The Fire INventory from NCAR (FINN): A High Resolution Global Model to Estimate the Emissions from Open Burning

The Fire INventory from NCAR version 1.0 (FINNv1) provides daily, 1 km resolution, global estimates of the trace gas and particle emissions from open burning of biomass, which includes wildfire, agricultural fires, and prescribed burning and does not include biofuel use and trash burning. Emission factors used in the calculations have been updated with recent data, particularly for the non-methane organic compounds (NMOC). The resulting global annual NMOC emission estimates are as much as a factor of 5 greater than some prior estimates. Chemical speciation profiles, necessary to allocate the total NMOC emission estimates to lumped species for use by chemical transport models, are provided for three widely used chemical mechanisms: SAPRC99, GEOS-CHEM, and MOZART-4. Using these profiles, FINNv1 also provides global estimates of key organic compounds, including formaldehyde and methanol. Uncertainties in the emissions estimates arise from several of the method steps. The use of fire hot spots, assumed area burned, land cover maps, biomass consumption estimates, and emission factors all introduce error into the model estimates. The uncertainty in the FINNv1 emission estimates are about a factor of two; but, the global estimates agree reasonably well with other global inventories of biomass burning emissions for CO, CO2, and other species with less variable emission factors. FINNv1 emission estimates have been developed specifically for modeling atmospheric chemistry and air quality in a consistent framework at scales from local to global. The product is unique because of the high temporal Correspondence to: C. Wiedinmyer (christin@ucar.edu) and spatial resolution, global coverage, and the number of species estimated. FINNv1 can be used for both hindcast and orecast or near-real time model applications and the results are being critically evaluated with models and observations whenever possible.


Introduction
Open biomass burning, which for this study includes wildfires, agricultural burning, and managed burns and not biofuel use or trash burning, makes up an important part of the total global emissions of both trace gases and particulate matter. According to the Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | biomass burning produced 51% of the global carbon monoxide (CO) emissions for 2000 and 20% of the oxides of nitrogen (NO x ) emissions. Current emission inventories estimate that open biomass burning (not including waste burning) emits 26-73% of global emissions of primary fine organic particulate matter (PM) and 33-41% of global fine black carbon (BC) PM emissions (Bond et al., 2004;Andreae and Rosenfeld, 5 2008).
Although episodic in nature and highly variable, open biomass burning emissions can contribute to local, regional, and global air quality problems and climate forcings (Crutzen and Andreae, 1990). The emissions of PM can degrade visibility (e.g., McKeeking et al., 2006) and cause health problems (e.g., Pope and Dockery, 2006). Gas- 10 phase components in fire plumes, including non-methane organic compounds (NMOC) and NO x , can react downwind of the fire location and contribute to the chemistry that forms ozone (e.g., Pfister et al., 2008). Carbon dioxide released to the atmosphere from largescale burning may have important implications for the carbon cycle (e.g., IPCC 2007; Wiedinmyer and Neff, 2007). Due to the importance of these emissions, 15 reasonable estimates of open burning emissions are critical to characterizing air quality problems, understanding in situ measurements, and simulating chemistry and climate.
