2.1 N_r and NO measurements by chemiluminescence

Total reactive N (N_r) was measured by catalytic conversion to NO, followed by O₃ chemiluminescence using an instrument described previously (Williams et al., 1998). N_r and NO were sampled from inlets inserted adjacent to the inlet-less open-path Fourier transform infrared spectrometer (OP-FTIR) instrument path during the stack burns (Selimovic et al., 2018) and from a platform approximately 4 m off the floor in the middle of the room during the room burns. The catalyst used for the N_r channel, described in detail by Stockwell et al. (2018), consisted of a 11 mm i.d. quartz tube, packed with 36 platinum screens, heated to 750 ^∘C. This tube was wrapped with high-temperature heating tape and insulated inside a 7 cm o.d. stainless steel tube that was fitted to a bulkhead placed through the wall of the stack. The N_r channel was diluted by a factor of 5:1 (±3 %) using a flow of zero air added immediately downstream of the Pt catalyst assembly. NO was sampled through a 6.3 mm o.d. stainless steel inlet tube, which was placed through the bulkhead directly into the free air stream of the stack and connected to a 50 mm Teflon filter holder immediately outside the stack. The transfer lines for the N_r and NO measurements consisted of 6.35 mm o.d., 1 mm wall thickness PFA tubing of approximately 20 m in length. N_r and NO data were acquired at 1 s frequency, but the flow rate through each inlet was 1 SL min⁻¹, resulting in residence time in each inlet of 14 s. This time delay was corrected in the data analysis. Any chemical effects of the inlet on the sampled air stream were negligible since the analytes consisted of only NO and NO₂, and those are known to be transmitted by PFA Teflon tubing with essentially no surface effects. However, there were possible effects of the inlets on the temporal features of the measurement through diffusion or turbulent mixing. Those effects were examined through comparison of the temporal variations in the NO signal with the NO measured by the OP-FTIR and comparison of the N_r signal under smoldering conditions with the NH₃ measured by the OP-FTIR. Both of these comparisons showed that the NO and N_r inlets had effective time constants of 4 s, somewhat slower than the diffusive relaxation time assuming solely laminar flow. Examples of the estimate of diffusion and dispersion on NO and N_r signals and the estimate of the effective time constant of these measurements are presented in the Supplement.

The inlet streams were sampled by the NO instrument either directly (NO channel) or after passing through a second catalyst of molybdenum oxide (MoOx) to convert remaining NO₂ to NO. The MoOx catalyst consisted of a molybdenum tube at 350 ^∘C to which a small flow of H₂ (0.8 % v∕v) was added to control the redox state of the surface. Both channels of the instrument were “detuned” to keep raw photon count rates below 4 MHz, by turning down the O₃ flows and photo-multiplier tube (PMT) voltages. Calibrations were performed with both a NO standard in N₂ (Scott-Marrin) and 10.1 ppmv standard of HCN in nitrogen (Gasco). The Pt catalyst was dismounted from the stack (or room) every few days and checked for conversion efficiency by the addition of the HCN standard to the inlet. Conversion efficiencies were found to be consistently high (>98 %) throughout the entire sampling period (5 October–12 November 2016). There were slight background signals (a few tens of ppbv) for both NO and N_r in both the stack and room air prior to and after the burns, and those were subtracted from the fires' signals prior to reporting the data. The overall uncertainties in the NO and N_r data were ±10 % for each measurement.

2.2 Other measurements

Measurements of individual species during the 2016 Fire Lab study have been presented in several previous publications. The OP-FTIR measurements were discussed by Selimovic et al. (2018), and the PTR-ToF measurements were discussed by Koss et al. (2018). In addition, some of the calibration methods and GC separation and identifications rely on additional analytical work presented by Sekimoto et al. (2017) and Gilman et al. (2015). We measured the mass and elemental content of the initial fuel and the mass of unburned fuel for all the fires, and we measured the mass and the elemental content of the ash during 21 room burns, which covered all the fuel types discussed.

