4 Discussion

Ecoregions within a given biome have similar fuel moisture thresholds (EFT) associated with observed landscape fires (Figures 1, 3). There is, however, significant variability in the biogeography of EFTs (Figures 2a, 3), which suggests that a single fuel moisture threshold cannot be interpreted as a universal determinant of flammability. This global variability is sensibly explained by eco-climatological variables like precipitation and temperature. In addition to contextualizing the EFTs in relation to these eco-climatological variables, they need to be considered in concert with the inflection point slope estimates (IPSEs) and fire activity as outlined below. These considerations are important in understanding the likely effects of climate change on fire regimes, as well as determining those ecoregions and biomes where the risk of fire activity is increasing. We propose identification and use of these EFTs is critical to understanding the future of global pyrogeography.

4.1 The biogeography of flammability thresholds

The global averages for our EFTs correspond closely to the extreme values of between 8 and 12% identified or used in prior research (Wotton, 2008; Slijepcevic et al., 2015; Flannigan et al., 2016; Nolan et al., 2016; Boer et al., 2017; Filkov et al., 2019; Ellis et al., 2022). This suggests that those commonly used EFTs as derived from the Canadian FWI system and ERA5 reanalysis data do, on average, reflect ecological switches controlling fire ignition and spread for many Earth environments. Despite this, relying on a global average oversimplifies local fuel-fire relationships and ignores our understanding of how the fibre saturation point can vary between landscapes and vegetation types (Fernandes et al., 2008; Alvarado et al., 2019). Biome-level differences in EFTs suggest that a generalized threshold will often under- or over-estimate the localised fuel moisture-fire relationship, with nearly 70% of the Earth’s surface holding an EFT above or below that range of 8 and 12%. The geographic distribution of EFTs (Figure 2a) as well as the differences between and within distinct biome EFT distributions (Figure 3) highlight where this oversimplification fails to capture the true association between fuel moisture and fire. This includes, for instance, all boreal forests (Q1: 13.9%, Q3: 16.6%) and tundra (Q1: 17.7%, Q3: 19.8%) ecoregions, as well as those North AmericanPinus banksiana and Pinus contorta forests where the FWI system originates (Van Wagner, 1987). For ecoregions or biomes such as these, the general threshold grossly under-estimates the true fuel moisture content shown to be associated with fire (Figure 3).
The need for biogeographically-variable EFTs is logical given the role water availability – including the moisture content of both live and dead fuels – plays in fuel accumulation within different types of fire regimes (Murphy et al., 2013). A universal threshold oversimplifies the distinct biology of vegetation and how vegetation in different ecoregions may have evolved with fire. A bespoke threshold can more accurately capture the existing variation in live and dead fuel moisture content associated with the phylogeographic and structural dimensions of vegetation and the fuel array (e.g., Pausas and Keeley, 2009; Keeley et al., 2011; Alvarado et al., 2019). For example, fuel-limited biomes like desert or tundra maintain arid conditions that prevent the growth of burnable biomass. Moisture-limited temperate forest systems, on the other hand, experience years- to decades-long lagged growth typical of the classic negative exponential fuel accumulation curve (i.e., Olson, 1963), while the most productive tropical forests tend to retain enough water that they rarely burn (Murphy et al., 2013). For these moisture-limited biome types, fuel moisture plays the key role in whether the landscape can ignite and maintain a wildfire.

