Carbon reporting in Canada's forests is broken down by Reconciliation Unit (RU), which is the intersection of terrestrial ecozones and provincial/territorial boundaries, as shown in Fig. . Forest carbon stocks in Canada are influenced by species composition, age class distribution, natural disturbance history, human-influenced management regime and climatic variables, among other factors, which are well-represented by the RU framework supporting Canada's NIR. Due to the current spatial imprecision of certain key model data (e.g. deforestation rates), analysis finer than the RU-level tends to misrepresent regional-variation in emission estimates.

The carbon pools represented in CBM-CFS3 can be broadly categorized as live biomass, dead organic matter (DOM) and are described in detail in . Key components that interact most with wildfire are summarized in Table . Though designed for forest GHG reporting, the pool structure of CBM-CFS3 does distinguish between softwood (conifer) and hardwood (broadleaf) that also have large contrasts in both foliage flammability and ability to survive low to moderate intensity surface fires . Tree biomass pools are further divided into merchantable tree biomass pools (foliage, stemwood, branches), with separate hardwood and softwood pools for smaller, subcanopy trees. Representative pool values (in t C ha⁻¹) per RU are produced and updated annually based on updated forest inventory and disturbance data. Pool size information by RU is provided in Table S1 of the Supplement. Upon the detection of a forest disturbance such as fire, pool C biomass (both dead and alive) can be transferred to other pools or the atmosphere (or remaining in the same pool) at proportions prescribed by a DM. Any live or dead biomass pools combusted immediately are tallied in the direct emissions sums reported here. Live biomass that is killed but not combusted by the fire is tracked, but experiences delayed decomposition on the order of years to decades, and therefore is not tallied within the direct emissions modelling shown below. See Appendix A for DM examples.

To simplify the process of creating DMs as a distillation of the complexities of fire severity and combustion patterns, the following statements are proposed and maintained throughout:

Disturbance matrices are to be in terms of mortality, not survival. Mortality here is defined as tree death by the end of the calendar year of the fire's occurrence. Tree mortality in subsequent years is not modelled here.

Of these, Crown Fraction Burned (CFB) is an important concept used primarily in fire behaviour science but not carbon accounting nor fire ecology. CFB was introduced in the 1992 Fire Behaviour Prediction System documentation , and provides a simple continuous 0–100  ariable for only the consumption of foliage (which we assume here is inclusive of both conifer and broadleaf). For our purposes, CFB is the desirable metric as opposed to ordinal and less precise systems like Canopy Fire Severity Index that allows the user to specify which pools of canopy biomass are consumed, but not the precise fraction of each given pool that is consumed.

Certain variables, like the partitioning of CO₂ : CH₄ : CO gas emissions, are constant throughout ecozones, but vary by flaming vs. smouldering combustion modes. The precise emissions ratios vary slightly between models and field studies, but for this initial algorithm assessment, we define these emissions ratios as being identical in the flaming phase to those used in Canada's operational wildfire smoke forecasting system, FireWork . Smouldering emissions factors used in this framework are taken as the arithmetic mean of smouldering and residual combustion phases from FireWork. Emissions ratios are provided in Table .

CO₂ is responsible for 86.8 % of emissions in the flaming phase, but only 70.3 % of emissions in the smouldering phase, with a doubling of CO emissions and tripling of CH₄ emissions (Table ). With a Global Warming Potential of CO equal to 1.0 and CH₄ of 28, the Global Warming Potential per unit of biomass consumption in the smouldering phase is 1.08 times higher in global warming potential compared to flaming, not including differential aerosol production and injection heights, however. With flaming and smouldering each contributing roughly equally to wildfire emissions, these distinct flaming and smouldering emissions ratios correspond well to prior emissions factors used in CBM-CFS3. Note that as currently described, the sum of CO₂, CH₄, and CO emissions from wildfires only represent approximately 95 % of the fire carbon mass emitted to the atmosphere, with 0.5 %–2.0 % of biomass emitted as particulate matter (e.g. PM_2.5, but also PM₁ and PM₁₀ classes of particulates at 1 and 10 µm diameters, respectively), and an additional 3 % to as little as 1 % composed of non-methane organic gases that have a large range in global warming potentials as compared to CH₄.

