This study adopted a ‘bottom-up’ emissions methodology to quantify CO2 and PM2.5 emissions from the 3–18 January 2013 Forcett–Dunalley fire in south-eastern Tasmania. We show that total CO2 and PM2.5 emissions from the fine scale analysis reached 1.125 ± 0.232 Tg and 0.022 ± 0.006 Tg respectively. A comparison of the fine scale (50 m) analysis that uses local fuel and fire severity estimates, and a coarse scale global emissions model GFED (0.25 degrees or ~ 28 km) showed that GFED had a good agreement with the fine-scale analysis regarding total CO2 emissions but not PM2.5 emissions. Naturally, fine scale analysis had more detailed spatial patterns of both emissions than GFED. Validation of the emissions estimates using the FullCAM model yielded 142 t CO2 ha− 1 (> 2 times the estimates from both inventories), suggesting further refinement of FullCAM is important, especially the parameters used in calibrating the model (e.g. debris pool) which are subject to large uncertainties .
Other wildfire emissions
A comparison of Forcett-Dunalley fire emissions with other Australian temperate fires showed similarities with some fires and considerable differences with other fires (see Table S1, Additional file 2). For example, the per-hectare CO2 estimate from this study was 55.7 t CO2 ha− 1 whereas Volkova et al.  reported emission of 105 t CO2 ha− 1 from a wildfire in a long-unburnt dry shrubby Eucalyptus forest in Victoria. However, our values are comparable to those reported by these authors from the areas within that wildfire that were previously fuel-reduced (42 t ha− 1 of CO2). The 2003 Canberra fire produced 20.2 Tg of CO2 emissions based on the Australian FullCAM model , translating to approximately 78 t CO2 ha− 1 from the 260,000 ha-fire size, assuming no unburnt patches. However, other studies have reported carbon emissions estimates of 40 M tonnes (or 40 Tg) from the same fire ; it is likely that CO2 emissions from that fire exceeded 400 t ha− 1 given that CO2 emission are 3.67 times more than carbon emission.
Previous studies in Australia have shown high agreement between GFED and other models/field observations in CO2 emissions e.g. Paton-Walsh et al. . This is despite GFED treating vegetation types, particularly Eucalyptus forests and woodlands, and fire behaviour in south-eastern Australia as the same as those found in the temperate biomes in Northern Hemisphere. The overall good performance of GFED’s CO2 estimates in this study also likely reflects an improved detection of smaller fires in GFED4 compared to previous versions of GFED .
Per-hectare estimates for PM2.5 in this study (1.1 t ha− 1) were inconsistent with emissions estimates from other Australian temperate fires (Table S1, Additional file 2). For example, Reisen et al.  reported emissions of 73.7–163.9 kg ha− 1 (0.07–0.16 t ha− 1) from prescribed fires in Victorian Eucalyptus forests while another Tasmanian study reported PM2.5 emissions of 7,789 tonnes (or 6.9 t ha− 1) from a high-intensity regeneration fire in a southern Tasmanian native forest . It should be noted that there is paucity of data on PM2.5 emission from temperate Australian forest fires; most of the studies have instead focused on PM2.5 concentration in urban airsheds for air quality purposes, involving a mix of emission sources. Beyond Australia, western US wildfires between 2011–2015 were estimated to have emitted 1,530 Gg (1.53 Tg) of PM2.5 annually . Similar to our study, the authors report that the emissions were three times higher than the estimates from the US national inventory. Further, in another study, the GFED3 PM2.5 emission estimate across contiguous US was lower by a factor of eight compared to the national emissions inventory , revealing a likely systematic underestimation of PM emission across jurisdictions.
Deficiencies in current fire emissions approaches
The discrepancy in GFED modelling in this study was the lower PM2.5 emissions by a factor of three, likely due to lower emissions factors (EFs) used for PM2.5 within GFED (12.9 g kg− 1). These EFs do not accurately reflect temperate Eucalyptus-dominated fuels in Australia, as they are averaged across the temperate biome globally. One of the main differences significantly affecting emissions amongst the temperate biomes is fire behaviour. For example, compared to other biomes, Australian forests and woodlands typically have a higher biomass of sclerophyllous leaves and bark, which burn intensely and support short-long distance transport and spotting of embers that spread landscape fire . Eucalyptus fuels have lower rates of decomposition (and therefore low/absent duff layer ) compared to northern hemisphere conifer/boreal forests that have a more-developed duff layer that supports smouldering combustion and can contribute up to 50–74% of fuel consumption . An upward revision of PM2.5 EFs to 16.9–38.8 g kg− 1  is therefore recommended to better accommodate typical fuels within these Australian ecosystems.
The accuracy of bottom-up approaches (such as the above inventories) that adopt fuel consumption estimates in emissions estimations has been a topic of debate relative to the more accurate top-down approaches that use satellite observations to directly estimate emissions within the atmospheric column [57–59]. Despite these limitations, two previous carbon emissions studies on the recent Australian Black Summer fires using top-down and bottom-up approaches revealed comparable CO2 estimates between the two methods [3, 13]. This highlights the importance of validating emissions estimations with diverse methods, including satellite and on-ground observations, to reduce the inherent uncertainties.
