Biological and geophysical feedbacks with fire in the Earth system

Roughly 3% of the Earth’s land surface burns annually, representing a critical exchange of energy and matter between the land and atmosphere via combustion. Fires range from slow smouldering peat fires, to low-intensity surface fires, to intense crown fires, depending on vegetation structure, fuel moisture, prevailing climate, and weather conditions. While the links between biogeochemistry, climate and fire are widely studied within Earth system science, these relationships are also mediated by fuels—namely plants and their litter—that are the product of evolutionary and ecological processes. Fire is a powerful selective force and, over their evolutionary history, plants have evolved traits that both tolerate and promote fire numerous times and across diverse clades. Here we outline a conceptual framework of how plant traits determine the flammability of ecosystems and interact with climate and weather to influence fire regimes. We explore how these evolutionary and ecological processes scale to impact biogeochemical and Earth system processes. Finally, we outline several research challenges that, when resolved, will improve our understanding of the role of plant evolution in mediating the fire feedbacks driving Earth system processes. Understanding current patterns of fire and vegetation, as well as patterns of fire over geological time, requires research that incorporates evolutionary biology, ecology, biogeography, and the biogeosciences.

Most of the 3099 articles published using the term 'fire regime(s)' between 1945 and 2017 addressed this topic at local or landscape scales (40%, compared with 14% for global and 15% for regional-see SI for methods). Global analyses often do not consider all components of fire regimes but focus only on the 'amount burned'-partly because of lack of data at these scales (for example often paleo analyses have only crude approximations of biomass consumed with which to work). However, advances in remote sensing (Archibald et al 2010), and detailed interpretation of paleorecords (Carcaillet et al 2001, Duffin et al 2008 are making it easier to consider the full complement of fire characteristics, thus shedding light on the evolution of fire in the Earth system.

Introduction
Fires have burned on Earth ever since terrestrial plants evolved and provided a source of fuel and oxygen (Glasspool et al 2004, Scott 2000, with subsequent profound impacts on the Earth system. Emissions from fire affect the cycling of carbon, phosphorus, and nitrogen (Crutzen et al 1979, van der Werf et al 2010, Wittkuhn et al 2017, and fires alter surface energy budgets and regional to global climate through emissions of greenhouse gases, aerosols and altering land surface albedo (Kaufman and Fraser 1997, Ward et al 2012. Fires also affect Earth system processes through their impacts on vegetation: altering the structure and composition of ecosystems (Pausas and Keeley 2009) and controlling the distribution and diversity of biomes (Bond et al 2005, Pausas andRibeiro 2017).
On the other hand, fire regimes (see Box 1 for a definition) are in turn controlled by atmospheric conditions, climate, and the type of vegetation available as fuel (Krawchuk and Moritz 2011, Pausas et al 2004, Swetnam and Betancourt 1990. These two-way interactions (here termed feedbacks) make it difficult to predict how climate interacts with fire and vegetation in a changing world (Archibald et al 2013, Bowman et al 2009. For example, high temperatures increase the probability of fire occurrence and the intensity of fire for a particular vegetation type (Flannigan et al 2009, Westerling et al 2006. However, increased temperatures can have negative or positive effects on the productivity of vegetation (Lu et al 2013, Pausas andRibeiro 2013), and can result in shorter fire return periods, which generally result in less fuel and thus less intense fires (Archibald et al 2013). The long term outcome of increased temperatures for fire regimes is a combination of both direct and indirect (via fuels) effects. The same caution should apply when assessing the consequences of human impacts on fire regimesdense human populations tend to reduce fire due to negative impacts on fuel, despite increased ignition frequency (Andela et al 2017, Archibald 2016. Vegetation is an integral component of the vegetation-fire-climate system, which implies evolutionary and ecological processes need to be adequately represented in any theoretical framework. There are dramatic examples in the paleo-record of changes in global patterns of vegetation and fire linked to the evolution of new leaf properties and plant architecture-e.g. early angiosperms, and subsequently the grass family-Poaceae (Belcher and Hudspith 2017, Bond and Scott 2010, Keeley and Rundel 2005. At shorter time-scales, changes in dominant plant species can also significantly alter fire activity and Earth system feedbacks (D'Antonio and Vitousek 1992, Girardin et al 2013. However, these biological processes tend to be overlooked when modelling current fire regimes, predicting changes in fire in the future, or assessing the atmospheric and biogeochemical impacts of fire. New developments in dynamic global vegetation models (DGVMs) are trying to account for these interactions (Scheiter et al 2013).
For example, global and regional data analyses have repeatedly indicated that burned area is maximised at intermediate moisture availability (Daniau et al 2012, Krawchuk and Moritz 2011, Meyn et al 2007-the explanation being that at low rainfall there is not enough fuel to burn, and at high rainfall the fuel is too moist to burn readily (Bradstock 2010). However, when examined for different vegetation types (figure 1), it becomes clear that this global pattern is largely driven by grass-dominated vegetation (>80% of the area burned currently is grassy (van der Werf et al 2006) so it dominates all global analyses when not specifically controlled for). Clearly, explaining global patterns of fire and interactions with climate requires that plant characteristics be explicitly considered (Pausas and Paula 2012).
Collation of the literature indicates that studies on feedbacks between fire and vegetation are less common than ones on feedbacks between climate and vegetation (figure 2). Publications that do consider vegetation are usually local in scope-only 40% of the global papers on fire feedbacks explicitly include vegetation.