2.3 PMF analysis

Trace gas measurements from multiple instruments involved in the Fire Lab study were combined and analyzed using positive matrix factorization (PMF). PMF is a numerical method that was used in this case to partition the compounds involved in a time-varying mixture of chemicals into a few groups, or “factors”, where a compound can appear in more than one factor. A factor represents a consistent profile of compounds that is representative of one of the sources contributing to the total signal. The sum of all the factors then ideally describes the total composition of the measurements, which in this case is the emissions of N_r compounds. By its nature, PMF assumes that the total signal is a linear combination of individual sources that have a consistent composition, the relative contribution of which is represented by the amount of each compound or category found in each factor (Paatero and Tapper, 1994; Ulbrich et al., 2009). We hypothesize that species with dominant fractions in the same factor are related to each other via the same formation processes. With knowledge of factor composition and the amount of each factor at any given time, the original emissions measurements can be reconstructed, and this approach provides an alternate source of profiles for fire emissions. PMF has also been used by a number of groups to explore how much various source profiles contribute to complex ambient measurements (see for example Ulbrich et al., 2009) and was recently used to analyze PTR-ToF-MS measurements from the Fire Lab (Sekimoto et al., 2018). Here, PMF was accomplished using the PMF Evaluation Tool v. 2.08A (Ulbrich et al., 2009).

Table 2Compounds and compound classes used in the PMF analyses and their corresponding errors.

Download Print Version | Download XLSX

The application of PMF to this data set is different than the instances where it is applied to data from a single instrument in which compound abundances are inherently scaled properly and error estimates are well defined and self-consistent. For example, when applied to mass spectral data from a single instrument, errors can be expected to scale as the square root of ion counts based on fundamental counting statistics (Sekimoto et al., 2018). In this work we are including nitrogen measurements from several instruments, and thus we chose to use mixing ratios as the unit of comparison. The error estimates required by the PMF analysis were taken from the reported combined uncertainties: the sum of the detection limit plus the estimated random error of the measured value. For example, the uncertainty in a NO mixing ratio of 500 ppbv was ±51 ppbv. The variables that were used in this PMF analysis and their units and corresponding errors are listed in Table 2. Where compound categories are specified (e.g., nitriles), the values were the sum of the measured compounds in that category as listed in the footnotes of Table 1. The data were further adjusted by subtracting the ambient air background before and after the fires, which was a relatively minor adjustment for most compounds and categories. Any negative numbers that resulted were very small compared to the fire emissions and were set to zero. In addition to the PMF analysis for the species listed in Table 2, several exploratory runs were tried with CO₂, CO added (in units of ppmv), and total N_r (in units of ppbv) added to the list in Table 2. No significant differences were observed in the results for individual N_r compounds and classes, so CO₂, CO, and total N_r were not included in this analysis.

We applied PMF to single-fire data as well as extended time series that included all fires of a particular fuel type, in line with the approach laid out by Sekimoto et al. (2018). By consolidating fuels from a particular vegetation type, the fire-to-fire variability largely driven by differences in the fuel (e.g., moisture content, structure, quantity) is constrained and the most representative fire conditions are captured. Two fuel groups were analyzed in this way: the western US coniferous ecosystem fuels, which included ponderosa pine, lodgepole pine, Douglas fir, Engelmann spruce, and subalpine fir, and the southwestern US chaparral ecosystem, which was represented by chamise and manzanita. The consolidated time series for the coniferous ecosystems included realistic mixtures, canopy only, and litter only, while duff and rotten logs were analyzed separately and not included in the time series.

3.1 Comparison of N_r and total carbon in fire emissions

The total N_r and total carbon emissions were measured for 75 stack fires in order to place the N emissions in the context of total carbon, which has been widely estimated for wildfires. Example time series of NO, N_r, ΔCO, and ΔCO₂ (CO and CO₂ corrected for their backgrounds) are shown in Fig. 2, for a fire burning a sample of ponderosa pine realistic mix (Fire 004). In addition to the chemical species, the modified combustion efficiency (MCE) was also plotted. MCE is defined as

