Large-scale fires have been making global headlines over the last couple of years, many of which highlight the importance of fire and it’s feedbacks within the Earth System. In November 2018 over 80 people were killed in Californian fires that also caused an estimated $27 billion in damages (1). Last August, South America saw the largest increase in fire count seen in nearly 10 years, probably driven by humans in regions experiencing widespread conversion of forest to savanna and agriculture (2). Hundreds of fires burnt throughout the 2019 summer in Siberia and Alaska, releasing over 150 Mega tonnes of CO₂, and large quantities of black carbon that could further accelerate local arctic ice melt (3). Even the UK has even seen some burning, including a peatland fire in north-east Sutherland that doubled Scotland’s carbon emission for 6 days in May 2019 (4). Perhaps the most devastating is the (at time of writing) ongoing fires in South Eastern Australia, which has seen unprecedented levels of burning in many regions, resulting in 27 deaths and the destruction of over 2000 properties, and produced smoke and particulate matter that has been deposited over the Pacific all the way to South America (5). Meanwhile, in tropical savannas, which experience regular, and in some cases annual burning (Fig. 2), there has been a substantial and sustained reduction in burnt area over the last few decades, largely driven by agricultural fragmentation of the landscape (6), affecting vegetation assemblages and altering carbon uptake. All this points to a dramatic shift in fire regimes worldwide, mainly driven by changes in climate, but also through mismanagement of the land (2, 6), with potentially widespread Earth System impacts on humans, vegetation dynamics, atmospheric composition and radiative forcing, and even ocean biogeochemistry and ice melt (7). 

However, disentangling the complex interplay of climate controls, human manipulation and land use has made modelling fire extremely difficult, and many global fire models struggle to reproduce even basic fire properties, even when driven by observed climate (8).  Trickier still is simulating the feedbacks of fire within the Earth System. Fire emissions and fire driven-mortality/recovery are poorly constrained in observations and show large disagreements between models  (8). As a result, CMIP5 only had a small number of ESMs simulating fire, all showing familiar problems in reproducing basic trends in global fire emissions, compounded by the sensitivity of fire regimes to even small climate biases, and making the assessment of wider Earth System feedbacks impractical (9).  

We are attempting to overcome many of the issues when introducing a coupled fire component to the UKESM through the development of a simple fire model, INFERNO, purpose-built for coarse-scale climate modelling, which is simple enough to be embedded into observations of fire and fire drivers using a novel inference optimization scheme. Here, we give a brief overview of the fire model and optimization development and highlight some of the key studies both have already been used in. 

Figure 1. INFERNO inputs (grey boxes) and key components (coloured boxes). 

 

INFERNO 

INFERNO (INteractive Fire and Emission algoRithm for Natural envirOnments) calculates flammability based on inputs of temperature, relative humidity, precipitation, and saturation vapour pressure, and simulated fuel density and soil moisture (Fig. 1). Anthropogenic and natural ignitions come from population density and lightning. Flammability and ignitions, together with an average burnt area by land cover type, simulate burnt area. Emission factors translate burnt area into emissions of CO₂, CO, CH₄, NO_x, SO₂, OC and BC, which can be fed back through to UKCA. INFERNO was originally implemented in JULES at version 4.6 (10), and recently updated to include mortality and combustion of vegetation and soil carbon at JULES version 4.9 (11), representing additional vegetation-fire feedbacks. It has since been updated to version 5.4 to include 17 PFTs and PFT-scaled fire mortality.  

Diagnostic fire, driven directly from UM climate, was included in UM at version 11.1 and subsequently incorporated in UKCA at UM version 11.2. At the moment, lightning is read in through an ancil, but there will eventually be an option to drive INFERNO from  UM lightning scheme. The final piece of coupling required for a technically working fire enabled UKESM is to incorporate biogeochemical feedbacks from fire when running the UM with a dynamic land surface. In the meantime, work is in progress to use output from UKESM data to drive future simulations of JULES using interactive fire-vegetation feedbacks. 

