Understanding how fire regimes change over time is of major importance for understanding their future impact on the Earth system, including society.
Large differences in simulated burned area between fire models show that there is substantial uncertainty associated with modelling global change impacts on fire regimes. We draw here on sensitivity simulations made by seven global dynamic vegetation models participating in the Fire Model Intercomparison Project (FireMIP) to understand how differences in models translate into differences in fire regime projections. The sensitivity experiments isolate the impact of the individual drivers on simulated burned area, which are prescribed in the simulations. Specifically these drivers are atmospheric
The seven models capture spatial patterns in burned area. However, they show considerable differences in the burned area trends since 1921. We analyse the trajectories of differences between the sensitivity and reference simulation to improve our understanding of what drives the global trends in burned area. Where it is possible, we link the inter-model differences to model assumptions.
Overall, these analyses reveal that the largest uncertainties in simulating global historical burned area are related to the representation of anthropogenic ignitions and suppression and effects of land use on vegetation and fire. In line with previous studies this highlights the need to improve our understanding and model representation of the relationship between human activities and fire to improve our abilities to model fire within Earth system model applications. Only two models show a strong response to atmospheric
Wildfires are an important driver of vegetation distribution and regulate ecosystem functioning, biodiversity and carbon storage over large parts of the world
Given the various impacts of fire on natural and human systems, and the large uncertainties, it is important to improve the understanding of what controls the occurrence of wildfires and to know how fire regimes might change in the future.
Based on current process understanding, the following drivers influenced burned area over the last decades to centuries.
Increasing atmospheric
There is still controversy about whether humans increase or decrease fire overall. Although there is broad agreement that humans suppress fires in regions with high population density, observational studies are less clear about what happens in areas of low population density and show both increases or decreases due to human activities
Fire was used to manage croplands in pre-industrial times (e.g.
Lightning is the main source of natural ignitions
Climate influences burned area through weather conditions and through its influence on vegetation
Fire-enabled vegetation models simulate fire regimes in response to the combination of individual forcings, including atmospheric
In this multi-model study we use the historical simulation to show the overall modelled response of burned area to changes in environmental and human factors. We then compare the sensitivity experiments of the five most commonly used driving factors to document how simulated burned area responds to the individual forcing factors and relate inter-model differences of the burned area response to differences in model assumptions or parametrization. We finally suggest implications of our results for model development and application.
The baseline FireMIP experiment (SF1) is a transient simulation from 1700 to 2013, in which atmospheric
Overview over the sensitivity experiments conducted by FireMIP models
Sensitivity experiments conducted by FireMIP models.
Our analyses of the SF1 and SF2 simulations focus on the simulation of burned area but are complemented by effects on vegetation carbon pools for the SF2_CO2 simulation.
We focus on the time series of global burned area over the historical simulation and the spatial patterns of differences in burned area between 1921 and 2013, as in this period all forcings are transient and different from the values used in the spin-up. Annual global values are an area weighted average using the grid cell area. We quantify the response of the models to each driving factor using the absolute difference in burned area between the baseline and the respective sensitivity experiment (SF1-SF2_i, with
Due to a postprocessing error, INFERNO lacks 2 years in SF2_CO2 (2001 and 2002).
To evaluate the simulations of burned area, we compare the simulated burned area with remote sensing data products. Global burned area observations from satellites still suffer from substantial uncertainty, as reflected by the considerable differences in spatial and temporal patterns between different data products
In spite of major advances in mapping burned area based on satellite data, these data products include major uncertainties. GFED4 and FireCCI50 provide uncertainty estimates for the burned area. Applying Gaussian error propagation, which assumes that errors are independent and normally distributed, yields uncertainty estimates of 0.01 % (GFED4) and 0.2 % (FireCCI50) of the global burned area, which is certainly an underestimation. The assumptions of normal distribution and independence are likely violated. The spread between global burned area datasets is probably a more realistic estimate. Since all the products rely on the MODIS sensor, this approach will not capture the full uncertainty. Nevertheless, to investigate the effect of data quality in the observations on the model–data comparison we use the burned area product uncertainty estimates (aggregated to model resolution assuming independence) to group the observations into points with low, medium and high uncertainty (low: within the 0–33rd percentile, medium: within the 33rd–66th percentile and high: within the 66th–99th percentile of the relative uncertainty; estimates
The models show magnitudes of annual global burned area between 354 and 531 Mha yr
Annual global burned area (BA) in megahectare per year (Mha yr
Global burned area averaged over 2001–2013 in megahectare per year (Mha yr
The simulated trend in burned area in the historical simulation differs between the models (see Fig.
