| 31 Jul 2026
Wildfire Hazard in Germany Under Worst-Case Conditions
Abstract. Climate change is amplifying wildfire frequency and intensity in temperate Central Europe, making proactive hazard mapping essential for landscape mitigation. This study evaluates wildfire hazard in Germany under worst-case (i.e., near-maximum) environmental conditions to identify high-priority management zones. The minimum travel time algorithm was used to simulate over 50 million independent fires at a spatial resolution of 100 meters. The landscape was divided into seven distinct climatic fire occurrence areas and evaluated across all combinations of three wind speed percentiles (80th, 90th, and 97th) and three fuel moisture scenarios. We quantified hazard by integrating conditional burn probability (CBP) and conditional flame length (CFL). Fuel moisture emerged as the dominant driver of wildfire hazard, with drier conditions dramatically increasing simulated fire sizes and variances, whereas increases in wind speed had a negligible effect. Shrub fuels produced the most extreme fire behavior and frequently generated the longest flame lengths under worst-case conditions. Grass and forest fuels burned less intensely on average, but maximum aridity caused high-hazard forest areas to expand tenfold. Across the landscape, CBP remained highly skewed, revealing that a small number of volatile hotspots burn repeatedly while the vast majority of the country burns infrequently. The hazard map of Germany is dominated primarily by low hazard ratings, which cover large, continuous forest stands with dense overstorey canopies. Elevated hazards are concentrated in the northeast and west. These are represented either as an intensity-driven regime, defined by severe flame lengths, or a frequency-driven regime, characterized by a mix of fast-spreading grass and shrub fuels. The highest hazard ratings occur in open landscapes such as military areas and subalpine parklands, where rapid fire spread greatly increased burn probability. High-hazard zones consistently correlate with low canopy cover and reduced crown bulk density. This structural vulnerability is particularly evident in disturbed regions, where canopy dieback has simultaneously increased surface fuel loads and reduced the sheltering effect that typically dampens wind speeds and evaporation. Long-term wildfire mitigation in Germany should focus on high-hazard hotspots where volatile fuel configurations overlap with containment constraints resulting from terrain or unexploded ordnance. As fuels are the only landscape component that humans can actively manipulate, management frameworks must prioritize reducing understorey shrub loads, maintaining closed overstorey canopies, and diversifying forest structure, in order to build resilience against accelerating climate pressures.
The main strength of this manuscript is its country-wide scale and the attempt to provide a consistent wildfire hazard assessment for Germany. However, the manuscript does not yet establish clearly how this national-scale analysis can support country-wide wildfire preparedness and management. In addition, several methodological choices require stronger justification and clearer links to comparable wildfire hazard studies. The following major comments concern issues that I believe should be addressed before the results can be interpreted with sufficient confidence.
Major Comments
The representation of the fuel landscape is central to the results of this study, and several choices require stronger justification and, ideally, sensitivity analysis. First, the statement that a spatial resolution of 100 m (1 ha) is sufficient to represent the typically small fires and fragmented forest landscape of Germany needs supporting evidence. In a fragmented landscape, this resolution may affect fuel connectivity, forest-agriculture boundaries and small fuel breaks, while several other input variables are originally available at much finer resolution and are subsequently resampled to 100 m.
Furthermore, agricultural land is treated as non-burnable even though the authors acknowledge that it may burn under extreme fire-weather conditions. The fact that the study focuses on forest-fire hazard does not require agricultural areas to be excluded as pathways for fire spread; even if hazard is ultimately evaluated only within forests, agricultural fuels may affect whether fire can propagate between forest patches. Treating these areas as non-burnable may therefore create artificial barriers and influence CBP, particularly in fragmented landscapes. The 100 m GR3 buffer introduced around forest edges partly addresses this, but the empirical basis for the buffer width and fuel-model assignment should be explained.
Finally, I am particularly concerned about the conversion of TU1 and TL3 to TU5 “to facilitate the simulation of extreme fire behaviour including crown fire.” Extreme weather or fuel-moisture conditions do not themselves imply a change in fuel type. TU5 represents substantially different fuel loads and structure, and the manuscript subsequently identifies TU5 as an important contributor to elevated CFL and wildfire hazard. The empirical basis for replacing the originally mapped TU1 and TL3 classes with TU5 should therefore be provided.
I also have concerns regarding the use of the CBPi metric. It is not sufficiently explained why CBP needs to be normalised by the maximum value of each individual scenario in order to enable comparison across scenarios. This is particularly unclear because the same set of ignition locations is used for all scenarios specifically to ensure comparability, meaning that the raw CBP values are already derived using a common ignition scheme.
Moreover, because one of the objectives of the study is to assess the influence of model parameters such as wind speed and fuel moisture, independently rescaling each scenario may obscure part of the differences in burn probability caused by those parameters. The resulting CBPi is a scenario-relative rather than a common-scale measure. Consequently, the five CBPi classes are also relative, and a “high” CBPi value does not necessarily represent the same underlying CBP in different scenarios. This makes comparisons of hazard between scenarios difficult to interpret. The authors should therefore explain why scenario-specific normalisation is necessary and demonstrate that it does not affect the conclusions drawn from comparisons among scenarios.
Finally, the integrated hazard matrix is adopted from a US decision-support system, and its transferability and usefulness for the German application also require justification.