Remotely sensed fire severity metrics (i.e. dNBR and related indices) are capable of determining if a specific location falls within a low, moderate, or high severity fire area by utilizing the change in the red and shortwave infrared band ratios after a fire . These remotely sensed fire severity classes can not alone inform estimates of fuel consumption down to specific biomass pools. Instead, this approach uses the change in biomass pools as measured at ground plots at varying satellite burn severity classes, using detailed and semi-standardized methods of burn severity ground plots with unburned controls e.g.. Burn severity ground plot measurements are often aggregated into a single “Composite Burn Index” (CBI) with a weighting scheme originating in southwestern US forests , which are classed into low, moderate, and high severity categories. Plot level measurements of individual forest biomass pools (e.g. conifer overstory mortality or canopy consumption) are recorded as part of the CBI methodology.

Individual burn severity plots were assigned a severity class based on ground observations, and an ecozone-level median biomass pool consumption fraction was computed for each severity class for each biomass pool. Field measurements of fire effects on individual ecosystem components (e.g. tree mortality, crown consumption, unburned surface litter area, etc.) taken as part of CBI plot measurements were aggregated to the plot-level severity class as determined by CBI. For instance, plot level data on tree mortality percentage was available across numerous plots within Canada's Boreal Plains ecozone; median values of tree mortality per severity class (low, moderate, high) were computed based the plot overall severity score as determined by CBI in order to compute ecozone-specific values. Field studies with data describing biomass consumption fractions for individual biomass pools were available from the northwestern boreal forest of Canada . Smaller datasets with eastern Canadian and Alaska data were also available , as were data from the northwestern conterminous United States in ecozones that overlap with Canadian forests . Detailed mortality and surface consumption data were not available for the Atlantic Maritime and Mixedwood Plains ecozones, so an estimate was made by subsetting data from to a moderate basal area range (25–50 m² ha⁻¹) mature eastern mixedwood and conifer-dominated stands. For the Boreal Cordillera ecozone, data were similarly subset to only sites with stand heights < 15 m. Field data for large diameter mature hardwood trees was only available from (primarily aspen, Populus tremuloides), and are used in all other ecozones without modification.

For ecozones without ground plot measurements, an adjacent and ecologically similar ecozone with observations was used. As such, Boreal Plains observations were used in the Prairies, Boreal Shield, and Hudson Plains Ecozones. Similarly, Taiga Shield West observations were used in the Taiga Shield East. The field survey data on biomass combustion and mortality rates as stratified by severity class, forms an improved data-driven and ecologically-informed approach to direct C emissions estimations from fire termed FireDMs (Fire Disturbance Matrix-severity).

Numerous process-driven or empirical tree mortality models are available that show significant skill in predicting tree mortality based on fire behaviour (i.e. flame length, rate of spread). Since the driving data in this model is the mapped fire severity over the landscape scale, fire behaviour metrics such as flame length or scorching height of bark are not available as a continuous mapped product. Instead, softwood and hardwood overstory mortality is calculated per ecozone as a function of satellite-observed fire severity using the aggregated ground plot data (Table ).

And since large-diameter, live trees killed by fire do not experience significant live stemwood consumption , the entirety of the live stemwood biomass pool that is killed is transferred to the snag pool. Note that the field data and disturbance modelling undertaken here only accounts for tree mortality within the calendar year of the fire, and delayed mortality of over one year has been documented in boreal low and moderate severity fires where less than half of total mortality occurs after the year of the fire. Thus, the modelling here does not account for delayed mortality that may extend upwards of 5 years after fire.

Crown Fraction Burned (CFB) speaks to the fraction of the live canopy that is itself consumed in the flaming front. The alternate outcomes being survival of the foliage, or the mortality of the tree without canopy consumption, resulting in the dropping of foliage onto the forest floor. From the definitions stated earlier, the CFB must be lower than or equal to the mortality rate. Using field studies that show any partial crown consumption is likely sufficient to result in high rates if not complete mortality , which is the case in Canadian trees, which are primarily species with thin bark. Due to the structure of CBM-CFS3, all high severity fires h 1.0, which is no more than a 5 % variance from observed values. Ecozone-specific CFB values from field studies of burn severity are summarized in Table .

