Global burned area increasingly explained by climate change

doi:10.21203/rs.3.rs-3168150/v1

Download PDF

Article

Global burned area increasingly explained by climate change

https://doi.org/10.21203/rs.3.rs-3168150/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 21 Oct, 2024

Read the published version in Nature Climate Change →

Version 1

posted

You are reading this latest preprint version

Fires are now raging longer and more intensely in many regions worldwide. However, non-linear interactions between fire weather, fuel, land use, management, and ignitions so far impeded formal attribution of global burned area changes. Here we show that climate change is increasingly explaining regional burned area patterns, using an ensemble of global fire models. Climate change has increased global burned area by 16% for the period 2003–2019, and raised the probability of experiencing months with above-average global burned area by 43%. Climate change-induced burned area increased in most regions, including those recently struck by extreme fire like Australia, South America and Siberia. Moreover, the contribution of climate change to burned area is rising by 0.22%year^− 1 globally, with the largest increase in Central Australia (2.5%year^− 1). Our results highlight the importance of immediate, drastic and sustained greenhouse gas emission reductions to stabilize fire impacts on lives, livelihoods and ecosystems.

Earth and environmental sciences/Climate sciences/Climate change/Climate-change impacts

Earth and environmental sciences/Climate sciences/Climate change/Attribution

Global mean temperature has already increased 1.1°C above pre-industrial levels¹. Concurrently, there is an ongoing change in fire regimes^2,3. Over the last decade, more frequent and intense fires have been observed in regions including California, the Cerrado in Brazil⁴, and Siberia⁴. In contrast, other regions experienced a reduction in fire activity, leading to an overall decrease in global burned area⁵. Several regions where fires were previously rare are now seeing an increase in the numbers and extent of extreme fires. The Siberian heatwave in 2020, for example, caused extensive fires across the Arctic and was one of the first extreme events shown to be almost impossible without climate change⁶.

Quantifying the effect of climate change on historical burned area is crucial to understand how global fire regimes will change in the future. Future fire projections remain uncertain^4,7, with high confidence of increases in fire frequency limited to a few tropical forests and forest-savannah transition zones, the Arctic and boreal regions, and the Mediterranean^2,4,8.

While the recent Assessment Report by the Intergovernmental Panel on Climate Change (IPCC AR6) concludes that anthropogenic climate change has ‘likely increased fire weather in some regions of all inhabited continents (medium confidence)’⁹, other studies have shown a decline in global total burned area dominated by land use change in savanna regions⁵. Some studies have attributed recent fire changes in boreal, temperate and tropical ecosystems to climate-driven changes in fuel dryness and availability^2,8. However, climate change’s overall contribution to global fire regimes has not yet been quantified^2,8.

Attributing changes in burned area to human or natural causes is complicated because multiple contributing factors drive fires, including fire weather, atmospheric CO₂ concentrations, fuel, land use, land management, and ignitions, which all may have human and natural components. People, for example, can drive fires both by causing climate change and providing ignition sources. Climate change also affects precipitation patterns, vegetation growth and species distribution, which can lead to a reduction in fire, even if fire weather increases. Therefore, it is not enough to only consider changes in fire weather, and coupled fire-vegetation models are required for a complete picture of change. Previous attribution studies focusing on vegetation fires have often used fire danger indices^10–22, drought indices^22–24, or individual components of fire weather such as temperature and preciptaton^25,26 which misses changes in ignitions and fuel. Observation-based approaches to modelling fire and its controls have proven useful for explaining how recent fuel and ignition changes influence burning^3,27,28. However, as these approaches rely on observations of vegetation cover, they cannot model fire changes in a counterfactual world without climate change.

The IPCC AR6 Working Group II report proposes an 'impact attribution' framework to disentangle impact drivers^29–31. In complex processes such as fire, direct human influences may counter the effects of a change in climate. For example, climate change may increase burned area, but socio-economic changes reduce it, resulting in no observed change. Impact attribution can quantify the effect of climate change on systems where multiple contributing factors result in little observed change in the real world³⁰. Following this definition, only the impacts of climate change are attributed, not the causes.

In this study, we attribute historical changes in burned area to changes in observed climate (meteorological conditions and CO₂) and direct human forcings (ignitions and land-use change) individually, within this IPCC framework^29,32. Two transient historical simulations were performed with seven dynamic fire-vegetation models: one experiment using reanalysis climate and historical atmospheric CO₂ forcing (“Factual”), and another employing de-trended historical climate and atmospheric CO₂ fixed at 1901 levels (“Counterfactual”). By definition, the counterfactual baseline here cannot be observed, and relies on fire model simulations with a stationary climate while other relevant impact drivers evolve according to historical conditions³⁰. By comparing both fire model ensembles globally and across IPCC regions³³, we can attribute changes in burned area to climate change. In our analysis, we use the relative anomaly of burned area, defined as the monthly anomaly of the Factual from the Counterfactual, divided by the Counterfactual climatology (see ‘Relative Anomaly and Probability Ratio’ in Methods), and consider fires to be any vegetation fire.

Climate Change increases Burned Area

We first evaluate the distribution, trends, and variance of the seven fire models on the global and regional level (Extended Data Fig. 1). Similar to the assessment of the previous generation of fire models³⁴, we find substantial differences in regional performance of the models. Therefore, we apply a weighting using their benchmarking scores against observations, and bootstrap monthly total burned area relative anomalies to obtain a large weighted ensemble for each region (see ‘Ensemble Weighting’ in Methods). We use Probability Density Functions^35–38 to show the effect of climate change on burned area.

We find a statistically significant (Table S1) shift in the distribution to higher burned area in the factual experiment compared to the counterfactual (Fig. 1), and no difference between the two distributions in the early industrial period (1901–1920, Fig. S1). This shows that we already experience increased global burned area compared to a world without climate change. The shift in the mean (vertical lines, Fig. 1) indicates that we experienced 16% more burned area globally due to climate change over the 2003–2019 period.

Over the same period, we find an increase in mean burned area due to climate change in most regions (Fig. 1). We highlight four regions with a strong and statistically significant signal of change that also perform well against observations in our evaluation (see ‘Burned Area Distribution’ in Methods; Tables S5-S6). In these regions, mean burned area for the 2003–2019 period increases by 125% in Central Australia (CAU), 50% in South-Eastern South America (SES), 43% in West Siberia (WSB), and 33% in Western North America (WNA) due to climate change (Table S2).

We also quantify the effect socio-economic factors of land use and population density have on burned area. To this end, we compare the early industrial period (1901–1920) to the present (2003–2019) in the counterfactual simulations only, where the climate data have no long term trend, and hence the only structural change between the two time periods is due to socio-economic dynamics (land-use change, land management, and population density, which we refer to as Direct Human Forcing (DHF)). Burned area has decreased by 16% in the present day compared to the early industrial period due to DHF globally (Fig. 1), and in most of the IPCC regions (Fig. 1), which suggests that DHF has damped the effect of climate change on burned area.

Climate Change effects not mitigated by DHF at high levels

We then calculate the Probability Ratio (PR)^32–35 of higher than average burned area, where the ratio is between the probability of being above the counterfactual mean (area above the blue vertical line, Fig. 1) under (i) the factual forcing simulation (orange shading), and (ii) the counterfactual simulation (blue shading; see ‘Relative Anomaly and Probability Ratio’ in Methods). This allows us to quantify how much climate change has increased the chance of above-average burning.