Humans have had varied and significant impacts on fire for many millennia. Humans generally increase ignition frequency (Syphard et al 2009), reduce area burned (Andela et al 2017), extend the length of the fire season, and reduce fire size (Le Page et al 2010). Human impacts on global climate, atmospheric properties, and vegetation distributions also feed back to alter fire regimes (Bowman et al 2011, Cochrane andBarber 2009 many proposed interventions to mitigate these global changes also involve managing fire and altering its occurrence/type (Moritz et al 2014). However, there are several research challenges to successfully assessing the Earth system consequences of our impacts on fire and to identifying opportunities for managing fire for geoengineering purposes. Solving these challenges requires better interactions between plant and Earth system scientists.
Here we present a conceptual framework (figure 3) to illustrate how the evolution of key plant traits (see box 2) has driven global scale changes in fire regimes. We start with several modern day examples of 'biogeographic conundrums'-where climate and soils are insufficient for explaining patterns of fire and vegetation. We then outline a conceptual framework of how plant traits influence the flammability of ecosystems, which then feed back to force both the evolution of plant traits and changes in fire characteristics. We explore how these evolutionary and ecological processes scale to impact biogeochemistry and Earth system processes and the potential role of global atmospheric chemistry as a regulatory mechanism influencing changes in fire over geological time. We address this using an approach that includes a collation of published literature.

Biogeographic conundrums
The interdependence of climate, vegetation and fire is illustrated by numerous biogeographic 'conundrums'-examples where abiotic conditions cannot be used without consideration of evolutionary and biogeographic history to predict the distribution of vegetation or fire (figure 4). Biogeographic conundrums are not narrow idiosyncrasies in world vegetation; rather they occur across the majority of the world's biomes and land surface (Moncrieff et al 2016). Indeed, these conundrums were recognised by . Fire acts through both geophysical processes (exchange of energy and matter between the biosphere and atmosphere) and biological processes (community assembly and evolution). Plant innovations that alter the degree to which vegetation is tolerant of fire and the flammability of the vegetation can alter fire regimes, and potentially impact global biogeochemical processes. With both biological and geophysical feedbacks, the type of processes impacted depend on the time scales involved.

Box 2. Plant functional traits.
Functional traits are defined as morphological, biochemical, physiological, structural, phenological or behavioural characteristics of organisms that influence performance or fitness. They are used to define the ecological roles of species in an environment (Díaz and Cabido 2001).
Traits can be considered in terms of their effect on ecosystem properties and the services that human societies derive from them, or as a response to environmental conditions (Lavorel and Garnier 2002). Thus fire-response traits are associated with fire tolerance and regeneration, and fire-effect traits influence flammability.
There are constraints on the combinations of traits that an individual organism can display, leading to trade-offs among traits. Trait syndromes describe groups of traits that are correlated; i.e. together result in an organism that is able to survive and reproduce in a particular environment, and have been used to define ecological strategies thought to be adaptive in those environments.
Functional traits are at the forefront of efforts to develop a mechanistic understanding of how species diversity influences ecosystem functioning, and are essential for scaling local ecological knowledge to the Earth system. For example, vegetation is represented in Earth system models by characterizing its functional traits such as plant height, maximum photosynthetic rates, and rooting depth, that together determine how plants cycle carbon, water, and energy through the Earth system, and interact with each other. Including fire-related functional traits in these models should likewise enable mechanistic simulations of fire regimes. many early ecologists (Myers 1936, Whittaker 1975 who noted that the 'climax' vegetation in a region was not always what climate could support, but rather the result of interactions with disturbances such as fire. However, deterministic relationships between climate and vegetation were assumed in the first global vegetation models, which arose in the 1980s (Prentice et al 2007, Woodward 1987, and are still employed in species distribution models for a wide variety of purposes (Merow et al 2014). It is only in the last decade that the interdependence between climate, vegetation and fire has been quantified at a global scale  and recognised as profound.
Biogeographic conundrums arise via two pathways. Divergent fire regimes refer to regions with similar climate but different fire regimes, arising through differences in community assembly (i.e. differences in the types of plants that dominate in two regions). Alternatively, convergent fire regimes refer to regions with different climate that converge on a similar fire regime by having plant traits that render a community similarly flammable. In each case, it is the biological properties of the vegetation that are critical in shaping the fire regime.
One example of divergent fire regimes arises from the difference in plant traits between boreal North America and Eurasia (figure 4). North American fires are largely stand-replacing crown fire events (Fire return interval (FRI) ranging from 80−400 years (Kasischke et al 2002)) whereas Eurasian fires are generally surface, litter-fuelled fires (FRI from 15−70 years (Sannikov and Goldammer 1996)). Boreal North American fires emit 53% more carbon (C) per unit area burned but cover one third as much area on an annual basis, emitting half as much C as Eurasia per year. They also have a stronger cooling effect (negative radiative forcing) due to the larger change in surface properties after fire (Rogers et al 2015). The climates of these two regions are too similar to explain these differences (Rogers et al 2015, de Groot et al 2013) and there is evidence that plant traits (box 2) of the dominant tree species of each region are The BUI is a measure of the moisture content of the litter fuels (duff) and is a function of temperature and precipitation (Van Wagner and Pickett 1985). key drivers of the fire dynamics. Fire-embracing black spruce (Picea mariana) has highly flammable evergreen needles and retains its dead branches, thereby facilitating crown fire regimes by allowing fires to climb up to the canopy. Spruce (Picea spp.) dominate in Alaska and Canada, but comprise only 15% of the tree biomass in Eurasia. By contrast, larch (Larix spp.) and pine (Pinus sylvestris L.) dominate in Eurasia and tend to resist high-intensity crown fires: larch is deciduous, and the two pine species shed their dead lower branches, so that when fire occurs it usually only spreads in the understory without reaching the canopy. As the same genera are present in both regions, it remains unclear how the differential dominance of particular functional types on each continent has occurred, or how easy it is to shift from one community state to another. It is likely that frequent surface fires in Eurasia kill juvenile Picea before they can reproduce (Black and Bliss 1980), and therefore contribute to maintaining the species composition and fire regime of the region. In fact, there are similarities with an even better studied example of divergent fire regimes within a similar climate: alternative stable state dynamics at forest/ savanna boundaries (Dantas et al 2016, Hennenberg et al 2006, Hoffmann et al 2012, Staver et al 2011, where regions with similar climate either burn frequently (due to a prominent grassy understory in savanna) or hardly ever (due to non-flammable litter and moist microclimate in tropical forests). While mostly studied in the tropics, this process also occurs in temperate ecosystems.