There have been many efforts to estimate the emissions of trace gases and particles from fires. Emissions from individual fire events have been calculated using site-specific information (e.g., the 2002 Biscuit Fire in Oregon, Campbell et al., 2007;20 the 2003 fires in Southern California, Muhle et al., 2007). Regional emission estimates have been created for specific time periods: for example, Michel et al. (2005) produced a fire emissions inventory for Asia for March-May 2001 in support of a major field campaign. Other inventories predict regional emissions over longer time periods (e.g., Lavoue et al., 2000;Soja et al., 2004;Wiedinmyer et al., 2006;Larkin et al., 2009). 25 At the global scale, several bottom-up biomass burning emissions inventories exist. Using literature emission factors and the biomass burned estimates of Yevich and Logan (2003) (obtained as described by Lobert et al., 1999), Andreae and Merlet (2001) estimated the total global emissions of many gaseous and particulate 2441 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | species representative of the late 1990s. Duncan et al. (2003) estimated global CO emissions from biomass burning for multiple years (1996)(1997)(1998)(1999)(2000) and evaluated the regional and interannual variability in the emissions. Using a combination of satellitederived datasets, Ito and Penner (2004)  Fire Locating and Modeling of Burning Emissions (FLAMBE) Program (e.g., Reid et al., 2009)    used carbon monoxide (CO) observations from space and an inverse model to constrain the CO emissions from biomass burning and other sources. 20 The Global Fire Emissions Database (GFED, Randerson et al., 2005;) is a widely applied global biomass burning emissions dataset. Now in its third version, GFED includes 8-day and monthly emissions of selected trace gas and particulate emissions from burning globally at horizontal resolutions as fine as 0. 5 • for 19975 • for -20095 • for (van der Werf et al., 2006Giglio et al., 2010;van der Werf et al., 2010). GFED is used by global chemical transport and climate modelers in efforts to understand the chemical composition of the atmosphere (e.g., Colarco et al., 2010;Nassar et al., 2009;Magi et al., 2009;Stavrakou et al., 2009). Other global inventories have been created for similar purposes. Recent work by Mieville et al. (2010) describes Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the Global Inventory for Chemistry-Climate Studies (GICC), which provides estimates of the emissions of CO 2 , CO, NO x , particulate black carbon (BC) and organic carbon (OC) from burning for historical and current time periods for use specifically in chemistry-climate modeling applications. Emissions of other key trace gas emissions from fires are also available from the GICC via the GEIAcenter.org website, but have 5 not yet been published. Despite all of these various efforts, the uncertainty associated with open burning emissions remains high, and often modelers do not have the spatial and/or temporal resolution needed to accomplish the required scientific goals.
Here we present a detailed description of the Fire INventory from NCAR version 1.0 (FINNv1) model, initial results from the model, a discussion of uncertainties, and a 10 comparison to other estimates. The FINNv1 provides high resolution, global emission estimates from open burning, which is an episodic phenomenon. Estimates from trash burning or biofuel use are not included, as they are expected to be less variable and they are only amenable to estimation using other methods. FINNv1 emission estimates have been developed specifically to provide input needed for modeling atmospheric 15 chemistry and air quality in a consistent framework at scales from local to global. The inventory framework described here produces daily emission estimates at a horizontal resolution of ∼1 km 2 . The product differs from other inventories because it provides a unique combination of high temporal and spatial resolution, global coverage, and estimates for a large number of species. Speciation profiles of the NMOC emissions 20 are provided for three chemical mechanisms based on a new compilation of biomass burning emission factors (Akagi et al., 2010).

Model description
FINNv1 is based on the framework described earlier by Wiedinmyer et al. (2006). FINNv1 uses satellite observations of active fires and land cover, together with emis-25 sion factors and estimated fuel loadings to provide daily, highly-resolved (1 km) open burning emissions estimates for use in regional and global chemical transport models.

2443
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | The emissions are estimated using the following equation: Where the emission of species i (E i , mass of i emitted) is equal to the area burned at time t and location x [(A(x,t)] multiplied by the biomass loading at location x [B(x)], the fraction of that biomass that is burned in the fire (FB), and the emission factor of 5 species i (ef i , mass of i emitted/mass of biomass burned). All biomass terms are on a dry weight basis. FINNv1 has been designed to use any fire detection data available. However, for the default model described here, the location and timing for the fires are identified globally by the MODIS Thermal Anomalies Product . This prod- • N and 30 • S, due to the observational swath path. To accommodate for this inconsistent daily coverage in the tropical latitudes, fire detections in these equatorial regions are counted for a 2-day period, following methods similar to those described by Al-Saadi et al. (2008). For each fire detected in this region, the fire is assumed to continue into the next day at half of its original size. Once the potential gaps in tropical fires are considered, multiple detections of individual fires 5 are removed as described next.
