The FSL has a large indoor combustion room described in more detail elsewhere (Christian et al., 2003; Burling et al., 2010). Briefly, the room is 12.5 m × 12.5 m × 22 m high with a 1.6 m diameter exhaust stack and a 3.6 m inverted funnel opening approximately 2 m above a continuously weighed fuel bed. The room can be pressurized to create a large constant flow that dilutes and completely entrains the fire emissions while venting through the stack. A sampling platform that can accommodate up to 1820 kg and sampling ports surround the stack 17 m above the fuel bed. Other instrumentation can be placed in adjacent rooms with sample lines pulling from ports at the sampling platform. Previous studies concluded that the temperature and mixing ratios are consistent across the width of the stack at the height of the platform, confirming well-mixed emissions that can be monitored by a number of different sample lines throughout the fire (Christian et al., 2004).

Our simulated fires used two configurations. In the first configuration, termed “stack burns”, fires were ignited below the stack and they burned for a few minutes to half an hour. As the fire evolved, the emissions, partially diluted and cooled by outside air, traveled up through the stack at a constant flow rate (∼ 3.3 m s-1). At the platform height, the well-mixed emissions were near ambient temperature, about 5 s old, and monitored by a large range of instruments in real time.

In the second scenario, referred to as “room burns”, most of the instruments were relocated to rooms adjacent to the combustion chamber and used sample lines that extended well within the combustion room. The stack was raised, the combustion room was sealed, and the fuels were burned for several minutes. After about 15–20 min, the smoke from the whole fire was well mixed vertically in the combustion room and was monitored under approximately steady-state, low-light conditions for up to several hours, though some infiltration and losses of gases and particles for some species occurred (Stockwell et al., 2014). Despite the losses, the configuration is useful for measurements requiring longer times. The OP-FTIR remained on the sampling platform during room burns, which helped to document the initial rise of flaming emissions and verified the overall mixing processes. Temperature and relative humidity in the combustion room were recorded for all fires and both stack and room burns were videotaped and stored in the NOAA archive.

A team of experts collected fuels that represent fire-prone western US ecosystems primarily from the Clearwater Wildlife Management Area (http://fwp.mt.gov/fishAndWildlife/wma/siteDetail.html?id=39754079) and the Lubrecht Experimental Forest (https://www.cfc.umt.edu/lubrecht/), which are managed by the state of Montana and University of Montana, respectively. Chaparral fuels and fuels for the Fire and Smoke Model Evaluation Experiment (FASMEE, https://www.fasmee.net/) were sampled by forest fire experts at locations in California and Utah, respectively, and shipped overnight to the FSL. A few fuels representative of prescribed fires were sampled by foresters at SE US military bases and burned for comparison purposes and for the FASMEE project. Sagebrush and juniper were sampled locally. Indonesian peat, aspen shavings (also known as “excelsior”), and dung were sampled and burned because of their global importance and/or to investigate the impact of fuel chemistry (e.g., N content) on emissions. Fuel components for the forest ecosystems included duff; litter; dead and down woody debris in different size classes; herbaceous, shrub, and canopy fuels; and rotten logs from two of the above ecosystems (ponderosa pine and Douglas fir). These fuel components were burned both on their own and in realistic three-dimensional mixtures to mimic the different fuel complexes for various ecosystems. The first-order fire effects model (FOFEM) (Reinhardt et al., 1997) was used to calculate the relative amount of each component that typically burns in coniferous ecosystems, while pure components were burned to probe how they affected the total emissions. The coniferous ecosystems modeled and burned included ponderosa pine (Pinus ponderosa), lodgepole pine (Pinus contorta), Engelmann spruce (Picea engelmanii), Douglas fir (Pseudotsuga menziesii), and subalpine fir (Abies lasiocarpa). Chaparral was represented by manzanita (Arctostaphylos) and chamise (Adenostoma fasciculatum). A full description of the fuels for each fire, including collection location; C, H, N, S, and Cl content; dry weight of each component; ignition time; etc. is included in Table S1 in the Supplement. Moisture content, ash data, and other details for fuels and fire duration were also recorded and are available in the NOAA archive or from the corresponding author.