Figure 2. 2001-2014 annual average burnt area (top row) and fire CO₂ emission (bottom row) from observations (first column) and simulated by JULES-INFERNO (2nd column). Observed burnt area is from GFED4s (12) and emission from GFAS (13). 

 

The key points of weakness of JULES-INFERNO, again shared with most other fire models, has been high fire incidence in densely populated and cropland areas, and a positive global trend in both burnt area and fire emissions (8). Capturing the observed decreased trends in fire carbon emissions are of particular concern as this should be a basic requirement of fire models used to in climate change studies (6), and yet models tend to only capture this at the expense of correct spatial patterns or magnitude of emissions (8). Very few models include specific representation of landscape fragmentation effects on burnt area, recently shown to be a major driver of decreased burnt area in most of the world’s fire-prone regions (6). Including the new representation of separate crop and pasture land use (14), enables us to reduce burning in agricultural areas for the first time, improving spatial patterns of burning (Fig. 2) whilst at the same time showing substantial improvement in trends in emissions. Although not yet incorporated into a JULES release, additional representation of landscape fragmentation that reaches beyond agricultural extent further improves trends in burnt area and emissions (Fig. 3). 

Since it was first implemented in JULES, INFERNO has been part of the global Fire Model Intercomparison Project (FireMIP) (15, 16) and has contributed to a number of multi-model analysis studies. A multi-model evaluation paper looking at fire model performance, and potential causes of model discrepancies, has recently been submitted by Hantson et al. (8), in which JULES-INFERNO compares very well to the other 8 models assessed. Another assessing the impact of fire-induced tree cover reduction on the carbon cycle was recently submitted by Lasslop et al. (17) which includes INFERNO as one of 7 other fire-enabled global vegetation models. The results show that fire accelerates the carbon cycle, and reduces tree covered area and carbon storage by 10% on average across models. INFERNO has also been used for scientific applications, assessing how fire changes through El Niño events such as the 2015/16 El Niño (18).   

Figure 3. 2001-2014 trends in fire carbon emissions. Maps show spatial variation in trends in observations from GFAS (13), and simulated by JULES. Time series shows annual CO₂ emission from GFAS, the newest version of JULES-INFERNO and the previous version of JULES_INFERNO (11) 

 

Bayesian Inference Methods 

The simplicity of INFERNO provides us with the opportunity to fuse the model with optimization techniques embedded in the ever-increasing wealth of satellite observations of fire. We have been developing a Bayesian inference scheme, aiming to quantify uncertainties in fire model parameters, while propagating uncertainties and (when driven by UKESM climate) errors in climate and observed fire. Mapping uncertainties in parameter space will provide two important pieces of information: it will find the overlap between observationally constrained parameters and parameters that will simulate fire under inevitable ESM climate biases; or otherwise determine areas in which simulation of fire is not possible, and whether it is due to fire model structure, limitations from types of inputs (i.e resolution or available variables), or insurmountable climate biases. 

 Figure 4. Modelled and observed fire count for South America. Reproduced from (2). First row: observed fire count from Brazil’s National Institute for Space Research (19), Aug 2002-2019 annual average (left), Aug anomaly in 2019 (centre), and the number of years from 2002 that 2019 fire counts exceed (right). Second row: as top row, as simulated by the model. Stippling represents where model uncertainties 5% percentile > half the 95% percentile. 

---

All this is very ambitious, and a fully integrated INFERNO Bayesian Inference scheme is probably a few years away. However, our work on parameter optimization has already attracted interest from other modelling groups (15), and early development has produced a number of interesting results. We have developed a simplified scheme already which uses a much less computationally demanding fire model that emulates INFERNOs structure, but runs on a monthly timestep and uses inputs readily available from observations (6, 20). This simplified scheme has been used to map significant and substantial recent changes in fire regime, and determine driving controls of these changes. While introducing the concept of flammability – an important step for converting to INFERNO (Fig. 1) – we were able to perform an assessment of drivers of fire activity in South America, determining which years were driven by variations in meteorological conditions and which are as a result of outside drivers, such as human land use (2) (Fig. 4). The framework has also been used to estimate the impact fires have on tropical tree cover (21), indicating that the reduction from fire-induced stress and mortality is much less than is implemented in JULES-INFERNO, and indeed most other fire models, though showing increased sensitivity in forested areas (Fig. 5). 

