Savannas are wrongly perceived as degraded ecosystems whereas they represent a highly valuable landscape and cover 20% of the Earth’s surface. In the USA, Cedar Creek Ecosystem Science Reserve (Minnesota, USA) focuses on temperate savanna, unique in the continent. A prescribed burning program has been established in the early 60’s to include this key structural process into the savanna system. In this study, we reconstructed for the first time the fire regime imprints in this ecosystem and analyzed the link with the savanna vegetation since its establishment, about 4200 years before present day. Replacing first the boreal forest, then the oak/pine forest, following the glacial retreat and the climate warming of the Holocene (at respectively ca. 8000 years and 7200 years before present), this savanna system experienced a mFRI of 156.5 +/- 158.2 years.
At its establishment, the savanna presented the lowest recorded abundance of pine species while the abundancy of Poaceae dramatically increased in the understory thus defining the landscape characterized by low density of oak that is observed nowadays. In parallel, the fire regime of the mixedwood forest was characterized by less frequent fires associated with relatively high charcoal accumulation rates (mean Fire Return Interval of 280.9 years +/- 160.6 years) which transitioned towards higher frequency of fire events associated with low charcoal influx, while moving towards the savanna vegetation. In addition, the change of fire frequency is associated with a change in the fuel type burned, with dominance of ligneous fuel during the forested phase (W/L ratio of charcoal particles < 0.5) shifting to a more diversified fuel type around 3750 years BP (WL ratio > 0,5). Our results thus suggest that savanna dominated landscape is associated with frequent fires of a mixed fuel composition, reflecting the more diverse vegetation and the establishment of the herbaceous layer in sparse oak trees in the landscape.
Ultimately, frequent fires have maintained savannas for over 4 millennia thus highlighting that prescribed burning is a practice to be maintained in order to protect this ecosystem.
How to cite: Hennebelle, A., Finsinger, W., Leys, B., Lapellegerie, P., Lestienne, M., and McLauchlan, K.: Fire as a key factor in the transition and maintenance of the oak savanna ecosystem in the Cedar Creek Ecosystem Science Reserve, Minnesota, USA, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18739, https://doi.org/10.5194/egusphere-egu25-18739, 2025.
The Mediterranean region hosts exceptional biodiversity shaped by millennia of interactions between climate, and disturbances: both fire and herbivores. This study reconstructs 8000 yrs of habitats combustibility and herbivores (domestic and wild) dynamics in the Crau Plain using paleoecological records.
Richness, evenness and turnover of vegetation dynamics were calculated to tackle the interconnexion with herbivores dynamics. Our results demonstrate a strong positive correlation between herbivores (indicated by coprophilous fungal spores) and palynological richness, determining the role of grazing by both wild and domesticated herbivores in maintaining ecological heterogeneity. The decline in grazing during the past millennium has coincided with an increase in woody vegetation, posing heightened fire risks under current climate change scenarios.
On the other hand, the palynological records has been converted into habitats and their relative combustibility. Early periods (7200–3900 cal. BP) exhibited lower habitat diversity dominated by less combustible vegetation correlated with cooler and wetter climate. However, from 3900 cal. BP, increased pastoralism and fire activity fostered the expansion of grasslands and fire-prone ecosystems.
By linking long-term ecological dynamics with modern conservation challenges, this study underscores the importance of integrating grazing management and fire regulation into biodiversity conservation strategies to sustain Mediterranean landscape resilience.
How to cite: Lestienne, M., Saurat, P., Mouden, G., Hennebelle, A., Latapy, C., Bajol, L., and Leys, B.: Fire and Herbivory as Architects of Mediterranean Biodiversity, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11573, https://doi.org/10.5194/egusphere-egu25-11573, 2025.
Insects outbreak and wildland fires are among the most relevant natural disturbances affecting forested ecosystems worldwide. Following the storm Vaia of 2018, many Norway spruce (Picea abies (L.) Karst.) forests of the Eastern Italian Alps have been affected by a severe outbreak of bark beetle (Ips typographus), leading to economic, management and social concerns. In this context, the interaction between bark beetle outbreak and alterations in wildfire behaviour is poorly analysed, especially for Italian forests. This research aimed to detect the effects of bark beetle proliferation in the alteration of potential wildfire behaviour in a forested area (Veneto region, northern Italy). The semi-empirical FlamMap software was used, based on ALS data processing for deriving the spatial distribution of forest attributes and fuels within the study area. The Minimum Travel Time (MTT) algorithm of FlamMap was adopted for wildfire behaviour simulations. The contribution of bark beetle in altering the spatial behaviour of wildfires was explored using ALS point clouds acquired before and after the proliferation of bark beetle within the study area (pre-beetle and post-beetle scenario respectively). From the ALS data 5 meters-resolution Digital Terrain Models (DTMs), Canopy Height Models (CHMs), topographic data and forest metrics were extracted for both scenarios, to model alterations of wildfire behaviour over time. Differences in Rate of Spread (RoS) and Burn Probabilities (BP) were assessed and their correlation with bark beetle effects on standing trees was investigated at the catchment scale. An increase in RoS over 25m/min and in BP greater than 0.5 were estimated in forested areas affected by bark beetle outbreak, confirming the key role of Ips typographus in altering wildfire behaviour. The relation between bark beetle impacts and changes in wildfire attributes was finally estimated by computing regression analysis that led to R2 of 0.78 and 0.82 respectively for RoS and BP. This type of analysis could be the starting point to inspect similar issues by the combined use of ALS data and wildfire behaviour models, with the ultimate aim of proposing effective management solutions and strategies for forest stands affected by natural disturbances.
How to cite: Mauri, L., Taccaliti, F., and Lingua, E.: Investigating the role of bark beetle (Ips typographus) in altering forest fire behaviour: a case study for Italian forests, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8130, https://doi.org/10.5194/egusphere-egu25-8130, 2025.
Fire is an Earth-system disturbance that occurs in most terrestrial ecosystems and has widespread impacts on biogeochemical processes and human life. The geographic and temporal patterns of fire activity reflect a strong interplay of climatic, human, and vegetation factors. However, due to the complex interaction between human activities and climatic factors, the direction and magnitude of human direct impacts on fire remain poorly understood.
Taking advantage of the unique setting created by shelter-in-place orders during the coronavirus disease 2019 (COVID-19) pandemic, this study causally estimates global changes in fire occurrences due to reduced human activities in the first half of 2020 compared to average levels from 2016 to 2019. Utilizing global satellite observations of active fire detections, we constructed an aggregated global dataset at 0.5° resolution by week scales, complemented by corresponding meteorological measures, lockdown policies, and mobility indexes.
First, we assessed the average change in global fire incidence and further examined its variability across spatial, temporal, and intensity dimensions. Lockdown measures led to an average reduction of 11.8% in fire incidence worldwide. Notably, there was significant spatial heterogeneity in the direction and magnitude of human impacts on fire incidence, with changes in individual countries ranging from a 6.3-fold increase to a 5.6-fold decrease. Within groups of countries where fire incidence decreased or increased, lockdown measures exhibited contrasting temporal effects. Additionally, a greater reduction in human mobility intensified these effects in the respective directions.
Second, leveraging the attribute information of fire detection locations, we conducted separate group regression to understand the diverse pathways of human influence across landcover types, protected areas, human footprint levels, and proximity to wildland urban interfaces (WUI). Among the four landcover types, fire detections exhibited a more pronounced decline in forests and grasslands compared to shrublands/savannas and croplands. Fire within protected areas showed a larger decline on average but also experienced greater variability. A striking relationship is that areas with lower human footprint levels demonstrated a more substantial reduction in fire incidence, highlighting the critical role humans play in fire occurrences in undeveloped or low-developed lands. This relationship was further corroborated by the observed trends in relation to the distance from the nearest WUI.
Third, we examined the time-lagged effects of human activities on fire occurrences. Areas that experienced a significant reduction in fire incidents during the early lockdown stage tended to see a subsequent rebound in the later stage, thereby delaying the local fire season. Conversely, areas with increased fire detections in the early stage may experience later declines, leading to an earlier fire season. Our results indicate that reduced human activity can influence the accumulation or consumption of local fuels, consequently delaying or advancing the fire season relative to normal conditions.
Our study provides an empirical quantification of the direct effects of human activity on fire occurrences and highlights human contributions to the complex interactions between human, climate, and fire.
How to cite: Zhang, Z. and Wang, J.: The Role of Humans in Fire Dynamics: A Natural Experiment from the COVID-19 Lockdowns, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6168, https://doi.org/10.5194/egusphere-egu25-6168, 2025.
Sub-canopy windspeed is a critical input variable in wildfire simulation modelling because it has a strong effect on the predicted rate of fire spread (ROS). In vegetated landscapes, windspeed reduction occurs due to the structural properties of vegetation, with the canopy height and forest/vegetation density being key structural attributes driving this effect. Wind reduction factors (WRFs) are used to represent this phenomenon in fire behaviour modelling. Significant variability in WRFs exist both laterally and vertically, however, this variation has been poorly represented in operational models for two key reasons: i) a lack of an operational-scale spatial dataset to characterise the key forest attributes and parameterise a WRF model spatially and vertically; and ii) the lack of a method to integrate these spatial parameters into an operational WRF model. We address these challenges by developing a novel WRF model using spatial inputs from the Global Ecosystem Dynamics Investigation (GEDI) spaceborne LiDAR dataset. This model relies on the Plant Area Index (PAI) vertical distribution and canopy height derived from GEDI to estimate WRFs at different heights within the forest profile and across landscapes. We utilise a spatiotemporally unique dataset of twenty-six sub-canopy WRFs measured at 2m height and derived from approximately five years of within-forest windspeed data across a range of vegetation types with diverse structural attributes. Model validation was conducted using observed (measured) vertical WRF profiles across 12 structurally diverse sites. The observed within-canopy WRF across the height range (1 – 75m) varied from 2.3 to 16.1. The new WRF model achieved a Kling-Gupta Efficiency (KGE) score of 0.8 and a coefficient of determination of 0.73, indicating very good agreement between the modelled predictions and the observed WRF data. The Mean Absolute Error (MAE) of the model was 1.36, and there was a slight bias towards overprediction of 0.43. The model represents an advancement in operational WRF modelling by explicitly integrating large-scale spatial datasets that characterise vertical forest structure. It demonstrates the feasibility of using GEDI data to model WRFs operationally, providing spatially and vertically explicit predictions. As a globally available dataset, GEDI enables this approach to be applied in forests/vegetation worldwide to better represent variability in WRF and therefore improve fire ROS modelling. This proof-of-concept establishes a scalable method to bridge critical gaps in WRF modelling for wildfire prediction.
How to cite: Sheridan, G., Keeble, T., Noske, P., Lyell, C., and Harrison, M.: A model of wind speed reduction (WRF) in forests that can be parameterised globally using spaceborne LiDAR, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14050, https://doi.org/10.5194/egusphere-egu25-14050, 2025.
Wildfires pose a significant threat to ecosystems, human lives, and property, especially in regions characterized by variable topography. This work delves into the complexities of wildfire behavior on sloping terrain, where the combined effect of crosswind and slope acting in the same direction substantially influence fire behavior, rate of spread, and fire intensity. Fire regime depends on Byram’s convective number that was modified to account for slope effect according to Eq. 1 (Morvan and Accary 2024), where I is the fire intensity, g is Earth’s gravity, α is the slope angle, ρ and Cp are respectively air density and specific heat at the ambient temperature T0, R is the rate of fire spread and Ue is the effective wind speed that includes the component of buoyancy characteristic-velocity acting in the direction of fire propagation. For steep slopes, this correction results in a convective number that is significantly different from the formulation proposed by Nelson (2015).