Smoke emissions analyses are constrained by the quality and representativeness of data on fuel types, requiring greater sampling of a broader range of vegetation, and fuels . Field protocols should include detailed inventories of vegetation characteristics, e.g. Prior et al.  and measurement of fuel loads across all fuel components, ranging from subsurface to overstorey fuels, and from fine to woody fuels. To date, coarse woody debris (CWD) estimation, being the less studied fuel component than fine fuels, is the most common source of emissions uncertainties in temperate Australian landscapes. This is because CWD is influenced in different regions by among other factors, the disturbance history (past fire or logging activities), forest age, and site productivity [18, 61]. More field inventories across Australia and particularly in Tasmania where there has been scarcity of fuel load data  are needed to provide confidence in emissions estimates. These inventories could make use of recent technologies such as LiDAR to increase the accuracy of fuel estimation, especially the amount of coarse woody debris, within a forest. Fire behaviour modelling in Australia has shifted from an emphasis on fine fuel loads, to a more realistic determination of fuel hazard scores across fuel types; nonetheless, we contend that there remains a need for accurate fine and coarse fuel load measurements to underpin fire emissions analysis .
Fire severity scales with fuel consumption, with high-severity fires typically associated with high consumption of vegetation; however, the general lack of empirical fuel consumption data can introduce variability in total emissions, despite the availability of fire severity information. This was evident in the spectral signatures (from satellite observations) in grassland areas of the Forcett–Dunalley fireground which exhibited very high severities despite their very low fuel loads and minimal biological impact. Fuel consumption estimates in this study were inferred from a few studies on temperate Eucalyptus forests (Table 2). Therefore, there is need to improve data collection of fuel consumption during wildland fires (supplemented by remote sensing), and measurement of residence time of flaming and smouldering to partition emissions into the different combustion stages. Although these attributes can be inferred from lab experiments, variability in fuel size, especially coarser fuels, are difficult to accurately characterised in the lab . There is also a need to clearly establish a quantitative link between severity measurements and fuel consumption for better applicability of fire severity data in future emissions studies.
Greenhouse gas accounting
Estimates of emissions from wildfires are of increasing interest given their contribution to climate change. Indeed, emissions from Australian wildfires are accounted for in the national GHG accounting to the Intergovernmental Panel on Climate Change, however, what constitutes a wildfire and a human-caused fire in the accounting is subject to debate and a number of pragmatic and often poorly justified ‘rules’. For example, the Australian Government accounting uses a burned area threshold (that is 16,950 ha in Tasmania) and fire emissions threshold (2 standard deviations above the mean of gross annual fire emissions) to exclude large fires or fire years, with the assumption that the fires were not human-caused and therefore are under no human control . These statistically large fires are therefore attributed as natural disturbances and are excluded in the final carbon accounting. It is therefore likely that the Forcett-Dunalley fire (with a burnt area of > 20,000 ha was excluded based on these criteria despite it being anthropogenically-caused. While there is some logic to this reasoning, there is uncertainty as to how to treat severe wildfires, such as the Dunalley disaster, that are human-caused, are exacerbated by anthropogenic climate change, burn over a highly human-modified landscape, and are subject to intensive human control efforts, yet they exceed the above threshold for defining anthropogenic fires.
Although, it is commendable that from the year 2019, the Australian government can report to IPCC fire emissions within the ‘natural disturbance’ provision , we recommend inclusion of all emissions from large, human-caused fires both at state and national levels in the final accounting, to prevent situations where net carbon credits are claimed despite insufficient fire management. Current accounting approaches can potentially lead to perverse outcomes where carbon neutrality could be claimed by reducing the extent of planned fires that are an important tool in mitigating uncontrolled bushfire and reducing emissions (Fig. 7). Current arrangements therefore provide disincentives to effective wildfire management to reduce carbon emissions from large fires that exacerbate climate change. Furthermore, the national policy is inconsistent because in north Australian savannas, there are carbon emissions abatement programs which reward pre-emptive early dry season burning to limit the high smoke emissions associated with late season burning .
The way wildfires are treated in Tasmanian government’s GHG reporting reveals that since 2012, forestry-related activities (LULUCF) have counteracted anthropogenic non-forestry GHG emissions , with an average removal of -9.17 Tg between the years 2012–2019, and an increased carbon sequestration from − 5.920 Tg in 2012 to -10.04 Tg in 2019. These estimates seem impressive; however, they are unaffected by major wildfires such as Dunalley disaster that according to the GFED model, accounted for one third of the state’s annual fire emissions. If severe-fire emissions were incorporated in the forestry-related GHG accounting for 2013 (-10.952 Tg in forest land), Dunalley CO2 emissions (1.125 Tg) could have reduced forest land CO2 sequestration (or removal) by 10%. These results suggest that if wildfire emissions are included, then Tasmania may not be actually achieving carbon neutrality.
An important consideration in the understanding and accounting of carbon emissions is the influence of climate change on, and feedbacks with, fire regimes. In the GHG accounting across many national jurisdictions, the emitted carbon from wildfires is assumed to be assimilated by forests in the following growing seasons via tree growth, and therefore carbon uptake post-fire can be substantial. However, it is not clear how the regrowth and carbon sequestration can be relied upon in a changing hotter or drier climate. For instance, a warming earth has increased the vulnerability of ecosystems to frequent and intense fires, which in turn emit large quantities of emissions, thereby creating a positive feedback loop where forests are converted to a treeless state . This calls for more investigation using diverse tools ranging from experiments, observations and models, to understand the complex interactions between climate, ecosystem structure and fire dynamics.