Daily CO₂ emissions from fires in North America were estimated for 2002 through 2006 using the methods described by Wiedinmyer et. al. 13. Annually, the average CO₂ emitted from fires in the lower 48 (LOWER48) states from 2002–2006 is estimated to be 213 (± 50 std. dev.) Tg CO₂ yr^-1 and 80 (± 89 std. dev.) Tg CO₂ yr^-1 in Alaska. There is substantial variation in the overall magnitude of annual emissions from states in the US, ranging from the average of 80 Tg of CO₂ in Alaska to < 0.01 Tg CO₂ in Rhode Island and Vermont. Emissions from the Northeastern and Midwestern US states tend to be very small; the annual emissions from the US are dominated by the Western and Southeastern US states. For many Western and Southeastern US States, there are large annual fire emissions of CO₂ averaging ~10 Tg CO₂ (with an average coefficient of variance of more than 50%). The Northeastern states have the least amount of emissions per area: Vermont, Rhode Island, Maine, and New Hampshire all have an average annual fire emission of <1 metric ton CO₂ km^-2. The Southeastern and Western states have the largest amount of CO₂ from fires: Alabama, Florida, Georgia, Louisiana, and Washington all have an average annual fire emission > 75 metric ton CO₂ km^-2.

The interannual variability in the annual emission estimates is substantial. In the LOWER48, the annual emissions from year to year vary as much as a factor of 1.8, and in Alaska, the annual CO₂ estimates vary by over an order of magnitude. Overall, the interannual variance of fire emissions in the Southeastern US is lower than in the Western US. This interannual variability could arise from several causes, including changes in meteorology and climate (e.g., drought) and land management practices that deal with agricultural and prescribed burning.

Fires occur within the US for a number of reasons, including wildfires started from both natural and anthropogenic causes, prescribed burning, and burning for agricultural purposes. An analysis of the fire emission estimates presented here shows that the majority of the emissions from fires in the US are from needleleaf forests. For 2006, needleleaf forests are estimated to emit 78% of the CO₂ emissions from continental US fires. This suggests that, although important, natural and prescribed burning in grasslands and burning in croplands for agricultural purposes does not contribute significantly to the overall annual US CO₂ fire emissions inventory. CO₂ emissions from grasslands account for 5% of the 2006 estimated fire emissions inventory, and emissions from croplands contribute <3%. In both the Western and the Southeastern US, 86% of the estimated 2006 CO₂ emissions come from needleleaf forests.

The amount of area burned for management practices (prescribed burns) varies by region. In the Southeastern US, the majority of acreage burned is via prescribed burns. According to the National Interagency Fire Center (NIFC; 27) less than one third of the reported area burned in 2006 in the Southeastern states was due to wildfires; two thirds of the area burned was the result of prescribed burns. In Alabama, 94% of the 2006 reported burn area was attributed to prescribed burns. Since prescribed burns in the Southeastern US tend to occur between November and April 28 and the majority of emissions in this region come from needleleaf forests, we assume that much of the emissions through the spring and fall months (discussed below) can be primarily attributed to prescribed burns in forested areas.

In the Western US, fire-related CO₂ emissions are dominantly related to wildfire activity. A report for the Western Regional Air Partnership 29 estimates that 57% of the acreage burned in 2002 in the Western US States was due to wildfires, 23% for agricultural purposes, and the remainder for land management practices. Although the percentage of agricultural burned area was significant, the amount of biomass burned, and therefore the emissions, were relatively small in the overall inventory.

There is strong seasonal variation in fire CO₂ emissions, with regional differences in the peak emissions across the US. Generally, the monthly emissions of CO₂ from fires in the LOWER48 have two peaks: a small peak during the spring months (March and April) and a larger peak during the summer months (Figure 1). These two peaks are driven by the timing of fires in two distinct portions of the US, with spring fire emissions dominated by fires in the Southeastern and Central US, and summer fire emissions driven by emissions for the Western US (Figure 2).

Large, periodic fires can cause massive fluxes of CO₂ to the atmosphere. Figure 3 shows monthly CO₂ release from fires from six states including Alaska, four western US states, and Mississippi. These results illustrate the extreme variability in emissions associated with large fire events, such as the Columbia Complex fire in Washington in August 2006, or the Biscuit fire in Oregon in July of 2002, during which more than 15 Tg of CO₂ was released from each of these states.

