We calculated carbon emissions from the forest fires in the western United States (Figure 1) between 1984 and 2020 and the largest fire in the continental United States, the 1910 Big Burn. Here, we group the “western United States” as the 11 states in the contiguous United States West (i.e., Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, Washington, and Wyoming). While there were other extreme forest fires in the 20th century (e.g., 1902 Yacoult Burn in Washington, 1933 Tillamook Burn), historic records of forest attributes were not available for analysis.

Extreme fires have continued to occur in recent decades (Figure 2). Availability of high-resolution fire perimeter and burn severity data allows for analysis of fires since 1984 (Eidenshink et al., 2007) through 2020. All wildfires >526 ha (1,000 acres) with >50% forest area within the burn perimeter were included in this analysis. In addition, we selected large, notable forest fires (or complexes of individual fires; referred to as “extreme” fires throughout the manuscript) that occurred between 1984 and 2020. Fires were selected based on how notable they were at the time for size, duration, volatile fire behavior and legacy of impact in the subsequent years. Wildfires included in the “extreme fires” list (Table 1) were chosen from the created emissions database based on high area burned (i.e., >40,000 ha), and overall significance of fire event (i.e., most were record events in some way such as: highest area burned in that state or human impacts).

Emissions have not been previously calculated for the Big Burn. We have calculated an estimate from the Big Burn not only because it is the largest known fire to have occurred in the continental United States, but also to serve as a baseline or reference for the range of emissions possible in the absence of fire suppression. While the Big Burn emissions estimate is calculated differently from modern fires due to lack of forest data from that time, the comparison between modern fire emissions and the Big Burn is still useful and has been completed with the best possible methodology given data availability. The 1910 Big Burn encompassed an area throughout Washington, Montana, and Idaho (Figure 2). We calculated fire-induced carbon emissions of the Big Burn using historical accounts and records (Koch, 1942). The fire perimeters used in this study were a cross-reference between Koch’s account and the 1910 Fire perimeters (USFS, 1978; Gibson, 2005).

Direct carbon emissions for contemporary forest fires (1984–2020) were calculated using, fire severity and area burned from the Monitoring Trends in Burn Severity database (MTBS; Eidenshink et al., 2007) for forest fires between 1984 and 2019, and Burned Area Emergency Response (BAER; Parsons, 2003) and National Interagency Fire Coordination (NIFC, 2020) data products for 2020 fires. Carbon emissions for the 1910 Big Burn were calculated using area burned from the Northern Rockies Fire atlas for the Big Burn (Gibson, 2005). All carbon stock calculations were from forest type and ecoregion-specific carbon data (Supplementary Figure 1; Buotte et al., 2019; Stenzel et al., 2019), and severity-specific combustion factors (Supplementary Table 1; Stenzel et al., 2019). Only fires that burned > 526 ha and in 50% or greater forested area within the burn perimeter (Ruefenacht et al., 2008) were used in this analysis. Aboveground carbon stocks were calculated for each forest fire area based on average carbon stocks for the forest type and ecoregion and area of the specific forest type within the burn perimeter (Supplementary Table 1). Aboveground carbon stocks were multiplied by the appropriate combustion factor for the fire severity value of that forested area to obtain carbon losses (Supplementary Table 1). Fires between 1984 and 2019 were calculated using MTBS severity classes. Smaller forest fires in 2020 were calculated using an average combustion factor. Big Burn carbon emissions were calculated using a range for the moderate-severe combustion factor which gives us a range (uncertainty) of emissions for this fire. Extreme 2020 forest fire emissions were calculated using BAER severity classes (which are precursors to MTBS severity classes). While we used contemporary forest structure data to calculate emissions from the Big Burn, our range of carbon emissions (Table 3) from this fire are robust because we used stem tree biomass and demographic data from pre-1910 timber cruise inventories (Koch, 1942) to identify FIA plots of similar structure (diameter and heights) and age classes to the 1910 inventory.

We calculated mean annual forest fire emissions for the western United States and each western United States state based on the 2009–2018 decade, to best represent the observed trends toward increased area burned under climate change (Abatzoglou and Williams, 2016). These calculations were completed using MTBS (Eidenshink et al., 2007) perimeter and severity data, carbon stock data, and combustion factors. All fire carbon losses were converted to Tg CO₂e (i.e., tera-grams CO₂ equivalent) for comparison with fossil fuel emissions (FFE). To normalize emissions on a per area basis, we calculated megagrams of carbon lost per hectare burned.

To calculate uncertainties for contemporary forest fire emissions, we used a propagation of error approach. We combined uncertainty estimates of each emissions calculation component to calculate total uncertainty for each individual extreme fire event (Table 4). We used the Combining Uncertainties Propagation of Error estimates from the 2006 IPCC Guidelines for National Greenhouse Gas Inventories (Eq. 1; IPCC, 2006).