Globally, we find a 43% increase in probability of above-average burned area due to climate change over the period 2003–2019 (Table S2). In the four selected regions, this probability increases by 56% in CAU, 37% in SES, 26% in WSB, and 16% in WNA.

In addition to the PR above the mean, we also calculate the PR above each percentile of the counterfactual burned area distribution from 20–90% due to climate change, DHF, and ‘all forcings’ combined, respectively. The PR increases across the distribution due to climate change (Fig. 2), meaning that the months with the highest global burned area increase the most in frequency. For example, the increase in probability of burned area being above the 85th percentile is approximately 2-fold (PR = 2, or a 100% increase), implying that global burned area occurring approximately twice per year in absence of climate change would now be expected around four times per year. Conversely, DHF reduces the probability of burned area by about 50%, although for the months with the highest burned area this effect is smaller, at around 25%. The net effect is that DHF counteracts the climate-driven increase in burned area at the lower end of the distribution, but in months with the largest burned area (> 80th percentile), climate change becomes the more important driver (Fig. 2). As the fire models do not perform well for the high tails of the distribution (see Extended Data Fig. 1), results above the 90th percentile should be considered with caution. However, our results suggest that these high fire months are the ones likely to be affected the most by climate change and least dampened by DHF (Fig. 2).

Climate Change-induced Burned Area on the Rise

We also analyse the trend over the historical period to understand how the impact of climate change on burned area is changing over time. We find a positive trend in climate change-induced burned area (“overburning”) in 38 out of the 43 IPCC regions (Figs. S2 and S3), with 14 regions showing an increase of over 0.5%year^− 1 for the 1980–2019 period (Table S2). This implies that the effect of climate change on burned area has increased over time in most regions. Currently, we find a global increase in overburning of 0.22 ± 0.03%year^− 1 (± 1 standard error) with respect to the counterfactual mean (Fig. S4). The largest increases occur in Central Australia (2.49 ± 0.46%year^− 1), the Tibetan Plateau (1.94 ± 0.34%year^− 1) and Southern South America (1.68 ± 0.23%year^− 1). In many regions, there is an increase in the rate at which these changes occur especially after the 1970s, with more recent periods undergoing faster increases in climate change-induced burned area (Figs. S2 and S3). Moreover, the overburning is highly correlated (Pearson r > 0.95) with the historic rise in atmospheric CO₂ concentrations in most regions (Fig. 3 and S2; Table S3). Globally, we find a correlation of 0.99. The correlations between overburning and (i) the change in temperature, (ii) the change in precipitation, and (iii) modelled vegetation carbon are high for most regions as well (Figs. S5 and S6; Table S3).

Using a weighted multi-model ensemble, we show that climate change caused an increase in global burned area over the historical period, and that months with the highest burned area are more likely. Many regions show a large (> 20%) increase, including all IPCC regions in Australia and many regions in South America, Siberia and North America.

We also find the relative contribution of climate change to a region’s total burning is increasing rapidly. In Western, Central and North-East North America, we show an increasing trend in the influence of climate change on burned area. This is in line with recent work³⁹ showing an increase in Californian summer fires due to anthropogenic climate change. We find this annual rate is also increasing over time, indicating of how trends could accelerate in the coming decades. Therefore, while DHF may have been mitigating the effects of climate change until now, our results indicates that fire regimes may be increasingly affected as the climate continues to change.

While we here show an increase in burned area due to climate change, several regions across the world have actually experienced a reduction in annual burned area⁴⁰. In line with previous research⁵, we find that socio-economic factors such as land-use and population have reduced climate change impact on burned area. This implies that present-day burned area would have been even higher without the mitigating influences of fire management and suppression, landscape fragmentation, and fuel reduction, which occurred through conversion of natural areas to urban, crop, pasture, and changing landscape management practices due to socio-economic development. However, the counteracting effect of DHF diminishes at higher burned areas, where climate change is having the largest influence (Fig. 2).

Our approach in this study allows us to account for model uncertainty, by utilising 7 different fire models (see ‘Fire Models’ in Methods). Individual model performance varies regionally, due to different representations of the land surface, human effects on fire and fire properties. To give the most robust results, we evaluate the models against observations in each region and subsequently weigh their contribution according to their performance in modelling observed trends. We highlight regions with a strong signal of change in burned area, and where the models perform well across all of our evaluation metrics.

Building on previous rounds of FireMIP, we use the latest versions of the state-of-the-art fire-enabled Dynamic Global Vegetation Models for this study. Evaluation of the previous FireMIP simulations⁴¹ suggested that direct climate change had minimal impact on the long-term global burned area, while increasing CO₂ increased burning. This is consistent with our results, where regions with the strongest increasing trends tend to be semi-arid, where CO₂ fertilisation has a stronger influence on vegetation productivity and, therefore, fuel load. Previous work also suggests that climate increases the variability of burned area⁴², consistent with the increasing spread that we find in global burned area distribution (Fig. 1).

By using coupled dynamic fire-vegetation models, we model the changes in burned area over time due to both changes in climate and fuel availability. Following the IPCC impact attribution framework, our approach also accounts for changes in land use and ignitions. While conducted with state-of-the-art global fire models, our simulations nevertheless remain characterised by a number of limitations. For example, the standard way of modelling anthropogenic ignitions based on population density makes it challenging to account for regional differences in fire management. In this study, the factual simulations were driven using one climate reanalysis dataset, and in future work this could be expanded to better sample observational uncertainty in the historical climate. Our approach to understanding the contribution of DHF to the change in burned area could be repeated using additional counterfactual experiments in ISIMIP, where socio-economic inputs are kept constant in a transient climate. Attributing changes in burned area is the first step in understanding the contribution of climate change to changes in fire impacts. Burned area is strongly correlated with fire emissions, and so our work provides a first indication of how emissions, air quality and health have been affected by climate change.

Our performance-weighted ensemble of global fire models paints a clear picture of competing drivers of fire occurrence, with climate change causing an increase in burned area globally. Further, it indicates that this increase is accelerating, which is consistent with reported increases in extreme climate-driven wildfire seasons around the globe. As decades of climate change are already baked in to our future⁴³, adapting to changes in fire regimes will be essential regardless of the degree of climate mitigation.

Fire has significant regional impacts on human lives and natural ecosystems as well as more far-reaching effects on atmospheric composition, albedo, hydrology, and carbon cycling, and those impacts can be expected to increase as the climate continues to warm. As for now, the effect of DHFs counters the effects of climate change on burned area in many regions of the world, leading to the observed stabilisation or slight decline in burned area in some areas. However, the strength of the climate change signal is increasing rapidly, suggesting that we will face stronger fire impacts on human well-being and ecosystem dynamics in the near-future. Therefore, both ambitious mitigation of climate change and adaptation to the specific impacts of increasing fire risk will likely be required in the future to manage the escalating impacts of climate change-driven fires on our natural and human systems.