(1)
To test the effectiveness of the proposed Byram’s number expression, Large Eddy Simulations of shrubland fires are carried out using a 3D fully-physical CFD fire simulator (FireStar3D) under various terrain slopes and prevailing crosswind speeds, covering both wind-driven fire regime (NC < 2) and plume-dominated one (NC > 10). Results show that the proposed modification of Byram’s convective number allows a better description of the obtained fire regime. In addition to the numerical simulations, a database consisting of 11 experimental fires carried out in shrublands was used to support the use of the new convective number formula. The heat transfer mechanisms governing fire propagation are described, highlighting in particular the role played by the convective cooling of unburnt vegetation in the case of a plume-dominated fire as the fire draws an adverse air flow in the opposite direction of fire propagation (see Fig. 1).
Furthermore, the development of fire-induced wind and its action on fire behavior is investigated and compared to field data gathered during an experimental shrubland fire on a sloping terrain. Simulations were carried out for three lengths of the ignition line: 30 m (as in the experiment), 90 m, and quasi-infinite fire line simulated using periodic boundary conditions. Results show that fire-induced wind is only significant in the case of a wide fire-front. As the length of the ignition line increases, the interaction between this induced wind and the fire front can change the fire regime from plume-dominated fire to a wind-driven one.
Fig. 1. Temperature field and streamlines obtained in the vertical median plane of a plume-dominated fire, 90 s after ignition, in the case of 10° slope, an initial crosswind speed of 0.5 m/s, and quasi-infinite fire front. Earth-gravity direction is indicated by an arrow.
How to cite: Accary, G. and Morvan, D.: Physics-based simulation of extreme wildfire behavior on sloping terrain, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1889, https://doi.org/10.5194/egusphere-egu25-1889, 2025.
Italy is one of the European countries most affected by wildfires and landslides.
To date, the research on wildfires is mainly addressed to evaluate best solutions for prevention, control, and mitigation. Nevertheless, not enough attention was given so far to study effects of wildfires in view of the analysis of the related geohazards.
In urban environments, the cascading effect of wildfires on landslides represents a clear example of multi-hazard. In this regard, wildfires can be regarded as a preparatory process for landslides triggering, as they significantly modify the local conditions of slopes for a not negligible time window.
In the FIRE (wildFire-related-landslide scenarIos for territorial planning and Risk managemEnt) project, funded by “Sapienza” University of Rome, a multidisciplinary research team experienced, since 2022, an innovative approach to derive quantitative scenarios of expected shallow landslides over burned areas, by evaluating the effectiveness of wildfires in preparing instabilities, also in view of defining best practices for fire extinguishing and land management plans with respect to the potential damage caused by fires at short- and long-term.
Two case studies have been selected in Italy: Mt. Epomeo at Ischia Island (Naples) and Camaldoli hill (City of Naples). These two sites suffered in the last decades a large number of wildfires, and, in the case of Camaldoli hill, consequential shallow landslides.
Since the project activity began, two severe wildfires struck Ischia and Camaldoli, on August 2023 and June 2024, respectively.
The physical, hydraulic, and mechanical properties of soil covers potentially unstable have been defined through field surveying, in situ determinations and laboratory geotechnical tests, performed on unburned and burned samples as well as in different seasons, from May 2023 to June 2024.
For the Ischia case study, simulations of multiple wildfire propagation scenarios were carried out, originating from the most probable ignition points. These scenarios incorporated spatially explicit fuel load distributions and seasonally varying meteorological conditions. The simulations were executed using a computational model rigorously calibrated with empirical data from the 2023 Ischia wildfire, ensuring scenario-specific precision. To further support the modelling of risk scenarios, a close survey of vegetation was conducted and then critically compared to data derived from official thematic cartography and the most recent systematic botanical studies available. This enabled an updated and detailed understanding of the identification and distribution of dominant plant species composing the habitats within the Ischian landscape system, whose dedicated documentary study on the morphology and development of root systems allowed for the construction of root profiles to be associated with the recognised plant communities.
Finally, after a specific parametrisation of the physical and mechanical properties of the burned and unburned soil covers characterised by different plant associations, for each scenario of wildfire propagation, the PARSIFAL approach was applied to obtain scenarios of shallow landslides by considering both rainfall and seismic triggers in a probabilistically-defined framework.
The abovementioned activities are here reported, together with some preliminary results of the FIRE project while further steps will allow a statistically-based analysis through data-to-model informed Artificial Neural Networks.
How to cite: Ferrarotti, M., Marmoni, G. M., Fiorucci, M., Esposito, C., Galuppi, M., Berardi, D., Salvi, F., Lombardi, M., Lei, A., and Martino, S.: The FIRE project: a multidisciplinary approach to provide innovative probabilistic scenarios of shallow landslides over burned areas, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13501, https://doi.org/10.5194/egusphere-egu25-13501, 2025.
Changes in wildfire regimes observed globally due to land use transformation, human activity and climate change are compelling the development of Forest Fire Danger Rating systems capable of accurately identifying spatio-temporal patterns of increased fire danger for effective wildfire risk management, with a focus on distinguishing extreme dangerous conditions for a proper resources deployment.
Many existing models primarily rely on weather conditions, often overlooking critical factors such as fuel and topography, which significantly influence wildfire behavior. However, these characteristics play a crucial role in wildfire activity by identifying areas where their interaction with fire-prone weather can result in extreme behaviors, leading to the majority of fire-related damage and civil protection emergencies.
This study analyzes RISICO, a fire danger rating system that explicitly incorporates fuel and other geo-environmental characteristics of the territory into its computations. Developed in the early 2000s for Italy, RISICO has been operationally used by the Italian Civil Protection system for decades, supporting the issuing of the National forest fires risk bulletin. The latest version of the model further enhances the integration of fuel in its calculations.
The model's performance has been evaluated over the past fifteen years of wildfire data in Italy, alongside other fire danger indices from the literature. The discrimination and detection capabilities of the indices have been assessed, along with their reliability, to ensure their suitability for operational use.
RISICO demonstrates strong performance in identifying wildfire-related conditions while reducing the extent of areas classified as high danger, thereby improving its applicability for efficient wildfire risk management. This study highlights the importance of incorporating fuel and other geo-environmental characteristics into fire danger models, moving beyond sole reliance on fire weather assessments, and enhancing their operational effectiveness in wildfire risk management practices.
Keywords: wildfire danger, wildfire risk management, fire weather
How to cite: Perello, N., Trucchia, A., Meschi, G., Ghasemiazma, F., D'Andrea, M., Degli Esposti, S., Fiorucci, P., Gollini, A., and Negro, D.: Fuel-aware Forest Fire Danger Rating System RISICO: a comparative study for Italy , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6955, https://doi.org/10.5194/egusphere-egu25-6955, 2025.
The vast latitudinal extent of continental Chile (approximately 17° to 55° South), combined with its contrasting anthropogenic land-use patterns and diverse altitudinal configurations, presents significant challenges for understanding fuel configurations. This study aims to define standard fuel models, based on the Scott and Burgan classification, to support stochastic wildfire spread simulations at the landscape scale.
In terms of landscape regions associated with forested or woody vegetation areas, three main regional units can be broadly identified. In the north (17°-30°), forest cover is sporadic or absent. In the center (30°-41°), the landscape is dominated by monoculture plantations of Pinus radiata and Eucalyptus globulus. In the south (41°-55°), extensive native forests prevail, with minimal or no human intervention.
The central Mediterranean zone presents the greatest challenge for defining fuel models, as this region has experienced the highest wildfire occurrence and damage levels in recent decades. Notably, two “firestorms” in 2017 and 2023 burned more than 900,000 hectares combined. In this area, forest monocultures undergo significant temporal changes due to both exploitation and wildfire impacts. Additionally, the lack of official data complicates the estimation of forest monoculture biomass and vertical structure, requiring the use of ancillary datasets to improve estimates of canopy structure and vegetation conditions.
In this study, fuel models were mapped across continental Chile producing a 100-meter resolution raster dataset standardized according to the Scott and Burgan methodology. The Vegetation Resources Cadaster produced by CONAF served as the foundational dataset, adapted and updated for fuel class assignments, with further refinements made by incorporating remote-sensed products such as canopy height data.
How to cite: Félez-Bernal, J., Barrao Simorte, S., Rodrigues Mimbrero, M., Galizia, L., and de la Riva Fernández, J.: Mapping fuel models across continental Chile, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12872, https://doi.org/10.5194/egusphere-egu25-12872, 2025.
Recent studies suggest that fire regimes will be altered in response to climate change, leading to increased frequency and intensity of wildfires in fire-prone areas, as well as expansion into previously unaffected regions. In the Mediterranean region, projected weather conditions, combined with existing vegetation patterns, are expected to contribute to more frequent and severe wildfires. Portugal has already experienced catastrophic consequences of these extreme circumstances in 2003, 2005, and 2017, with large wildfires causing extensive economic, environmental, and human losses. The characterization and mapping of fuels are recognized as critical factors in wildfire prevention and planning. Fuel management is a direct method for reducing fire risk, and fire behavior simulators (e.g., FARSITE, FlamMap) are valuable tools for supporting fire and fuel management decisions. However, the accuracy of simulation outputs depends heavily on the availability of precise fuel data. High-quality information on variables such as canopy height (CH), canopy cover (CC), canopy base height (CBH), canopy bulk density (CBD), and canopy fuel load (CFL) is essential for accurate wildfire management decisions. The overarching objective of this study had two main components: i) evaluating the utility of ICESat-2 data for estimating key fuel-related variables, and ii) creating a comprehensive map of these variables at a 25-meter resolution by integrating ICESat-2 data with other remotely sensed datasets such as Sentinel-1, Sentinel-2, ALOS2/PALSAR2, and SRTM. To achieve the first goal, a three-step approach was implemented: (i) modeling fuel-related variables using field-based vegetation measurements and ALS-derived metrics; (ii) generating ALS-based estimates of key fuel-related variables to provide ground-truth information across the study area; and (iii) assessing the utility of ICESat-2 ATL08 canopy height and cover metrics for estimating key fuel-related variables. An error analysis regarding the ICESat-2 derived estimates for the key fuel-related variables and the ICESat-2 standard CH estimates was performed to understand how different factors (e.g. land cover type, canopy cover, and slope) could affect the performance of the estimates. For the second objective, the Google Earth Engine cloud-computing platform was used to preprocess, mosaic, and retrieve Sentinel-1, Sentinel-2, PALSAR-2, and topographical data. Additionally, it was utilized to compute a suite of vegetation indices and textural metrics (GLCM). The Random Forest machine learning algorithm was then applied to predict each of the fuel-related variables using the aforementioned multisource satellite data. In this presentation, we will discuss the primary strengths and limitations of ICESat-2 data in providing useful and accurate information about key fuel-related metrics in a semi-arid Mediterranean landscape.
This presentation will be conducted under the scope of the FUEL-SAT project (PCIF/GRF/0116/2019) as the main findings here reported were collected and processed within this project.