Since observations from both MODIS instruments aboard the Terra and Aqua satellites are applied, the possibility of "double-counting" the same fire on a single day occurs. Therefore, for each day, multiple detections of the same fire pixel are identified globally and removed as described by Al-Saadi et al. (2008). 10 The type of vegetation burned at each fire pixel is determined by the MODIS Collection 5 Land Cover Type (LCT) product for 2005 (Friedl et al., 2010). The IGBP land cover classification (Table 1) is used to assign each fire pixel to one of 16 land cover/land use (LULC) classes. Additionally, at each fire point, the MODIS Vegetation Continuous Fields (VCF) product (Collection 3 for 2001) is used to identify the density 15 of the vegetation at each active fire location. The VCF product identifies the percent tree, non-tree vegetation, and bare cover at 500 m resolution (Hansen et al., 2003;Hansen et al., 2005;Carroll et al., 2011). The VCF data are scaled to 1 km spatial resolution to match the fire detection and LCT datasets. Inconsistencies between the datasets described above are resolved as follows. Any 20 fire detections in areas with the LCT classification for water, snow, or ice are removed (<0.2% of original annual fire points). When the total cover from the VCF product for any fire point does fully cover each pixel, the values are scaled to 100%. This primarily happens as a result of the scaling of the VCF product from 500 m to 1 km resolution. Those fire detections that fall in areas that are 100% bare cover or unclassified ac- 25 cording to the VCF product are reassigned vegetation coverage based on the LCT classification (typically <0.5% of original annual fire detections). In these cases, for fires located on LCT forest classifications, the percent coverage is reassigned to 60% tree cover and 40% herbaceous cover. For fires in LCT shrubland classifications, the 2445 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | percent cover is reassigned to 50% tree cover and 50% herbaceous cover. For fires in LCT grassland classifications that do not have associated VCF cover information, the percent vegetative cover is reassigned to 20% tree cover and 80% herbaceous cover. Fire points assigned by the LCT product as "urban" or "bare/sparsely vegetated" are assumed to be open vegetation burning and are reassigned a land cover type based 5 on the total tree and non-tree vegetation cover, as determined by the VCF product. For those fire points with less than 40% tree cover, the urban or sparsely vegetated land cover is reassigned to grasslands; for 40-60% tree cover, the point is reassigned as shrublands; and for tree cover greater than 60% tree cover, the point is reassigned as a forest. 10 The global LULC classifications of the MODIS LCT product are then further lumped into more generic land cover classifications that better match available information on global fuel loadings and emission factors. These generic categories include Savanna and Grasslands (SG), Woody Savannas and Shrublands (WS), Tropical Forest (TROP), Temperate Forest (TEMP), Boreal Forest (BOR), and Cropland (CROP) ( Table 1). The 15 evergreen, deciduous, and mixed forest land covers of the LCT are assigned as either boreal or temperate forest depending on the latitude of the point: if latitude is greater than 50 • N, the forest is labeled as a boreal forest.
At present, FINNv1 does not obtain the area burned at each identified fire pixel from burned area products since they are not rapidly available. Therefore, an upper limit 20 is assumed for the burned area. For each fire identified, the assumed burned area is 1 km 2 , except for fires located in grasslands/savannas: these are assigned a burned area of 0.75 km 2 Al-Saadi et al., 2008). This burned area is further scaled based on the percent bare cover by the VCF product at the fire point. For example, a forest fire detected at a point is assigned a burned area of 1 km 2 ; yet, if 25 that same pixel is assigned 50% bare cover by the VCF dataset, the assigned burned area is 0.5 km 2 . Fuel loadings (or the amount of biomass available that can be burned in each fire) for each generic LULC in the various world regions are assigned based on values from  Hoelzemann et al. (2004) and updates shown in our Table 2. For most classes and regions, the average of the GWEM v1.2 and v1.21 are used (Hoelzemann et al., 2004). Changes to the original values presented by Hoelzemann et al. (2004) include the following: (1) Temperate forests in Oceana are