Multi-decadal increase of forest burned area in Australia is linked to climate change

Trends in area burned

At a continental scale, total annual burned area (fire year defined as July to June to include the Austral summer of December to February) using the NOAA-AVHRR dataset (“Methods”: Burned area data), significantly increased over the past 32 years albeit with large interannual variability (Fig. 1a; Linear fit, p value = 0.04, Supplementary Table 1). The high variability is in part driven by large-scale modes of atmospheric and oceanic variability such as El Niño Southern Oscillation (ENSO) and the Southern Annual Mode^31,32 that influence fire weather conditions^16,22. Nine out of the 11 fire years, each with more than 500,000 km² (>50 Mha) burned, occurred since 2000.

Forest ecosystems also show increased burned area over time (Fig. 1b, linear fit, p value = 0.02, Supplementary Table 1; Fig. 2). The increasing trend is statistically significant with and without the 2019 fire year, indicating a robust increasing trend even before the extraordinary large burned area of that year (Supplementary Table 1). Forests in Australia experienced an annual average increase of 350% in burned area between the first (1988-2001) and second (2002-2018) half of the record, and an increase of 800% when including 2019. The 2019 fire year burned about three times (60,345 km²) the area of any previous year in the 32-year AVHRR-Landgate record (Fig. 3, Supplementary Fig. 1, “Methods”: Burned area). The burned area of the 2019 fire year was estimated at 71,772 km² based on State and Territory agencies (NIAFED) and 54,852 km² based on NASA-MODIS, with an average for the three products of 62,323 ± 8,631. Ten out of eleven fire years with at least 5000 km² (>0.5 Mha) burned have occurred since 2001. These trends are broadly consistent across the three burned area products (Supplementary Fig. 1).

We find a positive exponential trend of burned area in forest ecosystems during the cool season months of the Austral autumn and winter (March to August), with a mean growth rate of 14% yr^-1 (Fig. 1c; exponential fit, p value = 0.001; 9% increase without the 2019 fire season; Supplementary Table 1). These results indicate an extension of the fire season into the cooler months of the year, with more than a five-fold increase in the annual mean burned area in winter and three-fold increase in autumn between the first and second half of the studied period. However, spring and summer contributed about 10 times more to the increase in burned area than autumn and winter (Fig. 2; Supplementary Figs. 2 and 3). All seasonal linear trends were highly significant (p value < 0.001), and with p value = 0.07 for the summer season. The seasonal fractions of burned area for the 32-year period are 66% in summer, 24% in spring, 7% in autumn and 3% in winter.

Along the latitudinal increasing temperature gradient from South to North (Tasmania, Victoria, New South Wales, and Queensland), we find that the largest relative growth of burned area between the early (1988–2002) and later (2003-2019) periods of the record occurred in the southern coolest parts of the forest distribution (Tasmania) and the northern warmest parts of the forest distribution (Queensland), with Queensland showing the largest absolute difference of the two.

The AVHRR-Landgate data predominantly detects wildfires and misses low-intensity fires, including most prescribed (planned) burning³³, suggesting the trends presented here are very unlikely to be affected by changes in prescribed burning (“Methods”: Burned area data). This result is further corroborated by the lack of trends in the annual area of prescribed burning (see section below Fire risk factors and fuel load).

Fire histories compiled from State and Territory agencies (“Methods”: Burned area data), suggest that the extent of burned area for the 2019 fire year was also unprecedented since the 1930s when most agencies began to collect annual records (Fig. 3). Out of four forest megafire years (defined as the top 10 percentile years with most burned area) that have occurred since 1930, each with over 10,000 km² (1 Mha) of burned area (1930, 2002, 2006, 2019), three occurred since 2000. Historical records show that no other forest megafire years occurred during the 1900s but that a large fire year occurred in Victoria in 1851³⁴. Further disaggregation into States and Territories, and wildfires and prescribed burning is shown in Supplementary Fig. 4.

The increasing trend in annual burned area, and the exceptional 2019 fire year, become all the more significant against a concurrent diminishing of forest extent available to burn due to land clearing for pasture and agriculture³⁵, and increasing bushfire firefighting capacity in Australia^36,37.

