Predicting Wildfire Fuels and Hazard in a Central European Temperate Forest Using Active and Passive Remote Sensing

Round 1

Reviewer 1 Report

Considering expected climatic change, the assumption of the authors that wildfires in Central European forests will probably increase can be considered as a fact that is worth to study in order to focus on preventive management actions.

The approach is not original by itself but “by the book” based on field data, creation of custom fuel models, use of LiDAR and remote sensing data.

The results and discussion aim to bring a better understanding of wildfire risk of temperate forests and how it depends of qualitative and quantitative fuels combined with terrain and wind/weather.  

The interest of the proposed paper is not his originality but the fact is dealing with a subject that could be in a near future quite important in European Temperate forests.

The manuscript is well written and in a way exhaustive. 

The subject is logically presented and developed. The flow of the paper is logical and clear, it includes sufficient details on the background importance.

The objectives are clear and meaningful.

The approach is experimental and the proposed method is validated on study sites and can be generalized.

Conclusions are logical, mindful and sufficient at that stage of the demonstration.

Tables and figures are sufficient and necessary. The review of the literature is adequate.

The length of the presentation can be considered as appropriate in order to provide enough elements in the description of the methods.

However, a more concise presentation could be more interesting for potential readers and could increase its attractiveness.

The paper can be considered as original and well written and I propose its acceptation in the present form despite I would prefer if the authors could make an effort to take my previous remark in mind.

Author Response

Dear Reviewer 1,

we thank you for your positive feedback on our manuscript. We further agree with your point, that a more concise paper could be more interesting to a broader audience. With that in mind we evaluated the manuscript but were not able omit text passages or entire paragraphs without ignoring significant details. A central goal of this study is to provide a complete and reproducible workflow for forest management agancies, which are just starting to deal with the issue of wildfire. Therefore, we believe it is benefitial to include backgrounds on e.g. fuels sampling or fire hazard modeling.

Best regards in the name of all co-authors

Reviewer 2 Report

The authors designed a workflow to assess wildfire hazard for a small region in Germany. They also show how to define and estimate custom fuel properties by applying machine learning to remote sensing data. Finally, they used a fire spread model to estimate the conditional burn probability given random ignitions over the plot.

Major comments:

• Introduction: The authors should consider adding more references to similar studies and how they differ from this paper.
• Section 3.1: This section needs more description of what measurements were carried out and how were the locations decided?
• Line 255-256: The authors should consider adding what predictors were excluded.
• CBD prediction: Apart from the low number of training samples, the low CV training score compared to the higher score in the validation set indicates that the authors have a bias problem. Would you consider using a more flexible model to predict CBD? 

Minor comments:

• Line 13: The authors should consider replacing OA for overall accuracy.
• Table 1: Cumulative vertical profile variable name has an extra parenthesis. Also, just for checking, why the total number of returns is in meters?
• Table 2: Table 2 is labeled Table 1.
• Line 291: Missing a comma.
• Table 3: Table 3 is labeled Table 2.
• Figure 6-B: The authors should consider doing a plot similar to Figure 8-B, where the NB pixels are plotted as gray to better distinguish burnable pixels inside and outside.
• Line 562: then -> than

Author Response

Dear Reviewer 2,

we thank you for providing valuable comments and suggestions which clearly helped to improve our manuscript. Please see our in-line answers below.

Major comments:

• Introduction: The authors should consider adding more references to similar studies and how they differ from this paper.
  • We briefly presented additional similar studies conducted in common fire-prone regions. These can typically rely on valuable existing information relevant to fire spread modeling (e.g. surface/canopy fuel maps, fire history data). For our study area, like most European temperate forests, such information is lacking. Therefore, we aimed to document a complete workflow which can be adapted to areas, in which wildfires are also a newly emerging issue. We now cite papers covering individual parts of this process, but found none presenting the complete approach.
• Section 3.1: This section needs more description of what measurements were carried out and how were the locations decided?
  
  • In the revised manuscript, we elaborate on field measurements and on the random selection process for plot locations in more detail.
• Line 255-256: The authors should consider adding what predictors were excluded.
  • Corrected as suggested.
• CBD prediction: Apart from the low number of training samples, the low CV training score compared to the higher score in the validation set indicates that the authors have a bias problem. Would you consider using a more flexible model to predict CBD? 
  • The model we used indeed shows some bias. However, this is part of the Ridge Regression concept, where lower variance through constrained parameters is traded for the introduction of a bias. A nested cross validation may have overcome this problem. Alternatively, the suggested more flexible model (e.g. penalized GAM) could have been a solid option. We thank the reviewer for this suggestion and will consider a more flexible model in future studies, however, implementing and comparing different models for CBD, is beyond the scope of this study. We added a brief section on this issue to the discussion.

Minor comments:

• Line 13: The authors should consider replacing OA for overall accuracy.
  • Corrected as suggested.
• Table 1: Cumulative vertical profile variable name has an extra parenthesis. Also, just for checking, why the total number of returns is in meters?
  • Corrected as suggested.
  • Thank you for spotting this mistake. The total number of returns should have no units.
• Table 2: Table 2 is labeled Table 1.
  • Corrected as suggested.
• Line 291: Missing a comma.
  • Corrected as suggested.
• Table 3: Table 3 is labeled Table 2.
  • Corrected as suggested.
• Figure 6-B: The authors should consider doing a plot similar to Figure 8-B, where the NB pixels are plotted as gray to better distinguish burnable pixels inside and outside.
  • Corrected as suggested.
  • Besides the figure we adjusted the now displayed share of pixels that fall outside the AOA, described in the figure caption and the paragraph below.
• Line 562: then -> than
  • Corrected as suggested.

Best regards in the name of all co-authors