Satellite records show a decline in global burned area since 1996
No observations document the longer-term trends in burned area. Charcoal records
The understanding of the drivers on simulated trends that we give below provides insights on what causes the simulated trends and which assumptions control the trend. These insights will help to understand which observational constraints and process understanding is required to improve global fire models.
Trends (slope and standard error of a linear regression, megahectare per year, Mha yr
The response of burned area to the individual factors is determined by the changes in the driving factors and the sensitivity of the model to these changes. The population density forcing dataset has the strongest trend in the relative differences between the transient forcing and the year 1920 value followed by the land-use and land-cover change dataset. The trend in atmospheric
Absolute difference in annual global burned area (
The spatial patterns of trends in burned area are mostly heterogeneous (see Figs.
Area with a significant positive trend (red bar) or with a significant (Mann–Kendall test
In the following paragraphs we detail the inter-model differences and their causes for each sensitivity experiment.
The overall changes in burned area in individual simulations as a result of atmospheric
We use changes in vegetation carbon to understand changes in fuel load and composition because information on the amount of fuel used within the fire models was not available for individual plant functional types (PFTs). All models show an increase in total vegetation biomass (“total” is indicated by solid lines; see Fig.
Relative difference in global carbon stored in C
CLM and LPJ–GUESS–SIMFIRE–BLAZE include an interactive nitrogen cycle, and CLASS–CTEM includes a non-interactive nitrogen downregulation. Effects of atmospheric
Soil moisture is used by several models to compute fuel moisture (see Fig.
Models which include fuel load and moisture effects through threshold functions (see Fig.
Impact of fuel load on fire for CLASS-CTEM, INFERNO and CLM. Impact of fuel load on the probability of fire (
Annual average of the relative difference in volumetric soil moisture (CLM) and total soil moisture content (remaining models) between the baseline experiment SF1 and the sensitivity experiment SF2_CO2 (see Table
The population density forcing used for FireMIP increases in every region of the globe over time as well as in annual global values
All the models, except LPJ–GUESS-SIMFIRE–BLAZE, include the number of anthropogenic ignitions (
LPJ–GUESS–SIMFIRE–BLAZE does not include anthropogenic ignitions explicitly but rather treats the net effect of changes in population density, which was optimized using burned area satellite data
Variation in probability of fire due to human ignitions (
Suppression effects of population density on fire duration (
The models all agree that at high population density fire is suppressed. This leads to similarities in the spatial patterns of the effect of population changes (see Fig.
The land-use change imposed in SF2_FLA is characterized by a strong decrease in forested areas and an increase in pastures and croplands
The FireMIP models handle land-cover dynamics, the expansion of agricultural areas and fire in agricultural areas differently.
Some of the models (CLASS–CTEM, CLM, JSBACH–SPITFIRE and ORCHIDEE–SPITFIRE) prescribe the vegetation distribution so that the land-cover fraction for all PFTs does not change through time in SF2_FLA, while in the SF1 simulation the cover fractions of natural PFTs are reduced according to the expansion of agricultural areas. The other models simulate the distribution of the natural vegetation dynamically but prescribe the agricultural areas.
All models decrease the tree cover to represent the expansion of croplands over time. Land conversion due to the expansion of pasture is not represented in CLASS–CTEM.
Only CLM includes cropland fires, INFERNO treats croplands as natural grasslands, and all the other models exclude croplands from burning (see Table
Treatment of agricultural fires
In LPJ–GUESS–SIMFIRE–BLAZE pastures are harvested; this reduction in biomass leads to a decrease in burned area in addition to the decrease caused by exclusion of fire in croplands.
In JSBACH–SPITFIRE, the expansion of pastures occurs preferentially at the expense of natural grassland and does not affect tree cover until all the natural grassland has been replaced
LPJ–GUESS–SPITFIRE and ORCHIDEE–SPITFIRE react with an increase in burned area to the expansion of land use since they treat pastures as natural grasslands. The SPITFIRE fire module is very sensitive to the vegetation type with very high burned area for natural grasslands due to higher flammability compared to woody PFTs
Land use was already identified as a main reason for inter-model spread in the CMIP5 ensemble
Most of the models show a low response of burned area to lightning (see Fig.