1. Modelling Protocol

We follow the protocol from the Inter-Sectoral Impact Model Intercomparison Project phase 3a (ISIMIP3a), using reanalysis and de-trended climate data to attribute impacts to historical climate change³². We use seven fire models in this study, and each model performs a transient historical climate run with GSWP3-W5E5 reanalysis climate (“Factual”)³². ISIMIP3a also provides a detrended version of the GSWP3-W5E5 dataset where the long-term climate trend has been removed²⁹ that was used to drive the models in the counterclim experiment, along with constant atmospheric CO₂ and CH₄ at 1901 levels (“Counterfactual”)²⁹. We drive the models using 0.5° resolution daily climate (near-surface relative humidity, near-surface specific humidity, daily maximum temperature, daily mean temperature, daily minimum temperature, longwave downwelling radiation, near-surface wind magnitude, relative humidity, shortwave downwelling radiation, surface air pressure and total precipitation)^44–46. Lightning was prescribed as a monthly climatology from LIS/OTD data⁴⁷, and atmospheric CO₂ concentration was prescribed annually⁴⁸. Land use is derived from interpolating LUH2 and the Monfreda datasets⁴⁹, while population density, used by some models for ignitions and/or fire suppression, was provided annually⁵⁰.

Models were spun-up using constant CO₂ (284.73 ppm) and fixed land use (1850), before moving to the transition period between spin-up and start of the simulations (1850–1900) where constant CO₂, a transient climate and varying DHF were applied. Models provide their own test for equilibrium, as described in their documentation papers (see below).

2. Fire Models

In this study we use simulations from fire models taking part in the ISIMP3a fire sector, which is a continuation of the Fire Model Intercomparison Project (FireMIP)^51,52, plus VISIT which contributed simulations for ISIMIP3a's biome sector. This so far incorporates simulations from seven models, as follows:

CLASSIC

The Canadian Land Surface Scheme Including Biogeochemical Cycles (CLASSIC) simulates fluxes of energy, momentum, water, carbon, nitrogen, and fire⁵³. Land use is temporally varying where the fractional coverage of natural vegetation in the grid cell changes as the agricultural fractional coverage changes as derived from the Land-Use Harmonisation (LUH) project⁵⁴. The make-up of PFTs within the natural vegetation fraction is derived from the European Space Agency (ESA) Climate Change Initiative (CCI) land cover map⁵⁵. CLASSIC models both natural fires (caused by lightning) and anthropogenic fires (caused by human ignition) using fire ignition and extinguishing probabilities to calculate fire occurrence and area burned for each grid cell, assuming an elliptical shape for fire area. Ignition probabilities are based on above-ground biomass which is available as a fuel source, combustibility of the fuel which is based on soil moisture in the litter and vegetation root zone, and the presence of an ignition source (based on lightning strikes and population density). Extinguishing probability is based on population density.

JULES-INFERNO

The Earth System configuration of the Joint UK Land Environment Simulator (JULES-ES)⁵⁶ is coupled to the INFERNO fire model⁵⁷. JULES-ES includes dynamic vegetation from TRIFFID (Top-down Representation of Interactive Foliage and Flora Including Dynamics)⁵⁸ and incorporates vegetation changes from fire and land use change⁵⁹. Fire and background mortality varies by Plant Functional Type (PFT), of which there are nine natural PFTs (tropical and temperate broadleaf evergreen trees, broadleaf deciduous trees, broadleaf and evergreen needleleaf, broadleaf and evergreen shrubs and C3 and C4 grasses) and four crop and pasture (for C3 and C4) PFTs. Within each grid box, the fractional coverage of each PFT is determined by a competition hierarchy defined by TRIFFID⁵⁹, and is limited to just C3 and C4 crop/pasture in the prescribed agricultural regions.

LPJ-GUESS-SPITFIRE

The LPJ-GUESS dynamic global vegetation model^60,61 was coupled to the SPITFIRE global fire model⁶² to form LPJ-GUESS-SPITFIRE^51,63. The version of LPJ-GUESS used here has been updated to include nitrogen cycling and limitation on photosynthesis^51,61. Vegetation in LPJ-GUESS is represented by ten tree PFTs (for natural landcover), two grass PFTs (for natural and pasture landcover) and 5 crop PFTs (for cropland landcover). On natural areas, woody vegetation composition and structure is the result of competition for light and water between cohorts of identical tree PFTs individuals. Establishment, mortality and disturbance (including fires by simulated by SPITFIRE) occur stochastically, so replicate patches (here 20) are averaged to produce a grid cell mean. The SPITFIRE implementation simulates fires in pasture and natural land but no cropland burning. The version here also includes an increase of maximum fire duration to 12 hours complemented by a population density dependent reduction in fire duration⁶⁴. To account for landscape fragmentation by cropland and its resulting effect on fires in adjacent pastures and natural lands, fire size was scaled by the factor e^{− 4*gridcell_crop_fraction}.

LPJ-GUESS-SIMFIRE-BLAZE

The underlying dynamic vegetation model is the exact same version as described in LPJ-GUESS-SPITFIRE above. SIMFIRE-BLAZE is a two stage fire model implementation with SIMFIRE (SIMple FIRE model)⁶⁵ computing burned area based on human population density, maximum annual fAPAR, the maximum annual Nesterov index, and a set of fire biomes. Here, a climatology is used to derive an annual distribution of daily fire probability. The combustion model BLAZE (BLAZe induced biosphere-atmosphere flux Estimator)⁵¹ forms the second stage. In case of a successful stochastic ignition it computes fire-line intensity from available fuels and fire weather, the mortality of PFTs depending on fire-line intensity and allometry, and, eventually, the C and N turnover from the biomass pools (life and litter) to the atmosphere and between life and litter.

ORCHIDEE-MICT-SPITFIRE

ORCHIDEE is a process-based ecosystem model developed for simulating carbon, water and energy fluxes in ecosystems, from site level to global scale^66–68. ORCHIDEE-MICT is a version of ORCHIDEE with improved interactions between soil carbon, soil temperature and hydrology, and a fire module⁶⁹. For the ISIMIP3a simulations, a version based on ORCHIDEE-MICT with further detailed representation of grassland management^70,71 and northern peatland^72,73 was used. The fire module of ORCHIDEE-MICT is derived from the SPITFIRE model and was described in detailed by^74,75. The SPITFIRE fire module explicitly simulates open vegetation fires, including fuel load impacts on fire propagation and fire impacts on grassland and woodland dynamics. Fire and background mortality varies by Plant Functional Type (PFT). For the ISIMIP3a simulations, nine natural PFTs (tropical broadleaf evergreen forest, tropical broadleaf raingreen forest, temperate needleleaf evergreen forest, temperate broadleaf evergreen forest, temperate broadleaf summergreen forest, boreal needleleaf evergreen forest, boreal broadleaf summergreen forest, boreal needleleaf deciduous forest, and C3 and C4 natural grasses) and two grazing land PFTs (for C3 and C4 grazing land) can be burned, while cropland PFTs (C3 and C4 cropland) and norther peatland PFT (C3) were excluded from fire impacts.

