Impact of fire return interval on pyrogenic carbon stocks in a tropical savanna, North Queensland, Australia

Introduction

Pyrogenic carbon (PyC; char, black carbon) is naturally produced by the incomplete combustion of biomass during biomass burning (Bird et al. 2015). Once formed, it is largely resistant to mineralisation and will persist for centuries to millennia (Bird et al. 2015). The resistance of PyC to degradation and mineralisation has led to the suggestion that net carbon sequestration can be achieved by manufacturing PyC (biochar) and sequestering the carbon thus produced in agricultural soils, with attendant benefits to soil health and crop productivity (Woolf et al. 2010; Lehmann et al. 2021).

At the global level, Jones et al. (2019) estimated that 196–340 Tg of carbon was converted into PyC by wildfires annually over the 1996–2016 period, while Bowring et al. (2022) estimate 337 Tg of carbon converted annually to PyC. Both studies conclude that most PyC production from wildfires (49–73%) occurs in tropical mixed tree–grass savannas that are adapted to fire and burn more frequently with annual to sub-decadal return intervals, even though fuel loads in these biomes are relatively low compared to tropical or temperate forests. Tropical savannas therefore produce and sequester a quantity of PyC that is significant in the context of total terrestrial biospheric stocks and fluxes.

Savannas cover ~2 million km² of northern Australia, around one-quarter of the total Australian land area. Savanna fires have been estimated to be responsible for 65% of carbon emitted from biomass burning in Australia (Murphy et al. 2019). The timing and frequency of fires across northern Australian savannas is highly variable. This variability is partly linked to the three major types of land tenure – pastoral lands, conservation estates and Indigenous lands – but also dependent on rainfall amount and seasonality (Perry et al. 2019; Edwards et al. 2021). Fires are ignited naturally (via lightning strike) or human induced, either by an accident, or design, or through regular bushfire abatement practices such as cultural burning practices. The average fire return interval, collated over the last 22 years by the Northern Australian Fire Information (NAFI) database (Jacklyn et al. 2015) over the entire northern savanna region, return interval is approximately biennial (Edwards et al. 2021) but with a high degree of spatial variability. In some wetter, more coastal regions, fires are an annual phenomenon often concentrated in the late dry season (after August). A substantial proportion of the drier inland area and particularly large (~100,000 ha) pastoral leases have experienced a fire return interval of 5 years or greater. Large regions have not burned in the 22-year period. When fires do occurit is mostly late in the dry season (Edwards et al. 2021).

As a result of the large area of Australian savannas and their impact on the national carbon budget, the Australian Government developed a ‘savanna burning methodology’ in 2018. This enables payment for emissions reductions generated by implementing improved fire management practices in northern Australia. The methodology recognises the traditional burning methods of Aboriginal peoples by shifting the timing of burning to earlier in the dry season, resulting in smaller and lower intensity fires. This results in avoided emissions associated with reductions in the size and number of high intensity, late dry season fires, and attendant increases in the stock of dead organic matter in the landscape. The approach was tested before implementation through initiatives such as the West Arnhem Land Fire Abatement (WALFA) project (2006–present). THis project implemented traditional burning over an area of 28,000 km² in the Northern Territory and monitored the impact of changing the timing and frequency of burning on carbon emissions in an Australian tropical savanna area (e.g. Russell-Smith et al. 2013).

While the impact of changes in fire return interval on Australian above- and belowground carbon stocks has received some attention in the recent years (Richards et al. 2011; Hunt 2014), pyrogenic carbon stocks have received only a little attention, and pyrogenic carbon sequestration was not included in the savanna burning methodology for calculating emissions offsets. There have been attempts to map the spatial distribution of pyrogenic carbon stocks both nationally (as resistant organic carbon, ROC; Viscarra Rossel et al. 2019) and globally (Reisser et al. 2016), but the impact of changes in fire regime on pyrogenic carbon stocks has received less attention. This study assesses the impact of fire return interval on both total organic carbon (TOC) and PyC stocks across an area of similar climate, covered by similar open grassy savanna on similar soil type of similarly low relief.

