Unraveling the complexities in fire regimes by modeling of fire occurrence and burned area
Embargoed until 2025-08-07
Author
Date
2023Type
- Doctoral Thesis
ETH Bibliography
yes
Altmetrics
Abstract
Fire is an ancient phenomenon, with its presence on Earth dating back to the Silurian period. This long history of fire has caused ecosystems to co-evolve with fire, resulting in adaptations from individual species to the ecosystem level. However, human activities have begun to push fire activities around the world beyond natural regimes. Fires that exceed beyond natural ranges move from being a natural disturbance that is co-shaping the ecosystem to becoming perturba-tions that disrupt ecosystem functioning. Furthermore, these alterations are expected to be exac-erbated by anthropogenic climate change. It is therefore crucial to develop a solid understanding of fire regimes and how they will be affected by future changes.
The overall goal of this thesis is to deepen our understanding of the relationship between fire occurrence, burned area, and the environmental variables that influence them. Additionally, this thesis seeks to contribute to the improvement of modeling and prediction of these phenomena by investigating methodological aspects that can enhance modeling success.
In Part I, I investigated the importance of environmental variables and grain size (resolution) for modeling fire occurrence across multiple regions in the European Alps and the Mediterranean Basin. I investigated the importance of different groups of environmental variables in fire oc-currence models: climate, anthropogenic, forest, and topography at two grain sizes (100 m and 1 km). Finally, I assessed the spatial transferability by applying models between regions. I also created a cross-regional model that included all regions to investigate the effect of spatial scale on transferability. I found that grain size influences model performance. Fine resolution models consistently exhibited higher performance compared to coarse resolution models. The influence of grain size on the relative importance of environmental variables varied between regions. The highest transferability between models was found for the cross-regional model. However, it was also found that model transfer is possible between regions with similar environmental variables.
In Part II, I assessed the performance of the Cumulative Logarithmic Area Ranking Efficiency (CLARE), a novel measure recently developed for predicting fire size, across eleven regions in Europe that feature diverse environmental conditions. By incorporating CLARE in the model selection process in addition to AUC, we can better evaluate a model's capacity to predict both fire occurrence and burned area. I investigated the effects on CLARE of 1) different groups of input variables (meteorological variables vs. fire weather indices); 2) model complexity (multi-variable vs. single-variable models); and 3) modeling techniques (Generalized Linear Models vs. Maxent). Models exhibiting a high AUC performance in predicting fire occurrence may not necessarily feature high performance in predicting burned area. The use of multi-variable models is likely to result in higher CLARE performance compared to single-variable models. Such an approach also led to a better performance with multi-variable meteorological models com-pared to single-variable fire weather index models in specific regions. This approach may have particular advantages in areas where the calculation of fire weather indices is not feasible. The differences between modeling methods were primarily associated with the region or input variable groups that are in question. Maxent demonstrated slightly better performance in high fire activity regions for fire weather index models, whereas GLMs outperformed Maxent in low to intermediate fire activity regions for meteorological models. Overall, my findings highlight the potential for incorporating burned area into fire danger assessments.
In Part III, I tested how the performance of meteorological variables and fire weather indices changes with temporal resolution. In addition, I evaluated predictions for fire counts under a future climate change scenario to aid the interpretation of projections. The performance of fire occurrence models varied with respect to temporal resolution across study regions. The results also demonstrated that variations in temporal resolution affected model performance, primarily due to structural changes in the fire weather indices. This calls for caution when using fire weather indices at temporal resolution other than daily. Finally, fire occurrence predictions under a climate change scenario indicated that while models perform similarly under current conditions, models at lower resolution predict a higher number of fires. This suggests that although temporal resolution may not influence model performance under current conditions, it could distort future projections.
My thesis has provided insight into two key aspects of fire regimes: fire occurrence and burned area. Investigating the influence of environmental variables, spatial and temporal resolution, and modeling techniques on model performance has demonstrated the importance of considering these factors when assessing fire occurrence. In addition, my thesis has demonstrated the potential for incorporating burned area into fire danger assessments to provide a more comprehensive understanding of burned area. Through this work, I contributed to improving the modeling and prediction of fire occurrence and burned area, which is essential for informing fire management and policy decisions. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000625432Publication status
publishedExternal links
Search print copy at ETH Library
Publisher
ETH ZurichSubject
Forest fire; Fire regime; Fire occurrence; Fire weather; Fire dangerOrganisational unit
03535 - Bugmann, Harald / Bugmann, Harald
More
Show all metadata
ETH Bibliography
yes
Altmetrics