Equation 1: Combining Uncertainties (individual associated uncertainties):

Where U_total = the percentage uncertainty in the sum of the quantities (half the 95 percent confidence interval divided by the total, i.e., mean and expressed as a percentage).

x_i and U_i = the uncertain quantities and the percentage uncertainties associated with them. x_i refers to the specific fire emission calculation. U₁ refers to the uncertainty in biomass calculations (0.05) and U₂ refers to remote sensing uncertainty in area burned calculations (0.10).

Calculations and spatial analyses were conducted using R (R Core Team, 2017) and ESRI software (ESRI, 2020).

Fossil fuel emission numbers were sourced from the Environmental Protection Agency (EPA, 2020). The EPA provides yearly, state-by-state FFE from nearly all emissions sources (i.e., commercial, industrial, residential, transportation, and electric power; Table 4). We calculated an annual average for the most recent available data for each state (2009–2018), as well as an annual average for the entire western United States.

Carbon emissions from forest fires and FFE were normalized on a state-by-state basis by normalizing both average FFE (2009–2018) and record year fire emissions (1984–2020) with average fire emissions (2009–2018). A factor of both average FFE and record year fire emissions over the average fire emissions was calculated for each state.

Carbon emissions from 1984 to 2020 wildfire events varied considerably by fire severity (Supplementary Figure 1), forest type (e.g., varied carbon density), and size (Figure 2). As forest fire carbon emissions are a product of forest type (pre-fire aboveground carbon density per area, Figure 3) and fire severity (Table 4), it is notable that Colorado fires generally exhibit higher emissions per unit area (27.60 Mg C ha^–1) compared to other 2020 fire events in Oregon and California, although the 2020 Creek Fire in California had the highest emissions per unit area for a single contemporary fire (Table 4). This highlights how severely Colorado wildfires have burned in recent decades, given their lower pre-fire carbon density. By contrast, Oregon forests have much higher pre-fire carbon density and slightly lower area-normalized emissions compared to other western fires because they burned, on average, at a lower severity (Figure 3). When normalized by area burned to control for size, there is notable variation amongst the contemporary extreme fires in emissions per hectare (Table 4). For example, the 2020 Creek Fire in California had the highest emissions per hectare (29.7 Mg C ha^–1), and fires in Colorado all exceed the Idaho, Montana, and Wyoming fires, and many in the carbon-dense forests of Oregon when normalized by area. In addition, >90% of the burn area of extreme forest fires in 2020 were in low-to-moderate severity classes (Supplementary Figure 1).

Total average annual western United States total harvest emissions were lower than total average forest fire emissions (Table 4), however, actual harvest area is much lower than area burned (Berner et al., 2017). Actual harvest carbon losses vary greatly by state, with carbon-dense Oregon and Washington having the highest biomass harvest removals (Table 4). However, on a per unit area basis, hypothetical 100% harvest is 2–8 times greater than fire for the same perimeters across the entire region (Figure 4). We calculated hypothetical harvest carbon losses for the exact burn areas in each state to compare per area harvest losses. We found that for all states a 50–100% harvest would have led to greater carbon losses than fire for those burned areas, and even a 30% harvest led to greater carbon losses than fire for all but four of the western United States. Hypothetical harvest carbon losses continue to outpace fire carbon losses on a per unit area basis for most scenarios (Figure 4).

Anthropogenic fossil fuel emissions (AFFE) for each western United States and for the total western United States substantially exceed forest fire carbon emissions (mean annual and 2020), and average actual timber harvest (Figure 5 and Table 4). Mean annual AFFE in the western United States were over 15× higher than mean annual forest fire emissions and mean annual AFFE were 420% higher than forest fire emissions from the 2020 record fires across the west (Figure 5). Emissions vary widely by state, primarily due to population size (i.e., population and FFE are positively correlated) and large-scale, high-emissions industries within the state. Total western United States 2020 fire emissions were higher than the mean annual fire emissions (2009–2018), driven by large fire events in California, Oregon, and Colorado (Table 4). California, Oregon, Colorado, and New Mexico all had record-high forest fire emissions in 2020.

Public perception and existing overestimates of forest mortality and carbon emissions from wildfire feeds into the misconception that wildfire kills all live forest cover and combusts all forest carbon (Wiedinmyer and Neff, 2007; Mater, 2017; Zinke, 2018). The reality of actual fire emissions calculated from mixed-severity combustion rather than overestimates calculated from the false high-severity narrative highlights the need to disentangle ecological impacts of wildfire from societal impacts (i.e., loss of lives and houses). This will help to ensure that risk-reduction solutions can decrease wildfire disasters while still maintaining ecosystem services, such as live tree carbon uptake and wildlife habitat (Kolden, 2020).

As wildfire policy discussions increasingly include extractive forest harvest to mitigate forest fires (Executive Order, 2018; Newhouse, 2021; Senate Bill 762, 2021), a comparison of emissions from forest fire, timber harvest, and fossil fuels provides a more complete understanding of the relative contributions of emissions sources to anthropogenic climate change. Despite increasing area burned trends across the western United States (Parks and Abatzoglou, 2020), FFE still greatly outpace forest fire emissions in the last decade, including 2020. FFE are also significantly higher than intensive and large-scale land management operations like timber harvest in many United States (i.e., California).