SSiB4: SSiB4/TRIFFID-Fire consists of three components: a land surface model (Simplified Simple Biosphere Model; SSiB), a dynamic vegetation model (the Top-down Representation of Interactive Foliage and Flora Including Dynamics Model; TRIFFID), and a fire model of intermediate complexity, CLM-Li. The modelled plant functional types (PFTs) in SSiB4/TRIFFID include broadleaf evergreen trees, needleleaf evergreen trees, broadleaf deciduous trees, C3 grasses, C4 grasses, shrubs, and tundra shrubs. The vegetation competition in TRIFFID is based on the Lotka–Volterra equation and has been updated to represent the coexistence of grasses and shrubs. Fire impact on the terrestrial carbon cycle and vegetation fraction are updated every 10 days. SSiB4/TRIFFID-Fire is shown to reproduce the burned area and fire emissions globally from interannual to seasonal scales⁷⁶.

VISIT

The Vegetation Integrative Simulator for Trace gases is coupled to a semi-empirical fire model⁷⁷, mainly for estimating emissions from biomass burning (e.g., CO2, CO, CH4, and black carbon). Fuel accumulation and its moisture content are estimated by a carbon cycle and hydrological schemes of the VISIT. The fire component is briefly described in^78,79, where the original fire scheme is modified to simulate seasonal change in burned area by calculating fire season length at monthly time step. Burned area is calculated once a year (without ignition), and then weighted by (monthly fire season length) / (annual fire season length) at each month. A prescribed biome map (e.g., tropical rainforest and savanna) is used in the model, and therefore fire does not affect vegetation composition in a dynamic manner but influences biogeochemical processes such as net carbon budget. This is likely to affect seasonality and the simulation of fuel, and may explain the poorer benchmarking results compared to other models in this study. The VISIT results are only reported to the ISIMIP biome sector, but have been included in this study for comparison with other models reporting burned area.

3. Observations

We use two global burned area datasets (GFED5⁴⁰ and FireCCI5.1⁸⁰) as reference data for the analysis. Both datasets are MODIS-based but have fundamental differences in how they were developed. FireCCI maps fires at 250m resolution using the spectral information from MODIS in combination with the thermal anomalies⁸¹. However, it has been reported that global burned area products heavily underestimate the total magnitude of burned area⁸². GFED5 is based on the MODIS global burned area product MCD64A1, but the omission and commission errors were corrected by dynamic adjustment factors which were estimated based on Landsat or Sentinel-2 reference burned area maps⁴⁰. As such, global annual burned area in GFED5 is about double the burned area in FireCCI5.1 (Fig. S7). We use the burned area data from 2003 onwards, the first full year where both Terra and Aqua satellites are available⁸³.

4. Relative Anomaly and Probability Ratio

As shown in Figure S7, the spread in the absolute burned area is large amongst the observations (GFED5 has ~ 1.5 to 2 times more burned area than FireCCI5.1), models (350 to 750 Mha year^− 1), and regions (0 to 127 Mha year^− 1). Attributing changes in absolute burned area, therefore, has considerable uncertainty. We overcome this problem by focusing on relative changes in burned area, and use normalised relative anomaly (RA) rather than absolute burned area for our analysis. Using normalised relative anomalies also facilitates inter-regional comparison .

For the Probability Density Functions (PDFs) in Fig. 1, we calculate RA compared to the counterfactual mean burned area ($\overline{B{A}_{Cfact}}$). Subtracting the mean removes systematic biases, and dividing by the mean resolves the interannual variability. By comparing both factual and counterfactual experiments to the counterfactual mean, we are looking at the fractional increase in burned area driven by climate change compared to a baseline without climate change.

$${{R}{A}}_{{B}{A}}= \frac{{B}{A}- \stackrel{-}{{{B}{A}}_{{C}.{f}{a}{c}{t}}}}{ \stackrel{-}{{{B}{A}}_{{C}.{f}{a}{c}{t}}}}$$

Equation 1: Burned Area Relative Anomaly.

Where BA is the time series of the monthly (global / regional) total burned area for factual (orange in Fig. 1) or counterfactual (blue in Fig. 1).

For evaluation in Extended Data Fig. 1, we compare RA for factual simulations and observations, with anomalies calculated from their respective means ($\stackrel{-}{{\text{B}\text{A}}_{fact}}$ & $\stackrel{-}{{\text{B}\text{A}}_{obs}}$).

$${{R}{A}}_{{B}{A}}= \frac{{B}{A}- \stackrel{-}{{B}{A}}}{ \stackrel{-}{{B}{A}}}$$

Equation 2: Burned Area Relative Anomaly for the validation.

When constructing time series in Fig. 3, we are interested in the difference in burned area between the factual and counterfactual experiments through time. Interannual burned area is variable^84,85 and the counterfactual simulations have a negative trend in most regions. Therefore, we calculate this difference over a 21-year running mean centred on year Y, and treat the counterfactual as our baseline burned area for normalisation. By subtracting a mean, we reduce the effect of interannual variability and by choosing for a centred 21-year window mean, as opposed to the entire time series mean, we remove most of its long-term trend:

$${{R}{A}}_{{B}{A}}= \frac{\stackrel{-}{{{B}{A}}_{{f}{a}{c}{t}21{y}{e}{a}{r}}}- \stackrel{-}{{{B}{A}}_{{C}.{f}{a}{c}{t}21{y}{e}{a}{r}}}}{ \stackrel{-}{{{B}{A}}_{{C}.{f}{a}{c}{t}21{y}{e}{a}{r}}}}$$

Equation 3: Moving Window Relative Anomaly.

Probability Ratio (PR) is the ratio of the amount of months, over the same time period, the factual and counterfactual simulations are above the counterfactual monthly mean burned area.

$${P}{R}= \frac{\sum {({B}{A}}_{{f}{a}{c}{t}}>\stackrel{-}{{{B}{A}}_{{C}.{f}{a}{c}{t}}})}{\sum {({B}{A}}_{{C}.{f}{a}{c}{t}}>\stackrel{-}{{{B}{A}}_{{C}.{f}{a}{c}{t}}})}$$

Equation 4: Probability Ratio.

5. Evaluation Methods

We evaluated the factual burned area simulations compared to observations for the seven fire models to assess the ensemble's ability to represent changes in burned area in each region. We constrain the model and observation data to 2003–2019, the common period of simulations and observations.

For general evaluation, we used the Normalised Mean Error (NME) metric used in previous FireMIP benchmarking^51,52,86 for spatial comparisons⁸⁶ to quantify model performance in simulating spatial patterns of averaged burned area (Table S4), i.e. taking the mean over the temporal domain. NME₁ calculates the area-weighted absolute mean difference between observations and simulations, normalised by the mean variation in the observations. We also report NME Step 3 (NME₃) from the FireMIP benchmarking framework. FireMIP recommends using NME₁ and NME₃ as they are appropriate for comparing non-normal variables such as burned area. This approach removes mean bias and absolute variance before comparison and assesses the model’s ability to simulate regions of high and low burned area, and is therefore appropriate here as we are interested in relative anomalies.

We also add a ‘temporal NME’ metric (NME_t), where we take the sum over the spatial domain (per region) and calculate the NME on the resulting time series. This allows us to quantify the model’s performance in representing the observed regional variations over time, which we assume indicates a model’s ability to represent the influence of changing climate conditions on burned area. We asses each of the IPCC regions separately to match reported results and to inform model weighing (see below).

