The authors of “Understanding and simulating cropland and non-cropland burning in Europe using the Base model” use generalized linear models to develop a fire model capable of predicting cropland and non-cropland burned area in Europe. This model is likely suitable for use in land surface and climate models. To my knowledge, few land surface models include the ability to model cropland fire. This work is timely, technically rigorous, and falls within the scope of bio-geosciences. I have several comments which are listed below.

• Intro: The lack of land surface models capable of representing cropland fire is mentioned in the discussion. I suggest discussing it in the intro as well as the motivation for this work.
• Table 1: Adding the data source and citations could help better inform the reader
• L270: How were the data points sampled? Was anything done to uniformly distribute the sampling across space, or account for spatial autocorrelation?
• L460: Some text comparing and contrasting these models with mechanistic models could be interesting. For example in mechanistic models, wildland fire is influenced by wind and terrain which impacts spread, whereas cropland fire appears to be a more complex phenomenon perhaps better suited to description using a statistical model.
• L575: The analysis of the role GDP and HDI play in the model is interesting. Can the authors provide insight into whether this is corelative or causative? Do these relationships apply in time (i.e. moving into the future)? What if there were abrupt changes in these metrics due to for example a short-term financial crisis?
• L655: Did the authors consider other remote-sensed burned products that might include small fires like GFED4s?
• L690: Finally, can the authors address if this model is specific to this region or could be transferred to other regions of the world? How involved do they believe the process of doing this would be?

Minor comments:

• L26 here and elsewhere rephrase meteorological fire danger for clarity
• L34 remove “of” just before “state of the art”
• L39-40 suggest rephrasing these sentences
• L49-52 split and shorten this sentence
• L58 rephrase “coherent political level” for clarity
• L210 revise “artefacts”
• Figure 1: Here and through the figures would be clearer if the acronyms were defined in the axis labels and figure captions
• Figure 5: Here and elsewhere the single shared legend and brief caption could be clearer if they provided information about the mean lines, uncertainty regions, etc.

The authors present an important contribution to a pertinent issue in wildfire modelling, adding an important perspective to the hot topic of the differences between natural vegetation versus cropland wildfire. The model is well-described and rigorous, finding concrete differences between the drivers of burning in cropland and non-cropland vegetation. Performance statistics are largely better than fireMIP models, but similar-to-worse than the other GLM based studies cited in the introduction. However, as this is the only study focussed on Europe and is at a finer resolution than most of the other studies, the only conclusion that can be reached through this comparison is that BASE’s performance meets an acceptable threshold for publication. 

Whilst the paper’s conclusion emphasises predictive applicability to the projection of future burned area, the overall performance is not greater than existing global and regional modelling methods. The crucial contribution of this paper is the use of a twin-approach to cropland and “natural” vegetation wildfires. The authors demonstrate that this can substantially reduce confounding effects in specific drivers (e.g. population density or fire weather) that differ between land-cover types, and that trends in NCV/cropland fires can be disaggregated. The paper thus reflects an important contribution to the next generation of fire models, towards improving overall model performance and also to understanding the complex dual effects of climate warming and land-use change over time.

Specific Comments:

• L95: provide evidence/citation that existing global fire models do in fact perform worse in other biomes due to this training bias. Is this the case for all widely-adopted fire-enable DGVMs?
• L180/L190: explain why MEPI and PHI are defined as relative to GPP maximum month of prior year; as maxima are less stable against interannual variability than, for example, the study-period mean, what was the advantage to this formulation?
• L200/Table 1: In table 1 it is stated that FAPAR12 models fine fuel build up over twelve months. At lower values of fAPAR there is a roughly linear relationship with LAI (LAI ~ - 0.5 ln(1 - fAPAR)). This is only one component of leaf litter accumulation. But this neglects leaf mass per area and leaf lifespan, which can vary substantially between needleleaf/broadleaf/deciduous/evergreen. So what does fAPAR12 actually mean, and why is it selected over the more physical GPP? Could it be that the modelled GPP product does not give as spatially reliable a map as the remotely-sensed fAPAR? This could explain the decreased spatial (but improved interannual) performance when GPP replaces fAPAR (table 2). This issue can either be addressed with a good explanation of FAPAR12’s physical effect vs GPP12 in the method section, or by discussing the implications further in section 4.1.3.
• L214/223: Two sources of GDP data are cited, which was used in this study?
• L478: Could it be that these regions are too wet to seasonally burn?
• L568: Spain does differ, but there also appears to be more simulated cropland fire in France, Poland and the Baltics. It would be good to either acknowledge this difference, or to justify that this visual difference in the maps is not as significant as the Spanish case (e.g. due to GLM ‘smearing’ or colormap choice).
• L687: Consider changing “so is suitable for projecting changes in fire hazard over annual-to-decadal time scales” to “so is suitable for projecting differing changes in fire hazard between cropland and non-cropland vegetation over annual-to-decadal time scales”. Not to say it cannot be used in general projections, but that this is the unique value of the model compared to similarly performing fire models.

Technical Corrections:

• L245-250: give the list of actual land cover classes in the supplementary for better readability.
• L393: add full-stop.
• L490/491: correct “1)” and “2)” to (1) and (2) in sentence.