An example of a convergent fire regime, in which regions with very different climates converge on the same fire regime, occurs between pine savannas of the southeastern USA and tropical mesic savannas. These pine savannas are structurally similar to tropical savanna ecosystems, with a continuous grassy understory and an open tree canopy dominated by fire-tolerant longleaf pine (Pinus palustris). Like mesic tropical savannas, surface fires occur every ∼3 years (range 0.5−12) (Stambaugh et al 2011), and without fire they are soon replaced by broad-leaved forest communities that are much less flammable (Kane et al 2008). However, rainfall in long-leaf pine savannas is aseasonal (monthly rainfall never <70 mm), in contrast to tropical mesic savannas where seasonal drought is considered a necessary attribute because it both increases the probability of fire and reduces overall tree growth rates (Cole 1960, Lehmann et al 2011, Scholes and Walker 1993. How longleaf pine savannas with tropical savanna fire regimes persist in a system without seasonal drought is best explained by the traits of P. palustris (Ellair and Platt 2013, Platt et al 2016. Their leaves are 20−45 cm in length, more than twice as long as other pine species (Schwilk and Caprio 2011), and, in combination with a tussock-grass understory, they produce an aerated, flammable fuel bed that increases fire temperatures, duration of heating and combustion of fuels (Ellair and Platt 2013), and allows surface fires to burn during most weather conditions (Fill et al 2012). Thus, two quite different climate systems have converged to produce similar fire regimes through similarities in fuel structure that arise from biological characteristics of the dominant tree species in the USA.
Finally, the most telling biogeographic conundrum for global modellers are the numerous examples where exotic species (with novel plant traits) have invaded a particular ecosystem and fundamentally altered the fire regime, despite climate/weather conditions remaining unchanged (Brooks et al 2004, Pausas andKeeley 2014a). Here plant traits of the invading species are indisputably the factor driving the change in fire regimes (D'Antonio and Vitousek 1992, Fuentes-Ramirez et al 2016. These dramatic changes in fire regime are well described in the literature across a range of systems (Brooks et al 2004), but the conclusion − that including plant traits is essential for describing fire regimes − has not yet been adequately incorporated into Earth system research.
These biogeographic conundrums provide strong evidence for vegetation as a mediator of fire-climate relationships today. But the types of plants present on Earth have not been static over evolutionary time (Belcher et al 2013, Bond andMidgley 2012b), and thus it is necessary to understand both ecological and evolutionary processes when assessing feedbacks between fire and vegetation. Similarly, feedbacks with the Earth system also occur over short (albedo, climate) and longer (nutrient cycling) time scales. With the advent of global meta-databases of plant traits, and global and paleo-ecological datasets on fire, we suggest that it is now possible for evolutionary processes to be incorporated into the study of fire in the Earth system (He and Lamont 2017). Once we understand how plant traits mediate fire regimes and vice versa, we can then incorporate these effects into Earth system models of fire and model their changes through geological time. Below we synthesise current knowledge in this area, describing the pathways by which plant traits both respond to and drive fire regimes.

Ecosystem flammability
Flammability is the propensity of an ecosystem to ignite (given an ignition source) and then propagate a fire. Fire has therefore often been modelled as an infection/percolation process, with flammability being defined as the probability of spread (Cox and Durrett 1988, O'Neill et al 1992, Sullivan 2009). The likelihood of an ecosystem burning depends on both abiotic and biotic properties. These properties vary across temporal scales-from millions of years (O 2 concentrations of the atmosphere, evolution of flammable plant traits), to millenia (vegetation distributions, climate), to decades (fuel amount, community assembly), to days and hours (weather conditions and fuel moisture). Consequently, different disciplines emphasise particular aspects of flammability (Whitlock et al 2010). For example, paleoecologists focus on fuel type and O 2 levels, Earth system scientists emphasise climate, whereas foresters focus on weather and fuels. Current global predictive models of fire tend towards unidimensional indices of vegetatione.g. net primary productivity (NPP)-or none at all (Krawchuk et al 2009, Moritz et al 2012, Spracklen et al 2009, although models linked to DGVMs can define fuels more precisely Morton 2014, Thonicke et al 2010). NPP is used as a proxy for of the amount of fuel in a system, but once a certain minimum level of fuel is achieved in a system, how it burns is controlled more by other fuel properties such as arrangement, density, chemical properties, and moisture content (Schwilk 2015, Varner et al 2015. Conceptually, the flammability of a system is controlled by a number of factors: temperature, wind speed, relative humidity, O 2 concentrations, moisture, and the amount and structure of fuels, each of which can compensate or exacerbate others. For example, it is clear from experimental work that non-flammable fuel types become flammable as O 2 concentrations increase (Belcher et al 2010b), such that even moist forests would have been easy to ignite in the high O 2 atmospheres of the Carboniferous and Cretaceous periods (Belcher and Hudspith 2017). An example at much shorter timescales is the impact that a sudden rise in relative humidity can have by extinguishing a fire burning through a uniform fuel bed (Cheney et al 1993). Moreover, under constant environmental conditions, a change in fuel type can make a non-flammable system flammable (McGranahan et al 2012), or transition a surface fire into a crown fire event (Bradstock et al 2010). Because fire is a percolation process, it tends to show threshold behaviours (figure 5), implying that small changes in the flammability of the system can have large impacts on the area burned and on other fire characteristics (Archibald et al 2012, Cox andDurrett 1988). Thus, changes in plant structure and function that alter vegetation flammability potentially have large impacts on global patterns of fire at time-scales relevant to human management (Pausas and Keeley 2014a) as well as longer timescales (Belcher et al 2010a). Below we describe how fuel properties emerge from plant characteristics, and discuss various points of contention relating to the role of evolution in this process.