The lower the score for NME’s, the closer the match between observation and simulation. For the evaluation, we also use a randomly-resampled null model, which compares a dataset generated by sampling the observations without replacement, with the observations⁵⁹. As the resultant randomly-resampled score depends on sampling order, we repeat this procedure 1000 times to generate a "randomly-resampled" distribution. A model is better/worse than the randomly-resampled if its score is less than/greater than the mean +/- a standard deviation of this distribution⁸⁶. We use the scores in conjunction with burned area maps to give context to the performance scores (Figure S8).

Following⁸⁷, we also assess each model’s global and regional total burned area distribution compared to the two observational datasets using probability distribution and quantile-quantile analysis. This allows us to assess which parts of burned area temporal distribution each model can reproduce i.e., do models capture low, average, or extreme burned areas.

6. Ensemble Weighting

Analysis of the previous generation of this fire model ensemble³⁴, and the analysis shown in this paper highlights significant model performance differences among different regions. Therefore, we apply a region-specific weighting for each of the IPCC regions, based on model performance in that region. We apply a model weighting based on the distance of the modelled to the observed burned area temporal RA^88,89. We apply weights according to two aspects of temporal performance, which we assess for each observation (FireCCI5.1 and GFED5): (i) Do models capture the correct overall temporal variability in burned area? (ii) Do models capture year-by-year variations in burned areas? While land use and socio-economic factors can dominate long term trends in burned area^3,5, year-by-year variations are largely driven by climate and fuel dynamics⁹⁰. Therefore, we assume inter-annual performance is an indicator of a model’s ability to simulate climate impacts on burned area.

For both criteria, we perform an area-weighted sum of the pixels per IPCC region to get the regional total time series for the observations and the factual and counterfactual simulations. Then, we calculate the relative anomaly of the factual against counterfactual means, as per Eq. (1). For the first criteria, we perform these steps on annual total burned area NME_t against each observation in turn. For the second, we perform this on ranked monthly burned areas, and again use NME to compare against both observations.

This results in four NME scores (two criteria, two observations) for each model in each IPCC region (Tables S5-S6). We then sum these four NME scores to get a measure of a combined distance score (D_i). We apply D_i to Eq. 5, originally from⁸⁸, but without the denominator of the formula which deals with model (in)dependence and is relevant for large ensembles with e.g., multiple simulations of the same model. Lastly, we normalize each score per region.

$${{W}}_{{i}}= {{e}}^{-\frac{{{D}}_{{i}}}{{{\sigma }}_{{D}}}}$$

Equation 5: Model weight calculation. The weight for model i (W _i ) is calculated through the sum of its distances to the observations (D _i ) and the shape parameter σ _D. Here, the sum of the four NMEs (Tables S5-S6) are used as sum of distances, and σ_Dwas set at 0.5, adapted from⁸⁸.

There is no objective best method to determine a value for σ_d. However, a large σ_d corresponds to model democracy, whereas a small σ_d focuses all the weight on the few best model(s). Therefore, we apply a σ_d of 0.5 to calculate one weight per model per region, as this represents a general balance between democracy and performance⁸⁹.

We use this weight (Table S7) to bootstrap (with replacement) the monthly regional burned area RA 2003–2019 time series, where the 100,000 bootstrapped samples are assigned according to the weight for each model (number of bootstraps = 100,000 x weight). The weighted factual and counterfactual ensemble is then used to calculate the change in PDFs and the Probability Ratio. Bootstrapping the models using these weights implies that we sample the best performing models more times than models that perform poorly, adding confidence to our results.

7. Evaluation Results

Evaluation shows that no individual model captures the observations perfectly, but that at least some models in the ensemble capture the main spatial patterns and relative temporal anomalies in most IPCC regions. Most regional trends are appropriately modelled by multiple models, regions with the poorest performance are the Arabian Peninsula (ARP), West and Central Europe (WCE), South-West South-America (SWS), SAU (South Australia), and New Zealand (NZ).

7.1 General Benchmarking

Our metric scores indicate that the models reproduce the overall pattern of burned area, and perform consistently against each other globally for all products (Table S4), except VISIT. This indicates that the simulated biases in VISIT compared to biases in burned area observations are greater than the observed spatial variance. Considering the spatial distribution, VISIT simulates the highest burned area in Australia and minimal in Africa or South America (Fig. S8). SSiB4 more accurately captures the position and magnitude of Africa’s Sahelian (Sahara (SAH), northern Western-Africa (WAF), Central-Africa (CAF) and North Eastern-Africa (NEAF)) and Miomboian (Northern West Southern-Africa (WSAF) and East Southern-Africa (EASF)) fire regions (Fig. S4, NME₃ of 1.06–1.14). Meanwhile, CLASSIC reproduces the magnitude of these fire regions but over too extensive an area, simulating too much burning in some parts of Africa with little fire in the observations. JULES Sahelian fires are reproduced but are broader than in observations and shifted slightly northward, away from WAF, CAF and NEAF and into SAH. The two LPJ-GUESS models simulate too low Sahelian burning. ORCHIDEE captures the Sahel, but Miomboian burning extends too far into WSAF and EASF regions. However, importantly for this study, all models estimate the relative high and low fire regions fairly well, regardless of absolute magnitude, as indicated by lower NME₃ scores (Tables S5-S6). JULES, CLASSIC and the two LPJ models simulate more burning in the Cerrado and Caatinga (North-East South-America (NES) and South-East South-America (SES)) than in observations, while SSiB4, ORCHIDEE and VISIT. JULES, LPJ and (to a lesser extent) CLASSIC capture fires around the Amazon arc of deforestation (South-American Monsoon (SAM) and the southern part of Northern South-America (NSA)). CLASSIC has too much burning in South American Atlantic Forests (coastal NES). LPJ-GUESS-SIMFIRE-BLAZE captures the Northern Australian (NAU) burned area, which is underestimated but present in JULES and LPJ-GUESS-SPITFIRE. SSIB4 and CLASSIC burn little in NAU and have too much burning in Southeastern Australia (East Australia (EAU) and South Australia (SAU)), extending too far into inland Mallee and Acacia Shrublands (towards Central Australia (CAU)). CLASSIC, LPJ-GUESS-SIMFIRE-BLAZE, and SSIB4 simulate the observed burning in the Kazakh (East Europe (EEU) < West Siberia (WSB) and northern West Central Asia (WCA)) and Mongolian-Manchurian Steppes (North East Asia (EAS) and into Russian-Far-East (RFE)) in Central and Eastern Asia, though not at the observed magnitude. All models tend to burn more than observed in North America, though the exact locations are model dependent.

7.2 Temporal Variability

We evaluate the temporal NME to assess the capability of our models in representing the observed temporal patterns in burned area. An NME of 1 indicates that our models are, on average, a factor of 1 away from the observed burned area. For example, if the observed burned area relative anomaly is 0.7, then a modelled burned area relative anomaly of 1.4 would result in an NME of 1 ($abs(0.7-1.4)/abs\left(0.7\right)$), a modelled relative anomaly of 0 would also result in an NME of 1, and a relative anomaly of 2.1 (or -0.7) would result in an NME of 2. Here, we find that SSiB4 performs best globally, and ORCHIDEE-MICT-SPITFIRE has the highest NME (Extended Data Table 1). However, the global trend is dominated by the African savannahs, where ORCHIDEE-MICT-SPITFIRE has a poor(er) performance. The full list of regional NME scores are provided in Supplementary Tables S5 and S6. We find that the models tend to have low ranked monthly NMEs in the Boreal regions, indicating they capture the seasonality of fires in those regions relatively well. The four selected regions of interest (CAU, SES, WSB and WNA) are all modelled well (NME < 1) by at least multiple models in each of those regions, this holds for the majority of the 43 IPCC regions. With ARP, SAU, NZ, WCA and WCE and being the regions that have the poorest general performance. However, some of these regions undergo very little burning and their high NME scores are partially the result of our choice to focus on the relative anomaly.

