Anticipating Future Risks of Climate-Driven Wildfires in Boreal Forests

Round 1

Reviewer 1 Report

Comments for the Author:

The topic of the paper is important and useful in the current context. By examining the current climatic conditions and projecting future changes in wildfire risks and burned areas in the boreal forest, as well as analyzing the suppression efficiency of different regions, the paper establishes a theoretical foundation for future wildfire management. However, there are several major concerns that I would like the authors to consider and address.

1. Regarding population and lightning data, it is worth considering that the future population of the Boreal Forest regions may differ from that in 2020. Would utilizing population data from 2020 alone better serve the forecasting objectives? Similarly, Chen et al. (2021) suggested that temperature and lightning in the boreal region may increase in the future (e.g., Chen et al., 2021, https://doi.org/10.1038/s41558-021-01011-y). Could the use of lightning data from 2015 contribute to a more accurate prediction?

2. The style of Table 2 does not appear to be consistent with the common format in Fire journal. Additionally, in Table 2, the percentage of burned area includes units, whereas the average burned area does not. It is recommended to either include units for all values or place units solely at the top of the table. Similarly, if the contents of Table 2 represent the average of the two time periods 2001-2010 and 2011-2020, using “Mean ± SD” may be more appropriate for indicating.

3. In the article, it is not explicitly mentioned what size the grid pixels are in the figures. It is crucial to clarify this in the methodology section, especially in Figure 6, where the average forest burned area is depicted as up to 10000 ha per pixel. If the resolution is 0.5 degrees, the area does not seem to match correctly. Please address this discrepancy and provide the accurate information regarding the size of the grid pixels in the figures.

4. When describing the burned area of different RCP models in Figure 10, the author states, “......but differences in annual burned area become strongly apparent by the end of the century......”, it is noted that there is no correlation analysis, such as time series, presented in this paper to substantiate the trends of burned area. The description relies on the observation of the figure (Lines 381-389). Please consider addressing this by either incorporating correlation analysis.

5. In lines 461-463, it is mentioned that the change in precipitation is uncertain. However, later in the paper, it is stated that precipitation in the boreal region has largely increased. Additionally, Figure 9 also illustrates trends in precipitation in future boreal regions compared with historical periods. It is recommended to revise the description to address these apparent contradictions.

Minor editing of English language required

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The article is valuable, it addresses an important issue: "Risks of Climate-Driven Wildfires in the Boreal Forest".

The paper has a correct stucture, is easy to read, and is richly illustrated.

My main questions concern the information presented in Figure 12 and Figure 7.

Figure 12 – I understand that the smooth lines on this graph represent some kind of trend… Is this a non-linear regression? Is there information somewhere about how these curves were fitted? Please expand the legend to include these lines.

Figure 7 – Please explain in more detail where the "best fit" curve comes from. Since this curve is the best, because the correlation between best fit and FireCC151 is high (and statistically significant), it might be good to provide correlations in the remaining cases (visual assessment is good, but confirmation with a correlation coefficient, preferably non-linear, is better in my opinion).

Please comment where the values of 50% and 80% were taken from in the sentence: "Two scenarios of improved response times were produced, which showed that stopping a fire within 4 days or within 24 hours could reduce the average burned area by 50% and 80% , respectively, of the burned area sums in the future period without adaptation.” Are these average values?

Technical note - please draw the axes on all charts.

Please remove the title above Figure 7 - this text can be included in the title below the chart.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

General Comments:

Dear authors, you will read my questions, I have annotated them as I read through your manuscript. My general comments focus on my Question 13 and Question 15, the way you portray your model is confusing.

Using a best-fit curve together with figure 7 is confusing, because the figure itself is difficult to understand. Moreover, the best-fit curve proves that your 4 models based on different RCP scenarios match very well the data for the last decade. However, the discussion should revolve around the fact that, given you have reproduced the data for the last decade using one of the 4 RCP forecast at a given annual time (2011-2020), and that is why you have shown the best-fit curve. You can forecast the lower and upper bands in the upcoming years.

It is of more importance to depict the lower and upper bound of these 4 RCP, since within this threshold, the expected burnt area shall be predicted. The best-fit curve is a curve that has only meaning during the last decade, which is the one that you have data to compare. But the lower and upper bands, given that the last decade fits within (that it does), allows the scientific community to have a model to predict the future.

Questions:

Question 1. [line 37] Consider elaborating on why wildfires in boreal forests have received less attention compared to fires in the tropics for clarity and completeness.

Question 2. [line 38] Could you quantify or reference this increase, by percentage or absolutely? What is the time interval of this increase?

Question 3. [line 46] Specify the key findings of the review by [13] and maybe cite it as “… was done by Triviño et al. [13]”.

Question 4.  [line 115-117] Do you have any reference to support that variance of annual temperature and annual precipitation are the most important, and therefore, correlated variables to the probability of wildfire and to the extension of burned areas? Otherwise, could you elaborate more on this statement, on the comparison to other variables that may influence the wildfire probability and the burned area extension beyond the previous two?

Question 5. [line 132-133] Does it mean that you used different future conditions other than the ones from ISIMIP2b? Or that the ISIMIP2b future conditions are extracted from CMIP5?