Simulated burned area in FireMIP responds to changes in climate with strong inter-annual variability but only weak trends in burned area (see Fig.
The influence of climate on burned area is complex: it influences burned area through the meteorological conditions and through effects on vegetation conditions that influence fuel load and fuel characteristics
The impact of climate on the inter-annual variability, however, is strongly expressed in the simulated burned area. This is consistent with the finding that recent precipitation changes influence inter-annual variability in fire but have little impact on recent longer-term trends
Global vegetation models are an important tool for examining the impacts of climate change and are used in policy-relevant contexts
The main concern for model applications is the large spread of the historical simulated burned area. It remains difficult to evaluate and optimize the transient burned area simulations as the period observed by satellites is still short, and the trends are not robust
Our analysis shows which parts of the models are particularly important to simulate changes in burned area and need additional observational constraints or improved process understanding. In line with previous research
We identify land-use change as the major cause of inter-model spread. Only one model explicitly includes fires associated with land-use and land-cover change (cropland and deforestation fires), and all the other models only include such effects through changes in vegetation parameters and structure. The inclusion of cropland fires is certainly important to understand and project changes in emissions, air pollution and the carbon cycle
We also find inter-model agreement for certain aspects. For instance, burned area is suppressed at high population densities, which leads to a similar spatial response to population density (see Fig.
We show that, although all models show an overall increase in biomass as a consequence of increasing atmospheric
Climate and lightning have a much lower effect on the trends than the other factors. While this study focuses on the trends, research on the short-term variability and extreme events will be highly useful to investigate fire risks. The influence of climate and lightning on fire are therefore important research topics even if we find a comparably low influence on the long-term trends. Moreover the trends in climate parameters may increase for the future, and therefore the influence on burned area might increase.
In contrast to many model simulations that use a lightning climatology based on satellite observations, the FireMIP experiments were driven by a transient dataset of lightning activity created by scaling a mean monthly climatology of lightning activity using convective available potential energy (CAPE) anomalies of a global numerical weather prediction model. Since climate changes can be expected to cause changes in lightning, it will be important to develop transient lightning datasets for climate change studies on fire. The use of present-day lightning patterns, for example, will certainly lead to an overestimation of lightning strikes in regions with drier climate projected in the future. But not only spatial patterns of lightning are important, the covariation with climate, as well as the temporal resolution of the input dataset, determines the influence on burned area
Recent advances in remote sensing products have high potential to support model development. However, remotely sensed burned area datasets alone are not a sufficient basis to evaluate fire models as many model structures can lead to reasonable burned area patterns. The emergence of longer records of burned area and the increasing availability of information on other aspects of the fire regime considerably improve opportunities to evaluate and improve our models. The FRY database
The specific model application has a strong influence on judging the validity of a model. Our analyses of the controls on the variability of fire suggest that human activities drive the long-term (decadal to centennial) trajectories, while considering climate variability may be sufficient for short-term projections. Changes in the trends of the driving factors may change this balance. For instance, stronger changes in climate into the future may increase the relative importance of climate for long-term fire projections in the future.
This comprehensive analysis of the influences of climate, lightning, atmospheric
The analysis of the sensitivity experiments shows that the increase in atmospheric
Most models agree on a decrease in burned area due to increases in population density. Most models link the number of ignitions to population in a way that ignitions increase initially at low population densities. In densely populated regions, all models assume that the effect of anthropogenic ignitions is outweighed by fire suppression and the increased fragmentation of the landscape by anthropogenic land use. It would be useful to develop an approach that represents local human–fire relationships, but this will likely remain a long-term challenge and requires the synthesis of knowledge from various research fields.
The simulated response of burned area to land-use and land-cover change depends on how fires in cropland and pastureland are treated in each model. Most models simply exclude croplands from the burnable area; therefore the treatment of pastures causes the largest part of the model spread. Models that do not allow fire in croplands, and either harvest biomass in pastures or assume specific vegetation parameters, show a reduction in burned area. Models that treat pastures as natural grasslands and distinguish different fuel types or strongly increase burned area for grasslands show an increase in burned area. Improved knowledge on the effects of land-use intensity on burned area and the development of appropriate forcing datasets could strongly support model development.
