A national accounting framework for fire and carbon dynamics in Australian savannas

Description of the FullCAM model and its simulation of fire events

For any given pixel (approximately 25  × 25 m area of application of the model, as applied in the NIR) or ‘site’ (0.02–2 ha transects or plots measured in field assessments, as applied during model calibration), FullCAM simulates carbon dynamics of live and dead pools of both woody and grass components in response to regular mortality and turnover (and corresponding recruitment and recovery), as well as stochastic disturbance events such as fires (Fig. 1 and Supplementary Material A). Because FullCAM is a pool-based model and does not simulate populations of individuals, mortality, recruitment and recovery are simulated via increases and reductions in standing biomass. The model includes: (i) live biomass of trees and shrubs, simulated as woody Above-Ground Biomass (AGB); (ii) heavy woody fuel, including stags (elevated dead trees or shrubs) and Coarse Woody Debris (CWD, on-ground components of debris ≥5 cm diameter), simulated together in the model as the ‘standing dead’ pool; (iii) coarse woody fuel, including on-ground branch and bark litter of 0.6–5 cm diameter, simulated as 60% of branch and bark debris; (iv) fine woody fuel, including foliage litter and some of the ≤0.6 cm diameter branch and bark litter that is partially decomposed or broken down, simulated as foliage debris plus the remaining 40% of branch and bark debris; and (v) fine grass fuel, simulated as grass leaf biomass and grass litter.

Fig. 1.

Flow of carbon in a typical savanna woodland system as simulated by FullCAM. Flows of carbon associated with fire events are indicated in red.

A key spatially explicit input influencing productivity, and therefore predicted rates of post-fire recovery, is the maximum above-ground woody biomass, M, (Roxburgh et al. 2019) that a pixel or ‘site’ can support given long-term average site resources (e.g. rainfall), which for savanna regions is well validated (Supplementary Fig. S2; although this could still be further improved; Supplementary Material F). In the current implementation of the model, M is therefore constant, and thus FullCAM is insensitive to potential changes in M driven by changes in resource availability, such as a change to the hydrological regime leading to increases (or decreases) in soil water availability (Supplementary Material B). To further account for spatially explicit differences in sensitivity of biomass to fire, the model also accounts for seven different types of savanna vegetation (Supplementary Material A), as fire impacts these differently, largely because of their differing wood-to-grass composition (Table 1). Subsuming groundcover types (tussock, hummock, mixed grassed) into the vegetation structural classes outlined in Table 1 is a simplification of previous vegetation categories that accounted for differences in recovery rates of these groundcovers as it was assumed that over the extent of fire-prone savannas, such differences might be negligible.

Different types of fire are simulated, depending on the season (EDS, April–July inclusive; or LDS, August–December inclusive) and intensity (with subscripts of 1, 2 or 3 indicating fire intensity of <1, 1–2 and >2 MW m⁻¹, respectively, e.g. EDS₁ indicating early dry season fire of low intensity) (Russell-Smith and Edwards 2006). In FullCAM, the fire type influences combustion factors (C_F = CO₂-C released per unit of dry fuel C consumed, expressed on a dry weight basis and represented below as a percentage C loss term), and in the case of live pools of biomass, fire type also influences an additional loss of carbon from live pools of biomass accounted for via live-to-dead biomass pool transfer factors (T_F = proportion of carbon in live biomass pools transferred to standing dead pools in response to fire, expressed on a dry weight basis). To account for emissions of methane (CH₄) and nitrous oxide (N₂O) in response to changed fire management (IPCC 2006), the simplifying assumption was made that FullCAM-predicted CO₂-C emissions (Mg C ha⁻¹) are multiplied by the emission factors (E_F) listed in Supplementary Table S1 (Surawski et al. 2016). Differences in emissions between vegetation types were partly attributed to local variability in fire intensity and flame height (Meyer and Cook 2015).

The capacity for vegetative regeneration post-fire is well developed in savannas (e.g. Lacey 1974), and so although trees or shrubs may experience partial mortality due to fire, they typically survive and recover (e.g. Lonsdale and Braithwaite 1991; Williams et al. 1999). This is attributable to a combination of resprouting from epicormic buds and/or regrowth of saplings (Bond et al. 2012; Lawes et al. 2022). Hence, FullCAM simulations ensure fire-induced changes in AGB inversely impact standing dead mass (via the T_F), consistent with the creation of heavy fuel when trees die (Cook et al. 2015a) (Supplementary Fig. S3).