In 2002, the Biscuit fire burned in Oregon from mid-July to September. The emissions from this fire were exceptionally large and drove the peaks in CO₂ emissions for July and August 2002 for Oregon (Figure 3). Estimates using the methods described here (see Methods Section below) predict 4.9 Tg C (from CO2) and 5.3 Tg C (from CO2 and CO) from the Biscuit fire (in Oregon only). Law et al. 30 used a simple method, based on the reported burn area and an assumed carbon loading, to estimate 4.1 Tg C from the same fire. The sizeable difference in emission estimates emphasizes the large uncertainty associated with estimating C emissions from fires. Estimates of fire emissions of CO₂ depend on a wide range of factors, including the severity and type of burns, as well as the spatial heterogeneity of vegetation and fire intensity 26. Combined, these factors make it exceptionally difficult to accurately measure C emissions from field-based techniques, regardless of methods used. Unfortunately, remote sensing-based methods also result in highly uncertain C flux estimates for fire, and there is currently no clear method available to reduce these uncertainties 19. Given these complexities, the flux estimates for the Biscuit fire made by this study and the Law et. al. 30 study are probably about as similar as can be expected. Law et. al. 30 applied a reported burn area, while the method employed in this study applied a burn area based on remote sensing observations. Both methods used different fuel loading estimates and emission factors. The impact of inherent uncertainty in emission estimates is that the high degree of variability (e.g., >25% of the flux) in fire emission estimates is not likely to be reduced soon and has implications for both our understanding of fires in the global carbon cycle and our ability to monitor and assess the causes of biosphere-atmosphere fluxes at a regional scale.

A striking implication of very large wildfires is that a severe fire season lasting only one or two months can release as much carbon as the annual emissions from the entire transportation or energy sector of an individual state. While the long-term atmospheric implications of wildfire and fossil-fuel C release can be strikingly different, the pulsed emission releases from wildfire events can match or even exceed monthly or annual industrial emissions on a regional basis. To examine the role of wildfire in the context of industrial emissions, we compare national and state level emissions of CO₂ from fossil fuel combustion to our estimated fire emissions of CO₂.

Annually, for the continental US (not including Washington D.C.), the average CO₂ emissions from all fossil fuel burning (FFB) sources from 2000 – 2003 were 5738 Tg CO₂ 31. Annual average CO₂ emissions for 2002 – 2006 from fires in the continental US was 293 Tg CO2, corresponding to the equivalent of 5.1% of the annual FFB emissions from 2000–2003 (and 5.4% of the average from 1990–2003). Depending on the year, emissions from fires for the entire Continental US were equivalent to as little as 4% of the FFB emissions, and as much as 6%. However, tHowhis is for the entire U.S; on a state-level, the importance of fire emissions of CO₂ relative to FFB emissions is much different. There are eight states (Alaska, Idaho, Oregon, Montana, Washington, Arkansas, Mississippi, and Arizona) where the annually-averaged (2002–2006) fire emissions are equal to more than 10% of the state-level FFB CO₂ emissions, and eleven other states whose fire emissions equal more than 5% of the state-level CO₂ emissions (Figure 4, Additional Table 1). In the case of Alaska, annually-averaged fire emissions of CO₂ (2002–2006) are consistently greater than the annually averaged (2000–2003) emissions from FFB (Figure 4). For the states located in the Western and Southeastern US, average annual fire emissions of CO₂ range from the equivalent of 2–4% of FFB emissions in North Carolina, Colorado, and Wyoming, to 89% of emissions in Idaho. (It should be noted, however, that Idaho does not have any coal-fire power plants, which emit large amounts of CO₂). For the Western US States, fire CO₂ emissions on average are equivalent to 11 ± 4% of annual FFB CO₂ emissions, and for the Southeastern US fire CO₂ emissions are equivalent to 6 ± 2% of annual FFB CO₂ emissions.

The relative importance of CO₂ emissions from fires to regional C emissions varies seasonally and annually. For example, during particularly intense fire years, such as 2006 in Idaho, the emissions of CO₂ from fires in Idaho were 1.6 times higher than all of the annually-averaged (2000–2003) FFB emissions from that state, and nearly double the mean annual fire CO₂ emissions for the state for 2002–2006. Similarly, in 2006, Montana and Washington experienced CO₂ emissions from fires during the year that were equivalent to ~47 and 42% of the total annual state-level FFB CO₂ emissions, respectively. In addition to significant interannual variation, regional fires are typically active for just a few months of the year. The monthly emissions of CO₂ from fires for 2002 through 2006 for six selected states are shown in Figure 3. Alaska, Montana, Washington, and Oregon all show large summer peaks in wildfire CO₂emissions that are of the same magnitude or greater than the CO₂ from FFB sources during those months.