assigned the average of the fuel loadings for temperate and tropical forests assigned by Hoelzemann et al. (2004) 5 for that region. (2) The temperate forest loading for Australia, particularly in the eucalyptus forests of southeastern Australia, is typically much higher than the 7000 g m −2 assigned by Hoelzemann et al. (2004) (C. Murphy, personal communication, 2009 and is been replaced with a larger value. (3) The fuel loading assigned to croplands is 500 g m −2 for fires assigned to croplands . The one exception 10 to this rule is for croplands within Brazil, from latitude 20.36 • to 22.71 • S and longitude −47.32 • and −49.16 • W. In this region, sugar cane is assumed to be the crop type that is burned, and the fuel loading here is assigned as 1100 g m −2 (Macedo et al., 2008;E. Campbell, personal communication, June 2010). This is just one example of how the model can be modified at regional and local levels to include specific information. 15 The fraction of the biomass assumed to burn (FB) at each fire point is assigned as a function of tree cover, as described by Wiedinmyer et al. (2006) and taken from Ito and Penner (2004). For areas with 60% or more tree cover, as defined by the MODIS VCF product, FB is 0.3 for the woody fuel and 0.9 for the herbaceous cover. For areas with less than 40% tree cover, no woody fuel is assumed to burn and the FB is 0.98 20 for the herbaceous cover. Note that these are the upper limits as presented by Ito and Penner (2004). For those fires with 40-60% tree cover, FB is 0.3 for the woody fuels, and FB for the herbaceous fuel is calculated as: The amount of woody fuel available to burn at each fire is determined by the fraction 25 of tree cover and the fuel loading for the specific land cover type and global region; the herbaceous fuel loading is assigned the fuel loading for the grassland land cover in that global region. The amount of the fuel burned is equal to the biomass loading multiplied 2447 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | by the fraction burned and the fractional cover of each vegetation cover (by the MODIS VCF) for each pixel. For each LULC type, the emission factors for various gaseous and particulate species have been taken from available datasets (see Table 1 Andreae, personal communication, October 2008). However, to simulate NMOC chemistry in many models, particularly chemical transport models, many of the individual chemical compounds are assigned to "lumped" species in a simplified chemical mechanism. FINNv1 calculates both the total NMOC for each generic land cover type, and also lumps the vide the moles emitted of individual organic compounds or lumped species for the indicated chemical mechanism. Here we provide speciation profiles for the GEOS-Chem (Bey et al., 2001; http://www.geos-chem.org/), MOZART-4 (Emmons et al., 2010a), and SAPRC99 (Carter et al., 2000) chemical mechanisms. These speciation profiles are available for use not only with FINNv1 emission estimates, but also other NMOC 20 emission estimates from other models.
The primary differences in the methods described here, compared to the framework presented by Wiedinmyer et al. (2006), include the extension of the model domain from North and Central America to the globe; the removal of fire detections with confidence less than 20%; the "smoothing" of the fire detections in the tropical latitudes where satellite observations occur less than daily; the removal of multiple detections of the same fire for a given day; the use of the MODIS Land Cover Type dataset to describe the ecosystem burned; and the updated emission factors from the most recent datasets. For North and Central America, the model improvements in FINNv1 lead to significant differences between the results obtained from earlier versions of this model framework (e.g., Wiedinmyer and Neff, 2007).  Table 6.

Comparison of FINNv1 to other biomass burning inventories
On a global scale, the biomass consumption and total emissions predicted with FINNv1 are fairly similar to amounts from other global estimates. For example, the total global biomass burned in GICC (Mieville et al., 2010) for 2000 was estimated to be 5790 Tg, and the average annual global biomass burned from 2005-2009 estimated with FINNv1 is 5609 Tg ( Table 7). The global annual FINNv1 CO 2 emission estimates 15 are ∼5-30% larger than the GFEDv3.1 estimates ( Table 8). The agreement is variable by year and the differences result primarily from the different fuel consumption approaches that drive the two models.