Methods

Study area

The area of this study in tropical north Queensland lies within the broad transcontinental arc of Australian savanna between 600 and 1000 mm annual rainfall and intense seasonality. This is within the region proposed for implementation of a savanna burning methodology for the purposes of carbon sequestration (Whitehead et al. 2014; Fig. 1), where landowners are eligible for carbon credits under the Australian Carbon Credit Unit (ACCU) Scheme, using the Savanna Fire Management – Sequestration and Emissions Avoidance Method (Department of Climate Change Energy the Environment and Water 2018a) or Savanna Fire Management – Emissions Avoidance Method (Department of Climate Change Energy the Environment and Water 2018b). Eligible land areas under both methods fall within either high (>1000 mm p.a.) and/or low (600–1000 mm p.a.) rainfall zones. See examples by Cook et al. (2016, 2020) for mathematical determination of these zones.

Fig. 1.

Map of Undara and Forty Mile Scrub National Parks (grey boundaries) showing the locations of the seven transects (blue lines) in relation to the number of years burned between 2000 and 2022 (red shading) as identified by North Australia and Rangelands Fire Information (NAFI) (2023). Transects are labelled with the number of years burned. The inset map gives the location of this study area in relation to the low rainfall zone boundary (orange region) for determining eligibility for participation in the Australian Government Carbon Credits (Carbon Farming Initiative) Act 2011.

Sampling was conducted along transects with a range of fire return intervals in two small national parks in Einasleigh Uplands bioregion of North Queensland (Fig. 1). Undara Volcanic National Park was established from a former pastoral property in 1992. The park is located on a dominantly low relief landscape covered by relatively shallow stoney krasnozem soils (Grundy and Bryde 1989) developed on ~190,000-year-old Undara Basalt, part of the Late Cenozoic McBride volcanic province (Griffin and McDougall 1975), ~800 m above sea level. Climate is strongly seasonal with a mean annual temperature of 23.6°C (mean annual minimum/maximum temperatures 16.1/31.1°C) and mean annual rainfall of 795 mm, with 80% of this falling in the summer monsoon season from December to March (Bureau of Meteorology site number 030036, Mt Surprise township, 40–60 km west of Undara).

The basalt plains are covered by open, grassy Eucalypt woodland, traditionally the lands of the Ewamin Aboriginal peoples (QPWS 2023a). In an effort to combat invasive grass species (Vogler 2017) and return the park to something resembling an Indigenous fire regime with a mosaic of habitats for wildlife, the park area was aimed to be burned in rotation approximately every 2 years since at least 2002 (Abom et al. 2016). In practice, the Northern Australian Fire Information (NAFI) mapping service (Jacklyn et al. 2015) indicates that discrete areas in the Park have burned 5–13 times between 2000 and 2022. These high fire return intervals contrast strongly with the surrounding pastoral properties on similar soils and with similar vegetation, which have most commonly experienced 0–4 fires in the 2000–2022 period, equating to a return interval of ≥5 years (Fig. 1). The NAFI data is generated weekly by comparing MODIS satellite images 1–2 weeks apart and identifying areas which have been burned in the intervening period.

Forty Mile Scrub National Park is located 10–20 km northeast of Undara with similar climate and soils. While the park was established in 1970 to protect semi-evergreen vine thicket (QPWS 2023b), the park also protects areas of the same open grassy Eucalypt woodland that dominates Undara, and these areas were selected for sampling in this study (Fig. 1). Unlike Undara, this park has not had the extended history of intentional fire management. Like the surrounding pastoral properties, most areas have burned 0–4 times (Fig. 1), with unmanaged large fires having burned the open woodland and into the evergreen vine thicket on occasion in the 2000–2022 period.