Plant traits that impact flammability
There is a large and expanding literature that describes how vegetation traits scale to impact fuels, and ultimately fire regimes ( Relative humidity (%) Spread potential hot-flammable (easy to ignite with high heat release). Which of these flammability strategies dominates in an environment will not only affect the environment for other co-occurring species (via niche-constructione.g. Bowman et al 2017), but will also impact nutrient cycling (e.g. N-volatilisation, emission factors), soil properties (water infiltration), and other biogeochemical processes (Wittkuhn et al 2017).
Flammability is a property of the intrinsic chemistry and structure of plant parts, how they are arranged on the plant (or in the litter bed), as well as ecophysiological properties such as how quickly they cure, and how quickly they regrow (table 1). At a leaf level, the size and density of the leaf, as well as its nutrient content and presence of volatile oils and resins all affect its tendency to ignite (Cornwell et al 2015, Pausas et al 2016) (table 1). The arrangement of these leaves on a canopy (plant architecture) can strongly affect flammability: highly branched canopies with high surface area and low bulk density (g cm −3 ) are better aerated and easier to ignite, although sparse branching may also reduce flammability Ackerly 2001, Simpson et al 2016). Similarly, plants that retain dead leaf material have higher fuel loads and drier canopies and are therefore both more ignitable and have higher heat release (Jaureguiberry et al 2011, Schwilk 2003). Correspondingly, litter bed flammability is controlled by the packing ratio: the arrangement of individual leaves based on their morphology (de Magalhaes and Schwilk 2012, Engber and Varner III 2012, Kane et al 2008, Scarff and Westoby 2006. The different ways that plant traits alter canopy vs litter fuels (table 1) can reinforce the divergent flammability strategies mentioned above. For example, small leaves increase heat transfer through canopy fuels but pack more closely together as litter, and thus retard surface fire spread (Cornwell et al 2015). Traits that make canopy fuels flammable can therefore deter litter fires and vice versa . Moreover, the amount of fuel and its dryness are controlled by plant growth rates, phenology (deciduous vs evergreen) and decomposition rates, that represent the outcome of a range of plant traits and interact with many other aspects of plant ecological strategy (table 1).
Variations in these traits across species can alter ignition probability and fire spread rates by an order of magnitude (Burger and Bond 2015). Many environmental factors interact to select for leaf chemistry, specific leaf area, and plant architecture (Endara and Coley 2011, Wright et al 2004). Therefore these traits are not necessarily the product of selection for increased or reduced flammability, although there is a growing body of evidence to suggest that this can sometimes be the case-see box 3. Whatever the evolutionary mechanism by which different flammability strategies emerge and spread in populations, the impact that they can have on fire regimes is substantial (see box 4).

Plant traits that enable fire tolerance
Any organism occurring in an environment that burns needs to have traits that enable survival and/or regeneration after fires. Traits that determine plant performance in post-fire colonisation and regeneration depend on the particular fire regime (Keeley et al  2011). For example, serotiny (seed storage and postfire release) is common in ecosystems with stand-replacing canopy fires that have a return interval between maturation age and plant longevity (Lamont et al 1991, Table 1. Plant traits that influence flammability at leaf and whole plant levels, and in terms of life history/physiology. Leaf chemical properties impact heat release rate and propensity of fuel to ignite. Leaf geometry and plant architecture can alter bulk density and moisture content which affect how easily fuel ignites and propagates fire. Moreover, traits such as relative growth rates affect how quickly fuels accumulate after a fire, and can impact fire return times. Decomposition rates are key to defining fuel build up: generally more flammable leaves (higher C:N ratio) have slower decomposition rates, therefore flammability is increased at both the leaf level and through time. Such fire-plant interactions have sometimes resulted in convergent evolution: similar sets of traits evolving in different plant lineages exposed to similar fire regimes, so providing compelling evidence that they are adaptations. The mediterranean ecosystems of the world exemplify this. These ecosystems are geographically dispersed and do not share many plant lineages, but fire-related traits (e.g. serotiny, lignotubers, fire-cued flowering and germination) are common across this biome-as are canopy-fuelled fire regimes with moderate ( Fire regimes are generally determined by the few plant species that dominate a landscape, but cooccurring species-whether flammable or not-need to possess the functional traits that enable resistance to, or tolerance of the fire regime. A classic example is tropical savanna ecosystems where frequent fire is a product of the high annual growth rate of grasses that cure rapidly creating an aerated fuel bed enabling rapid fire spread. Woody species contribute little to the fuel, but all savanna plant species possess traits that confer survival under frequent fire. There is evidence that the most flammable members of a community may have disproportionate effect on the flammability of the ecosystem (Blauw et  It has been argued that many of the traits that enable species persistence under various fire regimes could have evolved as responses to other environmental stresses (for example, drought or herbivory) as no traits appear exclusively linked to fire (Axelrod 1980, López-Soria and Castell 1992). However, recent phylogenetic analyses linking the evolution of key fire-tolerant traits (such as bark thickness and serotiny) to changes in fire regimes lend support to the idea that fire has driven the emergence