Extensive instrumentation that monitored both the gas-phase and particle-phase emissions from BB was deployed during the FIREX laboratory study. A table of all the instruments can be found at this URL (https://www.esrl.noaa.gov/csd/projects/firex/firelab/instruments.html). We reiterate that for the first 31 stack fires the two PAXs were the only instruments measuring aerosol optical properties on the platform and only the 870 nm PAX measured optical properties on the sampling platform for the next 44 fires, which accounts for all the stack burns. The 401 nm PAX was deployed with a BC intercomparison that measured subsamples of smoke in a mixing barrel for fires 32–107. The 870 nm PAX was deployed with a large group of aerosol instruments that characterized aerosol subsamples from the room burns (fires 76–107). Other aerosol measurements on the platform during the stack burns included filter sampling with off-line analysis of non-methane organic compounds and AMS characterization of diluted smoke. Here we present the PAX (and FTIR) measurements on the platform and the other results will be described elsewhere.

We convert the time series of mixing ratios for each analyte (Fig. 1) into a form that is broadly useful to others for implementation in local to global chemistry and climate models. For this, we produce emissions ratios (ERs) and EFs. The process starts by calculating excess mixing ratios (denoted ΔX for each species “X”) for all 23 gas-phase species measured using the OP-FTIR by subtracting the relatively small average background mixing ratio measured before each fire from all the mixing ratios observed during the burn. The molar ER for each species X relative to CO2 (ΔX/ΔCO2) is the ratio between the sum of the ΔX over the entire fire relative to the sum of the ΔCO2 over the entire fire. A comparison of the sums is valid because the large entrainment flow ensures a constant total flow. Molar ERs to CO2 were calculated for all the species measured using the OP-FTIR for all 75 stack burns and the two most important room burns. For the other room burns, OP-FTIR data were generated only for CO2, CO, CH4, C2H4, C2H2, and H2O as losses in the room add uncertainty to the mixing ratios for many NMOGs, NH3, etc. The ERs to CO2 were then used to derive EFs calculated with the carbon mass balance (CMB) method, which assumes all of the burned carbon is volatilized and that all of the major carbon-containing species have been measured (Ward and Radke, 1993; Yokelson et al., 1996, 1999; Burling et al., 2010, Stockwell et al., 2014): EFXgkg-1=FC×1000×MMxAMC×ΔXΔCO2∑j=1nNCj×ΔCjΔCO2, where FC is the measured carbon mass fraction of the fuel, MMx is the molar mass of species X, AMC is the atomic mass of carbon (12 g mol-1), NCj is the number of carbon atoms in each species j, and ΔCj or ΔX referenced to ΔCO2 are the source-average molar ERs for the respective species. The denominator of the last term in Eq. (1) estimates total carbon. Based on many BB combustion sources measured in the past, the species CO2, CO, and CH4 usually comprise 97–99 % of the total carbon emissions (Akagi et al., 2011; Stockwell et al., 2015). Our estimate of total carbon in this paper includes these three species and all the rest of the C-containing gases measured with the OP-FTIR as well as the C in the particles (i.e., BC and OC) based on the PAX data. Samples of each fuel component were analyzed for moisture content by weighing until dry and for C, H, N, S, and Cl by a commercial (ALS, Tucson) and an academic laboratory, whose results agreed well with each other on several overlapping fuel samples. The fire-average carbon mass fractions for mixed fuel beds were calculated from the average of the relevant fuel component analyses weighted by dry mass (Table S1). The usually small error in the CMB method caused by neglecting char formation (Bertschi et al., 2003) tends to be canceled by more complete combustion of the higher-C components (Santín et al., 2015) and both these effects are ignored here but will be explored in more detail in a companion study.