7.3 Burned Area Distribution

Globally, the models tend to have a narrower and higher distributions than the observations (Fig. S9), especially in JULES, although the LPJ models and SSiB4 match the observations more closely. There is high regional variation in the distribution of burned area (Fig. S8), and the global trend is dominated by Africa’s Sahelian and Miomboian high burned area fire regions. Regionally, VISIT stands out of the distribution in most regions (Fig. S9). Excluding VISIT (Fig. S10), there are many regions where the models simulate the observed distribution well, including our four selected regions (Extended Data Fig. 1, top row).

Figure S11 shows the quantiles of the distribution for simulated and observed burned area RA. Globally, the models simulate the observed distribution well, especially for the lower and middle ranges of the distribution. At the very high end of the distribution, all fire models simulate too little burned area compared to observations (below the 1:1 line). Regionally, this is even more evident, where the models simulate much lower burned area than observed generally across most regions at the top end of the distribution. This reflects how fire models are designed to capture the mean burned area globally, therefore do not capture the stochastic nature of burned area and extremes outside of the normal distribution⁵². As a result, we focus our attribution analysis on the change in mean burned area rather than the change in large fires which the models cannot simulate well.

In our selected regions, the burned area at the lower and middle end of the distribution is well simulated compared to observations (Extended Data Fig. 1, middle row). In SES, all the models do well across the whole distribution. In other regions, especially CAU, models underestimate the observed burned area, with VISIT tending to underestimate burned area RA the most compared to observations. CLASSIC performs well in WSB, and LPJ-GUESS-SIMFIRE-BLAZE performs well in WNA. ORCHIDEE-MICT-SPITFIRE performs very well in West North-America (WNA) compared to FireCCI (solid lines) but over predicts burned area compared to GFED5.

We also assess the variability of the models against observations with a power spectra analysis (Fig. S12). Globally, the observations are within the model range, although at the upper end. Regionally, there is generally good agreement between the models and observations. In the four selected regions, the observations mostly lie in the middle to upper end of the model spread (Extended Data Fig. 1, bottom row).

Code availability

Scripts for the pre-processing and analysis are available at the GitHub repository.

Source data

Model and model input data is available from the ISIMIP data repository.

The observational FireCCI5.1 and GFED5 data are openly available.

Author contributions

*CB and SL contributed equally to the analysis and writing. CB, SL, DK, WT, SH, NC, LG, MF designed the analysis. EB, JC, HH, AI, SKG, WL, LN ran the model simulations and contributed data. YC and JR contributed observational data. CB, SH, FL, MF, GL, CR, co-ordinated the fire sector and simulations. All authors contributed to the final manuscript.

Acknowledgments and funding

CB was supported by the Met Office Hadley Centre Climate Programme funded by BEIS and the Newton Fund through the Met Office Climate Science for Service Partnership Brazil (CSSP Brazil). SL was supported by a PhD Fundamental Research Grant by Fonds Wetenschappelijk Onderzoek - Vlaanderen (11M7723N). Part of the resources and services used in this work were provided by the VSC (Flemish Supercomputer Center), funded by the Research Foundation - Flanders (FWO) and the Flemish Government. DIK was supported by the Natural Environment Research Council as part of the LTSM2 TerraFIRMA project. HH is supported by United States Department of Energy, Office of Science (Lab Directed Res & Dev (LDRD) 29IN290162:80941). The Pacific Northwest National Laboratory (PNNL) is operated for DOE by the Battelle Memorial Institute under contract DE-AC05-76RLO1830. The computations were enabled by resources provided by the Swedish National Infrastructure for Computing (SNIC) at LUNARC partially funded by the Swedish Research Council through grant agreement no. 2016-07213. FL is supported by the National Key R&D Program of China (2022YFE0106500). WL is supported by the National Key R&D Program of China (grant number: 2019YFA0606604). CPOR acknowledges funding from the EU Horizon 2020 research and innovation program under grant agreement 821010 (CASCADES). WT acknowledges funding from the EU Horizon 2020 research and innovation program (SPARCCLE). SH acknowledges support from the Max Planck Tandem group program and from Universidad del Rosario within the program of Fondos de arranque.