and spread of these plant traits across diverse lineages ( The idea that flammability (or non-flammability) might be acted on by natural selection has been hotly contested in the literature and there is a suite of theoretical, modeling, and empirical studies exploring this idea. Mutch (1970) originally suggested that species with the capacity to persist after recurrent fires might evolve traits that enhance the flammability of their own community. These arguments suffer from being group-selectionist (Snyder 1984), but can be reformulated within the framework of inclusive fitness theory: flammability-enhancing traits could be favoured in individuals because they increase the mortality of neighbours, and this creates space for the offspring of the flammable individual to recruit ('kill thy neighbour' hypothesis, Bond and Midgley 1995). This hypothesis applies quite narrowly to species for which recruitment is fire-stimulated, with soil or canopy-stored seed banks. Gagnon et al (2010) argued that high flammability could also increase individual-level fitness if it resulted in high spread rates: a short residence time would increase survival of below-ground organs and plant propagules. Midgley (2013) criticized these two explanations as being contradictory, but proposed that selection against flammability, i.e. selection for non-flammable plant traits, could be both selected for and evolve.  integrated all these ideas into a framework that shows that while flammability can be subject to natural selection, the type of selection and resultant plant traits depend on the environment and ecological strategy of the species. They identified three fire strategies 'hot-flammable', 'fast-flammable', and 'non-flammable' that include all examples mentioned above, and presented a framework to predict in which environments these strategies should occur.
There are several studies that demonstrate correlations between serotiny and flammable traits in ecosystems and across phylogenies (Schwilk and Ackerly 2001, He et al 2011, Burger and Bond 2015. Moreover, recent experimental evidence supports the idea that different fire regimes can generate within-species variation in flammability , that has a genetic basis (Moreira et al 2014). Considering the growing literature showing that plants can have a role in shaping the niches that they occupy (Laland and Sterelny 2006), it seems odd that there is resistance to the idea that plant traits related to flammability or fire suppression can be anything other than an exaptation. An analysis of all papers published in the last 40 years on this topic (see methods in SI) demonstrates that more papers have been published supporting than detracting from the hypothesis, especially those that make conceptual and empirical advances (table B1). Publications in support of selection for flammability traits are growing, and are moving beyond arguments to include modelling and data-driven tests of the theory. However, it is worth noting that most of these tests (77%) consider only serotiny and flammability, not the evolution of flammability more broadly, and that this burgeoning research field is dominated by a few names (65% of the papers are published by four authors). had a major role in shaping plant traits in fire-prone environments (He and Lamont 2017). Given that plant traits influence flammability, and fire regimes affect the traits of plants, we expect to see correlated evolution between traits that enhance flammability and traits that enable fire tolerance and regeneration. In spruce and some pines dead branch retention produces 'ladder fuels' enabling fires that start in the ground layer to reach tree canopies, facilitating stand-replacing crown fire regimes. Branch retention has been demonstrated to have co-evolved with serotiny, which is an effective strategy in crown fire ecosystems (He et al 2012, Schwilk and Ackerly 2001). Conversely, branch-shedding to prevent crown fires is also common in pine species, and is associated with thick bark and other fire-resistance traits. In the boreal forest (figure 3) the Eurasian larch species (Larix sibirica, L. gmelinii), which drop their branches and burn in surface fires, in many cases survive these fire events due to their high bark thickness. Larch also have longer life-spans than the North American spruce and pine species (Pinus banksiana, Picea mariana) which retain dead branches, are easily killed by fires, and display serotiny (de Groot et al 2013, Rogers et al 2015.

Geophysical feedbacks: short and long timescales
We have demonstrated above how plant traits and fuel characteristics affect the types of fires that occur globally. Because fire regimes differ in their biogeochemical impacts there are also consequences for the Earth system. Here we discuss some of these consequences and the relevant timescales at which they occur. In the short term the impacts of fire can be summarised into impacts on surface albedo, surface roughness, nutrient cycling, aerosols, greenhouse gas emissions, and the carbon cycle. In the longer term (i.e. millions of years), fire is important in regulating phosphorous weathering, geologic carbon sequestration, and atmospheric oxygen concentrations.

Short-term feedbacks
Annual to decadal scale impacts of fire are emphasised in current research on fire, and are the focus of efforts to manipulate fire to influence the Earth system (Landry and Matthews 2016). In terms of the land surface energy balance, fires initially reduce surface albedo due to the production and deposition of charcoal. However, their overall effect is thought to Box 4. Evolution of plant-fire feedbacks at geological timescales.
Over short timescales fire can influence community assembly: the types of plants and plant traits that occur in an environment. These, in combination with atmospheric conditions, climate and soils, influence the fire regime, which over longer time scales can result in evolutionary innovations. Therefore, while fire does not evolve itself, it can respond to changes in the composition of its fuel bed that, in turn, are partly a response to the fire regime to which plants are exposed.