In California, the annual FFB emissions inventory of CO₂ is the largest in the country behind Texas (362 Tg CO₂ yr^-1 averaged from 1990–2003). Even so, the annual averaged emissions of CO₂ from fires are significant (24 Tg CO2 yr^-1; equivalent to 6% of the FFB emission estimates). Although the ratio of annual state-level CO₂ emissions from fires to FFB sources is fairly low, and California does not have significant coal-fire power plant CO₂ emissions, this ratio is also subject to substantial variation. By the end of October 2003, wildfires burned more than 750,000 acres, producing the equivalent of 49% of the monthly CO₂ emitted by FFB sources for state. This occurred in more than one year that we investigated. The major wildfires in September 2006, including the Day Fire in Southern California, produced an estimated 16 Tg CO₂ for that month, equivalent to approximately 50% of estimated total monthly FFB emissions for the entire state. Thus, even in highly industrialized regions of the country with significant FFB CO₂ emissions, fires can contribute significant amounts of CO₂ to the atmosphere. These fires not only impact regional CO₂ fluxes, but can also impact visibility and air quality. Phuleria et. al. 5 shows how the emissions from the October 2003 Californian fires increased air pollutant concentrations, most notably particulate matter with diameters less than 10 μm (PM10), throughout the Los Angeles Basin.

Fires represent a potentially large short-term release of carbon that is largely offset over longer time scales (decades) by the uptake of atmospheric carbon associated with forest regrowth. From this standpoint, fires and fossil fuel emissions have entirely different effects on atmospheric CO₂ levels with the expectation that in the absence of changes in frequency or intensity, fire emissions would be balanced over a period of several decades by forest regrowth and C assimilation. To evaluate the magnitude of C released from fire in the context of annual plant C sequestration, we compared emission estimates from fire to annual estimates of Net Primary Productivity (NPP; gC m^-2 year^-1) derived from MODIS satellite observations (3132333435) for 2000 through 2005. Annually-averaged NPP (2000–2005) by state is estimated from these base datasets (Additional File Table 1). For the LOWER48, the annually-averaged NPP was estimated as 9369 Tg CO₂ yr^-1. On an annual basis, fires result in a release of the equivalent of 4% of the annual NPP flux in both the Western and the Southeastern US. However, this is highly variable. For example, average annual fire emissions of CO₂ range from 0.7–1.4% of estimated NPP in North Carolina, Colorado, and Wyoming, to more than 6% for Arizona, Idaho, and Louisiana. For the Western US, fires on average represent 3.8 ± 1.5% of annual average NPP, similar to the results for the Southeastern US, where CO₂ from fires is 3.6 ± 1.1% of annual average NPP.

The large conversion of terrestrial biomass to CO₂ during a fire is largely balanced over longer time scales by the uptake of C in regrowing forest. In North American Boreal ecosystems, there commonly is a period of several years to a decade during which C is lost from ecosystems, followed by several decades to a century of C uptake in regrowing forests 636. However, fire regimes and intensity are changing for at least some portions of the US 321, and following European settlement of the Western US, the fire frequency in some forests was reduced 37 leading to an accumulation of C in terrestrial systems. The relatively large fraction of NPP that is currently lost to fire in a number of Western US ecosystems represents, in part, the return of some of this historically accumulated C to the atmosphere, and sets the stage for future C uptake in these forested ecosystems. The historic and future impact of fire emissions on atmospheric CO₂ also depends on the frequency and intensity of fires in the 21^st century. A shortening of fire return intervals, increases in area burned, and/or increases in fire severity can lead to net emissions of CO₂, even on a multi-decadal times scale 61038. With changing climate and projected increases in burned area in the US 3940, there is a significant potential for additional net release of C from the forests of the United States due to changing fire dynamics in the coming decades.

A simple modeling approach, described by Wiedinmyer et. al. 13, was used to calculate the daily fire emissions of carbon dioxide (E_CO2) in North America from 2002 through 2006. E_CO2 was calculated as:

E_CO2 = A(x,t) * B(x,t) * EF_CO2

where A(x,t) is the area burned at location x and time t, B(x,t) is the biomass burned at location x and time t, and EF_CO2 is an emission factor, or the mass of CO₂ that is emitted per mass of biomass burned.

