations between burned fraction and burned biomass (Figure S11d), suggesting tight connections between these two parameters. The burned fraction is projected to increase continuously after the year 2040 (Figure S5b). Then why burned biomass showed such a large fluctuation over the same period?

Finally, the quality of result presentations was low. For example, the figures were very similar among the subplots of Figure 4e-4h, Figure, Figure S4e-S4h, Figure S6. Such information is useless as the readers could not tell their differences. Figure S28 showed the projected fire emissions of 33 trace gases by putting all the subplots together without any summary. It’s difficult to tell their differences and the main conclusions. The authors also spent great efforts in comparing results from the default and updated fire modules (e.g., Figures S4, S5, S9, S10, S11). While it’s important to understand how different the fire predictions before and after the improvement of fire modules, the authors could show some key results (e.g., Figure S11) and put more efforts in the validations of the updated model against observations, not only for fire activities but also some ecosystem parameters (e.g., tree height, biomass, NPP).

Figure 10: What’s the meaning of the shadings in (a)? What’s the meaning of the points in (b)?

Figure S5: What's the reason for the sharp decline of burned fraction around 2035 in (b)?