The authors of Sources of Uncertainty in the Global Fire Model SPITFIRE: Development of LPJmL-SPITFIRE1.9 and Directions for Future Improvements present an overview of the SPITFIRE model. They review the underlying theoretical basis of the model and attempt to better understand sources of error and model uncertainty. They make several revisions to the model to improve model performance and then evaluate these changes across the European modeling domain. The work appears rigorous from a technical and methodological standpoint. It will be of interest to both active users of the SPITFIRE model, as well as users of other fire models interested in incorporating similar improvements of carrying out model intercomparison.

Major points:

• The methods section is rather short as is the description of the model, the paper moves quickly into the result and, there are very few equations presented and a lot of references to model description appendices. This hurts the flow and clarity of the sections where the underlying theoretical basis is being reviewed especially for readers working with other fire models. Some examples include near line 105 where the methods are presented parameters, and values names, are difficult to follow. Near line 175 the prescribed fire starts could be more completely explained. I believe there are numerous other examples of this throughout the text. Getting information out of the appendices and into the main text would improve overall clarity and readability.
• There are also some sections (i.e. near line 545 and table 1) where there is information presented quite late in the paper in the results section that seems more methodological. Some re-organization here and elsewhere for clarity. Having this information presented earlier would improve the readability of the manuscript.
• Figure 8: This figure may not be referenced in the text. It also appears very late in the paper. But there are still some results sections that follow it. Organizing these sections to be a bit later in the text or making use of this figure earlier on could improve the manuscript.
• Line 403: Do the authors mean “inter-specific” here? PFTs tend to be coarse and wouldn’t represent differences between species. “Intra-specific” differences and “adaptation” would be on an even lower level and a lot of what’s in Appendix C3 seems to focus on “inter-specific” differences. Revising this for clarity and moving some text and references from C3 could address this.

Minor points:

• Abstract: reword “strong upward biases”
• Abstract: reword “moisture parameterization that results in substantial too low live grass moisture contents”
• Abstract: suggest revising the sentence stating “that bias SPITFIRE towards higher tree mortality” to better explain the mechanism. Is this tree mortality in grasslands?
• Line 83: when “assigning a uniform site” clarify if this means vegetation height is static or dynamically determined by the vegetation model.
• Line 613: reword “upward biases”
• Line 657: add information and reference for regions where shrubs are an important part of the fire regime for clarity.

In this study, Oberhagemann et al. conduct a thorough review of the SPITFIRE fire module, identifying sources of uncertainty and implementation errors, including two major errors in the model. The study also uses two new methods for validating LPJmL-SPITFIRE that allow for improved validation of different components of the fire model that extend beyond the usual comparison to the observed burnt area. The authors implement numerous improvements to the model and evaluate the improved version across a European domain, discussing uncertainties that exist and areas for future model developments. The work appears to use valid approaches and methods and presents important corrections for errors that have consequences for SPITFIRE-coupled models and potentially other fire models that were originally based on the Rothermal fire spread equations. It is therefore of interest to the fire modelling community and more generally the vegetation and earth system modelling community that include SPITFIRE or closely related fire models within.

Specific comments:

• The study starts with LPJmL version 4.0, but the improvements and model developments shown later in the study are using LPJmL version 5.7. The authors argue that version 4.0 was used to start as it is the most recently published global version of SPITFIRE, while the more recent version was used to include the most recent updates to LPJmL. Whilst it is understandable to use v4.0 to illustrate the errors and uncertainty in a published version, it would be nice to see a comparison between the model versions themselves if different versions are to be used. For example, how do figures 2 and 3 look using LPJmL v5.7? This would help make the results more comparable throughout the manuscript.
• Throughout the manuscript, where different models and approaches are mentioned, more details on differences would help inform the reader. For example, in line 31: Various DGVMs are listed, including several LPJ-based models. A brief statement on how the models differ from one another or their strengths and weaknesses would be more informative than a list. At least for the LPJ models, since it’s often confused what is included in each. Related to this, is the multi-day burning the same that is implemented in the LPJLM-fire DGVM?In line 81, "This description does not apply to other vegetation models in which SPITFIRE is implemented." More detail on why this is the case is needed to describe in which cases the description applies—for instance, is it that other DGVMs do not use the area-averaged approach but patch/cohorts, etc.?
• For Figure 2/3. It could be useful to have some spatial plots of simulated vegetation, as well as defined grass and tree grid cells, compared to observational to show the vegetative differences and eliminate them as a cause for differences in burnt area. Similarly, in the caption of Figure 2, examination of individual tree PFTs is mentioned but not shown. Whilst these plots may crowd the manuscript, they could be added into the SI.
• From line 175, my understanding is that LPJmL4-SPITFIRE is able to simulate smaller fires in large numbers well and the fewer but larger fires less well. This is also alluded to later in the paper (e.g., L583). Yet Line 180 seems to contradict this, stating that the incorrect implementation results in unrealistically large and severe fires. Is it that one is true on a global scale, while the other is true for just grasslands? Clarification is needed here.
• In Figure 4 and related discussions, ROS differences under low wind speeds have a large impact on fire size, but even larger differences in ROS result in minimal differences in fire size. This is due to the size of rate of spread difference compared to rate of spread values; however,  further clarity is needed for readers. Either giving an example in text of how the larger differences under high wind speeds result in little impact on fire size or ideally replacing Figure 4. Panels a) and b) with % differences instead.
• Why were the parameters and factors in section 3.3 chosen? For example, the minimum fire duration of 2 hours and a maximum of 7

Technical corrections

• Line 73: add the most relevant references on SPITFIRE there.
• Figure 3: swap around descriptions of b) and a)
• Figure 4: Clarify in the caption that it is SPITFIRE-Rothermal and mention of what the TL3/TU2 fuel classes are (or alter the panel titles to include).
• Line 438: “to rectify this,” what is meant by ‘this’? Reword for clarification.