The consumption of live bark biomass is calculated in FireDMs, and consumption rates can be defined by severity class. At the moment, lacking robust field data on bark biomass consumption rates across ecozones and severity classes (which are a small portion of the overall biomass), the bark proportional consumption rate is set to 34 % of the overstory mortality rate .

A major distinction is made between softwood and hardwood trees, where in Canada's boreal forests, a large fraction of hardwood trees (see Appendix B) are able to resprout even when the main stem has been killed by an intense forest fire . Accordingly, the root mortality rates differ greatly between softwoods and hardwoods, with softwood root mortality equal precisely to stem mortality, while in resprouting hardwoods, little root mortality is observed even after intense fire . Though spatially explicit versions of CBM-CFS3 can resolve a species list down to the pixel level, currently an ecozone-level regional average composition of hardwood species with resprouting traits (“ReSproutFactor”) is used and is shown in Table .

The fraction of fine roots contained within the combustible forest floor layers (the variable “Prop.Fine.Root.duff” in the model ) can be a close to or exceeding 50 % of the fine root biomass ; this fine root live biomass is assumed to burn alongside the organic soils, . Fine root allocation is between the organic soil (termed “aboveground” or AG in CBM-CFS3 vs. “belowground” or BG for mineral soils) is a parameter that can be ecozone level and is distinct for hardwoods (HW) vs. softwoods (SW). As a result, the calculation for softwood fine root consumption and stemwood mortality rate (HW.Mort, as observed from field plots) are as follows, using hardwood as an example: 1HWFineRootConsump=HW.Prop.Fine.Root.duff×Duff.Consump.Fract2HWFineRootMort.AG=HW.Mort×HW.Prop.Fine.Root.duff×(1-Duff.Consump.Fract)×(1-ReSproutFactor)3HWFineRootMort.BG=HW.Mort×(1-HW.Prop.Fine.Root.duff)×(1-ReSproutFactor). In contrast, the larger diameter of the coarse root biomass pool prevents its consumption during any smouldering of the duff layer, and the mortality rate of coarse roots is simply proportional to that of the stemwood overall.

Understory (or small diameter overstory) tree mortality is defined separately in the model in the “submerchantable” pool, but given the lack of data on diameter classes in the severity data, robust field data on differing mortality rates of smaller diameter trees is not available, and so the understory tree mortality rate is set equal to the overstory rate as defined in Table  above. Note that smaller trees with a top height less than 1.4 m are not considered in this pool, and instead are lumped into the “other” pool.

The litter layer forms the first biomass pool in which a spreading fire consumes fuel. In low-severity fires, the litter layer is consumed, but underlying duff material and live trees largely remain intact . In these low-severity patches, some areas of fully unburned forest floor are present, and are summarized from CBI plot data by severity class and ecozone in Table . Logically, since litter consumption is largely required for the ignition of the underlying duff layer, this unburned litter areas informs and constrains duff consumption.

Dead standing trees and branches on average is a biomass pool smaller in size compared to organic soils in much of Canada's forest, but in areas of extensive insect-killed trees, recent fires, and blowdown, can be a substantial biomass pool. Compared to live stemwood of the same diameter, the low moisture content of snags and snag branches allows for much greater consumption during the passage of an intense flaming front . The snag branch pool experiences almost complete combustion in high severity fires , and is highly correlated with CFB . The largest dead wood pool is the standing dead stemwood pool, which remains at most approximately 50 % consumed; the consumption of standing snags pool is not quantified precisely in the severity plot data available. In the absence of extensive field data, snag consumption is estimated as: 4SnagConsumptionfraction=(0.5×CrownFractionBurned)+0.05. And for snag branches, field observations reliably point to a higher rate of consumption for this smaller diameter dead biomass pool: 5SnagBranchConsumption=(0.9×CrownFractionBurned). Stumps are tracked as part of the “Other” pool in CBM-CFS3. As they are typically in contact with the upper forest floor, they are assumed to be consumed at the same rate as coarse woody debris.