Results

We discuss the data in two groups: (i) transects burned less than 5 times (transects 0–2 and 2–4 of Fig. 1, return interval >4.4 years; 18 samples), and (ii) transects that have burned 5 or more times in the 2000–2022 period (return interval <4.4 years; 48 samples). The results separated by individual transect are provided in Supplementary Tables S1, S2 and Fig. S1.

The only large-scale ecological difference between the two groups apart from fire return interval is that the <5 fire transects currently have a higher tree canopy cover (23–29% persistent green) than the transects that have burned ≥5 times (15.2–22%; Fig. 2). This difference may have developed as a result of the lower fire return interval, and/or through proximity to the vine thicket seed bank of Forty Mile Scrub itself, 2–3 km north of the <5 fire transects.

Fig. 2.

Box plots showing the percent woody cover (persistent green) within Undara and Forty Mile Scrub National Parks grouped by years burned (less than or greater than 5 fire years in the last 22 years) as identified from North Australia and Rangelands Fire Information (NAFI) (2023). Landsat seasonal persistent green estimates obtained from VegMachine (Beutel et al. 2019) have been averaged from the period of January 2000 to January 2021. The box encloses 50% of the central data with the mean (black diamonds), median (black lines), individual data points (circles), and maximum and minimum values (vertical black lines) also indicated.

Soil dry bulk density ranged from 0.55 to 1.55 g cm⁻³ over the 0–5 cm depth layer (Fig. 3), with the <5 fire transects having a significantly (t = −2.96, P-value <0.05) lower bulk density (µ = 0.95 g cm⁻³) than the ≥5 fire transects (µ = 1.18 g cm⁻³). There was no significant average difference (t = −1.10, P-value >0.05) between tree (T) and grass (G) pairs with respect to dry bulk density.

Fig. 3.

Box plots comparing sites of fewer and greater than or equal to 5 fire years in a 22 year period within Undara and Forty Mile Scrub National Parks with respect to (a) dry bulk density, (b) pyrogenic/organic carbon % and differences between their respective tree and grass pairs (c, d). The box encloses 50% of the central data with the mean (black diamonds), median (black lines), individual data points (circles), and maximum and minimum values (vertical black lines) also indicated.

Organic carbon stocks ranged from 10.3 to 27.2 MgC ha⁻¹ (Fig. 4), with no significant (t = −1.68, P-value >0.05) difference between <5 (µ = 15.22 MgC ha⁻¹) and ≥5 fire (µ = 17.46 MgC ha⁻¹), while there are differences between tree (T) and grass (G) pairs (t = −2.89, P-value <0.05) there is no significant difference between tree (T) and grass (G) samples between groups (t = −0.09, P-value >0.05). Pyrogenic carbon stocks ranged from 0.61 to 3.39 MgC ha⁻¹, with the mean stocks for the <5 fire transects (1.39 MgC ha⁻¹) being significantly lower (t = −5.66, P-value <0.05) than for the ≥5 fire transects (2.39 MgC ha⁻¹), and no difference in stocks between the T and G pairs in any group (t = −0.10, P-value >0.05; Fig. 4). The similarity in OC stocks and difference in PyC stocks drove a significant increase in PyC/OC ratio (t = −5.95, P-value <0.05). This ratio ranged from 4.1 to 23.6, with a mean PyC/OC of 8.94 for the <5 fire transects, and 13.74 for the ≥5 fire transects (Fig. 3). There was no significant difference between PyC/OC for T versus G pairs in either group (t = 0.59, P-value >0.05).

Fig. 4.

Box plots comparing sites of fewer and greater than or equal to 5 fire years in a 22-year period within Undara and Forty Mile Scrub National Parks with respect to (a) Organic and (b) Pyrogenic carbon stocks, and differences between their respective tree and grass pairs (c, d). The box encloses 50% of the central data with the mean (black diamonds), median (black lines), individual data points (circles), and maximum and minimum values (vertical black lines) also indicated.