Next  (Reid et al., 2009). The different estimates for the above species are consistent within the uncertainties of the model frameworks (see Sect. 3.5). clarification of the relevant terminology, which is also discussed in more detail in Akagi et al. (2010). Non-methane hydrocarbons (NMHC) are organic molecules that by definition contain only atoms of C and H such as alkenes and alkanes. In early biomass burning research, NMHC were thought to account for nearly all the organic compounds emitted by fires and it became commonplace to equate NMHC emissions to total or-10 ganic emissions. More recent work has shown that 60-80% of the identifiable organic compounds emitted by fires contain oxygen atoms in addition to C and H (Yokelson et al., 1996(Yokelson et al., , 2008Holzinger et al., 1999;Karl et al., 2007). A broader term for organic emissions that includes the oxygenated organic compounds (e.g. formaldehyde, methanol, etc.) is non-methane organic compounds (NMOC). An updated compila-15 tion of EF for NMOC by Akagi et al. (2010) is incorporated into FINNv1. However, in some other estimates the term NMHC is still used and the quantity represented by this term may vary. Sometimes NMHC refers to just the molecules with C and H (van der Werf et al., , 2010. In other studies, the term NMHC is not defined and thus unclear, but it may be intended to indicate the NMHC plus the other NMOC. In any case, 20 the intent of previous work was probably to estimate total organic emissions regardless of the terminology. Here we compare the amount of identified NMOC emitted by open burning, as derived by FINNv1, to previous estimates of total organic emissions. One other clarification is worthwhile. In studies that measure fire emissions by mass spectrometry or gas chromatography only about one-half of the NMOC peaks can be 25 identified as specific compounds. The large number of unassigned peaks confirms that fires emit a substantial amount of NMOC that have not yet been identified with present technology (Christian et al., 2003;Karl et al., 2007). The amount of these unidentified compounds is uncertain and, thus, not further discussed in this paper; however, estimates of total global NMOC that include the unidentified species can be found in Akagi et al. (2010). The emissions of NMOC from FINNv1 are a factor of 3.7 to 4.9 higher than the GFEDv3 NMHC emission estimates, as is to be expected due to the consideration of more species of organic emissions in FINNv1 (Table 8). 5 The open biomass burning emissions from FINNv1 for 2008 make up 27% of global particulate BC emissions, 33% of global CO emissions, and 62% of global primary particulate OC emissions; where the 2008 global totals were estimated by Emmons et al. (2010a) (Fig. 1).

Comparison of biomass burning to other sources
Emissions of individual organic species are estimated from the total global NMOC 10 emissions using the speciation profiles presented in Tables 3 through 5. The global annual totals of a few key organic compounds calculated with the MOZART-4 and SAPRC99 speciation profiles are shown in Table 9. Isoprene, the most abundant biogenic emission, is also emitted by open fires, but in small amounts compared to the 600 Tg emitted from undisturbed vegetation (Guenther et al., 2006). However, the fire 15 emissions of other individual NMOC species can be more important. Globally, the average annual methanol (CH 3 OH) emissions from open biomass burning are 2.8 times larger than the anthropogenic emissions of CH 3 OH, and the emissions of formaldehyde (CH 2 O) are a factor of 1.4 larger than anthropogenic emissions . Note that that the above ratios compare open biomass burning to other es-20 timates of anthropogenic emissions, and biofuel use (primarily cooking with biomass fuel) is included in the anthropogenic category.

High variability and major features of emission rates
The daily emissions of total identified NMOC for the Northern Hemisphere, the Southern Hemisphere, and the globe are shown in Fig. 2. The interannual variability can be 25 quite substantial, particularly at the hemispheric and regional scales. Weekly and daily 2451 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | emissions for specific time periods can vary by more than a factor of two from year to year. The NMOC emissions (as well as the other emissions of other species not shown) are also extremely variable day to day confirming the need for high temporal resolution for some applications. The major temporal features of the FINNv1 emissions are summarized next. Two 5 general peaks in emissions occur each year. The first peak is from ∼ mid-February through May and is primarily caused by the burning in the tropical and subtropical regions of the Northern Hemisphere. A second peak occurs from August through September, which corresponds to burning in the Southern Hemisphere tropics. Assuming that CO 2 and NMOC are emitted proportionally from open biomass burning,  are high during the spring months. There are also large emissions associated with burning throughout Central Asia. SH emission predicted for July are located mainly in South America and Africa, while NH July emissions are found mainly in North America, and Northern Asia. In October, open burning emissions are mostly produced in the Southern Hemisphere (Fig. 2)

Applications
The FINNv1 model was created to provide near real-time estimates of open burning emissions that can easily be incorporated into chemical transport models. FINNv1 2452 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | emissions or their predecessors have already been used successfully in several regional and global applications (e.g., Fast et al., 2009;Emmons et al., 2010b;Pfister et al., 2008;Hodzic et al., 2008). Methods for allocating the emissions to a diurnal cycle and incorporating plume rise can be found elsewhere (e.g., Frietas et al., 2009;WRAP, 2005). Because the FINNv1 emissions can be produced within a few hours 5 of each satellite overpass, they can be used for model forecast applications. FINNv1 emissions can be generated interactively and feedback provided by users will improve future versions of the model. The FINNv1 emission estimates have advantages over some other inventories when high spatial and/or temporal resolution or rapid availability is needed. For example, despite the limitations of using the daily fire counts (see Sect. 3.5), the daily estimates allow models to capture the highly episodic nature of fire emissions that could be missed with smoothing to 8-day resolution or using monthly fire counts, as is commonly done. Additionally, FINNv1 produces consistent emission estimates from coarse grid scales to local scales, which facilitates comparisons and is useful for nested applications.