Gulev, S. K. et al. Changing State of the Climate System. in Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (eds. Masson-Delmotte, V. et al.) (Cambridge University Press, 2021). doi:10.1017/9781009157896.004.
Jones, M. W. et al. Global and Regional Trends and Drivers of Fire Under Climate Change. Reviews of Geophysics 60, (2022).
Kelley, D. I. et al. How contemporary bioclimatic and human controls change global fire regimes. Nature Climate Change vol. 9 Preprint at https://doi.org/10.1038/s41558-019-0540-7 (2019).
UNEP. Spreading like wildfire: the rising threat of extraordinary landscape fires. A UNEP Rapid Response Assessment vol. 294 (2022).
Andela, N. et al. A human-driven decline in global burned area. Science (1979) 356, 1356–1362 (2017).
Ciavarella, A. et al. Prolonged Siberian heat of 2020 almost impossible without human influence. Clim Change 166, (2021).
Kloster, S. & Lasslop, G. Historical and future fire occurrence (1850 to 2100) simulated in CMIP5 Earth System Models. Glob Planet Change 150, (2017).
Canadell, J. G. et al. Global carbon and other biogeochemical cycles and feedbacks. in Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (2021).
Ranasinghe, R. et al. Climate Change Information for Regional Impact and for Risk Assessment. in Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (eds. Masson-Delmotte, V. et al.) (Cambridge University Press, 2021). doi:10.1017/9781009157896.014.
Goss, M. et al. Climate change is increasing the likelihood of extreme autumn wildfire conditions across California. Environmental Research Letters 15, (2020).
Touma, D., Stevenson, S., Lehner, F. & Coats, S. Human-driven greenhouse gas and aerosol emissions cause distinct regional impacts on extreme fire weather. Nat Commun 12, (2021).
Abatzoglou, J. T., Williams, A. P. & Barbero, R. Global Emergence of Anthropogenic Climate Change in Fire Weather Indices. Geophys Res Lett 46, (2019).
Abatzoglou, J. T. & Williams, A. P. Impact of anthropogenic climate change on wildfire across western US forests. Proc Natl Acad Sci U S A 113, (2016).
Du, J., Wang, K. & Cui, B. Attribution of the extreme drought-related risk of wildfires in spring 2019 over Southwest China. Bull Am Meteorol Soc 102, (2021).
Krikken, F., Lehner, F., Haustein, K., Drobyshev, I. & van Oldenborgh, G. J. Attribution of the role of climate change in the forest fires in Sweden 2018. Natural Hazards and Earth System Sciences 21, (2021).
Jan Van Oldenborgh, G. et al. Attribution of the Australian bushfire risk to anthropogenic climate change. Natural Hazards and Earth System Sciences 21, (2021).
Kirchmeier-Young, M. C., Zwiers, F. W., Gillett, N. P. & Cannon, A. J. Attributing extreme fire risk in Western Canada to human emissions. Clim Change 144, (2017).
Kirchmeier-Young, M. C., Gillett, N. P., Zwiers, F. W., Cannon, A. J. & Anslow, F. S. Attribution of the Influence of Human-Induced Climate Change on an Extreme Fire Season. Earths Future 7, (2019).
Williams, A. P. et al. Observed Impacts of Anthropogenic Climate Change on Wildfire in California. Earths Future 7, (2019).
Lewis, S. C. et al. Deconstructing factors contributing to the 2018 fire weather in Queensland, Australia. Bull Am Meteorol Soc 101, (2020).
Hope, P. et al. On determining the impact of increasing atmospheric CO 2 on the record fire weather in eastern Australia in February 2017. Bull Am Meteorol Soc 100, (2019).
Yoon, J. H. et al. Extreme fire season in California: A glimpse into the future? Bulletin of the American Meteorological Society vol. 96 Preprint at https://doi.org/10.1175/BAMS-D-15-00114.1 (2015).
Partain, J. L. et al. 4. An assessment of the role of anthropogenic climate change in the Alaska fire season of 2015. Bull Am Meteorol Soc 97, (2016).
Brown, T., Leach, S., Wachter, B. & Gardunio, B. The Northern California 2018 extreme fire season. Bull Am Meteorol Soc 101, (2020).
Gillett, N. P., Weaver, A. J., Zwiers, F. W. & Flannigan, M. D. Detecting the effect of climate change on Canadian forest fires. Geophys Res Lett 31, (2004).
Tett, S. F. B. et al. 12. Anthropogenic forcings and associated changes in fire risk in western North America and Australia during 2015/16. Bull Am Meteorol Soc 99, (2018).
Kelley, D. I. et al. Technical note: Low meteorological influence found in 2019 Amazonia fires. Biogeosciences 18, 787–804 (2021).
Moritz, M. A. et al. Climate change and disruptions to global fire activity. Ecosphere 3, art49 (2012).
Mengel, M., Treu, S., Lange, S. & Frieler, K. ATTRICI v1.1 - Counterfactual climate for impact attribution. Geosci Model Dev 14, (2021).
O’Neill, B. et al. Key Risks Across Sectors and Regions. in Climate Change 2022: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (eds. Pörtner, H. O. et al.) (Cambridge University Press, 2022). doi:10.1017/9781009325844.025.
Ara Begum, R. et al. Point of Departure and Key Concepts. in Climate Change 2022: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (eds. Pörtner, H. O. et al.) (Cambridge University Press, 2022). doi:10.1017/9781009325844.003.
Frieler, K. et al. Scenario set-up and forcing data for impact model evaluation and impact attribution within the third round of the Inter-Sectoral Model Intercomparison Project (ISIMIP3a). EGUsphere 2023, 1–83 (2023).
Iturbide, M. et al. An update of IPCC climate reference regions for subcontinental analysis of climate model data: definition and aggregated datasets. Earth Syst Sci Data 12, 2959–2970 (2020).
Hantson, S. et al. Quantitative assessment of fire and vegetation properties in simulations with fire-enabled vegetation models from the Fire Model Intercomparison Project. Geosci Model Dev 13, (2020).
Stott, P. A., Stone, D. A. & Allen, M. R. Human contribution to the European heatwave of 2003. Nature 432, (2004).
Stott, P. A. Attribution of regional-scale temperature changes to anthropogenic and natural causes. Geophys Res Lett 30, (2003).
Stott, P. A. et al. Attribution of extreme weather and climate-related events. Wiley Interdiscip Rev Clim Change 7, (2016).
Allen, M. Liability for climate change. Nature vol. 421 Preprint at https://doi.org/10.1038/421891a (2003).
Turco, M. et al. Anthropogenic climate change impacts exacerbate summer forest fires in California. Proceedings of the National Academy of Sciences 120, e2213815120 (2023).
Chen, Y. et al. Multi-decadal trends and variability in burned area from the 5th version of the Global Fire Emissions Database (GFED5). Earth System Science Data Discussions 2023, 1–52 (2023).
Teckentrup, L. et al. Response of simulated burned area to historical changes in environmental and anthropogenic factors: a comparison of seven fire models. Biogeosciences 16, 3883–3910 (2019).
Teckentrup, L. et al. Response of simulated burned area to historical changes in environmental and anthropogenic factors: a comparison of seven fire models. Biogeosciences 16, 3883–3910 (2019).
Lee, J.-Y. et al. Future Global Climate: Scenario-Based Projections and Near-Term Information. in Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (eds. Masson-Delmotte, V. et al.) (Cambridge University Press, 2021). doi:10.1017/9781009157896.006.
Lange, S. et al. WFDE5 over land merged with ERA5 over the ocean (W5E5 v2.0). Preprint at https://doi.org/10.48364/ISIMIP.342217 (2021).
Kim, H. Global soil wetness project phase 3 atmospheric boundary conditions (experiment 1). (2017).
Lange, S. ISIMIP3BASD. Preprint at https://doi.org/10.5281/zenodo.4686991 (2021).
Cecil, D. J. LIS/OTD 0.5 Degree High Resolution Monthly Climatology (HRMC) V2.3.2015. Preprint at (2006).
Büchner, M. & Reyer, C. ISIMIP3a atmospheric composition input data. Preprint at https://doi.org/10.48364/ISIMIP.664235.2 (2022).
Volkholz, J. & Ostberg, S. ISIMIP3a landuse input data. Preprint at https://doi.org/10.48364/ISIMIP.571261.2 (2022).
Volkholz, J., Lange, S. & Geiger, T. ISIMIP3a population input data. Preprint at https://doi.org/10.48364/ISIMIP.822480.2 (2022).