How to cite: Godinho, S., Guerra-Hernandéz, J., Benali, A., Barreiro, S., and Guiomar, N.: Assessing the potential of ICESat-2 data to retrieve fuel-related variables, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18219, https://doi.org/10.5194/egusphere-egu25-18219, 2025.
In the past decade, regions across the globe have experienced devastating fire years with far-reaching impacts including direct harm to communities, hazardous air quality, and high carbon emissions. We examine the role of antecedent and concurrent climate variability in enabling extreme regional fire years – herein defined as years with the highest forest burned area during 2002-2023 – across global forested lands. Extreme regional fire years typically coincided with years with extreme seasonal fire weather indices (FWI) and had an average four-fold increase in the number of very large fires emitted more than five-times the fire carbon emissions than non-extreme years. A majority of extreme regional fire years co-occurred with FWI metrics exceeding a 20-yr return period, whereas weaker FWI links were seen in the tropics where land-use and deforestation likely confound relationships. We show that the likelihood of FWI metrics exceeding a 20-yr return period is 50-150% higher for much of the globe under a contemporary (2011-2040) climate compared to a preindustrial (1861-1890) climate. These results suggest that human-caused climate change has augmented the odds of recent and near-term extreme climate-driven fire years across forested regions of the globe. While variability in fire years stems from the interplay of biophysical and societal factors, the exacerbating effect of climate change underscores the urgent need for proactive measures in mitigating risks and adapting to these extreme fire years.
How to cite: Abatzoglou, J., Jones, M., Kolden, C., Cullen, A., Sadegh, M., and Williams, E.: Climate change has increased the odds of extreme regional forest fire years globally, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2594, https://doi.org/10.5194/egusphere-egu25-2594, 2025.
The Mediterranean region is a focal point for wildfires. Climate change is projected to affect the Mediterranean hydrological cycle, resulting in intensified drought conditions and increased fire hazard. Even though northern Italy is rich in water resources, wildfires have become increasingly prevalent in recent decades, occurring not only during the summer but also in the winter season.
This study explores for the first time the climatic drivers influencing the monthly burned area (BA) during winter fire season in northern Italy from 2008 to 2022. To this end, we build multi-regression data-driven models that highlighted the main burned area drivers for the overall area. The GPS-based BA perimeters analysed here are provided by the monitoring campaigns performed by the Carabinieri Command of Units for Forestry, Environmental, and Agri-food protection. For winter (November - April) fire season, the monthly percentage of burned area at 0.11 degrees of resolution for the 2008-2022 period was obtained. A total of 150 daily precipitation and maximum and minimum ground station series were collected, converted at monthly scale, reconstructed, homogenised and spatialised at 0.11° resolution by mean of Universal Kriging with auxiliary variables. Subsequently, several climatic indices were computed for precipitation (Precipitation, Consecutive Dry and Wet Days (CDD and CWD)), temperature (Maximum and Minimum Temperature and Evapotranspiration (ET0)) and drought (SPI, SPEI and Water Balance (WB)). To find the best BA predictors, first we checked the pair correlations of BA with different temporal aggregations of climatic indices. The Pearson’s correlation test between the detrended and standardised monthly time series of BA and of climatic indices was performed for each pixel and only the strongest and significant correlations were retained. Based on the CORINE Land Cover map, the vegetation classes that were most susceptible to wildfires, and their typical elevation ranges, were identified. Then, for each pixel, we performed multilinear regressions models using every possible combination of the best predictors that exhibit the lowest correlations with each other. The selection of the best regression models was based on an out-of sample procedure, and the model performance was tested by comparing the predicted BA with the observed, analysing the explained variance and correlation.
This study shows that in northern Italy, fires are predominantly found in the Alps, Apennines, and pre-Alpine regions. In these areas, the fire return period ranges from 1 to 1.5 years, in contrast to the Po Valley, where it exceeds 7.5 years. Deciduous Broadleaf Forests appear to be the most fire-susceptible vegetation class in these fire-prone regions. Modeling results for the 2008–2022 period indicate that fires in northern Italy are primarily influenced by water stress rather than high temperature rates. In fact, the best predictors of BA were mainly precipitation and water balance recorded between December and March of the current fire year. Moreover even if burned areas are not directly correlated with drought, the study figured out the presence of a brief time window during winter months between the end of a prolonged drought and the onset of precipitation when fire risk is high.
How to cite: Baronetti, A., Fiorucci, P., and Provenzale, A.: Assessment of climatic drivers for winter wildfire burned area prediction in northern Italy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-175, https://doi.org/10.5194/egusphere-egu25-175, 2025.
There has been an increased prevalence of very large fires in European countries with Mediterranean-type climate where in wet winters vegetation productivity is promoted and in hot dry summers fuel flammability is enabled. Traditionally, the annual burned area (BA) is applied to summarize all attributes related to the fire regime in these regions, reflecting fire management success or failure. However, in Portugal, the most fire-prone country in Europe, no annual BA trend is observed. Many fire-related experts advocate for a shift in this paradigm toward using fire impacts instead of BA, thus, understanding burn severity (BS) is essential to assess impacts and form and implement better pre- and post-fire management plans.
In this study, we analyzed spatiotemporal BS trends for large fires (≥500 ha) in Portugal from 1984 to 2022 with the identification of BS drivers. BS estimates were obtained from the “Portuguese Burn Severity Atlas” using Landsat imagery (30 m) and estimating indices as the difference normalized burn ratio (dNBR), relative dNBR (RdNBR), relative burn ratio (RBR), and dNBR with enhanced vegetation index (dNBR-EVI) to check the coherency of any possible trend via different BS indices. Time series trend was analyzed using different statistics of BS indices. We incorporated climatic and environmental variables to identify the drivers of BS and any possible BS trend. For climatic drivers, all the seven daily Fire Danger Indicators (Fuel Moisture Codes and Fire Behavior Indices) and three hourly weather observation variables from the ERA5 dataset at 15 p.m. as total precipitation, temperature, and vapor pressure deficit were utilized. For environmental drivers, we used: i) the mean elevation obtained by the Digital Elevation Model (30 m) for each fire and ii) vegetation types, for which fires were divided into four macro regimes via the dominant land use and land cover according to Carta de Ocupação do Solo (COS) leading to Pastoral, Urban, Forest, and Agricultural fires. The statistical approaches to conduct both time series analysis and correlation with drivers were based on the simple linear regression, Mann-Kendall test using tau variables, Theil-Sen slope estimator, and Spearman correlation test.
Our analysis revealed a coherent and significant increase in BS over time across all indices except for agricultural fires highlighting that there is a BS evolution in Portugal. While no correlations between climatic drivers and BS trend were found, strong correlations emerged between fuel moisture codes and fire behavior indices with areas with high BS classification, suggesting drier fuels are a key driver. Importantly, annual BA showed no significant relationship with BS, emphasizing the need to shift focus from extent-based metrics to severity-based management.
How to cite: Jahanianfard, D., González-Pelayo, O., and Benali, A.: Spatiotemporal Burn Severity Evolution Analysis with Identification of most Influential Climatic and Environmental Drivers for Large Fires in Portugal (1984-2022), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-899, https://doi.org/10.5194/egusphere-egu25-899, 2025.
Fire is a natural hazard that is dependent on climate, vegetation, and human activities, and climate conditions and fuel availability in Portugal make it a fire prone country. Moreover, extreme weather conditions have led to three severe fire seasons in this century, namely in 2003, 2005, and 2017, which caused large burned areas, and economical and human losses. The relationship between weather conditions and burned area has been extensively studied, but the effect on fire intensity is less understood, despite this being an important variable for fire propagation and suppression.
In this work, the bivariate relationship between Fire Radiative Power (FRP) and two weather variables was studied, namely temperature and wind speed, using copula functions. FRP was used as a measure of fire intensity and was retrieved from the Global Monthly Fire Location Product (MCD14ML), a part of the MODIS Active Fire product. Daily temperature at 15h and hourly wind components were obtained from ERA5-Land. The analysis was performed for all cases, and separately for two main wind directions: northerly/westerly (NW) winds, and southerly/easterly winds (SE). The maximum daily wind speed and temperature for each direction was used, on the day of the fire (lag 0), and on the previous 1 to 3 days (lags 1 to 3), to account for preexisting weather conditions. FRP values occurring on the same day were summed. Only pixels identified as forests were considered in this analysis, and the study area is the North and Centre of Portugal. The analysis was performed in the period 2001-2023, on the months of March to October.
Copula functions were fitted to the variables and used to compute the conditional probabilities of high FRP values, under extreme temperature or high wind speed. The results show that SE wind on the day of the fire is the largest driver of high fire intensity, although SE wind in the previous days also yields high probability of high FRP values in the hotter months. NW winds do not significantly increase the probability of high fire intensity, compared to the case of non-windy conditions. The effect of temperature is very similar for both wind directions, and the cumulative effect of temperature on the days before the fire is higher than if considering only lag 0 on almost all cases.
Acknowledgments: This work is supported by the Portuguese Fundação para a Ciência e Tecnologia, FCT, I.P./MCTES through national funds (PIDDAC): UID/50019/2025 and LA/P/0068/2020 https://doi.org/10.54499/LA/P/0068/2020). This work was performed under the scope of project DHEFEUS (10.54499/2022.09185.PTDC), supported by national funds through FCT, and LA/P/0068/2020 (https://doi.org/10.54499/LA/P/0068/2020), “Fundos próprios para desenvolvimento de projetos de I&D” Project MEDCEX - reference: 100SPID8106.
How to cite: Páscoa, P., de Zea Bermudez, P., Pereira, S., Russo, A., and M. Gouveia, C.: Extreme weather conditions and Fire Radiative Power in Portugal: a probabilistic analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13683, https://doi.org/10.5194/egusphere-egu25-13683, 2025.
Wildfires are a significant natural hazard to European forest ecosystems and society. In recent years, increases in wildfire activity have been attributed to climate change, with escalating impacts on communities and ecosystems. While fire risk has been typically studied at individual locations independently, spatially compound events–where multiple wildfires occur simultaneously across different countries–have been overlooked so far. Such spatially compounding events can cause large aggregated impacts and pose severe challenges, particularly in the context of shared resources for wildfire response, as under the European Protection Agreement. To advance our understanding of spatially-compounding wildfires, we analyze the spatial dynamics of such large scale events across European countries. We use the daily-scale Burned Area dataset from the Global Fire Emissions Database (GFEDv4) for the period 2001-2015 and the Canadian Fire Weather Index (FWI) derived from ERA5 data for 1940-2023. By combining burned area with FWI data, during the May-October fire season, we find that the top 20% of days with the highest European area under FWI > 50 account for 60% of the total European burned area, all fires considered. By focusing on FWI data, we reveal that cross-country dependencies in fire weather enhance the likelihood of days affected by a larger fraction of Europe under extreme fire danger. Similar cross-country dependencies are observed for burned areas. The spatial dependencies in FWI can be linked to large-scale atmospheric patterns that favor fire-prone weather over different regions simultaneously. Typical meteorological conditions profiles for the most extreme FWI events across the continent indicate that persistent high-pressure systems, characterized by increasing temperature and decreasing relative humidity prior to the events, are key drivers for widespread FWI extremes. We also investigate recent trends in spatially compounding fire weather events using reanalysis data and CMIP6 climate model simulations. These findings improve our understanding of spatially compounding wildfires, serving as a basis for evaluating continental-scale risk and guiding the response to high-impact events in the context of shared resources.