Limitations and uncertainties
FINNv1 produces high-resolution (spatial and temporal) emissions from open biomass burning on a global scale relatively quickly (on the order of minutes to hours). Although useful for multiple applications, the estimates are very uncertain and have only begun to be compared to observations (e.g., Pfister et al., 2010). Uncertainties associated 20 with many aspects of the estimation process are described in detail by  and below. In summary, since most global fires are "small" it is likely that the largest uncertainties arise from (1) missed fires causing an underestimation of the number of fires and (2) overestimating the size of the small fires that are detected. These errors tend to cancel as discussed by Wiedinmyer et al. (2006) and Yokelson et 25 al. (2009). Additional uncertainty could arise from misidentification of the land cover, inaccurate fuel loading and parameterizations of combustion completeness, and both uncertainty and natural variation in the emission factors (Akagi et al., 2010). Other 2453 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | assumptions made in FINNv1 also add uncertainty, such as the smoothing of the fire detections in tropical latitudes to account for the lack of daily coverage by the MODIS instruments in this region, and the assumed burned area of each fire. For the global application described here, average values for variable phenomena are applied to broad regions. The average value may not always represent the real value for some fires or 5 some years. Next we discuss three of the sources of uncertainty in more detail. Satellite overpass timing and cloud cover may prevent the detection of fires. The need to estimate the number of fires on days without coverage limits the accuracy of any emissions model. Additionally, all remote sensing thermal anomaly products do not detect most of the fires less than ∼100 ha and some understory fires (e.g. Hawbaker et al., 2008), both of which can be a significant source of emissions to the atmosphere.
The land use/land cover (LULC) classifications assigned to the fires introduces some uncertainty to the emission estimates. For the results presented here, the satellitederived MODIS LCT and VCF products are used to identify the type and density of vegetation burned. These products were chosen specifically because of their consis-15 tency with the MODIS fire detections, easy access, and easy use. Yet, determination of ecosystem type can vary significantly from one land cover data product to another. For example, Wiedinmyer et al. (2006) showed that the use of three different LULC datasets to drive a regional fire emissions model for North and Central America led to 26% differences in annual emission estimates. Those authors ultimately selected the 20 Global Land Cover 2000 product (GLC2000; Fritz et al., 2003) to determine the land cover burned at each fire in North and Central America . Giri et al. (2005) detail differences in the MODIS LCT product and the GLC2000 dataset and show that the area totals of the generic land covers agree reasonable well globally, except for woody savannas/shrublands and wetlands. However, at the pixel level, agreement between the two datasets is not as good. A fire located in a forest will typically be associated with more emissions than a fire located in grasslands due to the higher fuel loadings. Thus, the determination of land cover and vegetation coverage can introduce significant error in the emission estimates.
Correct assignment of the vegetation does not prevent uncertainty due to the fuel consumption estimates. For instance, only one value for fuel loading is assigned to each land cover type in each region. A constant value is most likely not representative of a vegetation class within an entire region and will not reproduce the full heterogeneity of the landscapes. For example, Soja et al. (2004) found that disparities in the 5 amount of carbon stored in unique Siberian ecosystems and the severity of fire events can affect total direct carbon emissions by as much as 50%. However, a strength of FINNv1 is that it is relatively easy to introduce specific regional information to replace the generic information in an effort to reduce uncertainties in the emission estimation process. For example, as discussed above, specific fuel loadings for crop fires in a 10 small area of Brazil were applied to account for sugar cane burning.