Two major combustion processes are often recognized for open burning of biomass: flaming and smoldering, where smoldering is an approximate term for all non-flaming processes (e.g., glowing and pyrolysis) as explored in more detail elsewhere (Yokelson et al., 1996; Koss et al., 2017). Combustion efficiency is the fraction of fuel carbon converted to carbon as CO2, which is maximized by flaming combustion, but the modified combustion efficiency (MCE) is also a useful approach for characterizing the relative amount of smoldering and flaming combustion by comparing the fuel carbon converted to CO2 versus CO2+ CO. Although the two processes often occur simultaneously throughout a fire, a high MCE (near 0.99) is an indication of nearly pure flaming, while a lower MCE (∼ 0.8) is an indication of nearly pure smoldering (Akagi et al., 2011) and an MCE of 0.9 would indicate roughly equal amounts of flaming and smoldering (i.e., a flaming / smoldering ratio of ∼ 1): MCE=ΔCO2ΔCO+ΔCO2. In the PAX, the 870 nm laser is absorbed in situ by BC-containing particles only without filter or filter-loading effects that can be difficult to correct. We directly measured aerosol absorption (Babs, Mm-1) and used the literature-recommended mass absorption coefficient (MAC) (4.74 ± 0.63 m2 g-1 at 870 nm) to calculate the BC concentration (µg m-3) (Bond and Bergstrom, 2006), but the BC mass can be adjusted using different MAC values if supported by future work. Because the PAXs also measured light scattering, scattering and absorption values can be combined to directly calculate the SSA (the ratio of scattering to total extinction). SSA is a useful input for climate models, where an SSA closer to 1 indicates a more “cooling” highly scattering aerosol: SSA=scattering(λ)scattering(λ)+absorption(λ). To a good approximation, sp2-hybridized carbon (including BC) absorbs light proportional to frequency (Bond and Bergstrom, 2006). Thus, the Babs contribution from BC at 401 nm can be calculated from 2.17 times Babs at 870 nm, and any additional Babs at 401 nm can be assigned to BrC subject to limitations due to “lensing” by coatings discussed elsewhere (Pokhrel et al., 2016, 2017; Lack and Langridge, 2013; Lack and Cappa, 2010). Pokhrel et al. (2017) found that coatings typically accounted for much less than 30 % of absorption in room burn smoke 1–2 h old and coatings are likely much less important in 5 s old stack burn smoke (Akagi et al., 2012). Coating effects are very difficult to deconvolve from BrC effects even with additional instruments that were not available during the stack burns (Pokhrel et al., 2017). This adds some uncertainty to the BrC attribution (±25 %) but not to the absorption measurements themselves. Absorption by the BrC component of OA means that an approximate mass of OA can be calculated using an OA MAC of 0.98 m2 g-1 (Lack and Langridge, 2013), but the MAC for OA is variable because BrC chemistry and BrC / OA vary and the OA MAC is also highly dependent on the BC / OA ratio as described elsewhere (Saleh et al., 2014). We use the qualitative OA to calculate a small term in our CMB method that helps account for unmeasured C species (assuming OA / OC of 1.6), but we do not report OA or OC in the tables as quantitative species. Critically though, we do report the OA absorption due mainly to BrC at 401 nm, a poorly characterized term that needs to be improved in climate models to better estimate the radiative forcing of BB aerosol (Feng et al., 2013). The mass ratio of BC to the CO2 measured using the FTIR was multiplied by the EFs for CO2 to determine mass EFs for BC (g kg-1). The EFs for absorption and scattering optical cross sections at 870 and 401 nm (EFabs, EFscat) were calculated from the measured ratios of Babs and Bscat to CO2 and reported in units of square meters per kilogram of dry fuel burned. EFabs or EFscat are more precisely the optical cross-section (m2) due to the particles produced when a kilogram of fuel is burned if the emissions are mixed into a cubic meter of air. These EFs enable direct calculation of the absorption or scattering coefficient (m-1) by multiplication with a specified ratio of fuel burned to a volume of air (kg m-3) (Bond et al., 1999; Moosmüller et al., 2005). We also report the estimated portion of Babs at 401 nm due to BrC. Finally, the AAE (401 and 870 nm) can be calculated from the 401 and 870 nm data, where the AAE of pure BC is close to 1 and larger values are indicative of smoke absorption more dominated by BrC emissions: AAE=-log⁡Babs,1Babs,2log⁡λ1λ2. The AAE is useful as an indicator of BrC / BC, but in addition, the full aerosol absorption spectrum is often approximated with a power-law function (absorption = C × λ-AAE) and thus the AAE determined with any wavelength pair can be used to approximately calculate the shape of absorption across the UV–VIS range (Reid et al., 2005).