How to cite: Gauthier, E., Quilcaille, Y., Seneviratne, S. I., Zscheischler, J., and Bevacqua, E.: Cross-country dependencies in fire weather enhance the danger of extremely widespread fires in Europe, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10006, https://doi.org/10.5194/egusphere-egu25-10006, 2025.
Wildfires can spread rapidly, threatening ecosystems, human lives, and property. Real-time monitoring of fire dynamics is crucial for improving response times and firefighting efficiency. Meteorological models, particularly the WRF-Fire model, have been widely used to predict wildfire behavior by integrating factors such as weather conditions, topography, and fuel distribution. However, the radiative feedback from smoke, which is emitted during fires, can significantly influence fire spread, yet its impact remains underexplored. This study aims to explore the impact of smoke on wildfire spread by using the WRF-Fire-Chem model, a coupled version of WRF-Fire that integrates a chemical module. The research focuses on a typical U.S. forest fire case study and examines the effects of different smoke components (such as black carbon and organic carbon) and their radiative feedback on wildfire spread. Four simulation scenarios were designed: (1) wildfire spread without smoke; (2) wildfire spread with smoke emission; (3) wildfire spread with smoke but without black carbon; and (4) wildfire spread with smoke but without organic carbon. By comparing these scenarios, the study quantitatively investigates the role of smoke in influencing wildfire spread and examines how black carbon's heating effect and organic carbon's cooling effect contribute to the fire's dynamics. The results indicate that the heating effect of black carbon accelerates fire spread, while the cooling effect of organic carbon partially suppresses the expansion of the fire. This research not only deepens our understanding of the coupling effects between wildfire spread and atmospheric components but also provides important insights for improving and optimizing future wildfire prediction technologies.
How to cite: Fan, G., He, Z., and Luo, J.: Investigating the Impact of Smoke Radiative Feedback on Wildfire Spread, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1896, https://doi.org/10.5194/egusphere-egu25-1896, 2025.
The Santa Coloma de Queralt fire, a two-day wildfire in Catalonia, Spain (2021), exhibited exceptional wildfire behaviour. It spread four times faster than expected and continued burning into the night while maintaining significant intensity. Under normal circumstances, a wildfire would significantly reduce intensity during the transition to nighttime due to decreasing temperature and ambient turbulence in combination with increasing humidity. Additionally, the Santa Coloma de Queralt fire became extreme for a six-hour period, meaning that it became stronger than the extinguishing capacity of the fire service (10,000 kW/m). This combination of exceptional behaviour and extreme intensity makes it impossible to implement mitigation and evacuation measures timely (e.g. evacuation). To improve the predictability of future extreme wildfires, we investigated the Santa Coloma de Queralt fire, for which extensive documentation and measurements are available.
We hypothesised that exceptional wildfire behaviour could be explained by the wildfire modifying the local atmospheric conditions through its convective plume, thereby improving the burning conditions. Previous studies show that wildfires can significantly alter local wind patterns around the flaming zone by creating strong convective plumes. However, limited effort has been focused on fires' ability to change the local atmospheric conditions.
Hence, we simulated the first day of the Santa Coloma de Queralt fire with MicroHH, a three-dimensional large eddy simulation tool designed to resolve turbulent atmospheric convection, such as wildfire-induced plumes. To ensure realistic results, the simulation was validated against an in-plume sounding.
In line with previous work, we find that a convergence zone developed parallel to the fire front. Developing a convergence zone is typically associated with the acceleration of the wind upwind of the flaming zone. However, for the SCQ fire, our simulation shows the most acceleration inside the flaming zone instead of upwind. Furthermore, we find a significant reversal of the flow downwind of the fire, which leads to downdrafts from the overhanging plume towards the surface. These altered wind patterns downwind of the wildfire change the atmospheric stability up to 3 km downwind of the fire.
In conclusion, our results confirm our hypothesis that wildfires can create an environment with improved burning conditions surrounding the plume.
Acknowledgements: This study was part of the EWED project, funded by the European Union (Project no. 101140363).
How to cite: Roelofs, T., Castellnou Ribau, M., Janssens, M., Vilà‐Guerau de Arellano, J., and Van Heerwaarden, C.: Simulating Wildfire-Atmosphere Interactions during the Santa Coloma de Queralt Fire: An Extreme Wildfire Event Continuing into the Night, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11242, https://doi.org/10.5194/egusphere-egu25-11242, 2025.
Understanding and predicting wildfire dynamics is critical to mitigating their impacts. This is particularly relevant in regions experiencing increasing wildfire severity and frequency due to climate change. This study addresses the need for improved wildfire prediction by development of a system that uses Ensemble Fire Predictions (EFP), where we use a probabilistic fire model to model wildfire growth. Ensemble-based methodologies are particularly valuable for wildfire modeling as they account for the inherent variability in weather patterns, fuel conditions, and fire behavior that drive wildfire dynamics.
Our approach couples fire models with datasets on historical and future climate. Specifically, the system incorporates the Fine Fuel Moisture Code (FFMC) and Duff Moisture Code (DMC), (indicators of surface and deeper layer fuel dryness, respectively) from the Canadian Forest Fire Danger Rating System (CFFDRS) to estimate fuel moisture trends using time series analysis of historical weather station data. It also integrates high-resolution weather and climate datasets, including NASA NEX-GDDP-CMIP6, Ouranos ESPO-G6-R2, and CCRN CanRCM4-WFDEI-GEM-CaPA, to evaluate the impact of alternate climate scenarios. Stochastic time series of daily fuel moisture are probabilistically generated based on historical climatology to reflect seasonal variability and day-to-day fluctuations. Historical and modeled wind speed and direction data are used to construct joint probability distributions, enabling the stochastic generation of realistic wind conditions for simulations. This novel methodology allows us to capture a wide range of possible wildfire scenarios, improving the reliability and robustness of predictions.
This research contributes to advancing wildfire spatio-temporal modeling tools by enabling more accurate probabilistic forecasts that can support mitigation strategies and resilience planning. Future work will further develop these methodologies by incorporating the ensemble outputs into Burn-P3, enabling detailed probabilistic modeling of fire spread and burn probabilities, ultimately contributing to better-informed wildfire management and planning, improved resource allocation, and community protection during wildfire events.
How to cite: Reece, K. H., Eythorsson, D., and Clark, M. P.: Advancing Wildfire Risk Assessment Using Ensemble Fire Weather Predictions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7263, https://doi.org/10.5194/egusphere-egu25-7263, 2025.
Wildfires are a critical threat to both people and infrastructures. Although most wildfires in Italy are human-caused, their ignition and propagation are strongly influenced by wildfire-prone meteorological conditions, such as droughts, heatwaves, and strong winds, which are projected to increase in both severity and frequency in the coming decades due to ongoing climate change.
To effectively prevent wildfires and to forecast wildfire risk over a territory, it is essential to understand the meteorological situation in which they have ignited and developed in the past. In this work, we focus on calculating the meteorological wildfire danger through the Canadian Fire Weather Index (FWI) over two high resolution reanalyses for Italy, MERIDA HRES and MERIDA HRES OI.
The FWI represents an estimate of the meteorological wildfire danger of an area, combining 2m temperature, 2m relative humidity, 10m wind speed, and total rainfall fields; therefore, the more accurate the meteorological inputs are, the more accurate the FWI becomes. Meteorological reanalyses represent the most reliable source for such inputs, as they integrate observational data with numerical weather prediction models. This approach enables the detailed reconstruction of past weather conditions over extensive territories, including areas lacking direct observational data
In this context, we have investigated the added value of higher resolution reanalyses by comparing FWI computed over the coarser ERA5 reanalysis with the higher resolution MERIDA HRES and MERIDA HRES OI reanalyses. These two reanalyses, which use ERA5 as a meteorological driver, are downscaled through the WRF-ARW model with parametrizations specifically tailored to the complex geography of the Italian territory. MERIDA HRES covers the period from 1986 to 2021, while MERIDA HRES OI spans 2005 to 2021, integrating observational data for enhanced accuracy.
The comparison has been carried out through the analysis of several case studies and through the analysis of the datasets’ performances over all the wildfires that happened over Italy in the past decade, as well as through considerations over FWI climatological trends. While ERA5 is a robust and extensively validated resource, its coarser resolution poses limitations in accurately capturing the complex topography and local climatic variations of the Italian landscape. The MERIDA HRES datasets, with their finer resolution, consistently outperformed ERA5 in these scenarios, highlighting their added value for applications requiring detailed, high-resolution meteorological data.
In conclusion, MERIDA HRES and MERIDA HRES OI offer valuable tools for improving the characterization of wildfire danger across Italy, benefiting from their higher spatial resolution and parametrization specific for the Italian territory. These datasets contribute to a deeper understanding of the meteorological conditions associated with wildfire danger and provide robust resources for studying climatological trends. Additionally, they support a wide range of stakeholders by aiding in the development of more effective risk management and mitigation strategies in response to the growing threat of wildfires.
How to cite: D'Amico, F., Bonanno, R., Collino, E., Lacavalla, M., Sperati, S., and Viterbo, F.: Characterizing Wildfire Danger in Italy: The Added Value of High-Resolution Reanalyses, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15504, https://doi.org/10.5194/egusphere-egu25-15504, 2025.
This study presents a new differenced Automated Temporal Burn Index (dATBI) designed for mapping burned areas and assessing burn severity using Landsat data and Google Earth Engine (GEE). The dATBI can utilize atmospherically corrected surface reflectance pre- and post-fire images as well as single pre-fire and multi-temporal post-fire images to map burned areas accurately. The atmospheric correction was done using the Simplified Robust and Surface Reflectance Estimation Method (SREM). The effectiveness of dATBI was evaluated across various wildfire events, with its performance compared to the differenced Normalized Burn Ratio (dNBR), a key component in several initiatives such as the Burn Area Emergency Response (BAER), Monitoring Trends in Burn Severity (MTBS), and the Arctic-Boreal Vulnerability Experiment (ABoVE). The dNBR results were generated using Land Surface Reflectance Code (LaSRC) based surface reflectance images. The findings indicate that dATBI outperforms dNBR by accurately identifying fire-affected areas while excluding irrelevant pixels obscured by clouds, snow, water bodies, and other land features. In contrast, dNBR tended to misclassify these obscured features as burned areas, resulting in significant commission errors. The dATBI can generate seasonal or annual mean burned area maps using single pre-file and multi-temporal post-fire images. Overall, the results underscore the robustness of dATBI, demonstrating its applicability across diverse regions and its ability to manage large datasets effectively.
How to cite: Bilal, M.: dATBI: A New Remote Sensing Index for Burn Area Mapping Using Landsat Data and Google Earth Engine, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5093, https://doi.org/10.5194/egusphere-egu25-5093, 2025.
Wildfires are becoming increasingly more frequent and devastating across Europe. In recent years, wildfires consistently set new records, and have occurred in regions that are historically less fire-prone. It is still unclear how the drivers of wildfires vary within space and time across Europe, though understanding their composition is highly relevant for mitigating fire risk and exposure, especially with regards to climate change.
Here, we study the spatial and temporal patterns of wildfire drivers in eight distinct European climate regions by leveraging daily FireCCI burned area observations together with CERRA reanalysis data for hydro-climatic variables and MODIS gross primary productivity for fuel availability between 2001 and 2020. We develop random forest models for each region and season to identify the most important drivers of wildfire occurrence. To identify the time scales over which wildfire-favoring conditions develop, we analyzed the persistence of standardized anomalies before a fire event by using incrementally increasing temporal windows.