where ΔCO₂ and ΔCO are the CO₂ and CO levels above the ambient. MCE has traditionally been used to indicate the relative amount of flaming and smoldering combustion in a fire. The time series for Fire 004 (Fig. 2) shows a short initial smoldering/distillation phase (MCE 0.7 to 0.8) as heat pyrolyzes the fresh fuel and releases VOCs from existing pools in the fuel, followed after ignition by a relatively efficient mix of flaming and smoldering combustion (MCE 0.95 to 0.98) and then finally a subsequent period of essentially pure smoldering (MCE ∼0.80). The N_r and NO timelines had many features in common because NO is often the most abundant N_r compound (see below). As a result, it is useful to compare the quantities N_r−NO and (N_r−NO) ∕ N_r to the other measures of chemical species or combustion efficiency. As expected, (N_r−NO) ∕ N_r, in Fig. 2c is anticorrelated with MCE since N_r is primarily NO at high MCE. In addition to the anticorrelation, this non-NO fraction, like its approximate carbon analog CO∕CO₂, has a wider dynamic range than MCE and will often suffer less from background variability than carbon-based indices (Yokelson et al., 2013a).

The concentration profiles of the background-corrected measurements of N_r, CO₂, CO, and all the carbon-containing species measured by the FTIR (Selimovic et al., 2018) during the stack burns were integrated over the entire time of the burn to obtain total carbon, termed TC here, and total N_r. The additional carbon species included methane and a number of other gas phase VOCs as well as organic- and black-carbon aerosol. Altogether these carbon species should account for ≥98 % of emitted carbon (McMeeking et al., 2009). Total N_r is plotted in Fig. 3, versus TC (Fig. 3a) and versus nitrogen burned, which is calculated from the %N in the fuel times the mass of fuel consumed (Fig. 3b). The points in Fig. 3 are colored by the fuel N∕C mole % obtained from the elemental analysis of each fuel. The relationship between N_r and TC in Fig. 3a clusters around the 0.37 % line, and those points are from fuels most characteristic of the North American biomes impacted by wildfire. There are clear outliers in the correlation of N_r and TC; for example, yak dung and two samples of duff (Engelmann spruce and subalpine fir) were high due to the fact that they either have high fuel N∕C ratios (dung; see Table S1 in the Supplement) or burned with minimal flaming (whole-fire MCEs 0.86–0.89, duff and dung), hence experienced less denitrification. The fuels that were low in N_r∕TC in Fig. 3a, ponderosa pine rotten log, subalpine fir, and excelsior, had low fuel N∕C, so when plotted versus nitrogen burned in Fig. 3b, they cluster with the main group of characteristic fuels, i.e., they are no longer “outliers” in the distribution.

Figure 3Integrated N_r versus (a) integrated total carbon and versus (b) nitrogen burned, based on fuel nitrogen content and mass of fuel burned. The points are colored by fuel nitrogen to carbon ratio. Note that the x and y scales in panel (a) are different by more than a factor of 100.

Download

3.2 Estimates of denitrification

The removal of N to forms that are inactive in the troposphere, N₂ and N₂O, has importance in the biogeochemistry of forest ecosystems and also determines how much N takes part in wildfire plume chemistry. The points in Fig. 3a are all lower than the corresponding fuel N∕C mole ratio, due to the denitrification chemistry, shown in Fig. 1 and verified in lab studies described by Kuhlbusch et al. (1991), and the production of N₂O, which is also not measured by the N_r technique. The sum of N₂ and N₂O produced in the fires can be estimated from the difference between the fuel N∕C and the N_r∕total C emitted and the data on C and N content remaining in the ash. The mass balance equations used for these calculations are detailed in the Supplement.

Figure 4The histogram of the fraction of N loss to N₂ and N₂O estimated from the mass balance analysis described in the Supplement (52 burns).