When high-intensity fires are frequent, predicted AGB does not fully recover to M between fire events, with a fire-induced suppression of AGB (e.g. as occurs in 1972–1980 in Supplementary Fig. S3), ensuring simulations produce negative correlations between predicted AGB and fire frequency where fires are frequent and severe (e.g. Murphy et al. 2013). Hence, fire simulation in FullCAM temporarily decreases AGB below M, the climatically determined upper bound (e.g. Liedloff and Cook 2007; Murphy et al. 2015) (Supplementary Material B). Consistent with findings of Cook et al. (2017), a modelled assumption was that increased fire intensity results in increased fire impact on AGB (which in FullCAM is C_F + T_F), and hence, time to recover. Given Cook et al. (2017) reported recovery times of savanna biomass of 2–17 years, as fire impact on AGB increases from 2.5 to 15%, predicted recovery time increases from 1 to 5 years, with a maximum time to recover of 20 years (Supplementary Fig. S4).

As FullCAM can be applied spatially across the landscape, patchiness of fire is accounted for by simulating the fire event across only a proportion of the pixels within the total fire scar area (Paul and Roxburgh 2019). For EDS fires, this was 70.9 and 79.0% of the fire scar in high and low rainfall zones, respectively, whereas for LDS fires, it was 88.9 and 97.0% in high and low rainfall zones, respectively (Russell-Smith and Yates 2007; Russell-Smith et al. 2009a; Yates et al. 2015).

In summary, the type and frequency of fire event has an impact on carbon stocks in live and dead pools (both in total mass and in vegetative composition), which, in turn, impacts subsequent fire emissions owing to the availability of volatile components of live and dead biomass. Hence, the FullCAM structure allows for interactions and feedbacks between fire emissions and carbon stocks.

Model calibration

Calibration of FullCAM to savanna fires was a two-step process: firstly, constraining parameter values based on review of the collated datasets, and secondly, optimising remaining ‘unknown’ parameters by ensuring predicted pools of AGB and heavy, coarse, fine and grass fuel matched observations. Across these steps, calculations were required to estimate observed AGB and fuel pools from the available field data (details provided in Supplementary Material C). It was beyond the scope of this work to calibrate FullCAM for fire impacts on soil carbon.

Parameters constrained to observed data: data collected and assumptions made

Parameters calibrated via model optimisation and testing

Three parameter sets required fitting via optimisation given the paucity and/or high uncertainty of available data. The first was mortality, defined here as the regular annual rate of death of live stems not attributable to fire, and assumed to occur in November of each year (consistent with the seasonal observations of coarse litter production), but with resulting AGB loss being replaced within 1 year via regrowth that maintains the assumed steady-state condition. The other two were fire event parameters that determine, for different types of fire, C_F and T_F of different components of AGB. These parameters were calibrated together to ensure their combination resulted in dynamic predictions of AGB that matched those observed across the monitored transects (Supplementary Table S3 and Fig. 2).

Fig. 2.

Location of the transect-based surveys used to assess live biomass and heavy fuel in the different types of tropical savanna vegetation. Data source: Lynch et al. (2018), Cook et al. (2020), Murphy et al. (2023), S. Bray, pers. comm. (2020). WH, Woodland – High rainfall; WL, Woodland – Low rainfall; SH, Shrubland – High rainfall; SL, Shrubland – Low rainfall; PL, Pindan. Full vegetation type definitions are provided in Table 1.

Calibration stands were simulated to replicate field plot or transect areas (averaging 0.24 ha, range 0.03–2.00 ha) where: (i) measurements were made across a range of savanna vegetation types based on field assessments of tree stem density, cover and growth forms of dominant grasses, or when not available, vegetation maps of Thackway et al. (2014) and Lynch et al. (2018); (ii) recent (~5 years) fire history for each of these calibration stands was derived from aerial photos with on-site observations, or when not available and for earlier time-periods, time series of satellite-based observations ( 1-km resolution) as described by Meyer et al. (2015); and (iii) climatic conditions (monthly rainfall, temperature, evaporation over the period of simulation) for each of these calibration stands were derived from Commonwealth of Australia (2022).