We find strong anomalies of all drivers on fire days in comparison to non-fire conditions across all subregions and seasons - but the combination and strength of these drivers varies in time and space. Overall, drought conditions are the most important modulator of wildfire activity. Vegetation deficits are most relevant for wildfire occurrence in spring and summer, while long-term drought indicators, such as soil moisture deficits and the Standardized Precipitation Evapotranspiration Index, are most important in fall and winter. The seasonal cycle of gross primary productivity (GPP) before wildfire occurrence underlines the dynamic interactions between vegetation, drought, and fire. During spring and summer, wildfire events occur under seasonal GPP deficits, whereas in fall and winter fires occur under seasonal GPP surpluses. The persistence analysis highlights the time scales over which hot and dry conditions reduce GPP and increase fuel availability: In summer, dry conditions lead to less GPP and higher fuel loads on fire days in comparison to non-fire days, whereas in the fall high fuel loads originate from GPP surpluses of the previous spring that dry out during hot and dry summer weather.
Our findings illustrate the complex interdependencies of factors contributing to wildfire events across different climate regions in Europe and time scales, underscoring the need for targeted wildfire mitigation and adaptation strategies, especially in the context of climate change.
How to cite: Miller, J., Touma, D., and Brunner, M.: Spatial and temporal patterns of wildfire drivers across Europe, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3045, https://doi.org/10.5194/egusphere-egu25-3045, 2025.
Extreme weather conditions, such as heatwaves and droughts driven by climate change, have led to a surge of large wildfires across the globe. This trend is exacerbated by rapid urban expansion and increasing interactions between human societies and wildlands, making the prediction of wildfire risk an urgent research priority.
While the Fire Weather Index (FWI) has been widely employed to evaluate fire risk, it primarily considers meteorological factors including wind, precipitation, temperature, and relative humidity. Some classic machine learning algorithms, such as Random Forest (RF), and deep learning approaches, represented by Convolutional Neural Networks (CNN), have been utilized to better capture nonlinear characteristics of wildfires. However, Transformer models, although proven efficient in multiple tasks varying from natural language processing to weather forecasting, remain largely unexplored in the context of wildfire risk prediction. Moreover, few studies have attempted to predict fire regimes at a global scale.
Therefore, our research aims to predict the next day’s global wildfire danger with high accuracy. We first established a comprehensive global wildfire database covering years from 2001 to 2020. The database contains historical burned areas records, as well as 50 key variables influencing occurrence and spread of wildfires, categorized as ignition source, fuel availability, weather condition, human activity, and topography. We then employed the Earthformer model, a transformer-based model incorporates a space-time attention block, to effectively capture the complex interplay of factors affecting wildfire regimes. By utilizing the daily dynamic variables (e.g. relative humidity) for days t-1, t-2, …, t-10 and constant variables such as land cover type, we predicted the probability for wildfire on day t. Our results indicate that Earthformer performs well with an F1-score for the positive sample (which represents high fire risk) greater than 0.85, which outperformed RF and XGBoost according to the confusion metric. Additionally, we implemented explainable AI (xAI) techniques to rank the importance of each factor contributing to fire risk.
Our study re-evaluated and generated global fire risk maps since 2020, providing essential insights for resource allocation in fire prevention strategies. By enhancing the understanding of wildfire dynamics, we aim to facilitate a better coexistence between communities and wildfires, ultimately contributing to improved resilience and mitigation efforts in the background of climate change.
How to cite: Guo, Y. and Wang, J.: Predictability of global wildfire risk with transformer-based model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6174, https://doi.org/10.5194/egusphere-egu25-6174, 2025.
Effective detection and identification of wildfires are essential for efficient control and mitigation of their impacts. Satellite remote sensing is commonly used for hotspot detection, but its effectiveness is hindered in subtropical monsoon climates due to cloud and fog interference. Recently, smoke signals produced by wildfires have been successfully detected using weather Doppler radar, providing a valuable supplement to satellite-based monitoring. However, existing fire area segmentation techniques based on radar reflectivity data face significant challenges, including poor segmentation at target boundaries, limited adaptability to targets of varying sizes, and insufficient consideration of temporal correlations between data frames. To address these issues, we propose a novel wildfire segmentation approach that integrates a global-local attention mechanism with temporal correlation information. First, the Global-Local Attention (GPA) module is used to extract both key local features and global distribution patterns, thereby enhancing segmentation accuracy, particularly at target boundaries. Second, a Multi-Scale Fusion (MSF) module combines spatial features at multiple scales, enabling the model to better capture diverse spatial hierarchies of fire points and adapt to targets of varying sizes. Finally, a Temporal Feature Extraction (TEF) module is introduced to capture temporal dependencies, leveraging the correlations between consecutive data frames. Experimental results on the Fire-Radar Reflectivity (FRR) dataset demonstrate that our model outperforms baseline approaches. Compared to the Trans-UNet model, it improves pixel-level accuracy and target-level precision by approximately 3% and 4%, respectively, and by approximately 3% and 5%, respectively, compared to the state-of-the-art Evit-Unet model.
How to cite: He, Z., Fan, G., and Zeng, Z.-C.: Improved Wildfire Detection and Segmentation Using a Global-Local Attention Mechanism for Doppler Radar Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1893, https://doi.org/10.5194/egusphere-egu25-1893, 2025.
This study explores the potential for different fuel management strategies to mitigate future wildfire risks across Europe by leveraging advanced modeling techniques that integrate future climate and land-use change scenarios. Using the Lund-Potsdam-Jena managed Land model (LPJmL), coupled with the SPITFIRE fire model, we simulate the impacts of five fuel management interventions across four fuel classes, ranging from fine fuels (e.g., grasses and leaves) to coarse fuels (e.g., branches and mature trees). These scenarios are based on SSP1 (Shared Socioeconomic Pathway 1; "Sustainability") and SSP3 ("Regional Rivalry") pathways, aligned with Representative Concentration Pathway (RCP) 2.6 and RCP7.0, respectively. The study evaluates fire intensity, surface fire rate of spread, fuel bulk density, and biomass changes to assess how fuel-removal interventions (e.g., prescribed burning and mechanical removal) can influence burned area under varying future conditions.
Our findings highlight that fine fuel management is the most effective strategy for reducing wildfire spread in Europe, with especially potential burned area reductions in the Mediterranean. We thus suggest that in temperate and boreal Europe, retaining coarse fuels can contribute to ecosystem health through moisture retention, habitat conservation, and carbon storage. However, managing coarser fuels is critical near wildland-urban interfaces to mitigate fire risks and ensure accessibility for emergency responders in all parts of Europe. This is especially relevant given the large interannual variability in heat and precipitation which can create unpredictable conditions favoring severe fires in the Mediterranean region. We conclude that whether Europe’s future follows a more sustainable trajectory along the lines of SSP1 or a more tumultuous and nationalistic pathway such as SSP3, wildfire will remain a persistent threat with the potential to undermine climate change mitigation efforts. This highlights the need to view landscapes and priorities through a fire-focused lens, emphasizing targeted fuel treatments that optimize resource use and enhance fire resilience.
How to cite: Neidermeier, A., Billing, M., West, T. A. P., Thonicke, K., and Verburg, P. H.: Insights from Fuel Management Simulations for Wildfire Risk Mitigation in Europe under Future Climate Scenarios , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20715, https://doi.org/10.5194/egusphere-egu25-20715, 2025.
Fires in Chile are one of the main natural threats to society and the balance of ecosystems. As is common, the cause behind most of them is human action, and their occurrence is linked to numerous climatic, environmental, or accessibility factors. The aim of this study was to predict the daily probability of ignition for all continental Chile, with a spatial resolution of 100 meter, distinguishing between intentional and unintentional fires to study potential disparities in the role of fire drivers. To achieve this, a Random Forest machine learning model was trained and tested using CONAF's ignition data from 2009 to 2019.
A total of 100 model realizations were calibrated by combining randomly stratified samples of fire ignition with spatial variables related to accessibility, anthropogenic presence, infrastructure, and dead fuel moisture content. For each realization, we trained and evaluated a binary classification Random Forest model, aggregating their outcomes and predictions to account for uncertainty. Models were also evaluated in terms of prediction ability and residuals independence.
The results show differences between intentional and unintentional fires in terms of accuracy (0.86 and 0.83, respectively), but also regarding the role of the drivers. Notable differences in variable importance were also observed, with distance to power lines being the most important variable for intentional fires, while the Wildland-Urban Interface (WUI) played a larger role for unintentional fires. While the importance of WUI had been identified in previous studies, the significance of distance to power lines had not been widely considered, despite its potential impact on the accessibility of remote areas with high fuel loads. Interestingly, dead fuel moisture (DFMC) and fuel types were less important in both models, with DFMC showing surprisingly low relevance, contrary to expectations. The ignition probability maps generated displayed similar small-scale spatial patterns, with high ignition probabilities concentrated in central Chile, where most studies have been conducted. The southern and northern regions showed either negligible or low ignition probabilities, mainly due to a lack of fuel. At a local scale, intentional fire models were clearly associated with power lines and road networks, while unintentional fires were more influenced by proximity to buildings. Areas farther from human activity centers showed higher probabilities for unintentional fires, likely linked to recreational activities.
How to cite: Crespo Pérez, R., Rodrigues Mimbrero, M., Félez Bernal, J., de la Riva Fernández, J., Serrano Notivoli, R., Peña Angulo, D., Gelabert Vadillo, P., and Galizia, L.: Spatial modeling of fire ignition in Chile: comparing arson and accidental fires, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9110, https://doi.org/10.5194/egusphere-egu25-9110, 2025.
Wildfires are destructive to ecosystems and human life, exacerbated by climate change, yet deep learning models for fire forecasting lack interpretability, as they often rely on black-box models or post-hoc explainability methods only approximating the models' decision process. This limits their use for real-world applications and their potential to discover new scientific insights on wildfire regime shifts under climate change.
This study tests prototype learning as a per-design method for interpretable wildfire forecasting. The model selects real patches seen during training as prototypes and constructs the predictions based on the similarity between parts of the test region of interest and said prototypes. The dataset used is the SeasFire datacube, which forecasts wildfires with a lead time of eight days from eight environmental variables. We use a U-NET++ baseline and 10 prototype vectors per class: fires and no fires. A prototype layer computes the cosine similarity of the normalized output feature map pixels with all the normalized prototypes in a latent dimension space: D = 64. Then the 20 similarity scores are passed to a classification layer for all pixels. Three losses regularize learning by enforcing 1) clustering of the pixels around the prototypes, 2) orthogonality of the prototypes, and 3) a uniform use of prototypes across a batch. Our interpretable method achieved comparable performance to the non-interpretable baseline: U-NET++ (F1 score: 0.544, AUPRC: 0.590).
However, unlike images in RGB, the prototypes and their activations are not easily interpretable for spatial environmental inputs (here represented by 8 independent input channels). To address this issue, we propose two strategies for prototype summarization. First, through human-centered interpretability, we compute the 2D Wasserstein distance between each fire prototype activation and the environmental inputs for all patches with fires. For the three most common fire prototypes (located in Africa, Europe, and Australia), this approach showcases their similarity to the land surface temperature patterns but also, depending on the prototypes, different levels of proximity with the NDVI or relative humidity heatmaps as the second closest environmental variable. The second approach aims at approximating the model's non-linear relationships between environmental variables and prototype activations via a white-box model like Generalized Additive Models (GAMs) which predicts the prototype activations via a linear combination of smooth functions for all environmental variables independently. Predicting the prototype activation map leads to a R2 score of up to 0.682, and allows us to explain linear correlations, (such as between vapor pressure deficit and prototype activations) across the most common prototypes, or, depending on the prototypes, different functional relationships between NDVI and their activations.
