General Comments: 

The authors conduct 4 experiments, where different combinations of moisture codes (FFMC, DMC, DC) in the FWI system are replaced by outputs from the PEATCLSM model (adjusted for L-band assimilation; as per Bechtold et al., 2020). Hydrological model variable outputs are rescaled to match CDFs of moisture codes. The “experimental” variables are then compared against the ability of the original FWI components to predict fire occurrence in peatlands. 

The work is a much needed step towards developing fire risk prediction methods for peatland fires however the approach appears to diverge from the original purpose of the FWI system, requiring additional justification and discussion. The FWI was developed to estimate fire danger, whereas here the (PEAT-)FWI is evaluated in its ability to predict fire occurrence. As stated, the two are likely correlated but this is an important note that needs to be addressed more thoroughly before the conclusion. 

The discussion reads more like a further description of the results with some added justification, but no real discussion of the impact on predicting fire danger. E.g., Replacement of FFMC with sfmc results in smoother moisture values for the top fuel layer… is this supported by research of peat fuel moisture? What is the impact on predicting fire danger i.e., hits, misses, TPR, of EXP3 and 4 vs FWI and EXP1 and 2. It might be pertinent to focus on the net change or the misses to hits (green) % change for each EXP, as this is the aim of the study – to better predict fire danger (or in this case, occurrence). 

The title appears to be slightly misaligned from the work. I would suggest something like “Using an adapted Fire Weather Index to predict fire occurrence in Boreal peatlands.”  Or “Modifying the Fire Weather Index System for peatland wildfire occurrence using hydrological modelling”. 

Methodological Comments: 

-) Can you provide justification for your assumption that “soil moisture doesn’t control ignitions in early season”? 

-) What is the principle from Li et al., 2010, briefly? 

-) Can you clarify how matching CDFs per grid cell removes dry biases over the Boreal Plains? Li et al. states: “the dependence of the bias correction results on the choice of time period for training” hence, how does your training period affect BP projections? 

-) What is the potential impact of using MERRA data to calculate the (original) FWI? This is not how the FWI system is used operationally (interpolated weather station data is most common), and therefore you may be reducing the quality of the FWI output before you start. For example, limited ability to represent low RH has been found in some global climate models and Field et al. discuss issues with MERRA’s accounting of precipitation.

-) Can you justify the use of fire ignitions as a measure of fire danger? Or caveat the need for this simplification/assumption via the difficulty of modelling fire danger (intensity) for peatlands. 

-) What is the historic period for evaluating FWI? 30 years? 

-) How critical are those subjective thresholds of 70 and 90th percentile? Could a simple uncertainty analysis be conducted? 

-) It might be pertinent to focus on the net change or the misses to hits (green) % change for each EXP, as this is the aim of the study – to better predict fire danger (or in this case, occurrence; see above comment). 

-) What proportion of your study area is underlain by permafrost? How might that impact fire danger, and the hydrologic modelling (and therefore PEAT-FWI)? 

-) What is the total number of peat fire ignitions in the dataset? 

Line Edits: 

Line 17: References could be updated/expanded. 

Line 18: Consider “saturated” over water-logged. 

Line 62-65: Incorporate this into a paragraph.  

Line 70: “The fuel for early-season fires is mostly the dead plant matter of the previous growing season and thus less influenced by current soil moisture” Is this statement supported by literature? Although it is hypothesized (and there is some evidence) that peatland fire severity worsens throughout the fire season, there is much evidence that high peat burn severity is controlled on a much smaller scale (peat properties) and across scales (topographic position, hydrogeological setting), and that extensive peat organic matter can combust during early season fires. 

Line 70-75: Parisien et al., 2023 refers to boreal-wide ignitions and I’m not sure how context on anthropogenic vs lightning ignitions supports the assumption that soil moisture is having different levels of control over wildfire ignitions in early or late season. Note that I don’t disagree with this sentiment, just the supporting evidence. 

Line 76: “Furthermore, the fuel moisture of the latter depends on the peat moisture status (Harris, 2008)”. What does this sentence mean? Seasonal change or peatland type or something other? 

Line 100: “if the fine fuel is dry” is redundant since the fuel moisture is accounted for in the moisture index. 

Line 100: Please clarify what is meant by “without taking further vertical drying into account”. 

Line 103: I think this can be more concisely written as the “potential” fire danger i.e., if a fire were to ignite, or is already ignited.

Line 106: and historical data and context. 

Line 108+: Consider “original” rather than weather-based or actual, FWI calculations. 

Line 108+: Please provide specifications of the MERRA data – resolution, downscaling applied, etc. Similarly, please provide details on validation or accuracy measures conducted on the GFWED method. Somewhere here it should also be noted that most operational users of the FWI system use specific local weather (met) stations and interpolated station data to calculate FWI. 

Line 126: Please define DA. 

Line 150: fire presence, is that referring to an ignition or an active fire? Or a burned area?

Line 203: Could you comment on the role of human vs lightning caused fires altering the FWI threshold? 

Line 210: Is FWIEXP == PEAT-FWI? Please keep acronyms and notations consistent throughout. 

Figure 4: Would be fine as supplemental figure. 

Line 253: Despite accounting for X (i.e., relatively small) percentage of ignitions. 

Figure 6: Can you show which fires are hit by all FWIs and missed by all FWIs? Or give summary stats of these? 

Figure 7: What about which fires are hits to misses and which are misses to hits? 

Can you give numbers for the net impact of each EXP? 

Figure 8: I’d like to see FRP and TPR in a table. 