We sampled a total of 75 stack burns and 32 room burns at the FSL combustion facility during October and November 2016. Figure 1 shows temporal profiles for the excess mixing ratios of 21 gas-phase compounds (not including water) measured with the OP-FTIR for a complete juniper canopy fire (fire 75). Immediately after ignition, the fire is characterized by a large increase in CO2, corresponding to flaming, followed by a slower increase in CO from smoldering combustion. As is common to most fires, there is no clear distinction between flaming and smoldering but rather an evolving mix of the two processes. Fire-integrated ERs to CO2 and EFs were determined for all 75 stack fires based on the whole fire. For room burns, we calculated EF based on integrating the ΔX only up to the point at which emissions were well mixed to capture the whole fire but also minimize the effect of wall losses and infiltration (see Fig. 3 in Stockwell et al., 2014). The fire-integrated EFs for some of the most common western US ecosystem fuel complexes sampled in this study are summarized in Table 1. These are averages of the replicate fires (three to four replicate measurements for each fuel type). Table 1 does not reveal a strong ecosystem dependence across the coniferous ecosystems but does indicate lower EFs for many pollutants emitted by the chaparral fires. However, large wildfires often burn in multiple fuel types simultaneously. For instance, the Rim Fire burned in pine, pine–oak, and chaparral fuels simultaneously (Liu et al., 2017). These factors justify using a single set of EFs for all wildfires, unless detailed fuel data are available that warrant more precise EF estimates. The derivation of the best wildfire EFs is explored in more detail in the next section. A summary of all the EFs we measured with OP-FTIR and PAX can be found in Table S2, with the averages of specific fuel components and complexes found in Table S3. Numerous additional NMOGs that were measured using other instruments (e.g., H3O+ chemical ionization mass spectrometer (CIMS) and I- CIMS) will be presented elsewhere (Koss et al., 2017). These additional species are often reactive and very important in plume chemistry even though they have only a small effect on the carbon mass balance. A few species were measured with both OP-FTIR and MS and the preferred values depend on several issues such as S / N (often better on MS), interference (often worse on MS), “stickiness”, fragmentation, and proton affinity that are discussed in more detail elsewhere (Koss et al., 2017).

It is important to compare our FIREX laboratory fire emissions data to field measurements of real wildfires to assess how representative and useful the lab-based data are, especially for the many species measured in the laboratory but not the field. We assess representativeness by comparing the EF results for species measured in both the field and our laboratory fires. EF measurements on real wildfires are rare, but Liu et al. (2017) report recent EFs for three wildfires sampled during the 2013 Studies of Emissions and Atmospheric Composition, Clouds, and Climate Coupling by Regional Surveys (SEAC4RS, https://espo.nasa.gov/missions/seac4rs) (Toon et al., 2016) campaign, and the Biomass Burning Observation Project (BBOP, https://www.arm.gov/research/campaigns/aaf2013bbop) campaign.