In this study, we investigate the potential and limitations of per-design interpretability methods for wildfire forecasting with Earth observation data. In particular, we match the results of non-interpretable models, breaking a myth of the underperformance of XAI methods. Moreover, we propose two approaches to alleviate the lack of interpretability of prototypes via model approximations: GAMs or human-centered pattern matching with 2D Wasserstein distance. Both methods reveal interesting insights into the role of environmental predictors for wildfire forecasting.
How to cite: Porta, H., Kamoun, I., and Tuia, D.: Interpretable by-design wildfire forecasting via prototypes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15783, https://doi.org/10.5194/egusphere-egu25-15783, 2025.
Wildfires are frequently occurring hazards worldwide which are moving higher in elevation and threatening mountain regions. Each year, they result in substantial economic losses, fatalities, and carbon emissions. In addition, the interplay of climate change, land use changes, and socioeconomic factors is expected to increase the frequency and intensity of wildfires. In this context, developing reliable tools and early warning systems is critical to mitigate and reduce future impacts. At regional scales, data-driven analyses are commonly used to evaluate wildfire susceptibility based on static environmental conditions. However, the integration of the spatial and temporal domains remains challenging. Currently, there is evidence of an increasing trend in wildfires in the region of Trentino-Alto Adige, located in the northeastern part of the Italian Alps. Although this area has experienced limited impacts from wildfires in the past, new tools and applications are needed to prepare for worsening conditions.
This work aims to predict the occurrence of wildfires in space and time (i.e., the ‘where’ and the ‘when’) in Trentino-Alto Adige (13,600 km²). The analyses built upon a generalized additive model (GAM), multitemporal wildfire data from 2000 to 2020, and static and dynamic environmental controls (e.g., topography, land cover, daily precipitation, and temperature). The methodical framework involves filtering the wildfire inventory (wildfire presence data), sampling wildfire absences in space and time, extracting the environmental predictors, and removing trivial terrain and periods. The resulting predictions change dynamically as a function of static factors, seasonality, dynamic precipitation and temperature and are transferred into space under varying precipitation and temperature conditions to hindcast wildfire events. The model output is linked to known performance measures in order to estimate wildfire susceptibility thresholds that can be interpreted in analogy to commonly used empirical landslide rainfall thresholds. The validation routines confirm the high generalizability and predictive power of the model while providing insights into the interplay of environmental factors for wildfire occurrence in Trentino-Alto Adige. Application possibilities are presented.
The research that led to these results is related to the EO4MULTIHA project, which received funding from the European Space Agency (ESA).
How to cite: Moreno, M., Steger, S., Bozzoli, L., Terzi, S., Trucchia, A., van Westen, C., and Lombardo, L.: Space-time data-driven modeling of wildfire initiation in the mountainous region of Trentino-Alto Adige, Italy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17023, https://doi.org/10.5194/egusphere-egu25-17023, 2025.
Forest fires are a complex natural phenomenon, difficult to model because they depend many parameters, which vary in both time and space. It is therefore necessary to carry out research and prevention actions as part of the improvement and management of this risk. The severity of wildfires due to hotter and drier global climate conditions affects the ecological resilience and ecosystems at risk of deterioration following the failure of post-fire recovery. To properly prepare for wildfires, it is crucial to determine fire-sensitive areas, then locate fire suppression structures, and assess the spatial and temporal quantification of post-fire regeneration. The objectives of this study are, in the first stage, to introduce a new approach to fire detection using artificial intelligence, and in the second stage, to model the dynamics of regeneration and monitor the recovery of vegetation using satellite imagery and the post-fire stability index. This method is therefore based on the concept that the state of a disturbed system will be reflected by increasing or decreasing rates of change. While undisturbed or recovered system states are characterized by rates of change close to zero. This reflects the typical pattern of decreasing change rates in post-fire recovery trajectories. To do this, time series analyses of remote sensing images from Landsat and Alsat satellites between 2010 and 2023, both pre- and post-fire, were conducted in the Sidi Bel Abbés region, Algeria, to evaluate the post-fire stability index. Moreover, the rate of vegetation recovery after a fire was assessed using the normalized regeneration index. (NRI, RI). We therefore demonstrate the performance and relevance of the post-fire stability index compared to alternative approaches because this stability index provides a relatively simple and practical solution for consistent large-scale monitoring of post-fire recovery with satellite imagery, which, combined with standardized mapping of fire severity, thus offers numerous opportunities for further research on fires and landscape ecology.
How to cite: Zegrar, A., Bentekhici, N., Saad, A., and hadj shraoui, O.: Assessment and monitoring of post-fire recovery using satellite imagery and the stability index in the north west of Algeria, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-183, https://doi.org/10.5194/egusphere-egu25-183, 2025.
Accurate fire risk estimation requires accounting for fuel moisture and fuel load assessment. Satellite-retrieved vegetation parameters offer valuable insights into fuel characteristics but are challenging to integrate due to the complex and varying interactions between vegetation and bushfires during the pre- and post-fire stages. This study explores the potential of vegetation parameters to predict fire risk independently of fire weather data. Our focus is on the pre-fire stage where the fire risk rises from a minimum threshold, which is a key determinant in bushfire ignition. Using the McArthur Forest Fire Danger Index (FFDI) as a fire danger measure, we found that incorporating vegetation optical depth (VOD) into predictive models significantly enhances the performance compared to models that base past fire risk information alone. VOD was identified as a causal driver of FFDI in a significant number of fire-prone pixels in Australia, and the VOD-induced model outperformed the model that used only the past fire risk information over a 12-month lead span. These findings highlight the potential of vegetation dynamics as a standalone predictor of fire risk when the knowledge on fire weather is uncertain or unavailable. Future research will focus on enhancing this predictive framework by incorporating terrestrial water storage as an additional predictor, building on the recent studies that highlight the effectiveness of terrestrial water storage in explaining the vegetation dynamics variability and reflecting the moisture conditions during pre-fire periods.
How to cite: Liu, Y., Kankanige, D., and Sharma, A.: From vegetation dynamics to fire risk: a predictive framework, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-281, https://doi.org/10.5194/egusphere-egu25-281, 2025.
Wildfire smoke has raised concerns on air quality and public health with the increasing intensity, frequency, and duration of wildfires as a result of climate change. This study generates a global near real-time wildfire-related PM2.5 (i.e., fire PM2.5) concentration product by combining multisource data with machine learning algorithm. This is the first daily updated full-coverage high-resolution fire PM2.5 data products that allows timely tracking of the fast-growing fire PM2.5 globally. The gridded fire PM2.5 data at a spatial resolution of 0.1°×0.1° are estimated by fusing surface PM2.5 monitoring, satellite observations, meteorological fields, atmospheric composition reanalysis data, and population distribution through a three-layer random forest model. We found that during 2023-2024, wildfire smoke contributed 1.32 μg/m3 (4.7%) and 1.25 μg/m3 (4.7%) to population-weighted annual average PM2.5 worldwide, and caused 50,700 (95% confidence interval: 33,600-68,300) and 51,500 (34,100-69,400) all-cause deaths through acute fire PM2.5 exposure, respectively. Regionally, the record-breaking wildfires resulted to 1.42 μg/m3 (21%) and 2.53 μg/m3 (16%) increase in population-weighted annual average PM2.5 in Canada (2023) and South America (2024), respectively. We noticed that a relatively small number of extreme wildfire episodes could disproportionately impact regional public health, emphasizing the importance of timely monitoring of wildfire-induced PM2.5 pollution. The global fire PM2.5 data will be publicly available on the Tracking Air Pollution platform (TAP, http://tapdata.org.cn), to support promptly health impact assessment and policymaking for wildfire risk mitigation.
How to cite: He, C., Xiao, Q., Geng, G., and Zhang, Q.: Tracking Air Pollution: Global Near Real-Time Fire PM2.5 Retrievals from Multisource Data Fusion, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5476, https://doi.org/10.5194/egusphere-egu25-5476, 2025.
Accurate seasonal fire predictions can be decisive for mitigating wildfire risks, optimizing firefighting resources, and informing climate adaptation strategies. This study introduces an innovative hybrid approach that combines process-based seasonal climate predictions with a Random Forest (RF) climate-fire model to forecast burned area (BA) anomalies at a global scale. Utilizing the Standardized Precipitation Index (SPI) derived from both observations and ECMWF SEAS5 seasonal predictions, we demonstrate skillful fire forecasts up to four months.
Our findings indicate that observational data allows predictions of BA anomalies in approximately 68% of the burnable area globally, while skillful results are achieved in 46% of the area when incorporating seasonal forecasts. The RF model substantially outperforms traditional logistic regression models, capturing complex, non-linear relationships between climate variables and fire dynamics. The system achieves its highest skill in fire-prone regions, such as Australia and South America, leveraging antecedent and concurrent drought conditions to improve predictability.
This hybrid approach underscores the importance of integrating observational and forecast data to enhance the skill of seasonal fire predictions. By leveraging machine learning techniques, the system provides a flexible and robust framework for developing operational fire forecasts, paving the way for proactive wildfire management strategies under a changing climate.
Acknowledgements:
This work was supported by the project ‘Climate and Wildfire Interface Study for Europe (CHASE)’ under the 6th Seed Funding Call by the European University for Well-Being (EUniWell). M.T. acknowledges funding by the Spanish Ministry of Science, Innovation and Universities through the Ramón y Cajal Grant Reference RYC2019-027115-I and through the project ONFIRE, Grant PID2021-123193OB-I00, funded by MCIN/AEI/10.13039/501100011033 and by “ERDF A way of making Europe”. AP acknowledges the support of the EU H2020 project “FirEUrisk”, Grant Agreement No. 101003890.
How to cite: Turco, M., Torres-Vázquez, M. Á., Herrera, S., Gincheva, A., Halifa-Marín, A., Cavicchia, L., Di Giuseppe, F., and Montávez, J. P.: Hybrid Climate-Fire Models for Better Seasonal Wildfire Predictions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5938, https://doi.org/10.5194/egusphere-egu25-5938, 2025.
PROPAGATOR is a fire spread simulator designed as a stochastic cellular automaton model for rapid fire risk assessment. The model uses high-resolution data about topography and vegetation cover, accounting for different vegetation types. Key inputs include wind, fuel moisture, and the ignition point. Additionally, the model can incorporate firefighting strategies, such as modifying fuel moisture content or implementing firebreaks. The probability of fire spread is influenced by vegetation type, slope, wind, and fuel moisture content, while fire-propagation speed is calculated using a Rate of Spread model. PROPAGATOR generates independent realizations of a stochastic fire propagation process. At each time step, it produces maps showing the probability of each cell in the domain being affected by fire, along with the potential rate of spread and fire-line intensity.
The transition from low-intensity surface fires to burning in the vegetation canopy results in significantly larger flame heights, higher energy release rates, and increased rates of spread. Distinguishing between ground fire and canopy fire is therefore crucial for end users, impact evaluations, and the calibration of fire spotting submodels.
