Likely future(s) of global wildfires

UKESM-fire and JULES-ISIMIP contributions to UNEP rapid response assessment: “Spreading like Wildfire: The Rising Threat of Extraordinary Landscape Fires.”

The report will be made public in late January and available at

Douglas I Kelley1, Camilla Mathison2, Chantelle Burton2, Megan Brown1,3, Andrew Sullivan4, Elaine Baker5,6, Tiina Kurvits5

1UK Centre for Ecology and Hydrology, 2 UK Met Office, 3 Open University, UK, 4 CSIRO, Canberra, 5 UNEP/GRID-Arendal, 6 University of Sydney

“The risk posed by wildfire to people and the environment is increasing as a result of numerous factors, including but not only climate change” is a key finding from the upcoming UNEP report “Spreading like Wildfire: The Rising Threat of Extraordinary Landscape Fires[1].  This finding is despite the complex interactions between fire drivers – both bioclimate and humans’ profound impact on the landscape – that make simulating fire extremely difficult [2,3][1]. Add the uncertainty in future emissions and climate responses, particularly rainfall [4] to which fire regimes are extremely sensitive [1], and past attempts at projecting future burning and its impacts widely disagree [5].  For this report, instead of projecting future fire regimes directly, we simulated the likelihood of all possible future burning levels whilst accounting for these uncertainties in fire modelling and future climate. We demonstrate that many parts of the world show likely and robust future changes in burnt area – some of which may be avoidable with emission reduction. Ongoing development of this framework maybe helpful for translating climate model outputs – uncertainties and all – for climate-relevant impacts.

Finding likelihood through model uncertainty

For the report, we drove the ConFire Bayesian fire framework [6,7] with historic ISIMIP2b bias-corrected climate models and JULES-ES vegetation outputs [8,9]. ConFire optimises a simple fire model against GFED4s burnt area observations [10] and expresses burnt area and its drivers influence as probability distributions based on uncertainty and biases of the optimised model (Fig. 1). ConFire thereby preserves uncertainty when simulating future fires based on climate outputs from the four ISIMIP2b climate model projections [11]. The projections use emissions scenarios from RCP2.6  – the lowest emission scenario in CMIP5, targeting 2°C [12,13], and RCP6.0 representing a no-mitigation scenario. Developing the technique from Scholze el al [14], we identified areas where climate model projections show substantial changes in future burnt area, even if models disagree on the magnitude or direction of that change.

Likely occurrences of recent extreme events by 2100

We can assess how likely some of the recent, more extreme fire events will become in the future (Fig. 1). The severe 2019 Arctic fires were particularly concerning given the vast amount of carbon released – 150 Mt CO2, equivalent to half the UK’s annual emissions[2] and only beaten by 2020, where another 244 Gt CO2 was released into the atmosphere [15]. For Northern Siberia, which experienced some of the worst burning in 2019, the framework suggests an event of this magnitude or greater will become around 5 times more likely in the last 10 years the century – irrespective of fire emissions (Fig. 1).

[1] See previous newsletter:


Figure 1: Probability distributions of burnt areas for (black) 1996-2005 and 2090-2099 for (blue) RCP2.6 and (red) RCP6.0 for the 4 points marked in Fig. 2. Dotted lines show burnt area of recent extreme fire events. The likelihood of a fire event is the probability of a burnt area greater or equal to that event.

In Amazonia, 2019, fires burnt around 900,000 hectares of rainforest (an area larger than the island of Corsica). They exposed millions of people to increased air pollution, including in São Paulo, 1700 miles away. These fires have been attributed to increased deforestation [16], and anomalous meteorological conditions had only a 6-7% probability of impacting burning [6] – shown in a study using the same Bayesian technique as here. This explains why our framework – which does not capture human fire management policy changes beyond the training period- suggests 2019 fires would have been extremely unlikely in our 1995-2006 base period. However, future climate and land-use changes for the region alone mean fires of the magnitude or worse of 2019 could occur every 34 (RCP2.6) or 17 (RCP6.0) years (Fig. 1).

2019/2020 south-east Australian fires highlight remaining deficiencies in fire modelling. These fires were by far the most extensive since the start of the satellite record and almost entirely driven by anomalous meteorological conditions, including rapid stratospheric warming and long-term drought [1]. Despite this, residual uncertainties in our fire modelling framework suggest a 16.1% likelihood for 1995-2006 – a period that did not include such extreme meteorological conditions – equivalent to a fire event every six years. Better constraining this uncertainty is required before confidently assessing changes in likelihood in burnt in the future (1.4 x more likely based on our current understanding).