Download

The distribution of the N lost to N₂ and N₂O is shown in Fig. 4. Chemical analyses were not done for all fuels during the stack burns, and the analysis above assumes that the ash residues and ash ∕ burned fuel ratios from the stack burns were well represented by those for the same fuels used in the room burns, for which mass yields and chemical analyses were done. Data are missing for fuels that did not have a corresponding residue analysis. The median fraction of N lost to N₂ and N₂O for ash-corrected fires was 0.70, and the mean (± standard deviation) was 0.68 (±0.14). This fuel-based estimate is uncertain by approximately 25 % because of the above assumptions concerning the applicability of the residue analyses from the room burns and because fuel moisture corrections were assumed to apply to all of the materials burned, foliage vs. woody biomass (see the Supplement for details). The emission of N₂O relative to N₂ is approximately 10 % or less for a wide range of fuels (Andreae, 2019). Assuming the N remainder in our work is at least 90 % N₂ gives values that are somewhat higher than the N₂ values reported by Kuhlbusch et al. (1991) where N₂ accounted for 36 % of fuel N burned in flaming-stage fires. A closer inspection of Kuhlbusch et al. (1991) showed a range of N₂ yields of 40 %–54 % at highest MCEs of 0.94–0.97. Possible reasons for these differences are that the Kuhlbusch et al. (1991) fires were limited to grasses, hay, and pine needles, and the fires were confined to a closed container and so may not have experienced the convection and turbulence of typical biomass fires. In addition, the fires analyzed in our work were somewhat weighted towards the full canopy and higher temperature burning fuels, since ash analyses were not done for peat, dung, and many of the “litter” samples, all of which tend to burn less efficiently. Goode et al. (1999) estimated an N₂ emission of 45±5 % for MCE values of 0.95 in grass and surface fuels. The range of values determined in our work overlap with these literature values but are on average higher. It should be noted that such loss of reactive nitrogen can have implications for ecosystem N budgets, as discussed by Kuhlbusch et al. (1991).

Figure 5Timelines of (a) N_r and NO, (b) N_r minus NO and the sum of all measures N species except for NO, (c) residual of N_r minus all measured N species (N_r−NO−Sum N), (d) and MCE and (N_r−NO) ∕ N_r for Fire 047, subalpine fir realistic mix. The yellow box highlights the area of higher residual N_r that corresponds to more smoldering emissions.

Download

3.3 The budget of N_r and individual N-containing species

The composition of the N that does not get converted to N₂ or N₂O is of intense importance in determining atmospheric impacts of fires, since those compounds are involved in oxidation capacity (NO_x), radical production (HONO), and particle formation (NH₃). Emission factors (EF, defined as mass of compound emitted per mass of fuel burn) for all the individual N_r compounds identified in our work have been compiled and reported in previous publications (Koss et al., 2018; Selimovic et al., 2018), so this paper will focus on the N_r budget. The balance of N_r budget for Fire 047, subalpine fir realistic mix, is shown in Fig. 5, in which the timelines of N_r, NO, N_r−NO, sumN, and NVOC are plotted along with MCE and (N_r−NO) ∕ N_r. The quantity sumN is the sum of all other non-NO compounds, and NVOC is the subset of sumN that are organic nitrogen compounds measured by the PTR-ToF, as listed in Table 1. This fire had a mixture of flaming and smoldering combustion throughout the fire as indicated by MCE and nitrogen profiles (Fig. 5d). The comparison of N_r−NO with the sumN in Fig. 5b shows that much of the N is accounted for. The major contributors to sumN for this fire were HNCO, HCN, HONO, NO₂, and NH₃, while NVOC was a very small contributor to sumN (Fig. 5b). Note that while HNO₃ is measurable by FTIR with good sensitivity, no HNO₃ signals were observed above detection limit, which was 10 ppbv. Fig. 5c shows the residual left after NO and sumN are subtracted from N_r, corresponding to an integrated amount of 15.6±8 % of N_r. This residual is reasonable, considering typical published particle N_r measurements (Akagi et al., 2012, 2011; Liu et al., 2017; May et al., 2014), and consistent with there being some particle N_r from flaming, which is most likely organic nitrates or nitro-organics, and particle ammonium from smoldering with potassium or ammonium nitrate, potentially accounting for substantial N_r.