To achieve this, incorporating Canopy Fuel Characteristics—such as canopy base height, canopy fuel load, canopy bulk density, and foliar moisture content—while applying appropriate simplifications, will be a critical step. Implementing Crown Fire Initiation and Spread Models will complement those already used in PROPAGATOR for ground fire, with adaptations of well-established models where feasible. Additionally, Vertical Interaction Mechanisms will be introduced into the probabilistic rules of the cellular automaton to represent conditions under which surface fires escalate to canopy fires and vice versa.
These improvements will be validated using both synthetic and real case studies to assess their benefits for end users and practitioners. The development of these upgrades is driven by work conducted within the framework of the RETURN extended partnership (Multi-risk science for resilient communities under a changing climate) which aims at strengthening national research chains on environmental, natural, and anthropogenic risks while fostering participation in European and global strategic value chains.
Keywords: wildfire propagation models, cellular automata, crown fires, wildfire risk, wildfire risk management
How to cite: Trucchia, A., Perello, N., Meschi, G., D'Andrea, M., Ghasemiazma, F., Degli Esposti, S., and Fiorucci, P.: Expanding PROPAGATOR Cellular Automata based wildfire simulator to represent surface and crown fire transitions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6898, https://doi.org/10.5194/egusphere-egu25-6898, 2025.
Forest fires are one of the major forest disasters that pose various threats to both natural ecosystems and human societies, including biodiversity loss, large-scale destruction of forest resources, greenhouse gas and pollutant emissions, reduced tourism, and weakened ecosystem services. In the Republic of Korea, forest fires are primarily caused by human negligence. However, climate change factors, such as prolonged droughts and changes in precipitation patterns, also play a significant role in increasing the likelihood of forest fire occurrence. This study aimed to develop a machine learning-based forest fire prediction model using anthropogenic activity data, meteorological data, and climate extreme indices derived from SSP scenarios. PyCaret, a low-code machine learning library, was employed to compare and optimize various machine learning algorithms, maximizing predictive performance. The model can be utilized to identify high-risk areas in advance and assess forest fire risks under changing climatic and socioeconomic conditions. Furthermore, it is expected to provide scientific evidence for formulating forest fire prevention and management policies, thereby enhancing disaster response capacity and supporting sustainable forest management.
How to cite: Kim, S., Roh, M., and Lee, W.-K.: Forest Fire Risk Prediction Based on Machine Learning with Shared Socioeconomic Pathways (SSP) Scenarios, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9009, https://doi.org/10.5194/egusphere-egu25-9009, 2025.
In July 2022 the Iberian Peninsula was affected by an extreme heat wave, leading to multiple maximum temperature records. Extreme wildfire events also impacted northwest Spain under these conditions, with several fires exceeding 10,000 ha burned, some of them developing pyroconvection. Here we analyse, with the use of observations and the WRF-Fire coupled atmospheric-fire model, how the atmospheric environment have influenced the development of these extreme wildfire events, considering their ignition, spread and fire-atmosphere coupling. The results show the potential for these particular meteorological conditions to support the development of highly chaotic and severe fires that pose a challenge to suppression efforts.
How to cite: Romera Otero, D., Senande-Rivera, M., and Miguez-Macho, G.: Atmospheric conditions prone to extreme wildfire development: a modeling study for the July 2022 heatwave in the Iberian Peninsula, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9825, https://doi.org/10.5194/egusphere-egu25-9825, 2025.
Forest fire prevention, forecast, and control are becoming increasingly popular issues, in large part because of climate change. While several early warning systems use remotely sensed images collected by optical and non-optical sensors, as well as supervised AI (Artificial Intelligence) algorithms to detect fires early on, the development and dissemination of reliable, low-cost sensors together with the advancement of the IoT (Internet of Things) paradigm make it possible to apply monitoring techniques relying on widespread ground-based sensor networks.
This paper illustrates an innovative technique where smart CO2 sensors were used to capture smoke produced by combustion and discriminate an alert through AI techniques. In more detail, a small-scale field experiment was conducted where 44 CO2 sensors were deployed on a hillslope, triggering a small controlled fire. The sensors were connected via LoRaWan (Long Range Wide Area Network) technology and a gateway to an online platform that included an optimized database and an interactive management interface. Several environmental variables were monitored during the experiment, most notably wind speed and direction. In addition, 3 unsupervised AI algorithms were tested to discriminate alerts (Long-Short Term Memory - LSTM; AutoEncoder on CO2 absolute values and AutoEncoder on CO2 differences between two consecutive measurements) and compared with a classical alert system based on thresholds calibrated on each sensor, using the maximum CO2 recorded in the 5 days prior the experiment, in absence of fires.
Several sensors detected anomalies in CO2, particularly those placed downwind. The results highlighted the capabilities of AI to better discriminate the alert with respect to the classical no-AI system. More specifically, the application of AI-based methods could also bring the alert on many sensors forward with respect to the no-AI method. Future deployments of such a system will be carried out in a broader area, employing more than double the number of sensors and combining them with other detection technologies (e.g., remotely sensed RGB and IR images) and AI techniques.
Acknowledgments: This study was funded by The Next Generation EU—Italian NRRP, Mission 4, Component 2, Investment 1.5, call for the creation and strengthening of ‘Innovation Ecosystems’, building ‘Territorial R&D Leaders’, and Project Tech4You—Technologies for climate change adaptation and quality of life improvement, n. ECS0000009. This work reflects only the authors’ views and opinions, neither the Ministry for University and Research nor the European Commission can be considered responsible for them.
How to cite: Furnari, L., De Rango, A., Cortale, F., Senatore, A., and Mendicino, G.: Experimental Validation of a Wildfire Early Warning System Based on a CO2 Sensor Network, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15094, https://doi.org/10.5194/egusphere-egu25-15094, 2025.
The threat to life and property, and the relationship between fire regimes and biodiversity, are arguably the most significant ongoing challenges facing managers of parks and forests. Fuel moisture is a primary driver of fuel flammability and subsequent fires and varies spatially and temporally across landscapes. Vapour Pressure Deficit (VPD) is a measure of atmospheric dryness that strongly influences dead fuel moisture content. Previous research has established strong links between VPD, burned area, and fire severity at broader spatial and temporal scales. More recent work has found in-forest VPD is generally a stronger predictor of ignition and sustained burning than broader landscape variables. Understanding the spatial-temporal trends and variability of VPD is crucial for understanding future wildfire risk and estimating what (if any) management actions can reduce risk.
This study utilizes a long-term (40 year) dataset of in-forest VPD collected from weather stations established in 1984 in eucalyptus forest in southeastern Australia. Data were analysed to map temporal variations in VPD across different microclimates within the forest. Thresholds from previous research were used to determine availability for ignition and spread. Seasonal analyses were undertaken to examine the potential for wildfires (summer) or prescribed fire (spring and autumn).
The number of potential wildfire days has increased in the summer period over the duration of the study, with an acceleration in the last twenty years. Prescribed fire opportunities have also increased however these results should be cautiously interpreted as they may also represent an extension of the wildfire season compared to historic conditions.
Long term micro-climate studies are rare, and these data provided unique insight into the spatial-temporal variations of VPD within a eucalyptus forest in southeastern, Australia. These data support the notion that changes in climate are a much greater driver of fire regime changes, compared to land management decisions.
How to cite: Najera Umana, J., Penman, T., and Burton, J.: Forty years of micro-weather observations provide insights into variations of in-forest vapor pressure deficit (VPD) within a mixed eucalyptus foothill forest fire regime change, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21846, https://doi.org/10.5194/egusphere-egu25-21846, 2025.
How to cite: Falconieri, A., Filizzola, C., Mazzeo, G., Tramutoli, V., and Pergola, N.: On the portability of the RST-FIRES technique to higher resolution EUMETSAT systems for early fire detection, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8440, https://doi.org/10.5194/egusphere-egu25-8440, 2025.
Forest fires increasingly threaten human lives, properties, and ecosystems, with climate change amplifying their size, intensity, and simultaneous occurrences. In South Korea, where forests cover over 60% of the land and wildland-urban interfaces are extensive, mitigating wildfire impacts requires accurate and timely fireline predictions to optimize firefighting resource allocation. While existing process-based propagation models provide rough estimates, they face limitations in capturing the complex dynamics of wildfire behavior influenced by weather, fuel, and topography. Additionally, the scarcity of time-series fireline observations and data on firefighting interventions hinders the development of AI-driven predictive models. This study introduces a hybrid wildfire propagation model that integrates process-based algorithms with AI techniques. The system calculates the rate of spread (ROS) and fireline movement using a process-based approach, while neural networks refine model parameters using 5-meter-resolution topography, forest type maps, and hourly weather data. The model generates predictions at 1-minute intervals and is trained with diverse loss functions to assimilate process-based parameters, ROS calculations, and historical fireline data from 27 wildfire events. Validation on five wildfire cases demonstrated the hybrid model’s improved performance over traditional process-based models, achieving Intersection Over Union (IOU) scores ranging from 0.4 to 0.6, with an average improvement of 0.14. These results highlight the potential of the hybrid model to enhance prediction accuracy and bridge the gap between conventional and advanced modeling methodologies. Future work will focus on expanding the training dataset and refining the model to address uncertainties in ROS predictions caused by firefighting interventions.
How to cite: Jo, H.-W., Roh, M., Kim, S., Jeong, Y., Cha, S. E., Lee, B., and Lee, W.-K.: Development of Process- and AI-Based Hybrid Wildfire Propagation Prediction System for South Korea, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2152, https://doi.org/10.5194/egusphere-egu25-2152, 2025.
Due to climate change, there is an urgent call for scientific research into the prevention and mitigation of wildfires. Within the last 50 years, mathematical models for forest fire propagation have been developed to understand and predict the evolution of fire. In this work, we present a simplified Advection-Diffusion-Reaction (ADR) model that is physics-based and accounts for the effects of environmental conditions, topography, and the distribution and heterogeneity of fuel. The model consists of two equations: a partial differential equation for the conservation of energy and an ordinary differential equation for the evolution of biomass. It explicitly represents fuel moisture effects by means of the apparent calorific capacity method, distinguishing between live and dead fuel moisture content. Although simplified, the model is derived from the theory of two-phase porous flows and emphasizes a robust theoretical foundation. Using this model, we conduct exploratory simulations and present theoretical insights into various modeling decisions in the context of ADR-based models. We seek to understand the interplay between the different mechanisms involved in wildfire propagation, to identify key factors influencing fire spread, and to estimate the model's predictive capacity. We show that the model results are consistent with laboratory experiments and field observations by carrying out parametric analyses and qualitative comparisons. The rate of spread predicted by the model exhibits an exponentially decaying trend with increasing fuel moisture and a Ricker function-like behavior with changes in bulk density, which is consistent with previous literature. The results herein presented help build confidence in the model’s predictive capability and motivate further steps towards the application of the model to real-world scenarios.
How to cite: Navas-Montilla, A., Reisch, C., Díaz, P., and Özgen-Xian, I.: A physics-based simplified model for simulating wildfire spread in heterogeneous environments, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8223, https://doi.org/10.5194/egusphere-egu25-8223, 2025.