Years since the last fire

We further investigate trends in increasing burned area with an analysis of the number of years since the last wildfire (YSLF). The longer time series of the fire history records from State and Territory agencies enable us to construct a gridded database of wildfires at 250 m resolution to analyze decadal changes in the number of YSLF for all pixels that have burned at least once (“Methods”: Years since the last fire). The analysis shows that 48.8% of all forest area has burned at least once since the 1930s. The burned fraction of the total area for each of the forest classes consistent with the Australian National Forest Inventory (“Methods”: Forest extent and types) varied widely: eucalypt low-forest 61.1%, eucalypt tall-forest 60.6%, eucalypt medium-forest 58.2%, coastal non-eucalypt 54.3%, and rainforest 11.1%. Thus, YSLF estimates presented here do not represent a fire frequency for specific region or forest type, but they are correlated, as they have been estimated by a combination of observations and expert knowledge elsewhere⁹.

For context, some of the dominant vegetation types such as tall eucalypt forests have typical fire intervals of 20 to 100 years with extreme cases with more than 100 years⁹. Lower stature eucalypt forests have typical fire intervals between 5 and 20 years with extreme cases with a range between 20 and 100 years⁹. Tropical forests have fire return intervals above 100 years⁹.

The YSLF declined over the past four decades with decadal means (±SD) (all forests that had burned at least once since 1930s) of 70.6 ± 1.1 years, 68.1 ± 1.0 years, 53.2 ± 7.8 years, and 39.8 ± 8.1 years for the decades of 1980s, 1990s, 2000s, and 2010s, respectively (Fig. 4; continental figures in Supplementary Fig. 5). YSLF values for the last four decades of the record are the most robust due to the large amount of data from which they are estimated (cumulative), and therefore are the focus of this analysis. Fig. 4 reveals areas with rapidly declining YSLF, with some regions in Victoria with YSLF below 20 years. At the State and Territory level, the decade of 2010s shows a range from 46 ± 21 in Queensland to 33 ± 4 in Victoria, with a minimum of 18 ± 4 years for the Australian Capital Territory whose forests were largely burned in 2003 and again in 2020 (Supplementary Table 2).

Fire risk factors and links to climate change

There are four components that drive fire activity: weather that influences fire spread, ignition sources, biomass accumulation (fuel loads), and dryness of fuel loads (fuel availability)^2,7. In the next three sections we address each of these four factors and show trends of various indices and quantities which we link back to changes in fire risk over the past 40 years.

We use a daily dataset of McArthur Forest Fire Danger Index (FFDI) values gridded throughout Australia¹⁵. This index is specifically developed to indicate weather conditions across the “Danger” spectrum associated with wildfires in Australian forests³⁸. It integrates a combination of key weather factors known to influence the severity of wildfires including wind speed, humidity, temperature, and fuel moisture deficit, which is calculated from antecedent rainfall and temperature.

Multi-decadal changes towards more dangerous fire weather (as indicated by FFDI) are occurring for most of Australia (Fig. 5a). The largest increases occur in the southeast, including in association with increased temperatures (Fig. 5d) and decreased rainfall (Fig. 5e) in that region, as others have reported before^2,5,15,39. Temperature influences fire behaviour in a number of ways. First, through relative humidity (vapour pressure deficit) and its effect on the absorption of energy by the atmosphere and on the moisture content of vegetation. Second, by influencing atmospheric stability based on temperature lapse rates and humidity profiles, which influences the risk of extreme wildfire events including those that generate new thunderstorms^20,21,40. Third, through its contribution to the occurrence of heat waves¹⁸, and fourth by influencing the speed and intensity of drought development^41,42, along with other factors such as wind speed, specific humidity and soil moisture.

Over the last four decades, a trend has been observed towards drier conditions during the cooler months of the year (e.g., April to October) in many parts of southeastern Australia (Fig. 5e for annual change). This has been associated with a strengthening of the subtropical ridge and a decrease in rainfall from fronts and cyclones^13,14,43. The observed trends are consistent with projected future declines in cool season rainfall in the 21st century from global climate models, and are stronger than those produced in historical simulations for the period 1950-2005^44,45. Reduced Austral spring and winter rainfall influences the fuel moisture leading into the fire season, with drier fuels in spring being associated with a trend towards an earlier start to the fire season particularly in southeast Australia in recent decades^15,16,24.

In addition to the FFDI, which is based on weather conditions near the surface, we used the Continuous Haines (C-Haines) index⁴⁶ to represent dangerous weather conditions for wildfires based on higher levels of the troposphere (“Methods”: Fire weather indices and climate/weather variables). The C-Haines index is used to indicate a high chance of extreme wildfires such as those with thunderstorm formation in fire plumes (generally referred to as pyrocumulonimbus cloud, termed pyroCb)^21,40,46. PyroCbs are associated with extreme dangerous fire behaviour, including erratic and strong gusty winds near the surface and generating lightning in the fire plume that can ignite additional fires far ahead of the main fire front^47,48.