We compare our results from the FSL combustion studies to those reported by Liu et al. in two ways. In method 1, we plot the lab-measured EFs against their corresponding MCEs for all the fires and we fit the data with a linear regression relationship for each compound. Using the slope and y intercept of the linear regression, and the field-average MCE from Liu et al. of 0.912, we calculate a lab-based prediction of EF at the field-average MCE for each compound measured with the OP-FTIR. Figure 2 shows an example of the procedure for CH4, comparing the lab-predicted EF at the field-average MCE (4.76 g kg-1) to the average field-measured wildfire EF (4.90 g kg-1). In method 2, we compared the simple lab-average EF to the average field-measured wildfire EF. The results of these two methods are shown in Table 2 and Fig. 2. Method 1 is generally preferred because the laboratory fires had a higher average MCE (i.e., a higher fire-integrated flaming / smoldering ratio) than the real wildfires sampled to date, most likely due to some unavoidable drying of the fuels during storage and some underrepresentation of the largest diameter fuels (Table S1). The differences between the laboratory prediction at the field-average MCE and the field-average emissions are probably mostly due to the relative age of the smoke and the reactivity of compounds. The field study included smoke samples up to 2 h old and elevated OH, HO2, H2O2, O3, etc. have been observed in fresh smoke plumes (Hobbs et al., 2003; Yokelson et al., 2009). Thus the more reactive species (e.g., SO2, HCl, NOx, and some NMOGs) have lower EFs in the field data. For example, the lab / field ratio increases going from ethylene to propene to 1,3-butadiene in accordance with, though not directly proportional to, their increasing OH rate constants, and other chemistry, instrumental, and sampling challenges are relevant for some species (e.g., Finlayson-Pitts and Pitts, 2000; Apel et al., 2003; Fig. 7 in Hornbrook et al., 2011; Burkholder et al., 2015). A few reactive species were measured in two older airborne studies of fresh western US wildfire smoke and they agree significantly better with our lab-based predictions (Radke et al., 1991; Hobbs et al., 1996). For instance, Radke et al. (1991) report EFs for NOx as NO (2.0 g kg-1), NH3 (2.0 g kg-1), and C3H6 (0.70 g kg-1) for the Myrtle–Fall Creek wildfire that are all within 20 % of our lab-predicted EFs. Hobbs et al. (1996) report an EF for SO2 (0.79 g kg-1) that is closer to our value than the Liu et al. value is despite the much lower MCE (0.81).

Figure 3 shows the comparison for method 1 from Table 2 graphically. From Fig. 3 it is clear that for the main relatively stable compounds, including formaldehyde, methanol, and hydroxyacetone, the lab-predicted EF falls within 21 % of the measured wildfire EF and all the emissions except NOx and SO2 overlap within the observed variability. Also highlighted in Fig. 3, many compounds such as HONO, acetic acid, ammonia, phenol, glycolaldehyde, formic acid, etc. were measured only for our laboratory fires. The lab-measured EFs for these OP-FTIR species and the data for many NMOG species measured by MS and FIREX data in general can thus be used to generate representative EFs or other data for real wildfires. Many of these EFs are critically important to represent wildfire emissions well: e.g., NH3 (Benedict et al., 2017) and secondary organic aerosol or peroxyacetyl nitrate precursors (Alvarado et al., 2015; Müller et al., 2016). Other approaches to generate representative data that are not explored in detail here but should work well include reporting the average for the laboratory fires clustered around the field-average MCE (fires 8, 39, 45, 51, 59, and 66) or reporting ER to CO (e.g., Koss et al., 2017), where the latter can also be converted to EF by coupling with the field-average EF of CO. For example, if we take the average of six fires clustered around the field-average MCE in the new CH4 plot shown in Fig. 2, we get an average EF for CH4 of 4.67, which is close to the value of 4.90 reported by Liu et al. Alternatively, we can calculate a molar ER for CH4 to CO for all the laboratory fires (0.108), then utilize the wildfire-average EF of CO reported by Liu et al. (89.3 g kg-1) to calculate a new EF. Using this method, we get an EF for CH4 of 5.5, which is within 11 % of the field-average value. Either of these methods should help reflect the field-average flaming / smoldering ratio. In addition, positive matrix factorization was found to be useful to model field and laboratory EFs for NMOGs as discussed elsewhere (Sekimoto et al., 2018). Finally, given the small amount of field sampling, more field work is clearly needed.