Forest fires pose an escalating threat to biodiversity, particularly in ecologically sensitive regions like the Himalayas. Uttarakhand, with its unique ecosystems and high proportion of endemic and endangered species, has experienced a significant increase in forest fire frequency in recent decades, largely driven by climate warming. Despite growing concerns, research on the interplay between climate dynamics and forest fire events in Uttarakhand remains limited, with little quantitative analysis of how these events impact biodiversity hotspots. Using satellite observations, climate reanalysis, and forest survey reports, we investigated how climate warming has altered the dynamics of forest fire events in Uttarakhand over the past two decades and evaluated their effects on local biodiversity. Fire records from MODIS and VIIRS spanning 2000–2024 reveal a marked increase in both the annual frequency and spatial extent of forest fires. The frequency of these fires is significantly correlated with rising temperatures, reduced pre-monsoon precipitation, wind speed, and relative humidity. Pre- and post-fire imagery indicates that forest fires impact more than 10% of biodiversity hotspots annually. Involving local communities in fire reporting and management, alongside reliable early warning systems, can be essential to mitigate fire risks. Our findings provide a scientific foundation for policymakers and conservation practitioners to reduce biodiversity loss and enhance ecosystem resilience in the face of escalating fire risks and global warming.
How to cite: Chaudhary, K.: Escalating Forest Fire Events and Biodiversity Loss in Uttarakhand Under a Warming Climate, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17217, https://doi.org/10.5194/egusphere-egu25-17217, 2025.
Soil moisture is an important variable in wildfire risk assessment, influencing fuel moisture content, fire ignition potential and fire spread dynamics. In the existing fire danger rating systems, soil moisture is largely overlooked while at the same time, wildfire seasons are increasingly becoming longer with larger burnt areas. Studies have shown that using, e.g., in situ or model soil moisture information in fire danger ratings could better help forecast wildfires and lead to earlier warnings of wildfire dangers. On the other hand, the availability of ground-based monitoring stations providing reliable soil moisture information is limited.
This study focuses on validating our in-house soil moisture models against ground-based measurements, ensuring the model’s reliability for wildfire risk modelling. The large-scale, spatially resolved soil moisture and related variables were derived using thermal infrared remote sensing combined with surface energy and soil water balance models. Ground-based soil moisture data were obtained from networks such as the International Soil Moisture Network for diverse land cover types. Validation was carried out using statistical performance metrics and correlation coefficients to identify discrepancies and to improve model accuracy.
While the modelling and validation processes are still ongoing, preliminary results suggest there is an acceptable agreement for crop fields and ground-based data, whereas forest land cover validation remains challenging, showing the need for further refinements in the soil moisture model. This work also highlights the importance of access to reliable and frequent ground-based soil moisture data. Application to wildfire seasons in Australia show that soil moisture and evapotranspiration have high feature importance, emphasising their relevance in accurately predicting fire risk.
This research is an important step forward for bridging the gap between soil moisture science and wildfire risk modelling, also creating effective discussion on the topic, advancing our understanding and potentially improving fire danger ratings in the future.
How to cite: Byckling Smith, K., Chartrand, R., Werner, F., Ziliani, M., and de Leede, I.: Validating soil moisture models for wildfire risk assessment , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13222, https://doi.org/10.5194/egusphere-egu25-13222, 2025.
How to cite: Thoreau, R., Baggio, R., and Filippi, J.-B.: Emulation of Wildfire Rate of Spread Models in the Context of Surface Fire Spread Simulation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12173, https://doi.org/10.5194/egusphere-egu25-12173, 2025.
Accurate and rapid prediction of wildfire behavior is essential for effective management and mitigation efforts. However, the unpredictable nature of fire spread poses significant challenges to developing reliable simulators. Moreover, these models typically require parameter identification and adjustments based on real-time observations. Current physics-based simulations are mainly CPU-based, which can be computationally intensive and non-differentiable, making direct parameter calibration difficult. While deep learning surrogate models can enhance prediction efficiency, their generalizability to different ecoregions and climate conditions remains limited. This paper introduces PyTorchFire, an open-source Python library built on PyTorch that harnesses GPU acceleration. By utilizing a newly designed differentiable wildfire Cellular Automata (CA) model, the system achieves computational efficiency at the millisecond scale, outperforming conventional CPU-based wildfire simulators when applied to high-resolution, real-world fire scenarios. More importantly, real-time parameter calibration is enabled through gradient descent, allowing simulations to closely align with observed wildfire dynamics both spatially and temporally, thereby improving the realism of the results. By integrating real-world environmental data, PyTorchFire demonstrates enhanced generalizability compared to traditional supervised learning surrogate models. Its ability to simulate and adjust wildfire behavior in real time ensures a high level of accuracy, stability, and efficiency. Numerical tests have been conducted using simplified data from real wildfire events in California, specifically the Pier Fire in 2017 and the Bear Fire in 2020.
How to cite: Cheng, S. and Xia, Z.: PyTorchFire: A Differentiable Cellular Automata-Based Wildfire Simulator with GPU Acceleration, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13425, https://doi.org/10.5194/egusphere-egu25-13425, 2025.
Posters virtual: Wed, 30 Apr, 14:00–15:45 | vPoster spot 3
EGU25-15531 | Posters virtual | VPS13
Spatio-temporal analysis of forest fires in CroatiaWed, 30 Apr, 14:00–15:45 (CEST) | vP3.5
Although the European Forest Fire Information System (EFFIS), provided by the Copernicus Emergency Management Service, offers three different methods for determining forest fire danger, the Canadian method is usually used and accepted in Croatia. The Canadian Fire Weather Index (FWI) estimates the forest fire danger level based on meteorological parameters (air temperature, humidity, wind speed and precipitation amount) related to 12 UTC for the given day at the meteorological station or to a grid point of a numerical weather prediction model.
Thanks to the EFFIS statistics portal, it is possible to see the extent to which Croatia has been at risk from forest fires in recent years based on the areas burned and the number of fires. The Copernicus Climate Change Service (C3S) provides a much more detailed overview of the burned areas. The combination of data from the Climate Change Service and the Emergency Management Service can provide a better overview of forest fires in Croatia. The forest fire danger levels are analyzed spatially between different regions such as the continental, mountainous and Adriatic parts of Croatia. In order to find an appropriate duration of the fire season, the forest fires within and outside the fire season are listed. The aim of the spatio-temporal analysis is to show the most endangered areas and the seasonal trend of forest fires in Croatia.
How to cite: Škurić Kuraži, D. and Herceg Bulić, I.: Spatio-temporal analysis of forest fires in Croatia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15531, https://doi.org/10.5194/egusphere-egu25-15531, 2025.
EGU25-11004 | Posters virtual | VPS13
Modeling Human-Caused Wildfire Ignition Probability Across EuropeWed, 30 Apr, 14:00–15:45 (CEST) | vP3.6
This communication presents a unified modeling framework for human-caused wildfire ignitions across representative European regions (pilot sites, PS), aiming to enhance understanding of ignition drivers and support wildfire risk management. Our approach models ignition probability at a fine spatial resolution (100 m), identifies key influencing factors, and enables cross-regional comparisons.
We calibrated Random Forest models using historical fire records and geospatial datasets, including land cover, accessibility, population density, and dead fine-fuel moisture content (DFMC). Models were developed individually for each PS and compared to a comprehensive model integrating all PS. Spatial autocorrelation effects on model performance were also evaluated.
Model performance was robust, with AUC values ranging from 0.70 to 0.89. DFMC anomaly emerged as the most influential variable across all PS. Among human-related factors, proximity to the Wildland-Urban Interface was most significant, followed by distance to roads, population density, and wildland coverage. The full model achieved an AUC of 0.81, highlighting mean DFMC and anomaly as dominant ignition drivers modulated by accessibility and population density. Local model performance, however, dropped by 0.10 AUC in regions such as Southern Sweden and Attica, Greece.
These findings underscore the importance of integrating fine-scale spatial and environmental data for wildfire ignition modeling. The developed models provide valuable insights into wildfire ignition hazards and support the implementation of targeted mitigation policies in fire-prone European landscapes.
How to cite: Gelabert Vadillo, P. J., Jiménez-Ruano, A., Ochoa, C., Alcasena, F., Sjöström, J., Marrs, C., Ribeiro, L. M., Palaiologou, P., Bentué-Martínez, C., Chuvieco, E., Vega-García, C., and Rodrigues, M.: Modeling Human-Caused Wildfire Ignition Probability Across Europe, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11004, https://doi.org/10.5194/egusphere-egu25-11004, 2025.
EGU25-16116 | ECS | Posters virtual | VPS13
Daily Data-Driven Emulation of the Fire Weather Index: Deep Learning Solutions for Wildfire Risk PredictionWed, 30 Apr, 14:00–15:45 (CEST) | vP3.7
Wildfires are an intensifying global challenge, driven by climate change, which increases their frequency, severity, and spatial extent. Accurate wildfire risk assessment and forecasting are essential for effective mitigation, resource allocation, and long-term planning. The Canadian Fire Weather Index (FWI) is a widely used fire danger rating system that integrates four primary daily meteorological variables—24-hour accumulated precipitation, wind speed, relative humidity, and temperature—into six components representing fuel moisture, ignition probability, and fire spread potential. Its temporal "memory" feature, which tracks moisture changes over time, makes it particularly valuable for capturing wildfire dynamics.
However, the FWI reliance on specific daily input data at noon poses challenges for its application in regions or scenarios lacking such precise temporal measurements. To address this limitation, FWI proxies computed using daily mean data offer a practical alternative. Yet, these proxies often lack the fidelity required to fully replicate the FWI values.
This study focuses on enhancing the emulation of the original FWI using daily mean data and other proxy variables by leveraging advanced deep learning techniques. We explore a spectrum of architectures, ranging from conventional machine learning models to state-of-the-art approaches like convolutional neural networks (CNNs) and Convolutional Long Short-Term Memory (ConvLSTM) networks. These models are tailored to capture the spatial and temporal complexities of wildfire behavior while maintaining robustness in the face of variable data availability.
Our research centers on the Iberian Peninsula, a Mediterranean region highly vulnerable to extreme wildfire events. By utilizing high-resolution, geo-referenced datasets, we validate the ability of these models to emulate the original FWI with high accuracy. To enhance model interpretability, we integrate eXplainable Artificial Intelligence (XAI) techniques, providing actionable insights into the decision-making processes and addressing concerns about the "black box" nature of deep learning.
This work demonstrates how daily data, combined with cutting-edge deep learning methods, can effectively emulate the FWI, offering a scalable and reliable solution for wildfire risk prediction in regions where traditional inputs are unavailable. The proposed models bridge the gap between limited data availability and the growing need for precise fire danger indices, enabling improved assessment and planning for wildfire-prone regions.
By advancing the science of wildfire modeling through daily data-driven approaches, this study contributes to a deeper understanding of spatial and temporal wildfire dynamics. It highlights the potential of integrating geoscience, climatology, and artificial intelligence to develop practical tools for wildfire risk mitigation, resilience, and decision-making in a rapidly changing climate.
Acknowledgments: This research work is part of R+D+i project CORDyS (PID2020-116595RB-I00) with funding from the Spanish Ministry of Science, Innovation and Universities MCIN/AEI/10.13039/501100011033. O.M. has received the research grant PRE2021-100292 funded by MCIN/AEI /10.13039/501100011033.
How to cite: Mirones, Ó., Baño-Media, J., and Bedia, J.: Daily Data-Driven Emulation of the Fire Weather Index: Deep Learning Solutions for Wildfire Risk Prediction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16116, https://doi.org/10.5194/egusphere-egu25-16116, 2025.