The models are comparatively insensitive to changes in lightning, likely because lightning ignitions are not a limiting factor in many regions with very high burning activity. Previous studies however show the importance of lightning and changes in lightning for burned area in the boreal region. Therefore especially regional studies should pay attention to this factor.
None of the models shows a strong trend due to changing climate but all of them show a strong influence of climate on the inter-annual variability. Climatic and ecosystem parameters are only able to explain a rather small part of this variation, with stronger correlations for the ecosystem parameters on the longer annual timescale and a stronger relationship with climatic parameters on the monthly timescale.
Different drivers of burned area affect different timescales: the anthropogenic factors influence long-term variability, while climate and lightning affect short-term variability. Understanding the influence of climate and lightning is especially important for inter-annual variability and extreme events. On the other hand, understanding the impact of anthropogenic drivers is likely more important for the longer-term changes of fire, which is for instance needed in Earth system models. Changes in the trends of the forcing parameters might however affect the balance between them.
The uncertainties in global fire models need to be taken into account in model applications, for instance if model simulations are to be used to support climate adaptation strategies. Model ensemble simulations can give indications of such uncertainties. Therefore the results of this study provide a basis to interpret uncertainties in global fire modelling studies. The information content on the spatial variability of burned area has been well exploited in previous studies, and models reproduce the spatial patterns in a reasonable way. The temporal information of the satellite data is increasing with the increasing length of the record and has a higher potential to contain new information to support the improvement and evaluation of global fire models. Here we provide a summary of which model assumptions need additional constraints to efficiently reduce the uncertainty in temporal trends.
Processed data and processing scripts are available upon request to publications@mpimet.mpg.de.
Spatial distribution of annual burned area fraction (BAF) in percent for the baseline experiment SF1 and observation data, averaged over 2001–2013.
Spatial distribution of regression slopes for the baseline experiment SF1 over 1921–2013.
Spatial distribution of regression slopes for the difference between the baseline experiment SF1 and the sensitivity experiment SF2_CO2 (SF1–SF2_CO2; see Table
Spatial distribution of regression slopes for the difference between the baseline experiment SF1 and the sensitivity experiment SF2_FPO (SF1–SF2_FPO; see Table
Spatial distribution of regression slopes for the difference between the baseline experiment SF1 and the sensitivity experiment SF2_FLA (SF1–SF2_FLA; see Table
Spatial distribution or regression slopes for the difference between the baseline experiment SF1 and the sensitivity experiment SF2_FLI (SF1–SF2_FLI; see Table
Spatial distribution of regression slopes for the difference between the baseline experiment SF1 and the sensitivity experiment SF2_CLI (SF1–SF2_CLI; see Table
Spearman rank-order correlation coefficient for each grid cell over 1921–2013 for the difference between the baseline experiment SF1 and the sensitivity experiment SF2_CLI (see Table
Scatter plots for the GFED4 and FireCCI50 dataset without transformation, square root transformation and log transformation
Reference literature for FireMIP models.
Correlation coefficients between burned area simulated by the FireMIP models within the baseline experiment SF1 and the respective observation data. Due to the very skewed distribution of burned area, we use a square root transformation on both the models and the observations. Numbers in brackets show the Pearson correlation coefficients for not-transformed data. Only GFED4 and FireCCI50 provide uncertainty estimates; therefore GFED4s is not included. Correlation coefficients for 33 % show the correlation between all grid points that lie within the 0 % and 33 % percentile of the relative standard error. Values for 66 % lie within the 33 %–66 % percentile of the relative standard error, and values for 99 % lie within the 66 %–99 % percentile. Bold numbers indicate correlation coefficients that are significant (
LT and GL designed the study and performed the analysis with input from SPH, AH and SH. CY, GL, JRM, FL, MF and SH provided simulations. LT, GL and SPH wrote the paper with contributions from all authors.
The authors declare that they have no conflict of interest.
The authors are grateful for the support and guidance of Silvia Kloster, who initiated this work. We would like to thank Stephane Mangeon, who performed the simulations with INFERNO. Gitta Lasslop was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 338130981 and acknowledges the excellent computing support of DKRZ.
The article processing charges for this open-access publication were covered by the Max Planck Society.
This paper was edited by Akihiko Ito and reviewed by two anonymous referees.