Rabin, S. S. et al. The Fire Modeling Intercomparison Project (FireMIP), phase 1: Experimental and analytical protocols with detailed model descriptions. Geosci Model Dev 10, (2017).
Hantson, S. et al. The status and challenge of global fire modelling. Biogeosciences 13, 3359–3375 (2016).
Melton, J. R. et al. CLASSIC v1.0: the open-source community successor to the Canadian Land Surface Scheme (CLASS) and the Canadian Terrestrial Ecosystem Model (CTEM) – Part 1: Model framework and site-level performance. Geosci Model Dev 13, 2825–2850 (2020).
Chini, L. et al. Land-use harmonization datasets for annual global carbon budgets. Earth Syst Sci Data 13, 4175–4189 (2021).
Li, W. et al. Gross and net land cover changes in the main plant functional types derived from the annual ESA CCI land cover maps (1992–2015). Earth Syst Sci Data 10, 219–234 (2018).
Mathison, C. et al. Description and Evaluation of the JULES-ES setup for ISIMIP2b. EGUsphere 2022, 1–24 (2022).
Mangeon, S. et al. INFERNO: A fire and emissions scheme for the UK Met Office’s Unified Model. Geosci Model Dev 9, (2016).
Cox, P. Description of the ‘TRIFFID’ Dynamic Global Vegetation Model. (2001).
Burton, C. et al. Representation of fire, land-use change and vegetation dynamics in the Joint UK Land Environment Simulator vn4.9 (JULES). Geosci Model Dev 12, (2019).
Smith, B., Prentice, I. C. & Sykes, M. T. Representation of Vegetation Dynamics in the Modelling of Terrestrial Ecosystems: Comparing Two Contrasting Approaches within European Climate Space. Global Ecology and Biogeography 10, 621–637 (2001).
Smith, B. et al. Implications of incorporating N cycling and N limitations on primary production in an individual-based dynamic vegetation model. Biogeosciences 11, 2027–2054 (2014).
Thonicke, K. et al. The influence of vegetation, fire spread and fire behaviour on biomass burning and trace gas emissions: results from a process-based model. Biogeosciences 7, 1991–2011 (2010).
Lehsten, V. et al. Estimating carbon emissions from African wildfires. Biogeosciences 6, 349–360 (2009).
Hantson, S., Lasslop, G., Kloster, S. & Chuvieco, E. Anthropogenic effects on global mean fire size. Int J Wildland Fire 24, 589–596 (2015).
Knorr, W., Kaminski, T., Arneth, A. & Weber, U. Impact of human population density on fire frequency at the global scale. Biogeosciences 11, 1085–1102 (2014).
Krinner, G. et al. A dynamic global vegetation model for studies of the coupled atmosphere-biosphere system. Global Biogeochem Cycles 19, (2005).
Piao, S., Friedlingstein, P., Ciais, P., Viovy, N. & Demarty, J. Growing season extension and its impact on terrestrial carbon cycle in the Northern Hemisphere over the past 2 decades. Global Biogeochem Cycles 21, (2007).
Ciais, P. et al. Europe-wide reduction in primary productivity caused by the heat and drought in 2003. Nature 437, 529–533 (2005).
Guimberteau, M. et al. ORCHIDEE-MICT (v8.4.1), a land surface model for the high latitudes: model description and validation. Geosci Model Dev 11, 121–163 (2018).
Chang, J. F. et al. Incorporating grassland management in ORCHIDEE: model description and evaluation at 11 eddy-covariance sites in Europe. Geosci Model Dev 6, 2165–2181 (2013).
Chang, J. et al. Climate warming from managed grasslands cancels the cooling effect of carbon sinks in sparsely grazed and natural grasslands. Nat Commun 12, 118 (2021).
Qiu, C. et al. A strong mitigation scenario maintains climate neutrality of northern peatlands. One Earth 5, 86–97 (2022).
Qiu, C. et al. ORCHIDEE-PEAT (revision 4596), a model for northern peatland CO_2, water, and energy fluxes on daily to annual scales. Geosci Model Dev 11, 497–519 (2018).
Yue, C. et al. Modelling the role of fires in the terrestrial carbon balance by incorporating SPITFIRE into the global vegetation model ORCHIDEE – Part 1: simulating historical global burned area and fire regimes. Geosci Model Dev 7, 2747–2767 (2014).
Yue, C., Ciais, P., Cadule, P., Thonicke, K. & van Leeuwen, T. T. Modelling the role of fires in the terrestrial carbon balance by incorporating SPITFIRE into the global vegetation model ORCHIDEE – Part 2: Carbon emissions and the role of fires in the global carbon balance. Geosci Model Dev 8, 1321–1338 (2015).
Huang, H., Xue, Y., Li, F. & Liu, Y. Modeling long-term fire impact on ecosystem characteristics and surface energy using a process-based vegetation–fire model SSiB4/TRIFFID-Fire v1.0. Geosci Model Dev 13, 6029–6050 (2020).
Thonicke, K., Venevsky, S., Sitch, S. & Cramer, W. The role of fire disturbance for global vegetation dynamics: coupling fire into a Dynamic Global Vegetation Model. Global Ecology and Biogeography 10, 661–677 (2001).
Kato, E., Kinoshita, T., Ito, A., Kawamiya, M. & Yamagata, Y. Evaluation of spatially explicit emission scenario of land-use change and biomass burning using a process-based biogeochemical model. J Land Use Sci 8, 104–122 (2013).
Ito, A. Disequilibrium of terrestrial ecosystem CO_2 budget caused by disturbance-induced emissions and non-CO_2 carbon export flows: a global model assessment. Earth System Dynamics 10, 685–709 (2019).
Chuvieco, E., Pettinari, M. L., Lizundia Loiola, J., Storm, T. & Padilla Parellada, M. ESA Fire Climate Change Initiative (Fire_cci): MODIS Fire_cci Burned Area Grid product, version 5.1. Centre for Environmental Data Analysis Preprint at (2019).
Lizundia-Loiola, J., Otón, G., Ramo, R. & Chuvieco, E. A spatio-temporal active-fire clustering approach for global burned area mapping at 250 m from MODIS data. Remote Sens Environ 236, 111493 (2020).
Roteta, E., Bastarrika, A., Padilla, M., Storm, T. & Chuvieco, E. Development of a Sentinel-2 burned area algorithm: Generation of a small fire database for sub-Saharan Africa. Remote Sens Environ 222, 1–17 (2019).
Barnes, W. L., Xiong, X. & Salomonson, V. V. Status of terra MODIS and aqua modis. Advances in Space Research 32, 2099–2106 (2003).
van der Werf, G. R. et al. Interannual variability in global biomass burning emissions from 1997 to 2004. Atmos Chem Phys 6, 3423–3441 (2006).
Tang, R. et al. Interannual variability and climatic sensitivity of global wildfire activity. Advances in Climate Change Research 12, 686–695 (2021).
Kelley, D. I. et al. A comprehensive benchmarking system for evaluating global vegetation models. Biogeosciences 10, (2013).
Christidis, N., McCarthy, M. & Stott, P. A. The increasing likelihood of temperatures above 30 to 40 °C in the United Kingdom. Nat Commun 11, (2020).
Brunner, L., Lorenz, R., Zumwald, M. & Knutti, R. Quantifying uncertainty in European climate projections using combined performance-independence weighting. Environmental Research Letters 14, (2019).
Knutti, R. et al. A climate model projection weighting scheme accounting for performance and interdependence. Geophys Res Lett 44, 1909–1918 (2017).
Abatzoglou, J. T., Williams, A. P., Boschetti, L., Zubkova, M. & Kolden, C. A. Global patterns of interannual climate–fire relationships. Glob Chang Biol 24, 5164–5175 (2018).

There is NO Competing Interest.

Download PDF

Journal Publication

published 21 Oct, 2024

Read the published version in Nature Climate Change →

Version 1

posted

You are reading this latest preprint version

Global burned area increasingly explained by climate change

Status:

Journal Publication

Version 1

Abstract

Figures

Main

Results

Climate Change increases Burned Area

Climate Change effects not mitigated by DHF at high levels

Climate Change-induced Burned Area on the Rise

Discussion

Online methods

1. Modelling Protocol

2. Fire Models

3. Observations

4. Relative Anomaly and Probability Ratio

5. Evaluation Methods

6. Ensemble Weighting

7. Evaluation Results

7.1 General Benchmarking

7.2 Temporal Variability

7.3 Burned Area Distribution

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1