The ability to colonise land created fuel where there was previously none and vastly increased productivity and weathering rates on the Earth's surface. By accelerating weathering, early land plants also reduced atmospheric CO 2 to levels that were compatible with the evolution of leaves, leading to a boom in photosynthesis. This high productivity (and associated organic carbon burial) raised atmospheric O 2 to the point where ignition was possible (Belcher et al 2010b, Scott 2000, and the first vegetation-fuelled fires occurred. Approximately 300 million years ago plants evolved lignin for structural support (Kenrick and Crane 1997). This new form of organic carbon was difficult to decompose so organic carbon was buried, rather than respired, as evidenced by the huge deposits of coal that formed in the Carboniferous period. This further increased atmospheric O 2 (Lenton 2013). Moreover, these large woody structures increased above-ground fuel, so that with more fuels and higher O 2 concentrations the first forests were associated with a dramatic global rise in fire, and possibly the first 'hot-flammable'  fire regimes.
These fire-prone gymnosperm forests have now been pushed to more extreme climatic regimes on our modern planet. About 135 million years ago (MYA), angiosperm species started to invade the understory of gymnosperm forests. These plants had hydraulic pathways that increased productivity and ultimately led them to dominate most environments on Earth today. The rapid spread of angiosperms from 110-70 MYA is associated with a spike in inertinite levels (fossil charcoal) (figure B1)-i.e. a change in fire regime. Bond and Scott (2010) suggest that the angiosperm leaf traits that increase productivity created new fuels that accumulated very rapidly, and probably resulted in frequent, low-intensity fire regimes ('fast-flammable') that the prevailing gymnosperms were ill-adapted to survive (Bamforth et al 2014).
The spread and diversification of angiosperms has led to a range of differing fire regimes. Current forest angiosperms (except eucalypts) tend to have thin, flat leaves with rapid decomposition rates (Cornwell et al 2015, Grootemaat et al 2015. In contrast, gymnosperm leaves decompose slowly, so while gymnosperm litterbeds can accumulate several meters of fire sustaining 'duff' (non-decomposed plant material), in angiosperm forests there is usually very little litter to burn, even when weather conditions are suitable for fire. Interestingly angiosperms also seem to have evolved a unique relationship with fire, with several of the world's most flammable ecosystems being occupied by pyrophytic angiosperms. These include shrubs that have small but thick volatile-rich leaves that carry intense fires, such as chaparrel communities in California or the Cape Flora of South Africa. These pyrophytic plant communities include plants that have smoke-stimulated flowering or seed germination (Keeley et al 2012). Moreover, the expansion of tropical forest in the Palaeogene (Wing et al 2009) generated large areas with little fire, creating moist, non-flammable microclimates (Hoffmann et al 2011), and reducing rainfall seasonality (Lee et al 2012). Amazingly therefore, angiosperms appear to have evolved fire preferences across all fire regimes, from pyrophobic rainforests, to low intensity surface fire regimes in temperate forests, through to some of the most flammable pyrophytic plant communities in the world.
More recently the evolution of grasses (Poaceae) within the angiosperms, and particularly the evolution of open habit tropical savanna C4 grasses, with their fine fuels, low bulk density, rapid regrowth and curing rates have brought frequent and extensive surface fires (Hoetzel et al 2013, Keeley and Rundel 2005 to the planet. While the total biomass burned in these grassy fire regimes is small compared with the conflagrations of the Carboniferious (figure B1), the frequency with which these fires occur has resulted in massive ecological filtering, spreading short-statured grassy ecosystems into vast regions of the world that used to be forest. These fires account for ∼70% of annual burned area today, and are largely fuelled by grasses from one clade of Poaceae-the Andropogoneae. The impacts this had on Earth system processes have not yet adequately been addressed, but presumably they altered above-ground biomass stocks and C, N and P cycling rates, with consequent impacts on regional climates . Less? Figure B1. Changes in the abundance of inertinite (fossil charcoal) as preserved in coal deposits tells a robust and long-term story of global trends in fire-activity over the last 400 million years. The first impression is of dramatic changes over time, linked to the evolution of novel plant traits which fundamentally altered the fuel environment, the fire regimes, and the biogeochemistry of the earth. % charcoal taken from Glasspool and Scott (2010), who present the variations in charcoal abundance in a single depositional environment, in different aged rocks, found globally. be cooling (Landry et al 2015, Ward et al 2012, as the resulting successional dynamics typically include vegetation types that have higher albedos. This will depend strongly on the albedos of the pre-and postfire vegetation and land surface, and the rate at which different species recover after a fire (growth rates)i.e. stronger effects in boreal forest than in grasslands (Kaufman andFraser 1997, Ward et al 2012). Additional short-term land surface impacts result from the reduced surface roughness typical of shorter, more open, vegetation of burned ecosystems. Low surface roughness results in greater wind speeds and warmer surface temperatures (Foley et al 2003), which feed back onto greater flammability at local and regional scales (Hoffmann et al 2002).
Fires deplete local ecosystems of nitrogen (N) and phosphorus (P) (Pellegrini 2016), but these local losses result in regional and ocean fertilisation (Buendía et al 2014, Chen et al 2010, Crutzen and Goldammer 1993. The amount of N and P emitted by fires is globally significant (Vitousek et al 2013) and depends on the stoichiometry of the fuel: live fuel generally has higher nutrient content than dead fuel (due to translocation of nutrients at senescence), leaves have higher nutrient content than wood, and different plant clades (e.g. angiosperms vs gymnosperms) can differ by orders of magnitude in their nutrient content (Han et al 2005). Combustion efficiency (controlled by aeration of fuels and fuel moisture) also affects N loss in fire.