The modelling results are not sufficiently validated against observed fires in Germany. The manuscript refers to the major 2022 fires, but does not test whether the model reproduces observed fire locations, burned areas, or fire behaviour. Some comparison with historical fires, fire perimeters, or regional fire-occurrence data would substantially strengthen confidence in the resulting hazard maps.
The construction of the weather scenarios requires further justification. Wind direction is represented by only four cardinal directions, which seems rather coarse for a fire-spread model and may affect simulated spread patterns, particularly in complex terrain. The authors themselves acknowledge that using eight or more directions would improve robustness.
It is also unclear whether the wind percentiles and directional frequencies were calculated for the relevant fire season/fire-weather conditions or from all hours of the year. Finally, wind speed and fuel-moisture extremes are combined independently; the authors should clarify whether these combinations represent co-occurring conditions that can actually occur, or purely hypothetical worst-case scenarios.
The choice of a fixed 20-hour fire duration requires justification. Fire duration directly influences simulated fire size and burn probability, yet it is not explained why 20 hours was selected or how sensitive the results are to this assumption.
The operational relevance and management recommendations are insufficiently supported. The manuscript refers to operational, country-wide application and proposes specific management actions, but does not compare the resulting hazard map with existing German fire-management products, assess user needs, or evaluate management scenarios.
Recommendations such as maintaining closed canopies or increasing structural/species diversity should therefore be better supported by literature or softened to reflect that these interventions were not directly tested in the study.
Specific Comments
L3: “Maximum environmental conditions” is not a commonly used term. I suggest retaining simply “worst-case conditions”.
L11: This sentence is unclear. It could be rewritten to state explicitly that the forested area experiencing high CFL expanded tenfold.
L33-35: Additional references would be useful here. Did all of these factors contribute to difficult-to-contain wildfires? For example, for which fires did unexploded ordnance hinder firefighting efforts, and where were bark-beetle infestations relevant? Is there a distinct spatial distribution of these factors among the seven fire occurrence areas, or do they occur throughout Germany?
L39-41: The distinction between hazard and risk could be explained more clearly in relation to the scope of the paper. According to Scott et al. (2013), risk assessment additionally incorporates effects on highly valued resources and assets (HVRAs). Since the present study focuses on hazard rather than risk, this distinction and its implications could be stated more explicitly.
L42-43: The calculation of hazard does not explicitly require CBP and CFL, but rather likelihood and intensity metrics. Scott et al. (2013), for example, consider annual burn probability as a likelihood metric and mean fireline intensity as an intensity metric.
L44-45: The choice to integrate likelihood and intensity into a single hazard category requires further justification. It may also be useful to retain and interpret the two components independently.
L46-47: Which assessments are referred to here? This statement would benefit from references to comparable studies using alternative or extreme weather scenarios.
L48-49: It would be beneficial to mention what information or tools wildfire-prevention agencies actually employ in their management strategies. In particular, why is only fire intensity discussed here when likelihood is also treated as a component of hazard?
Data and Methods
L68: Even if the Scott & Burgan classification is commonly used in European studies, some examples should be cited to support this statement.
L75-76: The statement that a 100 m resolution is sufficient for the typically small fires and fragmented forests of Germany requires supporting evidence. See Major Comment 1.
L78: See Major Comment 1 regarding the treatment of agricultural land as non-burnable. Please also clarify whether all agricultural land or only irrigated agricultural land is ultimately treated as non-burnable.
L89-92: This section requires clarification. How were the reported transition thresholds derived, and what exactly is meant by the “ratio of fires” transitioning from surface to crown fire?
L101-102: See Major Comment 1. In particular, the empirical basis for converting TU1 and TL3 to TU5 should be provided.
L144-146: Four wind directions appear rather coarse for a spatial fire-spread simulation, particularly in complex terrain; many comparable analyses use more directional classes. Further justification is needed. Please also clarify how the wind-direction frequencies were calculated for each fire occurrence area. See Major Comment 4.
L169-172: The purpose of using random ignitions should be explained more clearly. Random ignitions may be appropriate for estimating conditional landscape fire behaviour, but they do not represent the spatial probability of real ignitions. This distinction is important for interpreting CBP.
L184-199: See Major Comment 2 regarding scenario-specific CBP normalisation, the resulting CBPi classes, and the integrated hazard classification.
Results / Discussion / Conclusions
L214: The comparison between the area classified as high hazard and the observed annual burned area should be interpreted carefully, as these represent different quantities.
L311: Historical ignition and fire-occurrence datasets could potentially provide an independent means of validating the spatial patterns produced by the model. It would be useful to discuss whether such data were considered.
L312: For the map to be described as operational, the needs of German fire-management and mitigation services should be considered, either through comparison with existing operational products or through consultation with relevant agencies. More generally, the manuscript should explain more clearly how the proposed hazard map would be used in practice.
Technical Corrections
L5: “spatial resolution of 100 × 100 m”
L12-13: “of the forested areas of the country”
L14: “elevated hazard values”
L33: “firefighting response”
L214: “by a factor of 10”
References
Scott, J. H., Thompson, M. P., and Calkin, D. E.: A wildfire risk assessment framework for land and resource management, Tech. Rep. RMRS-GTR-315, U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station, Ft. Collins, CO, https://doi.org/10.2737/RMRS-GTR-315, 2013.