With this method, fire location and timing is determined with the MODIS Active Fire product. The MODIS instruments aboard the NASA Terra and Aqua satellites each provide approximately twice-daily passes over North America. These daily fire detections were processed by the US Forest Service Remote Sensing Applications Center for 2002 through 2006 using the MODIS Active Fire data developed by the UMD Rapid Response team 41.

The fuel loading at each fire was determined using a combination of satellite products. The Global Land Cover 2000 (GLC2000) dataset is used to characterize the ecosystem type for each identified fire. The GLC2000 identifies 29 different land cover classes in North and Central America at a 1 km² resolution 42. For each land cover class, a total fuel loading has been assigned using a combination of values found in the literature 13. The fraction of woody and herbaceous fuels associated with each class was determined using information from the Fuels Characterization Classification System (FCCS; 4344). The fraction of forest, herbaceous cover, and bare ground at each fire was determined using the Vegetation Continuous Fields (VCF) MODIS product, scaled to 1 km² 4546. The amount of biomass burned was assumed to be a function of forest cover (where > 60% tree cover is considered forest, 40–60% tree cover is considered Woodlands, and <40% tree cover is considered Grasslands), following the methods applied by Ito and Penner 47.

For the results shown here, each detected fire was treated as an individual fire. Based on the nominal resolution of the MODIS instruments, the total possible area burned for each fire pixel was assumed to be 1 km². For each fire detection, the 1 km² was scaled to the amount of bare cover assigned at that spot by the VCF product. For example, if the bare cover was 20% at a fire point, the area burned was estimated to be 0.8 km². Using this methodology, daily fire emissions of CO₂ were estimated for 2002 through 2006. Only emissions from the US are presented in this paper.

The emissions of CO₂ from fires are highly uncertain due to the combined errors and uncertainties in the model framework and inputs. Uncertainties in the fire emission estimates may arise from the satellite detections of the fires, the assumptions made in the fuel loading and amount of fuel burned, the estimated area burned, and the assigned emission factors. The Active Fire satellite product produces daily fire detections. This product is not screened for missing data, and does not flag those areas obstructed by clouds. The timing of the satellite detections and the inability to detect fire through clouds can lead to missed detections and an underestimation of fire detections 4148. The area burned assigned to each pixel (1 km²) is considered an upper estimate. The fuel loadings associated with each general land cover classification are taken from few studies, and in reality are highly variable.

Wiedinmyer et. al. 13 were unable to assign a quantitative assessment of uncertainty on the emission estimates using the described modeling technique. However, they predict that the uncertainties can be over a factor of two. When compared to other estimates of CO₂ emissions from fires, these estimates are within this uncertainty. For the Conterminous US, the Global Fire Emissions Database, version 2 (GFEDv2, 2) predict emissions of CO₂ that are approximately two to five times lower than those estimates here. Other models used to predict emissions from fires are much closer to the values predicted here. A more comprehensive intercomparison of emission estimates of CO from fire emissions models for the US is described by Al-Saadi et. al. 12. In general, the emissions from the methodology used here are higher than those predicted by the GFEDv2, but lower than those predicted by a NOAA product 12. To consider the uncertainty associated with the emission estimates, we assign a factor of at least 2 to the estimates.

The validation of fire emission estimates is difficult, since the emissions from fire to fire are highly variable, and direct flux measurements from fires are extremely difficult. Inverse modeling of fire emissions using in situ measurements or satellite observations provides a means to constrain fire emission estimates: however, these methods can not provide a direct quantification of emissions from fires. The uncertainty in the fire emission estimates, along with the variability in the spatial and temporal allocation of these emissions, adds further complications for efforts to constrain C fluxes with monitoring and modeling techniques. Future work is needed not only to better quantify emissions from fires, but to better constrain the uncertainties associated with the estimates.

The Net Primary Productivity (NPP) is defined as the rate at which biomass grows in an ecosystem. It is often used as a measure of carbon uptake by vegetation, or carbon stored in vegetation. For this study, the annual NPP values determined from the MODIS Satellite instruments were used 3132333435. This product provided annual NPP values (gC m^-2 year^-1) with a spatial resolution of 1 km² for the continuous US Annual NPP values (TgCO₂ yr^-1) for each of the 6 years (2000–2005) were averaged for each state in the continuous US.

To evaluate the importance of biomass burning emissions relative to those from fossil fuel burning, the US Department of Energy report of annual CO₂ emissions from fossil fuel combustion for the country 31 is used. The annual total CO₂ emissions by state from 1990 to 2003, was published in April 2007 49. This inventory does not include all industrial sources, but is the most complete inventory of which we are aware.