As outlined in Supplementary Material B, the model input M is important in providing the upper limit for AGB increases following savanna fire management. The M applied to each calibration site was the maximum observed AGB at that site (i.e. maximum observed AGB in a time-series of observations for a given site, Supplementary Material B). A  500-year model spin-up period was applied (where satellite-derived fire histories for that site or region were assumed to also apply prior to 1988), such that by the time of interest (when predictions were compared with observations), predicted stand biomass and fuel pools were in a state of equilibrium with respect to the fire regime simulated, with simulated fire events at this time only influencing interannual variations in predicted stocks and fluxes. Where the fire intensity was unknown, it was assumed to be moderate, EDS₂ or LDS₂.

In addition to M, the predicted AGB of a stand is also influenced by the assumed rates of mortality. Annual rates of total mortality (fire- and non-fire-related) in stands of Australian savannas have been estimated to range from <0.1 to 4% per year (e.g. Lonsdale and Braithwaite 1991; Prior et al. 2009; Cook et al. 2016, 2020). There is a paucity of data to indicate how fire- and non-fire-related mortality interact, but an assumption was made that fire-related mortality would be between ~50 and 75% of these estimates of total mortality (Cook et al. 2020), but declining to a negligible contribution of total mortality under conditions of low fire intensities (Murphy et al. 2023). Based on these estimates, when calibrating typical rates at which AGB declines (and subsequently regrows) owing to non-fire-related mortality, values of 2.24% year⁻¹ (or 2.70% year⁻¹) for categories of vegetation from high (or low) rainfall zones were assumed to be the upper limit, and were implemented within FullCAM as percentage transfers from the above- and below-ground standing biomass pools to the standing dead pool.

When calibrating C_F and T_F parameters for AGB, it was ensured that the total impact of fire on AGB (= C_F + T_F) was consistent with that previously observed. Previous workers (Russell-Smith et al. 2009a; Yates et al. 2015) have estimated that total fire impact on AGB (previously defined as the ‘shrub fuel’ pool comprising individuals <2 m tall) in high (or low) rainfall zones was 0.29 (or 0.10) and 0.39 (or 0.11) in EDS and LDS fire respectively. These previous estimates were for the AGB as a whole, whereas FullCAM simulates a different C_F and T_F for stem, branch, bark and foliage components of AGB. Therefore, again, previous estimates provided only broad constraints on the calibrations.

Different combinations of parameter values for mortality and the C_F and T_F of components of AGB were tested to determine which provided the highest model efficiency (EF; Soares et al. 1995) and least bias of prediction of AGB for each category of vegetation while also ensuring that, across the different categories of vegetation, the extent of fire impact on AGB increased with the observed average proportion of total woody biomass that was heavy fuel:

where o2¯ is the mean square deviation of each observation from the mean of the observations and e2¯ is the mean squared residual. An EF of 100% indicates perfect match between observations and predictions; 0% indicates the predictions are no better than simply using the mean of the observations, and <0% indicates that residual variation is greater than the variation in the data.

In addition to testing the efficiency of prediction of AGB, performance of the calibrated model was also tested by comparing the predicted biomass with that observed for each of the various pools of fuel: heavy, coarse, fine and grass, including their rates of recovery from time since last burn.

Scenarios of fire management: implications

To demonstrate how FullCAM can be applied to predict the impacts of fire management on avoided emissions and sequestration of carbon, hypothetical savanna burning scenarios were simulated for each of the different savanna vegetation types (Table 1, Supplementary Material D). Project areas were assumed to be 25 ha, comprising 25 × 1 ha strata. Each stratum had a unique hypothetical fire history and was therefore separately simulated using a FullCAM plot file with a unique sequence of fire events to replicate the assumed fire histories, as indicated in Supplementary Fig. S12. For simplicity, only EDS₂ and LDS₂ fire types were simulated.

After a 500-year model spin-up period to bring carbon pools to equilibrium, the average frequencies of EDS₂ and LDS₂ fires across the project area were assumed to change between a 25-year early baseline period, and subsequent 5-year intermediate period (where fire management was assumed to commence, and pools of carbon begin to re-equilibrate) and followed by a 25-year project period. The strata 25-year average differences (and standard deviation) between baseline and project were calculated for: (a) carbon stocks, and (b) fire emissions using the vegetation types in Table 1: Woodland – High rainfall (WH_0.6, WH_0.3), Shrubland – High rainfall (SH) and Pindan (PL) scenarios (assumed to be in localities of relatively high M, and where the baseline fire frequency was relatively high), and Woodland – Low rainfall (WL_0.2, WL_0.1) and Shrubland – Low rainfall (SL) scenarios (assumed to be in localities of relatively low M, and where the baseline fire frequency was relatively low) (Supplementary Table S13).