Fire-emitted greenhouse gases such as CH 4 , CO and N 2 O have a warming effect (Ward et al 2012), but fire-emitted aerosols and their precursers scatter (organic carbon) and absorb radiation (black carbon), and the cooling effect is currently thought to predominate (Landry et al 2017, Ward et al 2012. The aerosols emitted also alter cloud condensation and can affect rainfall patterns regionally (Lohmann andFeichter 2005, Tosca et al 2014). Plant traits affect these processes because emission factors (g of particulate matter/chemical compound per kg fuel burned) are strongly influenced by moisture content of the fuel and how aerated it is. Traits such as leaf water content, phenology (deciduous vs evergreen), and architecture will all therefore influence fire emission factors, and consequently, the global energy balance. This is particularly significant in smouldering peat fires because moisture content, bulk density and aeration control the balance of smouldering combustion.
The combustion flux of CO 2 through fire is a major component of the carbon cycle in the short term-contributing to year-to-year variability in carbon storage (van der Werf et al 2006). However this carbon is quickly taken up again by regrowing vegetation, so fires will only alter the global carbon cycle if there is a mismatch between rates of burning and regrowth. It has been argued that a focus on fire carbon-fluxes from vegetation to atmosphere is misdirected, and that even at short timescales it is more relevant to consider how fire mediates fluxes from the passive soil carbon pool to the atmosphere (Landry and Matthews 2016). Anthropogenic burning of coal is a key example, but wildfires can affect the soil carbon pool negatively by burning soil carbon and peat (Giglio et al 2010, Turetsky et al 2011, and positively through char production (passive black carbon (Santín et al 2016)). The amount of biomass above-ground, and the amount of peat in the soil, are a function of both vegetation growth rate and decomposition rate (Grootemaat et al 2015). Therefore key plant innovations that have resulted in altered fire regimes (see box 4) are often associated with novel photosynthetic pathways/leaf anatomy Scott 2010, Bowman et al 2009) that will affect both photosynthetic rates and decomposition rates.
It is evident at pluri-millenial and millennial timescales that fire-climate feedbacks are vegetationdependent. In the extra-tropics biomass burning (figure 6(c)) shows the same increasing trend since the last glacial maximum as CO 2 (figure 6(a)), CH 4 (figure 6(b)) and temperature (figure 6(f)) (∼20 000 years ago to present) (Carcaillet et al 2002, Power et al 2008. This is probably due to higher temperatures increasing the opportunities for burning. However, grassy tropical and sub-tropical systems show an opposite trend (figure 6(d)) with peaks during periods of low CO 2 , CH 4, and temperature-i.e. glacial periods (Daniau et al 2013, Haberle and Ledru 2001, Nelson et al 2012, van der Kaars et al 2000, Wang et al 2005. The hydrology (Daniau et al 2013) and low CO 2 (Bond and Midgley 2012a) during glacial periods increases grassy fuel loads. Thus, despite low temperatures, tropical biomass burning increases. Thonicke et al (2005) illustrate how a change in the type of fuel can modify fire-Earth system feedbacks. Using a DGVM coupled to a fire model they demonstrate that the expanded grasslands during the last glacial maximum would have increased area burned in the tropics and sub-tropics and reduced it in higher latitudes (corroborating the charcoal data from figure  6). Although total fire emissions were not substantially different from current day, the NO x emitted in the tropics by grassy fires could have had a significant impact on the oxidative capacity of the atmosphere and might explain why atmospheric CH 4 was so much lower than it is today.

Longer-term feedbacks
Over longer (geological) timescales the fire-feedbacks on radiative forcing and atmospheric CO 2 are dwarfed by feedbacks related to the geochemical cycling of atmospheric O 2 (Mills et al 2016). Lenton (2013) suggests that fires are essential for maintaining O 2 within the relatively stable bounds (∼16% to ∼35%) necessary to sustain life on the planet for 100s of millions of years. When CO 2 is reduced to organic carbon through photosynthesis, O 2 is released. If parts of this reduced organic carbon escape consumption by heterotrophs it is not respired again and when buried in soils or sediments it represents a net flux of O 2 to the atmosphere. Weathering can draw down this atmospheric O 2 , but it goes to completion above ∼20% O 2 (present day concentration) meaning there must be some stabilising feedback such that as O 2 concentrations increase, rates of carbon burial decrease. Two main hypotheses have been proposed, both of which involve feedbacks with fire and their redistribution of phosphorus.
Fire is very sensitive to O 2 concentrations (see figure 5(a)). At 30% O 2 vegetation as moist as 80% dry weight will ignite (Watson and Lovelock 2013), which implies that even tropical forests could readily carry a fire. Fire exclusion experiments demonstrate that ecosystems which burn often have less above-ground vegetation than climate/soils can support -i.e. increased flammability results in less above-ground biomass. Therefore, under high O 2 we would anticipate more frequent and intense fires (Belcher and Hudspith 2017) that would suppress large land plant biomass, slowing biological weathering rates, and the rate at which key nutrients like phosphorus are released from rocks (Lenton 2001). This is because plant roots actively mine soils for phosphorus, and can increase weathering rates by an order of magnitude (Quirk et al 2012). This decline in phosphorus supply from the land to the ocean tends to limit carbon burial over the long-term (Kump 1988, Lenton 2001. Therefore, as O 2 increases fire activity feedbacks kick in causing a decline in terrestrial NPP that ultimately lowers the weathering flux of the nutrient phosphorus, slowing carbon burial, the long-term oxygen source. Plant traits that would affect this key regulating process include stoichiometry (C:P ratios), root characteristics, and all traits that alter flammability and fire behaviour in ecosystems. Currently fire probably limits forests to approximately half of their potential area . We know that higher O 2 concentrations would reduce this further, but the consequences for weathering rates and phosphorus cycling are yet to be quantified. Recent evidence for these proposed mechanisms links major carbon burial events (known as ocean anoxic events) to increases in atmospheric O 2 and fire (Baker et al 2017). As such, fires clearly respond to O 2, and higher resolution studies, combined with models of the final stages of these major disruptions to the carbon cycle, should be capable of exploring the nature of the fire-feedback to rising O 2 .