Several fuels had much lower NO emissions and higher unaccounted-for N_r. Yak dung was one such fuel, the emissions of which stand in contrast to the fire shown above. The nitrogen emissions from Fire 050, yak dung, are shown in Fig. S2 in the Supplement. This fuel produced mostly smoldering emissions as exemplified by the low NO levels relative to N_r (panel a) and the low MCEs observed (panel d). The sum of N_r species was somewhat correlated with the quantity N_r−NO but was substantially lower, and the residual N_r unaccounted for by the gas-phase measurements was 33.9±16 % of N_r (panel c). The majority of sumN was represented by HCN and NH₃, with acetonitrile (CH₃CN) higher than any of the other inorganics, HNCO, NO₂, or HONO. The NVOCs were also a larger fraction of N_r−NO than in the case of Fire 047 shown above, a feature that implies that more semivolatile organic compounds (SVOCs) survive these types of fires and could make a proportionally higher contribution to the N_r budget in this fire relative to Fire 047. Fire Lab results of particle organic carbon measurements (Jen et al., 2019) and field measurements in environments with a lot of dung burning (Jayarathne et al., 2018; Stockwell et al., 2016a) are consistent with a higher EF for particle organic carbon and by extension particle NVOC compounds. The quantity (N_r−NO) ∕ N_r was relatively high and had less dynamic range than for fires with more flaming combustion like Fire 047.

Table 3Summary of X_i∕N_r measurements for all stack burns^a.

^a Not every measurement was available for every fire; consequently the values do not add up to exactly 100 %.

Download Print Version | Download XLSX

Figure 6A histogram of the residual N for all the stack fires during the 2016 Fire Lab study for which there are FTIR, ACES, and PTR-ToF measurements (n=43). The median is 0.143, and the mean (± SD) was 0.15 (±0.10).

Download

An overall budget of N_r can be constructed for all of the stack fires through integrating the time profile of all the compounds and compound classes. The fire-integrated measurements of inorganic and NVOC species are listed in the Supplement as ratios to N_r for each stack fire (Table S1). The summary of all the fire integrated X_i∕N_r fractions (where X_i is the N_r species or quantity) is given in Table 3 for all the fires for which we have a complete set of measurements (43 fires). In general, NO was the major species, followed by NH₃, and the other inorganic N_r species, NO₂, HNCO, HONO, and HCN had individual contributions of 4.3 % to 9.4 %. NVOC species were less than 5 % of N_r on average. The unaccounted-for N_r, defined as $({\mathrm{N}}_{\mathrm{r}}-\mathrm{NO}-\mathrm{sumN})/{\mathrm{N}}_{\mathrm{r}}$ had a median value of 14.3 % and a mean (± SD) of 15 (±10) %. Overall, 85 % of N_r was accounted for by the gas-phase measurements. The distribution of whole-fire N_r residuals is plotted as a histogram in Fig. 6. We expect the residual N_r was composed of either semi- or low-volatility compounds or particle-bound N_r compounds, which we know are converted efficiently by the N_r catalyst (Stockwell et al., 2018) but not detected by the instruments included in this analysis. Along these lines, there is some indication that the residual has a systematic variation with whole-fire MCE, with higher residuals (up to 30 %) observed at lower MCEs and higher (N_r−NO) ∕ N_r (see Fig. S3a, b), which would be consistent with higher EF for SVOC at low MCE (Jen et al., 2019) and particle N_r having a higher contribution from ${\mathrm{NO}}_{\mathrm{3}}^{-}$ (May et al., 2014) and, perhaps, particle ammonium or reduced-N_r compounds. In general, there is more particulate organic material emitted from fires at low MCE (Jen et al., 2019), so we would expect more particle N at low MCE to go along with that.