Long-term changes in the C-Haines Index have occurred over the last four decades in many parts of Australia (Fig. 5b), with higher values representing more dangerous conditions for extreme wildfire events, particularly over forest ecosystems of southeast Australia. This is consistent with previous studies that have demonstrated significant increases in C-Haines Index in recent decades, particularly for regions of southern Australia⁴⁰. High C-Haines index occurred in previous wildfire disasters in southeast Australia including during the Black Summer fires in 2019/2020, the Black Saturday fires in Victoria in 2009, and the Canberra fires in 2003^40,49. Often high C-Haines values are accompanied by high FFDI values (Supplementary Table 3), but there are exceptions with a notable case for the Canberra fires in 2003 where the C-Haines had a higher percentile value than FFDI, although both were high⁴⁰.

Modelling shows that increasing anthropogenic greenhouse gas emissions can increase the C-Haines in some regions including southeastern Australia^20,21. These climatological changes are primarily associated with an increase in the moisture-related component of the C-Haines (representing increased dewpoint depression at 850 hPa) rather than its stability component as seen based on historical reanalysis data⁴⁰ and projections of future climate²⁰. Increased dewpoint depression (and similarly relative humidity or vapour pressure deficit) is primarily caused by increased temperatures, but countered to some degree by an increase in atmospheric moisture content that is as expected consequence of a warmer world⁵⁰.

Fire ignition factors in Australia include multiple anthropogenic and natural sources. The primary natural cause is lightning, particularly that known as “dry lightning”, which occurs when the lightning is not accompanied by substantial rainfall. A threshold of rainfall of about 2.5 mm is used to define dry lightning. Rainfall less than this quantity on a day with lightning is associated with a higher than average chance that the lightning will result in a wildfire⁵¹. Lightning can be responsible for more than 50% of the area burned in some regions of Australia including the temperate forest regions of southeast Australia⁵². Although lightning observations are not available in a homogenous form over many decades, environments associated with dry lightning events show an increased frequency of occurrence in recent decades in near-coastal regions of southeast Australia (Fig. 5c)⁵³. Specifically, there is an increase of about 50% in the frequency of occurrence of those environments for the recent period of 2000–2016 compared with the previous two decades of 1980–1999.

Burned area and fire weather

To understand the influence of fire weather on the observed trends in burned areas in forest ecosystems, we use uni- and multi-variate regression analyses with the fire risks factors of FFDI ≥ 25, FFDI ≥ 50, C-Haines and dry lightning as predictor variables. Note that we find no significant relationships with biomass variables (see section Fire risk factors and fuel loads, and Fig. S7), and therefore they are not included in this analysis.

We found both FFDI equal or exceeding 25 (Very High fire danger classification starts at FFDI ≥ 25) and equal or exceeding 50 (Severe fire danger classification starts at FFDI ≥ 50) to be significantly increasing over the past four decades for the forest areas (Fig. 6a,b; linear fit, p value = 0.01, Supplementary Table 1). The five years with most severe or greater fire danger (FFDI ≥ 50) all occurred since 2002 (2002, 2006, 2009, 2013, 2019).

For the last 32-year period (1988–2018, section Trends in burned area), burned area in forest ecosystems increased exponentially with the number of days in which FFDI was equal to or greater than either 25 or 50 (Fig. 6c,d, exponential fit, p values < 0.001; Supplementary Table 1; Supplementary Fig. 6 for exponential fits). These trends equate to 21% increase in the burned area for every additional day of FFDI ≥ 25, and about 3 to 5 times increase in the burned area for every additional day of FFDI ≥ 50.

Using a multivariate regression analysis, we show that FFDI is a strong predictor of burned area for both FFDI ≥ 25 and FFDI ≥ 50, as FFDI alone is able to explain more than two-thirds of the variance in both models (Supplementary Table 3). Including C-Haines and dry lighting into the regression model improves the explained variance only marginally, most likely due to fact that these two variables are highly correlated with FFDI (Supplementary Fig. 7).

We use the exponential regression of burned area against days of very high FFDI ≥ 25 for the period 1988–2018 (Supplementary Table 1, regression no.8) to test whether it could have predicted the record burned area of the exceptional 2019 fire year (June 2019 to July 2020). The regression model predicts a total area burned of 143,151 km² against the three estimates available (71,772 km² Agencies-NIAFED, 60,345 km² AVHRR-Landgate, 54,852 km² NASA-MODIS) (“Methods”: Burned area data). The regression model was not developed as a predictive tool for burned area under future climate conditions, however, it clearly predicted a record megafire year for forests for the 2019 fire year, as occurred. These results suggest the very likely possibility of further increases in burned area in response to the predicted higher FFDI values under future warming scenarios^2,6,20,21,54.

