the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 23 May 2024
Modelling Current and Future Forest Fire Susceptibility in north-east Germany
Abstract. Preventing and fighting forest fires has been a challenge worldwide in recent decades. Forest fires alter forest structure and composition, threaten people’s livelihoods, and lead to economic losses, as well as soil erosion and desertification. Climate change and related drought events, paired with anthropogenic activities, have magnified the intensity and frequency of forest fires. It is crucial to identify the conditions that cause the emergence and spread of forest fires to improve prevention and management. We applied Random Forest (RF) machine learning (ML) algorithm to model current and future forest fire susceptibility (FFS) in the federal state of Brandenburg (Germany) using topographic, climatic, anthropogenic, soil, and vegetation predictors. FFS was modelled at a spatial resolution of 50 metres for current (2014–2022) and future scenarios (2081–2100) considering different shared socioeconomic pathways (SSP3.70 and SSP5.85). Model accuracy ranged between 69 % (RFtest) and 71 % (LOYO), showing a moderately high model reliability for predicting FFS. The model results underscore the importance of anthropogenic parameters and vegetation parameters in modelling FFS on a regional level. This study will allow forest managers and environmental planners to better identify areas, which are most susceptible to forest fires, enhancing warning systems and prevention measures.
- Preprint
(21244 KB) - Metadata XML
-
Supplement
(942 KB) - BibTeX
- EndNote
Status: open (until 04 Jul 2024)
-
RC1: 'Comment on egusphere-2024-1380', Anonymous Referee #1, 15 Jun 2024
reply
Thank you very much for submitting this very interesting manuscript. The authors modelled current and future forest fire susceptibility in Brandenburg using a random forest approach. They analyzed variable importances of topographic, climatic, anthropogenic, soil and vegetation predictors, highlighting the influence of human factors for fire ignitions in Brandenburg. Overall, one strength of this article is the comprehensive description of methods and its well written nature. I very much enjoyed reading the study. Well done! So far, I only have only one major comment regarding the temporal selection of climatic variables, the rest is minor.
Major comment:
Over that whole manuscript I am wondering why only a selection of months (here June) was used to build your random forest models. I agree that human factors have a strong influence on fire susceptibility in Brandenburg and I can also follow your discussion on explaining the rather weak influence of climate variables given your analysis and missing extreme events in the data. However, in general, I miss more details/justification why the selection of only few summer months was done here. In the variable description the authors refer to a publication by He et al. (2022), which however, modelled Australian bushfires, thus the climate-fire-susceptibility relations might be different from those compared to forest fires in Brandenburg. Therefore, I kindly ask the authors at least to better justify the rather strong assumption to select only certain months for their analysis. I also kindly ask the authors to test if your random forest analysis yields very different results if you include all the months of the year (from Table 1 I assume that monthly resolution is given also for the future data). After all, I think including more months in your analysis is highly valuable, because this could also improve our predictive outcomes and messages you could convey for your future projections (as you discussed in section 4.2). I suspect the main reason why your future predictions are weakly diverging from the present day, might not only be due to a limited representation of extreme events in our future data, but rather the fact that the only changing variables in your predictions are climatic - and those have a fairly weak importance our RF-models.Minor comments:
Line 5: Please shortly define fire susceptibility already in the abstract.
Line 16: Consider to check the reference for the increasing number of fires in Germany. I guess it should be rather the study by Gnilke et al. from 2021 not 2022.
Line 54: Consider to check the reference Gnilke & Sanders 2021. I think here it should be rather the Gnilke et al. 2022 publication.
Line 62: Could you please cite some of the few studies that you found, which have analyzed current and future FFS at a high spatial resolution?
Line 79: I could not find A2 in the supplement. Maybe it should be S2 here.
Line 115: Here the authors state that climatic variables where aggregated to 3 months, but in line 90 is written that only June was selected. Please indicate which months were used to train the models. (see also my major concern) If only June was selected to built the RFs, I recommend to check if the peak fire season might be shifted under future climate conditions - and if so, shortly discuss this point in the discussion.
Line 320: I agree to the points you raised to explain the weak importance of climatic variables. However, I miss a discussion what would happen if more months (and therefore more intra-annual variability) were considered in your approach (see major concern). How would that change your results?
Line 369: I highly acknowledge that you outline forest fire prevention strategies in Brandenburg. Please add references for the lines 369 – 371.
Citation: https://doi.org/10.5194/egusphere-2024-1380-RC1
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 99 | 19 | 10 | 128 | 15 | 7 | 5 |
- HTML: 99
- PDF: 19
- XML: 10
- Total: 128
- Supplement: 15
- BibTeX: 7
- EndNote: 5
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1