Parameters constrained to observed data

Regular turnover of carbon from live to dead pools of carbon

When the resulting data-constrained turnover rates were applied in the model (Table 2), the model-predicted turnover-to-AGB relationship matched that observed (Fig. 3). Moreover, the FullCAM-predicted contributions of branch and bark to total litterfall were 25 and 16% respectively, which was similar to those expected (Supplementary Material C).

Fig. 3.

FullCAM-predicted and observed relationships between total turnover (Mg C ha⁻¹ year⁻¹) and the above-ground biomass carbon (AGB, Mg C ha⁻¹) of a stand. Source of observed data is given in Supplementary Table S12, but the observed litterfall data have been multiplied by a factor of 1.33 to account for additional turnover typically attributable to larger litter components not generally captured in traditional litter traps monitoring. FullCAM-predicted turnover and AGB are the average across years for which fire history data were available (1988–2018). EF, model efficiency.

Seasonality of turnover is also important when accounting for the impact of fire management on fuel dynamics, with higher peaks of total standing litter in LDS cf. EDS, contributing to LDS fires being of relatively high intensity (Yates et al. 2020). Therefore, monthly turnover parameters were also constrained by datasets on seasonality of turnover collated from 11 stands within the high-rainfall zone (Williams et al. 1997; Cook 2003; Cuff and Brocklehurst 2015; Yates et al. 2020), and in four stands within the low-rainfall zone (McIvor 2001; Cuff and Brocklehurst 2015; Yates et al. 2020). These datasets showed foliage litterfall was typically highest in the months May to September in high-rainfall zones (Fig. 4b), but slightly later (August to November) in low-rainfall zones (Fig. 4c). In contrast, there was large variability in the seasonality of branch and bark litterfall (relatively large error bars in Fig. 4a cf. Fig. 4b, c), probably because a majority of branch and bark litterfall tends to be associated with stochastic severe wet season storms (i.e. October to December) (Yates et al. 2020).

Fig. 4.

Seasonal turnover (proportion of total annual turnover that occurs each month) for: (a) branch and bark litter as observed in either high or low rainfall zones; (b) foliage litter as observed in the high rainfall zone; (c) foliage litter as observed in the low rainfall zone, and (d) grass die-back as assumed in the FullCAM application in the NIR (Commonwealth of Australia 2022). Error bars represent the standard deviation across: (a–c) observations, or (d) years 1970–2018 in FullCAM inputs of monthly estimates of monsoonal perennial grass die-back. Source of data: Williams et al. (1997), McIvor (2001), Cook (2003), Cuff and Brocklehurst (2015), Yates et al. (2020), Commonwealth of Australia (2022).

Although most of the grass curing in Australian savannas occurs in April and June (Meyer et al. 2012), it was beyond the scope of the present study to collate datasets on seasonality of turnover of grass, with default monthly parameters and seasonality for grass die-back being applied (Fig. 4d, Commonwealth of Australia 2022). Similarly, default values were applied for turnover rates of grass roots (Commonwealth of Australia 2022).

Model optimisation and testing

The highest model efficiency for prediction of AGB was found with mortality rates between 1.120 and 2.025% year⁻¹, total fire impacts (=C_F + T_F) on AGB that were either very low or low for woodland or open forest vegetation types, but moderate to high for shrubland vegetation types, and T_F that varied between components of AGB, increasing in the order: stem, branch, bark and foliage (Table 3). Applying these calibrations gave model efficiencies of AGB prediction between 82 and 93%, demonstrating goodaccuracy (Fig. 5).

Fig. 5.

Relationship between observed and predicted AGB that represented the woodland and open forest vegetation types in: (a) Woodland – high rainfall zones (WH); (b) Woodland – low rainfall zones (WL), and (c) shrubland vegetation types, either in high rainfall zones (circle symbols, SH) or low rainfall zones (square symbols, SL). Predictions of AGB were based on the fire intensity assumptions that provided the highest model efficiency (EF) (Table 3). Dashed line represents the one-to-one line. Axis units are Mg of dry mass (DM) per hectare.