We burned individual fuel components (duff, litter, canopy, etc.) in addition to mixtures of major components found in widespread western US coniferous ecosystems for insights into fuel effects on emissions and to what degree specific emissions were enhanced by a certain component. For example, Fig. 4 shows the EFs of 21 trace gases from the Douglas fir ecosystem fuel mixture burns side by side with the EFs from burning pure Douglas fir components in separate fires. Emissions of furaldehyde, formaldehyde, and methanol were enhanced when burning a pure rotten log component, while acetylene, ethylene, and propene, as well as other non-methane hydrocarbons (NMHCs), were more prevalent in emissions from Douglas fir canopy. We did the same analysis for a ponderosa pine ecosystem (Fig. 5). While the canopy component in ponderosa pine produced enhanced emissions of NMHCs, the rotten log did not contribute to the same level of enhancement in furaldehyde, formaldehyde, and methanol because of a transition to flaming combustion during the second half of the fire. Additionally, we observed an enhancement in NOx emissions from the litter and canopy components in ponderosa pine. Coniferous ecosystem values are fairly similar for both fuels and agree within 30 % for the majority of compounds, excluding methanol, furan, and NOx. We also note that while the mixed Douglas fir and ponderosa pine ecosystem fuel mixtures that we burned contained canopy, litter, and woody components in varying diameter classes, they did not contain a rotten log since the latter component is not included in FOFEM. We further investigate fuel variability by taking pure components from several ecosystems and comparing them to one another. Figure 6 shows species emitted by duff from three different coniferous ecosystems. Acetic acid and methanol are strongly emitted by all three duff fuels, but ammonia enhancement occurs in only Engelmann spruce and subalpine fir fuels. Jeffrey pine duff had a lower EF for NH3 despite similar fuel N. This could possibly be due to the age of the fuel as it was contained in storage longer than other fuels and not fresh. Additional results for other fuel components (rotten log, canopy, litter) are in Figs. S1, S2, and S3, respectively.

As mentioned previously, we measured absorption and scattering coefficients directly at 401 and 870 nm. For the first 31 stack fires, which included most of the studied fuel types, we have both 401 and 870 nm data. For the remaining 44 stack fires, we only report data at 870 nm as we used our 401 nm PAX for intercomparison studies that will be reported elsewhere. Figure 7 plots the AAE and SSA at both wavelengths of 31 stack fires as a function of MCE. High AAE is an indicator of BrC and relates to smoldering, which is denoted by low MCE and high SSA values. Smoldering is also associated with higher EFs for OA, most NMOGs, and other gases such as NH3. Low AAE, along with low SSA and high MCE values, indicates more flaming combustion, which is also generally associated with higher EF for BC and “flaming compounds” such as CO2, NOx, and SO2. The lab-based average fire-integrated optical properties for some of the most common western US ecosystems are listed in Table 3. Table 3 does not reveal a strong ecosystem dependence among coniferous ecosystems tested for optical properties but does indicate that chaparral fire aerosol has consistently lower SSA than coniferous fire aerosol and that there are significant contributions of absorption by BrC at 401 nm among all ecosystems. The absorption by BrC is responsible for at least half and up to two-thirds of the absorption at 401 nm even at higher MCE. The laboratory-average AAE of 2.80 ± 1.57 across all 31 fires confirms a role for BrC, while the lab-average SSA at both wavelengths indicates the fresh lab-fire aerosol would have a net warming influence in the atmosphere (SSA < 0.9; Praveen et al., 2012), although SSA can increase with smoke age (Yokelson et al., 2009). The absorption of BrC at 401 nm has several implications in atmospheric chemistry, including impacts on UV-driven photochemical reactions producing ozone, and the lifetime of NOx and HONO. Furthermore, because of its absorbing nature, factoring in the BrC could mean the net radiative forcing of BB is not cooling or neutral as often assumed, but warming if the BrC is sufficiently long-lived as probed in other FIREX studies and previous papers (e.g., Feng et al., 2013; Forrister et al., 2015).