Discussion: 

Line 325 - 333: I don’t find the following analysis useful to the discussion as study regions/ ecosystems are not comparable. Perhaps replace with discussion on the challenges of mapping both peatlands and fires and then discuss why your approach is well-suited. 

Line 338: Merge this with above paragraph. 

Line 361: missing bracket

Line 361+: Showing data for one time period for one specific geographical location is helpful for readers to understand the differences in temporal dynamics of the FWI and PEAT-FWI however it doesn’t stand up when discussing the overall efficacy of the PEAT-FWI, in particular because fire ignitions are so stochastic and therefore you could just as easily show a time/place where high PEAT-FWI does not co-occur with fire occurrence. Specifically, the wording: “As these two values reach their maximum, five fires are observed, indicating that they estimate fire danger relatively well.” is not justified in this context. 

Line 361+: Further, discuss the role of lightning/ignition in fire occurrence. 

Line 373 – 385: Some of this could go in the Methods; description of FWI moisture codes. 

Line 384/6: Replace “spiky” with “sensitive” or “highly fluctuating” 

Line 386 – 394: Comment on the potential for hydrologic disconnection of water table and peat surface under deep water table conditions (and/or frozen ground).  

Line 425: “The fact that EXP3 mainly shows misses to hits and even fewer hits to misses than EXP1and EXP2 for the early fires indicates that these early fires are not so much driven by hydrology.” Is this the correct way around? I see many more red bars (hits to misses) for EXP3 in early fires. In general I find this terminology confusing and wonder if there’s a more concise way to describe it. 

Line 444: “Overall, the PEAT-FWI performs worst over Canada, which could be related to the more aggressive fire management compared to e.g. Siberia, which influences the fire behavior overall. It can cause fires with a very high PEAT-FWI to be extinguished before being detected by satellite remote sensing, eventually lowering the performance of the PEAT-FWI.” Do you have evidence of this occurring? 

Line 449: “They also found that the Canadian fire season peaks earlier than the Russian fire season (April-May for Canada versus May-June for Russia; De Groot et al., 2013”. After reviewing the paper I do not agree with the above statement, see Table 7 for monthly fire stats. See Hanes et al 2019 and Parisien et al., 2023 for more up-to-date analysis of the Canadian fire season. Note its own binomial distribution (which does not align with the GFA), and the peak in July. 

Review

The fire weather index improved for boreal Peatland Using Hydrological modelling and satellite based-L band microwave observations

By Mortelmans et al.

The paper investigates the use of hydrological model outputs  specifically designed for peatland to replace part or all components of the FWI danger rating systems. It responds to a known limitation of the system that, while developed for above ground fires, has been, and still is, commonly used in global fire forecast systems.

It is well known that the FWI does not correlates very well  with fire activities where the fuel is very dissimilar to  forests. So I believe it is a valuable idea to address this problem. The paper is well written and explores in very much details the source of predictability that can arise by employing the water table and/or the soil moisture content. Certainly I recommend publication of the manuscript as it provides an useful framework to improve fire danger prediction.

However given the results that clearly highlight that the direct use of the water table is a better predictor than the FWI for peatland, I am asking the author if suggesting to rescale this variable to match the value of the FWI is really the right thing to do? Is it going in the right direction wanting to retain the infrastructure of the FWI at all costs?  The water table is a physical measure  that could be measured and even adjusted through the use of satellite observations while the FWI is an empirical transformations of the fire intensity that its calibrated on a specific ecosystems. To CDS-match the FWI seems a weird think to do if you have a metric (the FWI) that is not very correlated to the fire activities in peatland. Even more so as you indead then evaluate against fire activity and not fire danger.

The question for me would be why do not directly rescaling the water table by training it to detect actual fire activities ? Along those lines I developed the FOPI which was trained on observed fire activities and did not attempt to rescale the FWI while still using the FWI as a driver for the fire weather component.

Importantly, when you train for fire activity, your output is somehow a probability which is more intuitive to understand. Another benefit of using directly the water table would be that when these variables are improved by the model or the assimilation system, this improvement shoud  benefit  the fire danger indicators in cascade. With the empirical structure of the FWI if you improve the moisture content estimation of the duff layer do you really improve the DMC ?

Indeed the motivation to be using the FWI infrastructure is provided in the paper. The FWI  is easily interpretable by fire agency. This is a good motivation. Still I am asking the author to elaborate a bit  as  I think they should discuss what would be the benefit of also shifting  toward a more physical based indicators of fire danger.

Few more comments:

1. I would put figure 4 and all the detailed explanation of the ROC curve derivations in an appendix. I think is quite a standard metric and while is nice to have a refresh in the paper it is distractive to have it in the methids section.
2. I would not mention satellite L-band Microwave Observation in the title. The focus here is not much how the water table is calculated but the use of the Hydrological model. Probably the same results would hold  with another model.
3. How easily available would this prediction of water table be for the forest agency to be able to calculate the new FWI-PEAT Index? Maybe a short discussion of the complexity of creating this index could be provided.  
4. You use fire ignition as a fire activity indicator but FWI expresses a measure of fire intensity. I am not sure if this has an impact on your results.
5. I haven’t quite worked out why a CDS matching would remove a bias ? Could you give me more datails ?
6. Finally I think the discussion is a bit fragmented. For exemple, the limitations of the GFA for peatland are nicely discussed but are not put in the contest of how (or if) they could affect the results presented. Similarly for the discussion about the new index. A clear statement of in how many more cases you are likely to get a good prediction compared to the use of the standard FWI (which you can read from the ROC curves) would certainly benefit the readibility of the conclusions.