Consistent with Cook et al. (2020), mortality of AGB was higher in the vegetation from high cf. low rainfall zones (Table 3). Within the low rainfall zones, this non-fire related death rate was predicted to be higher in vegetation types dominated by relatively short-lived species (e.g. Acacia and Grevillea), consistent with calibrated mortality rates being higher in stands of Shrubland – Low rainfall (SL) and Pindan (PL) cf. Woodland – Low rainfall (WL).

When C_F and T_F parameters of AGB (Table 3) were applied in FullCAM, the model predicted that a total (= C_F + T_F) of between 0 and 25% of AGB is lost through either combustion or transfer to non-living pools, which was within the order of the 1–23% estimate for Australian savannas, albeit for fires of fairly high intensity (Cook et al. 2015a). Using Woodland – High rainfall (WH) calibration sites as a case study, it was demonstrated that the fire impact parameter values were appropriately optimised, as decreasing C_F + T_F below the calibrated value resulted in an increasing positive bias in predicted AGB (Supplementary Fig. S15).

The relative sensitivity of different vegetation types to fire (Table 3) was generally consistent with the observed average proportion of total above-ground woody biomass that was heavy fuel (Fig. 6). There was evidence that the average proportion of total above-ground woody biomass that is heavy fuel was in turn related to the proportion of AGB that was attributable to fire-susceptible small trees and shrubs (Supplementary Material E). The relatively high total fire impact on AGB for Shrubland – High rainfall (SH) and Pindan (PL) (and presumably also Shrubland – Low rainfall (SL)) vegetation was expected given they have a much greater proportion of shrubs or small multi-stemmed acacias (e.g. Acacia, Calytrix, etc.) compared with the other vegetation types (Supplementary Fig. S11). Shrubs have relatively small diameters, heights and bark thicknesses, which renders them more fire-sensitive than trees (Williams et al. 1999; Lawes et al. 2011; Bond et al. 2012). For example, in plots of SH vegetation, 94% of live stems had a stem diameter at 130cm height (D₁₃₀)  < 20 cm cf. only 67–89% for other vegetation types (data not shown).

Fig. 6.

Relationship between the relative impact of fire on above-ground biomass (AGB) (Table 3) and the observed proportion of total woody biomass that was heavy fuel (Fig. 7). WH, Woodland – High rainfall; SH, Shrubland – High rainfall; WL, Woodland – Low rainfall; SL, Shrubland – Low rainfall; PL, Pindan.

The calibrated total fire impact on AGB was higher for vegetation from high cf. low rainfall zones (Table 3), perhaps for two reasons. First, high rainfall zone vegetation had quite high proportions of stems that were relatively large. Across the calibration dataset (Fig. 5), the high rainfall zone vegetation category had 0.3–2.8% of live stems with D₁₃₀ > 50 cm, whereas the percentage of stems of this size was ≤0.1% for the other categories of vegetation. Larger trees in Australian savannas are often fire-sensitive as they can burn from the inside when hollow owing to damage from termites or previous fire (Williams et al. 1999; Cook et al. 2005). Secondly, high rainfall zone vegetation has relatively high proportions of fire-sensitive non-eucalypt species relative to the less fire-sensitive eucalypt species (Bond et al. 2012). Although calibration sites used in the present study were not a random sample, across the calibration dataset, stands from high rainfall zones had relatively high proportions of the more fire-sensitive non-eucalypt trees such as Callitris intratropica (e.g. 16% of species found in WH_0.6) and Erythrophleum chlorostachys (e.g. 17% of species found in WH_0.3) (Supplementary Fig. S11).

The final test of model performance was the assessment of how well predicted pools of fuel matched those observed, with results again indicating model performance was good (Fig. 7). Moreover, with the calibrated parameters given in Table 3, FullCAM-predicted recovery (Supplementary Fig. S4) was consistent with observations (Russell-Smith et al. 2009a; Yates et al. 2015, 2020; Lynch et al. 2018), with increases in fuel with time since last burn being least pronounced with the coarser components (e.g. deadwood, Supplementary Fig. S17a) and most pronounced with the relatively fine components of litter (e.g. foliage litter, Supplementary Fig. S17b, c).

Fig. 7.

Comparison between observed and predicted fuel biomass (in Mg C ha⁻¹) in: (a) heavy (stags and CWD), (b) coarse, (c) fine, and (d) grass pools, within differing categories of vegetation. Error bars represent standard errors. Numbers and percentages above the bar represent the sample size (N), and proportion of ‘observed’ heavy fuel calculated to be stags. Bars with the same letters were not statistically different according to the Tukey test at 95% confidence interval. Data source is provided in Supplementary Tables S4–S8. Vegetation type definitions are given in Table 1.