3.4 Systematic dependences of N_r composition on combustion processes

The features noted in fires shown above, as well as the anticorrelation of MCE and (N_r−NO) ∕ N_r lead to the question of whether there are systematic dependences in N_r-compound composition on fire stage that can be used to formally classify and/or potentially predict the relative emissions of N_r compounds. MCE has been used as a rough indicator of the relative amounts of flaming and smoldering combustion in a fire, with high MCE (∼99 %) being “pure” flaming, low MCE (∼80 %) being “pure smoldering,” and an MCE of ∼0.9 being roughly equal amounts of both (Sect. 2.1.1 in Akagi et al., 2011). It should be understood that “smoldering” in this framework is a lumped term that includes all non-flame processes such as pyrolysis, glowing, and distillation, which are the processes that produce gaseous fuel to support flaming (Yokelson et al., 1996). In addition, pure flaming is essentially the efficient oxidation of smoldering products before they enter the atmosphere. However, for MCE to predict flaming and smoldering N_r species well, the variable fuel N must be considered. For instance, NO_x is clearly produced by flaming based on its temporal profile, but fire-integrated EF${}_{{\mathrm{NO}}_x}$ may not correlate with MCE due to variable fuel N. In these cases, EF${}_{{\mathrm{NO}}_x}$ ∕ fuel N or ΔNH₃∕ΔNO_x may still correlate (or anticorrelate) well with MCE (e.g., Fig. 4 in Burling et al., 2010, or Yokelson et al., 1996). Finally, the flame chemistry involving NH₃, HNCO, and HCN both produces and destroys NO in a fashion that does not conserve N_r. This chemistry is explored in Fig. 7, in which NO_x, NH₃, HNCO, HCN, HONO, and CH₃CN ratios to N_r are plotted vs. real-time MCE for Fire 047 as a typical example for fires that have a substantial range of MCEs (e.g., from 0.8 to above 0.98). The relationship between NH₃∕N_r and MCE confirms that NH₃ is primarily a smoldering emission, and NO_x∕N_r increases with increasing MCE in a nonlinear fashion that confirms it is primarily a flaming compound. Such a nonlinear dependence has also been seen for other flaming-related quantities such as elemental carbon/TC or EF_HCl (Christian et al., 2003; Stockwell et al., 2014). Most importantly, the variations in HNCO∕N_r, HCN∕N_r, HONO∕N_r, and CH₃CN∕N_r versus MCE do not arise dominantly from either regime, as these are species that are likely produced by multiple pathways (e.g., “incomplete flaming”, pyrolysis, or possibly glowing). By incomplete flame chemistry we mean the production of incompletely oxidized products in flames such as the complex system of reactions shown in Fig. 1. These reactions involving HNCO, HCN, and NH₃ both produce and destroy NO, while HONO is produced from reactions of NO and NO₂ that are faster at slightly lower temperatures, for example the three-body association reaction of NO with OH radical (Manion et al., 2015). Variable turbulence in the turbulent diffusion flames that are characteristic of open BB likely contributes to varying temperatures and, therefore, varying amounts of incomplete oxidation of the fuel N (Shaddix et al., 1994).

Figure 7(a) The relationships between NO_x∕N_r and NH₃∕N_r vs. MCE and (b) the HNCO∕N_r, HCN∕N_r, HONO∕N_r, and CH₃CN∕N_r vs. MCE for Fire 047.

Download

Figure 8(a) The measured N_r signal for Fire 063 (lodgepole pine) (blue line), the sum of the signal reconstructed by the PMF (purple points), and the three PMF factors – combustion (grey), high temperature (green), and low temperature (red) – plotted in a stacked fashion (i.e., added on top of one another). (b) The “residual” of the PMF fit consisting of the measured N_r signal minus the N_r signal reconstructed by the PMF, as a percentage of the N_r signal.