Figure 5. Decrease in tree cover caused by fires (21). Stippling area indicates low uncertainty on tree cover impact. 

 

Conclusions 

Coupling of fire to UKESM has progressed very rapidly in the last couple of years. Technical coupling is almost complete, with only a finalised version of UKCA and implementation of biogeochemical fire coupling in the UM now outstanding. What might be trickier though, is achieving a simulation of fire that provides us with useful information about the change in fire regimes over time – something which most modelling groups struggle with even when running fire models offline. Our offline runs, however, show promising results which have already been useful in a number of important studies. And, while a long way from full incorporation of INFERNO, our optimization scheme has already proven useful for assessing the drivers behind some of the recent signs of fire regime shift over much of the world. This has all been driven predominantly by early career researchers – Chantelle only recently graduated from her PhD and Joao is still half way through his studies, which makes the recent progress and high impact science we’ve achieved even more impressive. 

 

References 

Italics indicate studies produced in the development or application of INFERNO model or ConFIRE optimization scheme 

1. Nauslar, N. J., Brown, T. J., McEvoy, D. J. & Lareau, N. P., in State of the Climate 2018, J. A. D. S. A. Blunden, Ed. (Bulletin of the American Meteorological Society, 2019), p. 195.
2. C. Burton, D. I. Kelley, C. Huntingford, M. Brown, R. Whitley, N. Dong, Low Climatic Influence found in 2019 Amazonia Fires. Environ. Res. Lett. (submitted).
3. K. Patel, Arctic Fires Fill the Skies with Soot. NASA Earth Observatory (2019), (available at https://earthobservatory.nasa.gov/images/145380/arctic-fires-fill-the-skies-with-soot).
4. J. Wiltshire, J. Hekman, B. F. Milan, “Carbon loss and economic impacts of a peatland wildfire in north-east Sutherland, Scotland, 12-17 May 2019” (ED12990–3, WWF-UK, 2019).
5. BBC, Australia fires: A visual guide to the bushfire crisis. BBC news (2020), (available at https://www.bbc.co.uk/news/world-australia-50951043).
6. D. I. Kelley, I. Bistinas, R. Whitley, C. Burton, T. R. Marthews, N. Dong, How contemporary bioclimatic and human controls change global fire regimes. Nat. Clim. Chang. 9, 690–696 (2019).
7. D. M. J. S. Bowman, J. K. Balch, P. Artaxo, W. J. Bond, J. M. Carlson, M. A. Cochrane, C. M. D’Antonio, R. S. Defries, J. C. Doyle, S. P. Harrison, F. H. Johnston, J. E. Keeley, M. A. Krawchuk, C. A. Kull, J. B. Marston, M. A. Moritz, I. C. Prentice, C. I. Roos, A. C. Scott, T. W. Swetnam, G. R. van der Werf, S. J. Pyne, Fire in the Earth system. Science. 324, 481–484 (2009).
8. S. Hantson, D. I. Kelley, A. Arneth, S. P. Harrison, S. Archibald, D. Bachelet, M. Forrest, T. Hickler, G. Lasslop, F. Li, S. Mangeon, J. R. Melton, L. Nieradzik, S. S. Rabin, I. Colin Prentice, T. Sheehan, S. Sitch, L. Teckentrup, A. Voulgarakis, C. Yue, Quantitative assessment of fire and vegetation properties in historical simulations with fire-enabled vegetation models from the Fire Model Intercomparison Project, , doi:10.5194/gmd-2019-261.