Future changes in global wildfire

We forecast substantial shifts in fire regimes of much of the world by the end of the century (Table 1; Fig. 2). Tropical savannas show the most potential change in burnt area, though climate models disagree on change direction. Where models agree, they suggest an increase in burning, though increases are within fire model uncertainty. This means that, while it looks likely fire regimes will change in tropical savannas, we cannot say how based on our understanding of future climate or fire drivers. An increase in fuel availability through decreased fuel moisture will cause more burnt area in North Australia, though not significant against uncertainty associated with present-day veg/fire model performance. The overlap between historic and future probability distributions indicates how likely a significant change in burnt area. This highlights a significant agreement for decreased burning in Southern Brazil, Uruguay and northern Argentina and the US east coast under RCP2.6, but not RCP6.0.

However, all climate models implicate fuel drying as likely to increase burning significantly greater than veg/fire model uncertainty in Indonesia and many Arctic areas across RCPs. These areas are of particular concern given the potential to release the high carbon content of peatlands and their irrecoverable carbon [17].

[1] Met office coverage of Australian fires:

Projected 2090-2099 changes in burnt area compared to (top) historic burnt area from GFED4s [10] 1996-2005 for (middle row) RCP2.6 and (bottom) RCP6.0 ISMIP2b climate model ensemble.

Figure 2: Projected 2090-2099 changes in burnt area compared to (top) historic burnt area from GFED4s [10] 1996-2005 for (middle row) RCP2.6 and (bottom) RCP6.0 ISMIP2b climate model ensemble. (left) areas in red are where we project burnt area will increase, while the blue shows areas where it will decrease. Purple is where some climate models project an increase while others a decrease. (right) significance in changes in burnt area against fire model uncertainty – defined as the % dissimilarity between historical and future probability distributions (example above maps).

Worldwide, there will be an increase in extreme fire events (which we define as a one in 100 annual burnt area from the last 10 years; Table 1). This increase is likely to be small in the next 10 years – 8-14%. By the end of the century, though, this increase is likely to be much more substantial – up to 36-57% under RCP6.0. Emissions reduction will mitigate some extreme events, though we will still see a considerable increase in events (21-52%).


Table 1: Global increase in extreme fire events, defined as a 1 in 100 annual burnt area for 2010-2020.


Near-term – 2020-2030

Medium-term – 2040-2050

Long-term – 2090-2100



20-33% 31-52%
6.0 8-14% 21-27%


Ongoing work

Not all sources of uncertainty are accounted for in this framework yet, and there are still avenues of fire model development that may further constrain uncertainty. We simulate fuel properties using JULES-ES, whereas there are multiple vegetation models that sample the range of our understanding of the land surface [18]. Moreover, by optimising on the relatively short satellite period, we assume human fire management will continue as present day, which the 2019 Amazonia fire demonstrates is not always the case [6]. There is active research on human-fire interactions and management decisions within JULES [19] that may help incorporate changes in humans future behaviour in the landscape

Finally, incorporating fire-vegetation feedback within the Bayesian framework will account for additional fire-feedback uncertainties. It will also allow us to assess the likelihood of fire regime changes’ impacts on local ecosystems. Some of these impacts could be substantial. If global warnings pass 2°C, Burton et al.  [20] suggest increased burning in South America, mainly in Southern and Eastern Amazon – plausible in our analysis (Fig. 2) – might emit as much carbon as half the USA total emission since pre-industrial times. However, this is a single model, with just four possible futures. Finding the likelihood of these and less or even more catastrophic futures is required before we can adequately plan mitigation and determine likely temperature and emission targets to avoid the worst impacts of fire in our warmer world.