There are very few field measurements of the optical properties of smoke from US wildfires, but we can compare our results from the laboratory studies to the initial aerosol optical properties for one wildfire (the Rim Fire) reported by Liu et al. (2017) and Forrister et al. (2015). An AAE of 3.75 at an MCE of 0.923 for the Rim Fire is reported between these two studies. With the linear regression of the laboratory data shown in Fig. 7, we can predict an AAE of 3.31 at the wildfire field-average MCE (0.912) and an AAE of 2.91 at the Rim Fire MCE (0.923) using prediction method 1 described in Sect. 3.2. At the wildfire field-average MCE, our calculated AAE represents 88 % of the reported Rim Fire AAE, while at the Rim Fire MCE, our calculated AAE represents 78 % of the reported Rim Fire AAE. Although our calculated values are relatively close to the reported value, a small change in AAE implies a big change in the BrC / BC absorption ratio, but only a small change in the percentage of absorption by BrC. Our AAE values imply that BrC accounts for 77 to 82 % of the absorption at 401 nm. The average of the AAE from the single Rim Fire measurement (3.75) and the AAE predicted from the more extensive laboratory fires (3.31) is ∼ 3.5, which may be a reasonable best guess at the AAE of fresh US wildfire smoke and implies that ∼ 86 % of absorption at 401 nm is due to BrC.

In Fig. 8, we plot the initial percentage of absorption by BrC at 401 nm for the Rim Fire measured AAE and for our lab-estimated AAE at the field-average MCE. Figure 8 also shows the lab-measured total EFabs401 and the BrC contribution to EFabs401 for 31 laboratory fires. BrC dominates absorption at 401 nm at low MCE values, and as MCE increases, BrC absorption remains a significant but variable component of overall absorption. The variability is likely due to realistic “natural” fire-to-fire variability in fuels, moisture content, etc.

In Table 4 we report the study averages for BC mass EF, absorption and scattering EFs, SSA, and AAE. The quantities that require 401 nm data are averages for the 31 stack fires for which 401 and 870 nm data were obtained, while the quantities that need just 870 nm data are averages for all 75 stack fires. We also show the comparison of our lab-average and lab-predicted AAEs to the AAE in Forrister et al. (2015) and our lab-average and lab-predicted BC EF to the unpublished BC EF calculated as part of Liu et al. (2017). Table 4 also presents a set of equations that can be used to fit lab-measured optical properties and make predictions at any MCE. However, more measurements of wildfires in the field and the laboratory (including aging) are needed to asses wildfire aerosol optical properties.

Burning individual fuel components in addition to mixtures found in typical, widespread western US ecosystems allows us to investigate the extent to which optical properties are either enhanced or diminished by certain components. Table 5 lists the study-average BC EF and optical properties for all the coniferous ecosystems shown in Table 3 and the study-average BC EF and optical properties for the individual fuel components averaged across all the coniferous ecosystems. The averages and standard deviations for each reported quantity indicate that there is large variation among specific components and a large coefficient of variation for the coniferous ecosystem average. The variability could potentially depend on ecosystem type, fuel components, fuel moisture, or other things as discussed for trace gases in section 3.3. While there is considerable variation within each ecosystem type, the individual ecosystem averages in Table 3 all agree within 38 % of the study average for all the coniferous ecosystems shown in Table 5 and the AAEs are all within 20 %. However, Table 5 also shows that the average AAE for some ecosystem components is very different from the average AAE for all the coniferous ecosystems (2.26). For instance, the largest contribution to a high AAE per fuel component consumed comes from duff, where BrC accounts for almost all of the absorption at 401 nm (AAE 7.13). The rotten log component also contributes an anomalously high average AAE of 4.60. Thus, these components contribute more BrC relative to BC in proportion to their fuel consumption to the mixed ecosystem results, where AAE is 2.26 and BrC accounts for just over half of the absorption at 401 nm. Conversely, litter consumption would tend to lower a fuel mixture's AAE. However, AAE is a measure of the shape of the aerosol absorption cross section and the absorption EFs are a measure of total emissions of absorbing material. In this respect, litter produces more BC absorption and more BrC absorption per unit mass than duff though at a lower BrC / BC ratio than duff. This is consistent with the lower SSA for litter. We conclude that the variability in mixed ecosystem optical properties was likely due to variable consumption of pure components, with a weaker dependence on the dominant tree species. For example, much of the variability in ecosystem-average AAEs and the study-average AAE is linked to the varying amount of duff consumed in the mixed fuel beds (Table S1). (The variability in actual duff consumption is likely larger than the variability in duff loading shown as the amount of residual material also varied.) Duff consumption in the field is increased by drought conditions, which would contribute to variability in real fires (Davies et al., 2013).