δ¹³C_OC values ranged from −26‰ to −16.6‰, a range typical of savanna environments (Fig. 5). As a result of the higher canopy cover, the average δ¹³C_OC (−24.1‰) in the <5 fire transects was significantly lower (t = −14.27, P-value <0.05) than for the ≥5 fire transects −18.3‰) with the T samples in the ≥5 fire transects having a lower δ¹³C_OC (by an average 1.1‰) than the G samples, with no difference between T and G samples in the <5 fire transects (t = 0.72, P-value >0.05; Fig. 5). δ¹³C_PyC values ranged from −15‰ to −27.1‰, with average values 1–3‰ higher than δ¹³C_OC values and the average value for <5 fire transects (−21.1‰) being significantly lower (W = 0, P-value <0.05) than the average value for the ≥5 fire transects (−16.3‰; Fig. 5). In both groups the average difference between the T and G δ¹³C_PyC pairs was close to 0 (µ_T–G<5 = 0.65, µ_T–G>5 = −0.67; Fig. 5).

Fig. 5.

Box plots comparing sites of fewer and greater than or equal to 5 fire years in a 22-year period within Undara and Forty Mile Scrub National Parks with respect to (a) Organic and (b) Pyrogenic carbon δ¹³C value, and differences between their respective tree and grass pairs (c, d). The box encloses 50% of the central data with the mean (black diamonds), median (black lines), individual data points (circles), and maximum and minimum values (vertical black lines) also indicated.

Cross validation results of PLSR models for TOC concentration were excellent with an R² = 0.95, root mean squared error (RMSE) = 0.024, bias = 0.001 and Ratio of Performance to InterQuartile distance (RPIQ) = 4.10 (Fig. 6a). Results for PyC concentration were less certain with an R² = 0.40, RMSE = 0.033, bias = −0.001 and RPIQ = 1.56 (Fig. 6b). Inspection of the PLSR model forms (i.e. variable importance) indicates a mix of similar and distinct features (Fig. 6c). In particular, the two models share positive features related to aromatic carbon (1650 cm⁻¹) but only the TOC model includes aliphatic carbon (2925 and 2850 cm⁻¹)

Fig. 6.

Mid-DRIFT model results for total organic carbon (a) and pyrogenic carbon (b). Variable importance for both models is given as the regression coefficients when p for each wavenumber is less than 0.25 (c).

Discussion

Across both the <5 and ≥5 fire transects, OC stocks are on average higher in the -T (tree) than the -G (grass) samples (Fig. 4), consistent with previous research that has shown higher soil carbon stocks beneath trees in Australian tropical savannas (Wynn et al. 2006). In contrast, there is no difference in the PyC stocks associated with current tree cover, due to the long residence time of PyC in the soil (e.g. Lehmann et al. 2008, 2021; Bird et al. 2015), and hence PyC stocks are not closely related to current tree cover distribution.

The δ¹³C_OC values for all transects (Fig. 5) are within the range previously reported for Australian tropical savannas and -T samples have lower δ¹³C_OC values than the -G samples also consistent with previous research (Bird and Pousai 1997; Wynn et al. 2006). The lower δ¹³C_OC values for the <5 fire transects are consistent with the higher tree cover on those transects (Fig. 2). The δ¹³C_PyC values tend to be 1–3‰ higher than the corresponding δ¹³C_OC values, suggesting preferential incorporation of C₄-derived PyC into the soil. As with the PyC stocks, there is no significant difference in δ¹³C_PyC values between the -T and -G samples due to the long residence time of PyC in the soil.

The results clearly indicate that PyC stocks are higher in all areas burned five or more times (5–13 times) in the 2000–2022 period, compared to unmanaged areas that burned between zero and four times (<5 times) in the same period. The difference in PyC stocks amounts to an average of ~0.25 MgC ha⁻¹ more PyC at sites burned ≥5 times. The increase did not come at the expense of soil organic carbon stocks which average ~16–17 MgC ha⁻¹ for the 0–5 cm layer (Fig. 4). It is not a result of the minor differences in bulk density between the <5 and ≥5 fire transects (Fig. 3) as the PyC/OC ratio, which is not influenced by soil bulk density, is also significantly higher in the more frequently burned transects. Given the recalcitrant nature of the pyrogenic carbon component measured by hydrogen pyrolysis, the pyrogenic carbon stock in the soil can be considered sequestered for centuries to millennia (Bird et al. 2015).