4.2 Interpreting the ecoregion flammability threshold

The EFT method was applied using global scale data and is therefore an ecological generalization. Although there is variability within each ecoregion due to terrain, vegetation, and fuel structure, EFTs are still biologically meaningful across ecoregions and should be interpreted asrepresentative thresholds. That is, when the average DFFMC nears or falls below our identified EFT for a given ecoregion, that reflects an average state of susceptibility across that landscape rather than a strict measure of the moisture content of all fuel particles. It’s also important to acknowledge some EFTs are inaccurate or imprecise. Sources of error include the short length of the underlying remotely sensed fire records and the differential importance of fuel moisture as a fire trigger across the productivity gradient (Figure 5a-c). For instance, the modelled P(BA < DFFMC) relationships for some ecoregions feature gradual shifts in the cumulative proportion of burnt area over the full range of DFFMC (supporting information). In other cases, we identified EFTs well above the upper limits of the fibre saturation point (i.e., > 30%: Fernandes et al., 2008). Both these cases could reflect a lack of available fire data within that ecoregion, differences in local land-use practices (e.g., savanna or agricultural lands: Le Page et al., 2010; Andela et al., 2017), that the role of fuel moisture for fire ignition and spread in that ecoregion is less important (e.g., Alvarado et al., 2019), or a combination of these factors. Furthermore, the ERA5 grid scale resolution (0.25°) may be too coarse to accurately capture and differentiate the climatology of small ecoregions from surrounding regions. This includes most flooded grasslands and savanna and tropical and subtropical coniferous forests given their small size, number, and limited geographic range (supporting information).
Our results show that eco-climatic factors explain a large proportion of the variation in the EFTs amongst the ecoregions (Figure 4a-b, Table 1 and supporting information). Overlaying biome means onto the NMDS ordination, for example, shows that the lowest EFTs in desert environments (Q1: 4.2%; Q3: 8%) are strongly associated with precipitation seasonality and low overall precipitation variables. The highest EFTs in tundra environments (Q1: 17.7%, Q3: 19.8%) are notably associated with high percent herbaceous cover and lower temperatures. Higher-latitude temperate forests – particularly those with higher EFTs like boreal forests (Q1: 13.9%, Q3: 16.6%) and coniferous forests (Q1: 9.3%, Q3: 17.2%) – are associated more strongly with a combination of precipitation, temperature, and seasonality. It is important to note, however, that the precipitation NMDS axis includes the effects of NPP and percent tree cover, which conforms to pyrogeographic theory between productivity and global fire activity (Pausas and Bradstock, 2007; Pausas and Ribeiro, 2013; Jones et al., 2022). For example, low-productivity deserts and xeric shrublands feature the most extreme combination of EFTs, IPSEs, and productivity on Earth (median of 0.17 t C ha-1 year-1; Figures 5b-c). However, fire activity in these environments is controlled by intermittent periods of high productivity rather than moisture content (Archibald et al., 2009; Bradstock, 2010; Kelley et al., 2019).
The relationship between fire activity, fuel moisture, and fuel availability invites consideration of climate change. Climate change can manifest itself directly (via fuel moisture) and indirectly (via production of phytomass). Those environments with abundant fuel and higher EFTs are often those most vulnerable to climatic change. Tropical and subtropical rainforests, for example, feature high moisture and high fuel and are at risk of catastrophic change, largely driven by human land-use impacts on fire regimes (Le Page et al., 2010; Canadell et al., 2021). Mediterranean forests are among the least productive forests in the world (Figure 5a-c) and could shift towards savanna or desert environment as drying trends continue and leave fuel accumulation unable to keep up with increasing fire intensity (Pausas and Paula, 2012; Pausas and Bond, 2020). The most arid desert landscapes rarely support enough fuel for fire spread (Bradstock, 2010; Murphy et al., 2013; Bedia et al., 2015), and are unlikely to have fire regimes driven by anthropogenic climate change except along biome transition lines (Archibald et al., 2009; Senande-Rivera et al., 2022). At the upper limits of the intermediate fire-productivity zone, the most important intersection supports higher productivity (> 4.5 t C ha-1 year-1), higher fire activity (> 0.5 FAI), and extreme values of both EFT and IPSE (Figures 5b-c). This area reflects some of the most at-risk environments under climatic change, including many higher-latitude temperate broadleaf and boreal forests susceptible to ecological collapse under drying climate trends (Ellis et al., 2022; Senande-Rivera et al., 2022), as well as more productive, moisture-limited tropical savanna ecoregions (e.g., Alvarado et al., 2019).
One important advance our identified EFTs provide is the elimination of the persistent uncertainty in defining fire season onset – a key pyrogeographic parameter that further defines fire regimes and pyromes. At the continental scale, for example, Australia’s fire-prone tropical and temperate ecoregions feature fire season onsets driven by a distinct latitudinal climate gradient (Murphy et al., 2013; Williamson et al., 2016). The evident drying trend in fuel moisture for wetEucalyptus forests along this gradient places those ecoregions at risk of a potential collapse due to increasing fire frequency (Bowman et al., 2014; Furlaud et al., 2021; McColl-Gausden et al., 2022). Worldwide, boreal forests, Mediterranean forests, and both temperate and tropical broadleaf and coniferous forests are all at risk due to increasing burnt area, fire frequency or both (Westerling, 2016; Forkel et al., 2019; Kelley et al., 2019; Abatzoglou et al., 2021; Ellis et al., 2022). Our EFTs can be used to detect the onset of a fire season in real time in environments like these, which is of primary importance to tracking the effects of climate change on different landscapes’ fire activity, as well as determining the allocation of forest management and firefighting resources.
Specific EFTs will be useful in understanding changes in fire seasons for individual ecoregions, biomes, and specific vegetation types globally. The identification of changing length, intensity, and extremes of fire seasons using meteorological data has been an important line of evidence of changing global fire risk. However, previous analyses have used generalised or assumed thresholds to which this study provides a key innovation. Additionally, the expansion of the active fire season is not just occurring at the seasonal boundaries, but into the night-time – a period previously assumed to provide relief from rapid wildfire spread (Balch, et al., 2022). To advance our understanding of shifting fire regimes under anthropogenic stressors, identifying the biogeography of EFTs as we have done is a prerequisite for defining the bounds of the fire seasons, as well as conducting trend analyses sensibly informed by the local fuel-fire relationship. Such trend analyses can be used to identify and mange those most at-risk environments for ecological collapse under predicted future fire regimes.