  1. Sullivan, A., Kurvits, T. and Baker E. (eds.) (2021). Spreading like Wildfire: The Rising Threat of Extraordinary Landscape Fires. United Nations Environment Programme and GRID-Arendal
  2. Hantson S, Kelley DI, Arneth A, Harrison SP, Archibald S, Bachelet D, et al. Quantitative assessment of fire and vegetation properties in simulations with fire-enabled vegetation models from the Fire Model Intercomparison Project [Internet]. Vol. 13, Geoscientific Model Development. 2020. p. 3299–318. Available from:
  3. Hantson S, Arneth A, Harrison SP, Kelley DI, Prentice IC, Rabin SS, et al. The status and challenge of global fire modelling. Biogeosciences. 2016 Jun;13(11):3359–75.
  4. Langenbrunner B, David Neelin J, Lintner BR, Anderson BT. Patterns of Precipitation Change and Climatological Uncertainty among CMIP5 Models, with a Focus on the Midlatitude Pacific Storm Track. J Clim. 2015 Oct 1;28(19):7857–72.
  5. Kloster S, Lasslop G. Historical and future fire occurrence (1850 to 2100) simulated in CMIP5 Earth System Models. Glob Planet Change. 2017 Mar 1;150:58–69.
  6. Kelley DI, Burton C, Huntingford C, Brown MAJ, Whitley R, Dong N. Low meteorological influence found in 2019 Amazonia fires. Biogeosciences. 2021;18(3):787–804.
  7. Kelley DI, Bistinas I, Whitley R, Burton C, Marthews TR, Dong N. How contemporary bioclimatic and human controls change global fire regimes. Nat Clim Chang. 2019 Sep 1;9(9):690–6.
  8. Sellar AA, Jones CG, Mulcahy JP. UKESM1: Description and evaluation of the UK Earth System Model. in Modeling Earth … [Internet]. 2019; Available from:
  9. Mathison C, Burke E, Robertson E, Hartley A, Burton C, Williams K, et al. Description and Evaluation of the JULES-ES setup for ISIMIP2b. in prep;
  10. van der Werf GR, Randerson JT, Giglio L, van Leeuwen TT, Chen Y, Rogers BM, et al. Global fire emissions estimates during 1997–2016. Earth Syst Sci Data. 2017 Sep 12;9(2):697–720.
  11. Frieler K, Lange S, Piontek F, Reyer CPO, Schewe J, Warszawski L, et al. Assessing the impacts of 1.5 °C global warming – simulation protocol of the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP2b) [Internet]. Vol. 10, Geoscientific Model Development. 2017. p. 4321–45. Available from:
  12. Rogelj J, Luderer G, Pietzcker RC, Kriegler E, Schaeffer M, Krey V, et al. Energy system transformations for limiting end-of-century warming to below 1.5 °C. Nat Clim Chang. 2015 Jun 21;5(6):519–27.
  13. IPCC, 2018: Global Warming of 1.5°C.An IPCC Special Report on the impacts of global warming of 1.5°C above pre-industrial levels and related global greenhouse gas emission pathways, in the context of strengthening the global response to the threat of climate change, sustainable development, and efforts to eradicate poverty [Masson-Delmotte, V., P. Zhai, H.-O. Pörtner, D. Roberts, J. Skea, P.R. Shukla, A. Pirani, W. Moufouma-Okia, C. Péan, R. Pidcock, S. Connors, J.B.R. Matthews, Y. Chen, X. Zhou, M.I. Gomis, E. Lonnoy, T. Maycock, M. Tignor, and T. Waterfield (eds.)].
  14. Scholze M, Knorr W, Arnell NW, Prentice IC. A climate-change risk analysis for world ecosystems. Proc Natl Acad Sci U S A. 2006 Aug 29;103(35):13116–20.
  15. Witze A. The Arctic is burning like never before – and that’s bad news for climate change. Nature. 2020 Sep;585(7825):336–7.
  16. Silveira MVF, Petri CA, Broggio IS, Chagas GO, Macul MS, Leite CCSS, et al. Drivers of Fire Anomalies in the Brazilian Amazon: Lessons Learned from the 2019 Fire Crisis. Land. 2020 Dec 14;9(12):516.
  17. Goldstein A, Turner WR, Spawn SA, Anderson-Teixeira KJ, Cook-Patton S, Fargione J, et al. Protecting irrecoverable carbon in Earth’s ecosystems. Nat Clim Chang. 2020 Apr 31;10(4):287–95.
  18. Fisher RA, Koven CD. Perspectives on the future of Land Surface Models and the challenges of representing complex terrestrial systems. J Adv Model Earth Syst [Internet]. 2020 Mar 10; Available from:
  19. Millington J, Perkins O, Kasoar M, Voulgarakis A. Advancing representation of anthropogenic fire in dynamic global vegetation models. In: EGU General Assembly Conference Abstracts. 2021. p. EGU21–9502.
  20. Burton C, Kelley DI, Jones CD, Betts RA, Cardoso M, Anderson L. South American fires and their impacts on ecosystems increase with continued emissions. Climate Resilience and Sustainability [Internet]. 2021 Jul 31; Available from:
Download this Newsletter as a PDF (2015KB)