Saiz et al. (2015) quantified PyC production using the same analytical technique (hydrogen pyrolysis) for small experimental burns also on Undara Volcanic National Park. They found that each fire produced an average of 0.11 MgC ha⁻¹ that remained on the ground, with some additional PyC exported in the very fine fraction (as ash and other particulates). Given likely post-fire losses of PyC through recombustion in subsequent fires (Saiz et al. 2014), via water transport and through wind erosion, particularly of fine grass-derived PyC (Saiz et al. 2018), the measured production of PyC at Undara in the experimental burns is broadly consistent with the extent of measured additional sequestration of ~0.25 MgC ha⁻¹ in the ≥5 fire transects.

For at least the last ~11,000 years (Bird et al. 2024) but ceasing with the establishment of pastoral stations in the late 1800s, the tropical savannas of northern Australia were subjected to Indigenous fire management involving mosaic burning with a relatively high fire return interval (e.g. Russell-Smith et al. 2013). Under European management, fire was not actively managed or was suppressed for much of the 20th century. The reimposition of an Indigenous fire regime over the last two decades at Undara Volcanic National Park has enabled an assessment of the impact of the re-introduction of Indigenous fire management on pyrogenic carbon stocks, through comparison with an adjacent area that has not seen the return of Indigenous fire management.

The <5 fire transects would have been burned at similar frequency to the ≥5 fire transects before European settlement (e.g. Russell-Smith et al. 2013) and other things being equal, the PyC stocks for both groups should have simply remained as they were, that is, similar to the stocks ≥5 fire transects now. Therefore, the lower PyC stocks in the 0–5 cm soil layer of <5 fire transects indicate that there has been a relative loss of PyC from the 0–5 cm soil layer over the last century for the <5 fire transects. The PyC component quantified in this study is resistant to degradation on centennial to millennial timescales (Bird et al. 2015), implying this loss has been the result of physical translocation out of the 0–5 cm layer. Some lower temperature formed PyC may have been lost via mineralisation, and some PyC may have been lost through deflation from the soil surface. But it is likely that the majority of the decrease has resulted from the movement of PyC below the 0–5 cm layer investigated in this study. This likely occurred by a combination of bioturbation and illuviation during rain events over the last century. Downward movement of PyC has been reported in experimental studies (Nguyen et al. 2009; Major et al. 2010) and must occur given the fact that significant PyC stocks are reported well below the soil surface (Koele et al. 2017; Hobley 2019). With a lower rate of ‘replenishment’ of PyC to the soil surface as a result of a low fire return interval, the PyC stocks in the 0–5 cm layer of the <5 fire transects have decreased as the PyC has been redistributed to depth over the last century. Reimposition of a higher fire return interval has ‘replenished’ the PyC stocks in 0–5 cm soil layer of the ≥5 fire transects over the last two decades.

The mid-DRIFTS analysis was included in this study because if soil or pyrogenic carbon sequestration due to increased fire return interval, were to be traded on the carbon market, low-cost methods for quantifying the increase in TOC and PyC are needed (Department of the Prime Minister and Cabinet 2021). In this regard, mid-DRIFTS is an allowable alternative to traditional analysis for TOC in the Australian Government’s soil carbon credit methodology (Department of Climate Change Energy the Environment and Water 2021). The excellent results for TOC are consistent with other studies (Baldock et al. 2013; Sanderman et al. 2021b). The models for PyC had greater uncertainty but importantly were unbiased, (Fig. 6) suggesting that with increased sample numbers detection of even small changes in PyC over time is possible. The PyC model performance was lower than in some other studies that focused on developing spectral-prediction models for PyC (Baldock et al. 2013; Cotrufo et al. 2016; Hobley et al. 2016), but the samples from Undara had a much smaller range in PyC concentrations than those other studies, suggesting that model performance could be significantly improved with a more diverse training set.