Fire risk factors and fuel load

In addition to weather and climatic conditions, and the availability of ignition sources for a fire to occur, fuel amount, structure, continuity and condition are also key components of fire risk^26,29. The availability of fuel plays a key role in determining the intensity (peak energy output) and severity (impacts related to the amount of canopy scorch, canopy loss, tree death, biomass loss) of fires. Understanding the role of fuel loads in fire activity is important as it also provides the knowledge base to reduce fire risk through the management of fuel quantity and distribution in the landscape^26,30. Fuel loads are composed of aboveground biomass (particularly leaf biomass) and litter, including coarse woody debris (e.g., dead branches, logs, standing dead trees) and fine litter (eg, leaf, twigs).

There is no continental forest observatory available that tracks fuel loads which could enable a similar analysis to the one above for climate/weather factors and burned area. However, regression analyses with modelled fuel loads show no statistically significant relationships with burned area (Fig. S7). This result does not rule out the possible role of fuels in influencing burned area, but it is the reason for not including fuels in the multi-variate analysis above.

Here we analyze two datasets that characterize processes influencing fuel loads and therefore fire risk: (1) changes in the area burned by hazard reduction fires; and (2) the influence of climate change and increasing atmospheric CO₂ concentrations on biomass production and fuel loads. We hypothesize that changes in trends leading to reduced fuel loads will reduce fire risk and burned area, while the opposite will hold true for changes leading to increased fuel loads.

First, we address hazard reduction burns, also known as prescribed fires, planned fires or fuel reduction burns, including cultural burns by indigenous people. These fires are deliberately human-lit during the cooler months of the year leading to reduced fuel amounts. They have multiple purposes including the prevention or reduction of damage to human life and infrastructure, biodiversity conservation, reducing future fire intensity and increasing hunting for First Nations communities. We extracted the prescribed burning areas in forest ecosystems over the past 32 years from the State and Territory fire history databases (“Methods”: Burned area data). Prescribed fires burned an average of 3071 ± 732 km² per year, or about 1% of the current area of forest ecosystems. There is large inter-annual variability due to the year-to-year variability in suitable weather conditions to conduct prescribed burns, but the data show no trends over the past three decades (1988–2018) in forest ecosystems (Fig. 7).

Second, we address the influence of climate change and the associated accumulation of anthropogenic CO₂ emissions in the atmosphere on biomass production and fuel loads through changes in temperature, rainfall and the elevated CO₂ effect on vegetation growth (i.e., the CO₂ fertilization effect)^55,56.

Here we use a highly parameterized and benchmarked biospheric model for Australia (CABLE)^57,58,59 for the estimation of fuel types and loads⁶⁰. CABLE was forced with observed CO₂ concentration and climate over the past four decades, but does not include the effects of fire and other disturbances, hence we refer to the simulated rates as the potential rates of fuel production in the absence of fires. The simulation shows a small positive trend of potential aboveground biomass and coarse woody debris (Fig. 8a, b; Supplementary Table 4), but a declining trend of fine litter associated with declining canopy leaf biomass over Australian forests (Fig. 8b). This declining trend could be associated with the reduction in rainfall and soil moisture of the past four decades (Fig. 5e, f). In addition to declining fine litter and its associated fire risk, we find a decrease of rainfall and soil moisture down to 23 cm depth over the past 40 years, as seen in the historical record (Fig. 5e) and simulated by CABLE-BIOS as a proxy for reduced water availability at the forest floor (Fig. 5f) (see “Methods”: Biomass and fuel loads). This suggests a likely increase in dryness of fine litter, which is associated with increases fire risk⁶¹. Fine fuels are a key determinant of fire risk and initial spread²⁵, while coarse fuels are more associated with total energy output and overall fire severity²⁶.

We isolate the effects of the CO₂ fertilization from those of climate change (e.g., changes in temperature, rainfall, wind) on the observed biomass and fuel trends with an additional simulation (“Methods”: Biomass and fuel load). The elevated CO₂ effect led to an increase in potential aboveground vegetation (biomass and litter) of 11% during the period 1960–2018 but with an overall net increase of only 9.5% due to the offsetting effects of changing climate (Fig. 9).