Limited data is available on the fraction of woody debris consumption alongside fire severity measurements from wildfires. Coarse woody debris of overstory stems that makes up 60 %–80 % of woody debris biomass in Canada's boreal and temperate forests , with its moisture and consumption patterns largely following the moisture regime of the Drought Code . In this modelling framework, the proportion of coarse (>7.5 cm diameter) and medium (>0.5 and <7.5 cm) woody debris consumption is estimated based on detailed measurements of consumption from experimental fires. Coarse woody debris (CWD) is responsible for approximately 50 %–75 % of the total woody debris load in most ecozones, and approximately 60 % of the total woody debris consumption. Ecozone-level CWD consumption rates are summarized in Table .

Note that where historical burn severity data from experimental fires is not available, the fire recorded fire classification types of surface, intermittent crowning, and active crown fire are used instead as proxies for low, moderate, and high severity fire, respectively. Fine woody debris < 0.5 cm in diameter is consumed at the exact same rate as the litter pool (see section above).

Unlike canopy consumption metrics that are reliably estimated via burn severity remote sensing, the depth and magnitude of consumption in the forest floor is only weakly related to burn severity metrics, since increasing burn depth in organic soils typically yields little spectral response in remote sensing imagery . Furthermore, there is at times a temporal disconnect between consumption in the canopy and the forest floor, with lower-intensity surface fires with a flaming residence time of a minute or less without canopy involvement at times leading to deep smouldering occurring over hours to days afterwards . Consumption of fine fuels in the litter layer of the forest floor is nearly complete for any given surface fire intensity , consumption of deeper organic soil horizons (F + H layers in upland forests and upper peat layers in wetlands) is more dependent on the interaction between moisture content of the soil organic layer and depth of the organic soil . While the moisture content of the soil organic layer is not directly observable nor parameterized in many numerical weather models , it is well-estimated by simple water balance models within the Canadian Fire Weather Index System , the Drought Code and Duff Moisture Code, as well as their composite, the Buildup Index. Given this lack of direct observability of forest floor consumption, this model uses fire weather and fuel loading as inputs to the forest floor consumption model, without consideration of the remotely sensed fire severity observations used in canopy consumption modelling. This approach mirrors many established and widely-used approaches in Canada such as CanFire (e.g. ) as well as Consume in the United States.

In the fire literature in Canada, the soil organic layer is sometimes termed the Forest Floor Fuel Load (FFFL) and is dominated by the equivalent organic soil Slow pool (Aboveground Slow Dead Organic Matter, or AGSlowDOM) in CBM-CFS3. Historically, attention has been paid to the absolute value of Forest Floor Fuel Consumption (FFFC) ; however in the case of carbon modelling, it is the relative fraction of consumption (FFFC/FFFL) that is of interest. Given the strong empirical evidence that the amount of organic soil present strongly influences the absolute and relative combustion rates , modelling of forest floor consumption was conducted on field measurements of the relative fraction of organic soil consumption (Fig. ). In this scheme, we utilize a composite of wildfire data from alongside the ABoVE duff consumption data , with an alternative modelling approach to compute the relative amount of consumption (scalar from 0 to 1) rather than directly modelling an absolute value in kg m⁻² or cm as otherwise done in the literature. A logit transform is used on the scalar data to make it suitable for the fitted non-linear least-squares modelling: 6logitFFFCFFFL=3.911-e(-0.008BUI)+-0.53log⁡e(AGSlow) where BUI is the Fire Weather Index System's Buildup Index, and AGSlow in the CBM-CFS3 (given in Mg C ha⁻¹ in this equation), and also synonymous with the the Forest Floor Fuel Load (with ecozone averages given in or site-level data where observed) (Figs.  and ). Assuming an average carbon content of 50 %, CBM-CFS3 values in Mg C ha⁻¹ are converted to fire behaviour units of kg of biomass per squared metre by dividing by 5 (Fig. ).

While ultimately this scheme can be used on individual fires with estimated or measured fuel loading and specific BUI values, as an example, an ecozone-averaged fuel load and a multi-year average of BUI during active fire satellite detection can also be used to provide representative values to showcase the model framework. Specifically, a median BUI of detected fire hotspots in Canada from 2003–2021 using the same data as the Canadian CFEEPS-FireWork wildfire air quality model of is presented in Table , along with proportional consumption values of the forest floor by ecozone.