Noting that any fire, regardless of fire regime, will lead to some PyC delivery to the soil, our results suggest a managed return to ≥5 early dry season fires per 1–2 decades will yield sequestration of PyC in the 0–5 cm soil layer amounting to ~0.25 MgC ha⁻¹. Therefore, sequestration of 1250–2500 MgC annually is potentially achievable for an average pastoral property of 100,000 ha. With the current spot generic Australian Carbon Credit Unit (ACCU) price sitting at ~AUD$32 (Clean Energy Regulator 2023), in fiscal terms this equates to almost AUD$1 million per decade (~AUD$40,000–80,000 per year) in potential revenue. While this is obviously an illustrative figure only, it does indicate that a return to a managed, early dry season, higher return interval fire regime can provide significant long-term carbon sequestration given the large area of northern Australian savannas. Given the recalcitrant nature of the sequestered carbon compared to OC, this may be attractive to the buyers of carbon credits. We also note that an increase of ~2 MgC ha⁻¹ of OC occurred in the more frequently burned plots. Although this increase is not significant, it is consistent with the modelling of Richards et al. (2011) who found that a statistically significant increase in soil organic carbon stocks of 4 Mg ha⁻¹ over a century accompanied a change in fire regime to cool burns with a 5-year return interval, equivalent to ~0.8 MgC over the period considered in this study.

It is likely that the PyC sequestration measured in this study is conservative as only the 0–5 cm layer was sampled, and PyC translocation to deeper soil layers is occurring. It is also possible that PyC stocks will continue to increase to an unknown extent if the higher return interval is maintained in the future. There is a suggestion that optimal PyC sequestration in the soils of the study area occurred for transects burned 5–8 times in the 2000–2022 period (H = 8.12, P-value <0.05; Fig. S1), equating to a return interval of 3–4 years. From a fire frequency of 5 in 22 years, there is a significant (z = −2.49, P-value <0.05) downward trend in PyC stocks with increasing fire frequency. At the highest return intervals, PyC sequestration may be limited by a combination of reduced biomass available to burn along with increased potential for recombustion of PyC by subsequent fires before incorporation into the soil. It should also be noted that PyC sequestration may be augmented or offset by changes in the quantity and composition of greenhouse gas releases, and changes in dead biomass, soil carbon and nutrient cycling accompanying a change to more frequent, cooler fires (Cook et al. 2016).

The potential for PyC sequestration in savanna soils across northern Australia will also vary as a function of at least (i) climate – through control of tree–-grass distribution and biomass, (ii) livestock management – through control of above ground biomass, and (iii) soil type and landscape position, through control of tree–grass distribution, total biomass and the rate of PyC erosion from the soil surface and/or protection through illuviation to deeper soil layers or lateral movement by water erosion.

There is clearly potential for significant, ongoing soil PyC sequestration on decadal timescales through the broader reintroduction of a fire regime similar to that imposed for millennia by Aboriginal peoples in northern Australian savannas. This sequestration is supplementary to the avoided emissions that are currently allowable for carbon credits through the Australian Government's savanna burning methodology (Edwards et al. 2021) and additional biodiversity co-benefits may also accrue over time (Evans and Russell-Smith 2019; Sangha et al. 2021; Greenwood et al. 2022). Management of fire as a tool for improving carbon sequestration, avoiding emissions and improving biodiversity outcomes has the advantage of being relatively easy to implement and to monitor remotely (Jacklyn et al. 2015). However, the benefits of more frequent, cooler fires must be weighed against the opportunity costs of forgoing other land use options, broader changes in ecosystem carbon stocks and greenhouse gas emissions (Cook et al. (2016), positive (Greenwood et al. 2022) and potentially negative biodiversity impacts (Abom et al. 2016) as well as the potential health impacts associated with increased smoke pollution (Jones et al. 2022).

Supplementary material

Supplementary material is available online.