Question 6. [line 138] It is just a minor and personal comment, the way it has been phrased it seems that figure 2 corresponds to historical and figure 3 to future, however both are historical. It is also true that “respectively” is not included, and therefore it does not imply that figure 2 is historical and figure 3 future.

Question 7. [line 156 and table 1] According to figure 4, the output should be the expected burned area. In 2.4 you describe the FLAM input data, and these input variables are given in table 1. Should it be that “Burned area” is an output variable that you have used to compare the results? I suggest renaming the section to “2.4 FLAM source data” and the table to “Input and output data required by the flam model”.  The table then is self-explanatory and the reader can understand what source of data have been used to calibrate your model.

Question 8. [line 184-185] Your model assumes a static evolution of biomass. First, is it static or steady? By static, I mean no change with respect to time and steady means that there might be change with time but with a low rate (no transient phenomenon involved). Of course, a forecast of biomass is not a trivial analysis to be done. However, you need to either elaborate why you have assumed it static, whether it is the lack of information or it is the most sensical/moderate assumption. Otherwise, if not elaborated, you should add a comment in the conclusions regarding the limitations of your results based on this static assumption.

Question 9. [line 298-211] Similar to Question 8, these constant assumptions should be either discussed as why they are the best assumption or include this as a limitation in the conclusion.

Question 10. [line229-239] The suppression efficiency is introduced as variable “q”, also there is a variable suppression efficiency threshold. You should be clearer with your nomenclature. I suggest that after equation 1 you explicitly define what is exactly qij and qa and after equation 2 what is q.

Question 11. [line229-239]  More specifically, the matrix Q of NxM rows and columns, what exactly are the rows and columns for this matrix? Is a matrix of a geospatial grid?

Question 12. [Figure 6] It is just a minor and personal comment, maybe separating this figure in two figures. First one with two subfigures showcasing a) 2001-2010 and b) 2011-2020. The second figure the comparison, and state clearly how the comparison is done (2011-2020 minus 2001-2010 or otherwise) or issue a comment similar to [line 273-274] specifically for this new figure. With this modification, the reader would see better the information depicted and be able to understand better the comparison.

Question 13. [line 287-292] It is a bit difficult to understand the figure with the explanation. To my understanding, you are trying to explain that the different RCP do not fit the ground-truth (FireCCI51) for a fine grid of time (daily). Therefore, you select a scenario which is the best fitting curve. What should be explained is how the best fit scenario is derived. First, is this for a given RCP? Basically, how the red line correlates with the dashed and dotted lines? And given that relationship, how you derive the best fit exactly? For this last question, I believe that in figure 7, FLAM RCP 2.6 at 2012 coincides with the blue and red line. So the way the best fit works is that it is the closest value to the ground-truth among the four RCP models you have, right?

However, the way it is explained and showcased is confusing, since the purple dashed line completely disappears and leads to strange assumptions for the reader. I suggest, maybe, to show the predicted FLAM curves on a subfigure on the left. And on the right, a subfigure with the ground-truth, the best fit, and also adding a lower and upper band limit, maybe with a transparent area in between the limits, which would represent the range of the FLAM's RCP predictions. That would be much easier to understand, also to improve the explanation by focusing on the left subfigure and to explain the derivation of the best fit curve by referencing the right subfigure.

Question 14. [line 412-446] I understand the discussion here, it is what should be expected from figure 4, if suppression efficiency is an input to your expected burnt area, of course, figure 11 would yield less expected burnt areas if the suppression is more efficient. Also from a realistic point of view. Nevertheless, you have shown for RCP 8.5, the expected burnt area for different suppression efficiency models and figure 11 focuses on those for the decade 2011-2020. But this is only the model results, how those compare with the reality during the last decade? Which RCP reproduces the expected burnt area the best? RCP 8.5? I believe the maps are very interesting by themselves, specially to understand the dynamics of wildfire and to find potential hotspots rather than pointing out what is expected. Basically, the higher the suppression efficiency taken into account in the simulations, the lowest the expected burnt area. Also, the results may be of interest to find hotspots that, regardless of the suppression efficiency, they have a significant and immutable expected burnt area.

Question 15. [line 474-475] Similar to question 13. The critic I make here is the following: the best fit curve is alright, it shows that within the 4 RCP scenarios that you have used, depending on the RCP chosen, you can reproduce the ground-truth values. However, reality does not account for a sudden change on the RCP model (an annual change), it may vary if climate policies are enforced the right way. However, it follows a tendency, it seems clear that the real world has a determined RCP, the one that best forecasts the future. It seems, as 2023 has ended, we are closer to the RCP 6.5. I believe that the best fitting curve is a proof that the 4 RCP you considered are somewhat representative of reality. However, rather than claiming:

“was possible to select the best RCP scenario for each year, enabling FLAM to capture the inter-annual variability provided by the FireCCI51 dataset. This best fit demonstrated a high correlation with the observations (Figure 7).”

You should point out that given that the best fit can reproduce the inter-annual variability, therefore the forecast of your model shall fit inside the lower and upper bounds for the given 4 RCP. Making the whole argument based on the best fit undermines the potential of your model. Since the best fit can only be fitted for the known data from 2011 to 2020, and makes no sense to just pick one model over other by convenience on a specific year. However, the lower and upper bands of your model can forecast the past, present, and future under some degree of confidence.

Author Response

Please see the attachment

Author Response File: Author Response.docx