9. S. Kloster, G. Lasslop, Historical and future fire occurrence (1850 to 2100) simulated in CMIP5 Earth System Models. Glob. Planet. Change. 150, 58–69 (2017).
10. S. Mangeon, A. Voulgarakis, R. Gilham, A. Harper, S. Sitch, G. Folberth, INFERNO: a fire and emissions scheme for the UK Met Office’s Unified Model. Geoscientific Model Development. 9, 2685–2700 (2016).
11. C. Burton, R. Betts, M. Cardoso, R. T. Feldpausch, A. Harper, C. D. Jones, D. I. Kelley, E. Robertson, A. Wiltshire, Representation of fire, land-use change and vegetation dynamics in the Joint UK Land Environment Simulator vn4.9 (JULES) (2019), doi:10.5194/gmd-12-179-2019.
12. G. R. van der Werf, J. T. Randerson, L. Giglio, T. T. van Leeuwen, Y. Chen, B. M. Rogers, M. Mu, M. J. E. van Marle, D. C. Morton, G. J. Collatz, R. J. Yokelson, P. S. Kasibhatla, Global fire emissions estimates during 1997–2016. Earth Syst. Sci. Data. 9, 697–720 (2017).
13. J. W. Kaiser, A. Heil, M. O. Andreae, A. Benedetti, N. Chubarova, L. Jones, J. J. Morcrette, M. Razinger, M. G. Schultz, M. Suttie, Others, Biomass burning emissions estimated with a global fire assimilation system based on observed fire radiative power (2012) (available at http://dare.ubvu.vu.nl/handle/1871/48600).
14. E. Robertson, The Local Biophysical Response to Land-Use Change in HadGEM2-ES. J. Clim. 32, 7611–7627 (2019).
15. S. Hantson, A. Arneth, S. P. Harrison, D. I. Kelley, I. C. Prentice, S. S. Rabin, S. Archibald, F. Mouillot, S. R. Arnold, P. Artaxo, D. Bachelet, P. Ciais, M. Forrest, P. Friedlingstein, T. Hickler, J. O. Kaplan, S. Kloster, W. Knorr, G. Laslop, F. Li, J. R. Melton, A. Meyn, S. Sitch, A. Spessa, G. R. van der Werf, A. Voulgarakis, C. Yue, The status and challenge of global fire modelling. Biogeosciences. 13, 3359–3375 (2016).
16. S. S. Rabin, J. R. Melton, G. Lasslop, D. Bachelet, M. Forrest, S. Hantson, F. Li, S. Mangeon, C. Yue, V. K. Arora, Others, The Fire Modeling Intercomparison Project (FireMIP), phase 1: Experimental and analytical protocols. Geoscientific Model Development. 20, 1175–1197 (2017).
17. G. Lasslop, S. Hantson, S. Harrison, D. Bachelet, C. Burton, M. Forkel, M. Forrest, F. L. Li, J. Melton, C. Yue, S. Archibald, A. Arneth, T. Hickler, S. Scheiter, S. Sitch, Global multi-model estimates of fire-induced tree cover reduction and acceleration of the terrestrial carbon cycle. Glob. Chang. Biol. (submitted).
18. C. Burton, R. A. Betts, C. D. Jones, F. Feldpausch, M. Cardoso, L. Anderson, El Niño driven changes in global fire 2015/16. Frontiers  (submitted).
19. Brazil’s National Institute for Space Research, Burn Database. INPE – Burned Program – Support (2019), (available at http://queimadas.dgi.inpe.br/queimadas/bdqueimadas).
20. D. I. Kelley, thesis, Macquarie University (2014).
21. D. I. Kelley, I. Bistinas, R. Whitley, C. Burton, F. Gerard, R. Ellis, G. Li, T. Marthews, E. Veenendaal, G. Weedon, N. Dong, J. Lloyd, in EGU General Assembly (2019), pp. EGU2019–17591.