The uncertainty in total emissions that can arise from coupling all the inherent uncertainties is illustrated briefly with a few examples. Al-Saadi et al. (2009) reported that monthly fire emission estimates generated for the contiguous US over several months in 2006 by various remote sensing-based techniques varied by an order of 15 magnitude. Chang and Song (2010) used two different burned area products to derive their open burning emissions for Southeast Asia: the L3JRC and the Collection 5 MODIS (MCD45A1) burned-area products. They found that the average annual burned area estimates for the two products over Asia from 2000-2006 were almost a factor of 2 different, and the interannual variation in the burned area estimates differed as well. 20 When compared to the annual average burned area estimates of GFEDv2 for the same time period, GFEDv2 was 50% greater than the MCD45A1 burned area estimates and almost a factor of 2 higher than the L3JRC estimates.
In light of the above discussion, we examine some of the sources of uncertainty in FINNv1 in more detail. For a sensitivity test, we used the GLOBCOVER global 25 vegetation map (downloaded from http://ionia1.esrin.esa.int/index.asp, January 2010), to assign land cover for detected fires in FINNv1. Globally, the annual (2006) total emissions of CO did not change significantly (2%) between the default case and the GLOBCOVER run, although emissions of NO x and NH 3 change by as much as 24%.

2455
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | However, the amount of some landcover types that is assumed to burn globally in each run changes substantially when the different land cover datasets are applied, which can have large impacts on the estimated emissions from a region since different land covers have varying emission factors and fuel loadings that can lead to variations in emission estimates. For example, globally for 2006, the GLOBCOVER simulation es-5 timates that more than 3 times the amount of temperate forest burns compared to the default run driven by the MODIS LCT data. Additionally, the LCT data implies that more shrubland and grasslands burn globally. Regionally variation is also high. For example, the GLOBCOVER assigned 70% more forest fires to the contiguous US, Mexico, and Central America, leading to 20% higher CO and 24% higher NMOC emissions than the default simulation in these regions. In this case, the MODIS LCT assigned more shrubland, cropland and grassland fires. The total emissions estimated using GLOB-COVER were only 10% lower in Canada and Alaska than the default simulation, due to fewer forest fires assigned in these areas.
In summary, a quantitative assignment of uncertainty is difficult, due to the uncertainties associated with the land cover classifications, the fire detections, the assumed area burned, the biomass loading, the amount of fuel burned, and emission factors. At this time, we follow other efforts (e.g., Wiedinmyer et al., 2006;Mieville et al., 2010) and assign the uncertainty as a factor of two for the FINNv1 estimates. We continue to apply in situ measurements, satellite observations, and model simulations to evalu-20 ate the accuracy of the estimates provided here. Future versions of FINN will contain updates intended to reduce these uncertainties.

Conclusions
Open biomass burning injects significant amounts of particulate matter and trace gases into the atmosphere. The Fire INventory from NCAR version 1.0 (FINNv1) estimates 25 these emissions daily on a global basis at a resolution of 1 km 2 . The inclusion of a number of important individual trace gases and particulate species is presented. FINNv1 also provides speciation profiles for several lumped chemical mechanisms used by some chemical transport models. FINNv1 includes updated emission factors for nonmethane organic compounds, which may have an important impact on the way in which global and regional chemistry is simulated. Strengths of FINNv1 include: it quickly provides modelers with reasonable, high temporal/spatial resolution fire emission esti-5 mates based on updated emission factors, and it is easily adapted to more accurately target specific regions of interest. For many chemical species, the FINNv1 emission estimates agree well with other inventories; specifically the GICC and the GFEDv3. However, this does not in itself establish the absolute accuracy of the estimates because open burning emission esti-10 mates are subject to inherent limitations that lead to large uncertainties. Thus, the uncertainty assigned to the FINNv1 estimates is a factor of 2. Future work will compare the FINNv1 estimates to in situ measurements, satellite observations, and chemical transport models.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Table 1. Land use/land cover classifications as assigned by the MODIS Land Cover Type, assigned generic land cover class, and emission factors (g kg Biomass Burned −1 ). Sources of emission factors are by color.