Biological vs geophysical feedbacks and geoengineering
Given that global fire regimes are the outcome of both biological and geophysical processes, understanding the impacts of fire over time requires us to understand these feedbacks and the time scales at which they work. A conceptual framework that summarises these ideas is presented in figure 3: 1. plant traits aggregate to produce fuel properties that drive ecosystem flammability (bulk density, fuel structure, packing ratio, fuel moisture, stoichiometry); 2. ecosystem flammability interacts with climate and weather to affect fire regimes; 3. this is expected to feed back on plant traits in the community both by ecological filtering of species, and through evolution by natural selection (e.g. serotiny and branch retention associated with crown fire regimes); 4. these processes help explain current global patterns of fire, vegetation, and climate (including biogeographic conundrums); 5. the fire regimes that emerge from the climate and vegetation traits present will, over short time scales, impact atmospheric CO 2 levels, surface albedo, and aerosols-which in turn influence atmospheric properties, weather, and plant trait distributions; 6. over longer time scales, carbon burial rates, weathering rates of phosphate-bearing rock, and atmospheric O 2 are all impacted by fire regimes, and this influences the biological and geophysical processes mentioned above.
The relevance of the conceptual model presented here becomes clear as we increasingly consider managing Earth system processes through geoengineering. In the boreal forests, Girardin et al (2013) demonstrated that vegetation feedbacks with fire would dampen the impacts of changing climates on the fire regimes if spruce was replaced by deciduous broad-leaved species because it would disrupt the feedbacks between increased temperature and more frequent/intense crown fires. Strategic manipulation of fire through fuels is clearly feasible (Hirsch et al 2001, Krawchuk and Cumming 2011, Terrier et al 2013, and comprehending the full impacts of these interventions would rely on accurately understanding the feedbacks presented in figure 3. The potential for mitigating high atmospheric CO 2 levels by manipulating fire regimes is similarly being explored in several other regions (Grace 2011, Russell-Smith et al 2013). Information presented here demonstrates how this short-term intervention in the carbon cycle needs to be assessed against alterations of the surface energy balance via albedo and aerosols, as well as considering the inevitable biological feedbacks that will occur.

Research challenges: the role of fire in the Earth system
Research on fire in the Earth system has not yet provided the cross-scale and quantified insights needed to predict future fire regimes when novel ecosystems will undoubtedly emerge, and elucidate the role of fire in shaping the Earth system. We work in a world of unprecedented access to data. The overarching challenge here is to reconcile the scales and types of data to produce useful insights. Our research challenges require interdisciplinary collaboration but also centre on how bringing new data and ideas to a field can replace our understanding of the place of fire in the Earth system.
Research challenge (RC) 1: Linking traits to ecosystems (up-scaling). We need models describing how the fuel properties that define ecosystem flammability emerge from the combination of plant traits. These can be combined with information on species distributions to produce spatially explicit maps of fuel properties.
RC2: Biological vs geophysical drivers. Defining the relative roles of fuel versus weather in a changing climate will be critical to predicting future fire, and to managing unwanted changes in the fire regime. Information on fuel properties (above) will enable modellers to move beyond simple measures of NPP to assess the complex and non-linear responses of fire to changes in climate and species composition associated with global change.
RC3: Evolutionary history. Meta-analyses linking plant phylogenies/chronograms, functional traits and patterns of fire (from remotely sensed imagery or fossil charcoal) can be used to examine the influence of fire on the evolution of plants and to test biogeochemical models of O 2 . This requires collaborations across a wide range of research areas.
RC4: Biogeochemical feedbacks. It is becoming clear that geochemical models need to incorporate variations in fuel properties as well as fuel amount and atmospheric properties, as these interact to determine both short and long term C cycling feedbacks. Information from RC3 above can be linked with modelling and fossil based studies to improve our understanding of evolutionary changes in vegetation, climatic drivers and regulation of O 2 through time.

Implications
Human impacts on fire include ignition and suppression, connectivity of landscapes, alteration of geophysical drivers such as temperature, and alteration of biological drivers such as species composition and fuel structure. Often it is the impact on fuels that is most poorly understood and managed. The Anthropocene is a time of unprecedented biotic mixing, where novel plant trait combinations and ecosystems are emerging. Any prediction of future fire regimes that uses climate alone would paint a disturbing picture of wholesale increases in fire across all regions, but we know the reality is more complex than that-largely because the response of plant traits and plant communities to global change can act to both increase and decrease vegetation flammability. Moreover, focus on one or a few time-scales when assessing Earth system feedbacks can mask important processes and result in inappropriate interventions. A longer term view would help to separate transient dynamics from major trends in Earth system drivers. Understanding current patterns of fire and vegetation, as well as patterns of fire over geological time, requires research that integrates evolutionary biology, ecology and the biogeosciences.