Scenarios of fire management: implications

Land managers can use planned EDS fires to reduce the frequency of high-intensity LDS fires and, to a lesser extent, overall fire frequencies (e.g. Russell-Smith et al. 2013). When such fire management scenarios were simulated with FullCAM using the calibrated parameters (Tables 2 and 3), on average across all vegetation types, 65% of total abatement (range 45–87%) was attributable to sequestration of CO₂-C, with the remainder being attributable to avoided emissions (Fig. 8). In regions of high (or low) rainfall, net abatement over a 25-year simulation period was between 0.06 and 0.43 (or 0.02 and 0.05) Mg CO₂-e ha⁻¹ year⁻¹ (Fig. 8). But even within a region, variation in this abatement was evident among vegetation types owing to differences in C_F of AGB (Table 3) and the productivities of the woody and grass components of the stand (i.e. M and grass cover).

Fig. 8.

Summary of the impact of changed fire regime (project compared with baseline) on annual average rates (over the 25-year baseline and project periods) of: (a) sequestration of CO₂-e and annual average rates (over the 25-year baseline- and project-periods) and (b) avoided fire-induced CO₂-e emissions, with these being attributable to five pools: total biomass of woody vegetation, namely above- and below-ground pools, heavy fuel, coarse fuel, fine fuel and grass fuel. Vegetation type definitions are given in Table 1.

The results from these analyses illustrate potential outcomes across a range of scenarios with differing baseline fire frequencies, and with assumed impacts of management reflected as changes in the relative frequencies of EDS and LDS fire. It is important to note these scenarios are hypothetical and are thus not validated, and do not relate to specific locations or situations. Cook et al. (2015b), using the individual-based FLAMES model, also investigated the impacts of changed fire management on total above-ground carbon stocks and rates of sequestration, with model parameters based on the ‘Three Parks’ monitoring study (Russell-Smith et al. 2009b). Although the simulations of Cook et al. (2015b) explored a greater range of fire frequencies and seasonal fire timings than those considered here, changes in carbon stocks and sequestration rates were broadly similar to our results in terms of both direction and magnitude.

Consistent with the other recent studies (Levick et al. 2019; Werner and Peacock 2019; Murphy et al. 2023), our results indicate that sequestration of carbon in live biomass is a key driver of abatement following savanna fire management, noting that predictions are highly sensitive to the assumed upper limit of AGB, or the M input layer. Although verified for savanna vegetation (Supplementary Fig. S2), the M input layer remains a key source of uncertainty for any given stand, given M may be inaccurate depending on fine-scale spatial variability associated with position within the landscape, and hence, soil nutrients and depth (and thus, water holding capacity) (Supplementary Fig. S6). Also, not all categories of savanna vegetation were represented in the calibration of this input layer (e.g. Pindan). As outlined in Supplementary Material F, in addition to improving predictions of AGB, there are opportunities to collate additional datasets to further constrain calibration of model parameters that are currently highly uncertain owing to the paucity of available data, e.g. rates of decomposition of pools of standing dead and debris. Parameters that could not be constrained owing to negligible available data (mortality, C_F and T_F of AGB) were optimised in this study, and further work is required to verify these, and where required, constrain them to collected datasets.

As a result of this work, modelling capability has been developed to facilitate the development of a comprehensive methodology encompassing the impacts of different fire types on both avoided emissions and sequestration of carbon, thereby overcoming limitations of existing methods that only account for avoided emissions (Commonwealth of Australia 2015), only account for sequestration in dead (rather than also live) pools (Commonwealth of Australia 2018), or do not distinguish between effects of fires of different intensities (Voluntary Carbon Standard 2015). However, the scenario analysis (Fig. 8) provides only simplistic hypothetical results on abatement (both sequestration and avoided emissions) from implementation of savanna fire management. Work is currently under way to facilitate the spatial application of FullCAM at a project scale (as currently, spatial FullCAM application is limited to national-scale application via the NIR; Commonwealth of Australia 2022). This capacity will provide opportunities to work with project proponents to test FullCAM-predicted abatement implications of actual savanna fire management projects, and compare results with existing methodologies (Commonwealth of Australia 2015, 2018).