Note that the maximum upland Forest Floor Fuel Load is approximately 30 kg m⁻² (150 Mg C ha⁻¹) ; higher values are typically seen only in peat ecosystems, where the above scheme is still valid, as approximately 33 % of the data used to fit this model is for organic soil pool values over 30 kg m⁻² (>150 Mg C ha⁻¹). Fully 10 % of the data used in the FFFC model is for sites with over 60 kg m⁻² of organic soil. Absolute consumption values are similar between deep forest floor organic layers and peatlands , though relative consumption rates decline rapidly in deeper organic soil layers, as demonstrated in the model selected here. For Canadian peatlands, the CaMP model is instead used in spatially explicit version of CBM-CFS3. Within CaMP, a separate peatland water model driven by Drought Code determines the thickness of the unsaturated peat layer, and an amount approximating 12 % of the thickness of the unsaturated peat is consumed as smouldering consumption. The peat-specific carbon pools and fire disturbance matrices are fully described in ; large peatland trees will still utilize the DM scheme described below. Since deeper forest organic soil and peat layers show approximately similar absolute consumption rates, for the purpose of this general model description no peatland-specific components of the fire DMs are required.

MCE observed by aircraft from during the peak burn period (when the majority of fuel consumption and area burned occurs) largely corresponded with modelled values (Table ). This suggests the model provides a fair representation of the balance of flaming and smouldering during large active wildfires. Subsequent smoke plume observations during periods of greatly reduced spread and intensity (late evening and after rain) showed a substantially reduced MCE of 0.82–0.83 that indicates a near lack of flaming combustion. Even when represented as 100 % low severity fire in the model, the modelled MCE only declines to 0.90, suggesting FireDMs does not adequately capture carbon emissions ratios during marginal smouldering conditions. Since the fire DM model is based on area burned, fire activity such as smouldering but with minimal actual area burned increased is going to show an MCE far lower than areas of low severity fire spread, which are typically still a low sub-canopy flaming front that features a mix of smouldering duff and woody debris alongside flaming consumption of litter . During the same ARCTAS campaign over a larger sample of 39 fires, observed a mean MCE of 0.89 for smoke plumes within 2 km of a large wildfire.

The forest floor emissions modelling scheme used here builds upon the CanFIRE model which incorporates a similar estimate of absolute forest floor emissions based on fuel type and various Fire Weather Index inputs (including but not limited to Buildup Index). While the CanFIRE model is also capable of estimating tree mortality, the premise of CanFIRE lies in the use of the FBP System to model fire intensity and a physiological-thermodynamic model of tree damage and mortality. CanFIRE is able to run entirely in scenario or forecast mode, i.e. no fire severity map is needed to run the model, unlike the focus here on a mechanism to estimate carbon fluxes and pools after fire severity is mapped. Ultimately, models like CanFIRE can be used for near-real time emissions estimates of direct and indirect fire carbon emissions (and also for prescribed fire planning and scenario testing), while this modelling framework is best used solely operational carbon accounting and reporting given its strong dependence on severity maps. While fire severity can be estimated empirically based on geospatial inputs without observation using multispectral satellite data such as Landsat, estimation using coupled fire behaviour and ecology models such as CanFIRE is a more direct approach. Under severe drought conditions and light winds, very large fires often create more wind and energy than is available in gradient winds as measured at nearby airports and fire weather observing stations . Thus, fire behaviour and fuel consumption models solely driven by wind inputs from distant meteorological stations can under-predict fire intensity and therefore likely fire severity. While fire-atmosphere models are available for research and some forecasting and planning purposes , modelling resources are typically not made available for the extensive and largely unsuppressed fires in northern Canada. Fire severity mapping after the fact is able to account for high fire severity even when fire spread occurs under light winds more associated with lower fire intensity .

While the fire DMs presented here are designed for use in NFCMARS for GHG estimation and reporting, their simplicity and similarity to many other forest carbon schema allows them to be used elsewhere such as in other forest dynamics and earth system models that require estimates of Canadian forest carbon emissions from wildfire. showed that severity-informed fire C estimates (with snags) in the US are only 30 %–40 % as large as fixed or variable severity data using their LANDIS-type models. Large over-estimates of in-fire bole consumption in models were not observed to correspond to measurements in the field. In the model presented here, our estimates of snag consumption are based on simple but logical relationships with CFB, but more comprehensive data collection and field observations would be required to extend the accuracy of the snag consumption scheme.