Much of United States fire management and policy has been shaped by specific, previously unprecedented wildfire events. The Big Burn of 1910 was the first massive fire event for the fledgling United States Forest Service (Koch, 1942), and is still the largest wildfire complex that has occurred in the contiguous United States. Fire suppression as the main form of fire control persisted until the late 20th century, when ecological restoration efforts began seeking to reduce hazardous fuels and increase ecologically beneficial fire effects (Parsons et al., 1986). However, these efforts have not yet altered the fire suppression culture instilled by 1910 (Stephens and Ruth, 2005; Kolden, 2019; McWethy et al., 2019).

Like past extreme fire events, the 2020 and 2021 fire seasons have accelerated fire policy and forest management discussions at all levels of government – federal, state, and local – including recent bills introduced in the United States Senate (S.4625, S.4331). Many new policy discussions on fire and forest management are being based upon the misconception that harvest will protect forests from mortality and carbon loss (Executive Order, 2018; Zinke, 2018; Infrastructure Investment and Jobs Act, 2021; Newhouse, 2021), and decrease fire risk (Forest Climate Action Team, 2018; Figure 1) despite substantial uncertainty over long-term impacts to forest climate resilience (i.e., forest treatments may decrease forest resilience in the era of climate change). Our results and the majority of full-carbon accounting studies conclude that any type of harvest (logging or commercial thinning) decreases forest carbon storage (Law et al., 2013), and this research shows harvest emits more carbon per unit area than fire at all scales (Figure 5).

To mitigate climate change, it is key we understand exactly where emissions are originating. While increased intensity and size of fires are increasing overall fire emissions, these emissions are still substantially less than FFE. This is true even in record forest fire years (Figure 5). While the 2020 fire season was unprecedented in many ways (record forest area burned in California, Oregon, and Colorado; societal devastation including fatalities, thousands of homes consumed, dramatic evacuations, and regional hazardous air quality events), ecologically, most of the forested area burned in extreme fires was in a low-to-moderate severity class (>90%, Supplementary Figure 1). Moreover, while there is assumed high tree mortality in these forest fire perimeters, many of these burns were mixed-severity fires (Supplementary Figure 1 and Figure 3) meaning many live trees will persist across most of the low-to-moderate severity burned areas. Locations with high-harvest rates and carbon-dense forests, such as the Pacific Northwest United States, see higher carbon losses from harvest than fire compared to areas in the Southwest United States with low harvest rates and carbon-sparse forests (e.g., Oregon versus Arizona). Forest management needs to be specific to forest type and region; old-growth and wet forests in the Northwest are best left preserved while dry, fire-prone forests or areas in the Wildland Urban Interface benefit from fire risk-reduction strategies like small-diameter thinning and prescribed fires (Law et al., 2018; Case et al., 2021). Inclusion of specific diameter limits in policy for public lands could help prevent large-diameter tree removal and subsequent unintended consequences.

Forest management strategies that are site-specific and balance the immediate protection of life and property with long-term preservation of existing and potential carbon stocks in forests are critical to mitigating the negative impacts of climate change. The most effective forest management strategy to protect forest carbon stocks on public lands is to preserve forests through decreased harvest and thinning, lengthened harvest rotations, increased proportion of long-term wood products, reduced harvest and mill waste, and working toward afforestation and reforestation (Hudiburg et al., 2013; Law et al., 2018; Buotte et al., 2020; Figure 1). Prescribed burns reduce fire risk while minimizing carbon losses and amplify tree growth and carbon sequestration in large-diameter trees in fire-adapted forests (Hurteau and North, 2009). In western United States forests, 33 to 46% of aboveground live biomass is stored in the large diameter trees (>60 cm; Lutz et al., 2018; Mildrexler et al., 2020). Carbon-smart treatments on public lands need to be specific about diameter limits to avoid large-diameter tree removal.

Here, we have shown that FFE for the western United States are 7 times greater than emissions associated with timber harvest and fire (Figure 5 and Table 4). As more forest-fire policy and management plans are expanded, written, and discussed following extreme fires of the recent decades, and especially the extreme forest fires of 2020 (DNR, 2020), it is crucial that these policy changes focus on the largest driving factor of these fires – anthropogenic climate change. In practice, large-scale extractive forest management efforts will hamper climate mitigation and may be futile for decreasing fire risk. To be most effective, policy will need to focus on fire-wise adaptations for homes and property and disentangle ecologically-good fire from destructive fires (Kolden, 2020). Protecting forests with ecologically sound principles, rather than increasing extractive management, may be the best scenario for the mitigation of climate change (Law et al., 2018), and protecting humans, biodiversity, and forests (Walsh et al., 2019; Buotte et al., 2020; Law et al., 2021). The continued escalation of fires throughout the 21st century is evidence of climate change-mediated intensification of fire regimes in the United States (Williams et al., 2019). Fire catastrophes will continue to occur and worsen if we do not focus on decreasing FFE, the primary driver of climate change (IPCC, 2018).