We can compare our duff results to previous measurements of optical properties of duff-fire aerosol by Chakrabarty et al. (2010). These authors identified tar balls as a major BrC species produced by duff combustion and they measured an AAE of 4.2 (405 and 532 nm wavelength pair) for a ponderosa pine duff sample from MT. Including their other duff sample (Alaskan feather moss duff), they obtained a study-average duff-combustion AAE of 5.3. We measured AAE on two much larger burns (∼ 4 times more fuel mass, fire nos. 12 and 26) in Engelmann spruce duff, with different wavelengths, and at much lower MCE (0.843 ± 0.036 versus ∼ 0.91). We obtained a study-average duff combustion AAE of 7.13 (0.057). Both studies observed a high AAE for duff combustion. Their lower AAE values could be related to different wavelengths used, the possibility of some BrC absorption at 532 nm (Bluvshtein et al., 2017), the different duff type, and/or their higher MCE, which they attributed to sampling some flaming combustion during the ignition process. The AAE calculated from our AAE versus MCE fit (for all fuels) at their MCE of 0.91 is relatively closer to their value.

In summary, the results presented indicate that, in all cases, burning a typical ecosystem mixture of components produces a significant amount of BrC. As mentioned previously, this has several implications in regional atmospheric chemistry and radiative forcing. Additional instruments were deployed on room burn experiments, in which the fuels were also purposely changed to investigate the effect on optical properties and the results will be reported elsewhere (e.g., Manfred et al., 2018).

We also measured emissions from several fires of peat, rice straw, and dung due to their widespread burning in Asia and their value as extreme examples of fuel impacts (e.g., high smoldering-to-flaming ratio or high N or Cl content). Peat, which is especially important in Southeast Asia (Stockwell et al., 2016a) is similar to duff found in the western US in that it is often consumed by pure smoldering combustion and produces high AAE (Pokhrel et al., 2016), high HCN emissions, and low BC emissions. Although we did not measure the AAE for our peat fire aerosol, we do report an MCE of 0.83, where a low MCE likely indicates a high AAE. We also report EFs for CH4 (10.39 g kg-1), HCN (3.97 g kg-1), acetic acid (4.44 g kg-1), and BC (0.003 g kg-1). We compare these values to the field measurements reported in Stockwell et al. (2016a): CH4 (9.51 ± 4.74 g kg-1), HCN (5.75 ± 1.60 g kg-1), acetic acid (3.89 ± 1.65 g kg-1), and BC (0.006 ±0.002 g kg-1) and find that our values agree well (EF of BC extremely small compared to most BB (Akagi et al., 2011) and gases within 31 %) between peat fire measurements in the laboratory and the field. (A more detailed comparison will follow planned field measurements.)

Additionally, we compare our dung MCE (0.90), CH4 (6.63 g kg-1), HCN (1.96 g kg-1), acetic acid (6.36 g kg-1), and BC (0.01 g kg-1) values to those based on field work in Nepal reported in Stockwell et al. (2016b): MCE (0.90), CH4 (6.65 ± 0.46 g kg-1), HCN (2.01 ± 1.25 g kg-1), acetic acid (7.32 ± 6.59 g kg-1), and BC (0.004 ± 0.003 g kg-1). We find excellent agreement between our values (15 % for trace gases and a very small EF of BC) and those reported from field measurements in Nepal.

Rice straw was burned because of its global importance in agricultural waste burning and to probe the extremes of fuel chemistry (Akagi et al., 2011). Grasses are usually very high in chlorine content (0.61 %, Table S1; Lobert et al., 1999) and our EF for HCl of 0.65 g kg-1 for rice straw was the highest of any fuel measured during the FIREX campaign. Furthermore, our rice straw EF for HCl is comparable to Stockwell et al. 2015 (0.43 ± 0.29). The findings briefly summarized in this section further suggest and reinforce the idea that simulated laboratory fires can probe fuel effects and provide an accurate representation of measurements in the field, even outside the scope of western US wildfires. More comprehensive, recent discussions of these fuels can be found elsewhere (Stockwell et al., 2016a, b; Jayarathne et al., 2018a, b).