While fire emissions within the year of the fire are majority of the GHG flux, enhanced post-fire decomposition of dead biomass persists for many years after a fire. The increase in post-fire tree mortality between the year of the fire and the year following the fire is as much as 30 % in low and moderate-severity stands . This extended mortality is not captured in year-of fire severity mapping but is likely better assessed using fire severity data captured the year following fire, as is operational practice in Canada. Similarly, the transition of fire-killed stems (snags) from upright to the forest floor woody debris takes between 5–8 years for 50 % of stems to fall with stemfall rate highest in low-severity fire. Field observations show that initial assessments of fire severity are the primary predictors of snag fall and decomposition rates, meaning the mapped severity method used here provides a useful input for multi-year fire carbon modelling of the snag pool in CBM-CFS3. For organic soil pools, fire has been shown to suppress decomposition rates in upland soil pools where drier post-fire conditions are present , while in permafrost upland and peatland systems, fire has profound impacts on soil carbon pools by rapidly increasing thaw depths and subsequent decomposition rates. Currently in CBM-CFS3, decomposition rates are calibrated against field decomposition litterbag experiments and annual climate metrics that do not take into account how disturbances such as fire (and fire of varying severity) modify decomposition rates in the absence of a change in climate.

Currently, FireDMs only accounts for regionally-averaged soil carbon stocks, meaning that RUs with high peatland areas will show higher organic soil slow pool size, but peatlands themselves are not spatially represented in the model. Instead, peatlands influence the reported Aboveground Slow pool size. Currently, fire disturbance in peatlands is performed by the CaMP model, which uses the closely related Drought Code fire weather metric to estimate peat layer consumption and uses the overstory carbon schema of CBM-CFS3 to estimate overstory tree carbon pools and fluxes. Direct satellite monitoring of upland forest organic soil and peatland carbon loss via burn severity monitoring is possible but remains a challenge to implement with accuracy for monitoring purposes . Despite the large carbon stock (c. 500 Mg C ha⁻¹ from ), the mean C emissions per area in peatland due to fire (65 t C ha⁻¹) as modelled by are comparable to regional emissions from peat-rich ecozones such as the Boreal and Taiga Plains under moderate to severe drought conditions. Indeed, it is under these moderate to severe drought conditions that peatlands burn more frequently and severely .

While Canadian wildfires shown an overall selection bias towards preferentially burning older conifer stands , fuel selectivity against less flammable portions of the landscape decreases strongly under severe drought conditions . The severe drought conditions experienced in Canada in 2023 are an ideal setting to compare an emissions model using spatially referenced biomass data against observed emissions. The 2023 fires were observed to largely burn boreal forest ecosystems at the same rate as their abundance on the landscape, owing to severe drought and fire weather conditions . Emissions model evaluation against individual fires or in years of less profound drought would be more likely to highlight the limitation of the regionally-averaged carbon pool (i.e. fuel load) values used here. Ultimately, the mapped severity used here is best paired with mapped carbon pools at a similar scale, such as the 1 ha forest carbon modelling for analysis and scenario testing used by . Future work will extend this fire disturbance algorithm to finer-scale carbon reporting using gridded data as it becomes available for annual reporting purposes.

Species-specific traits in trees with overlapping ranges, such as resprouting in aspen but not other broadleaves such as oak or maple, are not explicitly handled in this model. Instead, the relative abundance of trees with contrasting traits that influence ecological outcomes such as fire survival are lumped at the ecozone scale from the aggregation of the plot data. While ecozone-level contrasts in key considerations such as overstory mortality do strongly vary by ecozone (Table ), the same DM (e.g. hardwood root mortality) is applied in adjacent stands at the same severity class despite the potential for strongly contrasting traits that are readily tied back to mapping at the stand level by leading species. Alternately, models such as CanFIRE can account for species-level contrasts in fire ecology traits, but compiling the relevant ecophysiological properties and traits for all tree species across the spectrum of fire intensity in Canada remains incomplete at this time.