Download

3.5 The PMF analysis of coniferous fuels

The complexity of the dependence of N_r speciation on combustion chemistry suggests that MCE is an insufficient model to use for applying lab results to real-world fire emissions (Stockwell et al., 2016a; Yokelson et al., 2013b). Accordingly, we employed the positive matrix factorization (PMF) method (see Methodology section) that has been used by a number of groups to probe the sources contributing to complex mixtures (see for example Ulbrich et al., 2009; Sekimoto et al., 2018). Our PMF results showed several general features, irrespective of the inclusion or exclusion of CO₂, CO, and N_r. The emissions were best fit by three factors (with approximate descriptive names justified below and prime species): (1) a combustion (flaming) factor (abbreviated Comb-N), (2) a high-temperature pyrolysis factor (HT-N), and (3) a low-temperature pyrolysis factor (LT-N). We use these terms in part to harmonize our discussion with the VOC results discussed by Sekimoto et al. (2018). An example time series for the PMF analysis of a coniferous fuel with just the N_r species included is shown in Fig. 8 for a realistic mix of lodgepole pine (Fire 063). In this case, several different F_peak values were tried (−1, 0, +1) and runs with 100 different seeds (initial factor profiles) were performed. The results of those analyses (Fig. S4) show that a three-factor PMF result is robust. A PMF analysis was performed on the consolidated time series of all coniferous fuels fit using just the N_r species, as shown in Fig. S5. In this case F_peak=0 was used, and the Q∕Q_expected showed an inflection for the three-factor solution at a value of 5.3. The three factors successfully describe the majority of the N_r emissions where the difference between the measured and calculated mass is on average 5.1 % for coniferous fuels and 4.6 % for chaparrals as indicated in Table 4.

Figure 9(a) The contributions of nitrogen species to the factors that simulate the emissions from coniferous fuels shown in Fig. S2 and (b) the fraction of each compound or class found in each factor.

Download

Table 4Residuals of the PMF analyses by fuel, as percent of total signal.

Download Print Version | Download XLSX

Several metrics of the PMF analysis quantify how the compounds or compound classes contribute to each factor. The “loadings” of the three different factors, i.e., the contribution of compounds to each factor, for coniferous fuels are shown in Fig. 9a, and the distribution of a given compound or compound class amongst the three factors is shown in Fig. 9b as normalized fraction. Normalized fraction is equal to the PMF-determined contribution of a compound to a factor, divided by the sum of the contribution of the compound to all three factors. The Comb-N factor contained NO, NO₂, and HONO; the HT-N factor had mostly HCN, HNCO, and nitriles, with contributions from NO₂ and nitro compounds; and the LT-N factor contained NH₃, amines, amides, and heterocyclics. Within the Comb-N factor there is some evidence that the relative amounts of HONO and NO_x depend on fuel moisture. For example, the ratio HONO∕NO_x for whole fires shows some correlation with needle moisture in coniferous fires that were canopy fuels (foliage and small woody biomass), as shown in Fig. S6. This may be due to flame processes that interconvert NO_x and HONO in the presence of water vapor or OH (see Fig. 1).

Literature values from studies where flame temperature was measured are typically in the range of 1100–1200 ^∘C (Taylor et al., 2004; Wotton et al., 2012), so we would assume that would constitute the upper range of our Comb-N factor. The radical chemistry involving HCN, HNCO, and NH₃ starts to shut down below about 800–900 ^∘C, according to the modeling of Glarborg et al. (2018), so we set 800 ^∘C as a lower limit for the Comb-N factor. The HT-N factor species are known to be produced by the intense pyrolysis of fuel N_r compounds (Hansson et al., 2004; Liu et al., 2018; Ren et al., 2010), which for these compounds becomes important at temperatures around 500–600 ^∘C. Accordingly, we estimate the temperature range for the HT-N factor at 500–800 ^∘C. The remaining LT-N factor results from mild pyrolysis and pertains to fire conditions of roughly 500 ^∘C and below, and it was dominated by NH₃, amines, amides, and some of the more complex organics (Koss et al., 2018). The names and temperature ranges are approximate and likely include processes that occur inside flames as part of the flame proper, as turbulent diffusive flames are highly variable in space and time.

Figure 10Comparisons of the N-PMF combustion factor (Comb-N) with (a) CO₂ and MCE for Fire 037 (ponderosa pine). Panel (b) shows the scatter plot of the Comb-N factor versus CO₂, and panel (c) shows the scatter plot of Comb-N factor versus MCE.

Download

3.6 The comparison of N-PMF factors to other fire parameters and VOC emission factors

It is useful to explore the correlation of N-PMF factors with other fire indicators to determine relationships for parameterizing N_r emissions together with carbon and VOC emissions, in order to simplify how emissions might be parameterized in models. The Comb-N factor for coniferous fuels, which consisted of NO_x and HONO, would be expected to correlate with CO₂ but not as well with MCE since the latter includes an indicator of incomplete combustion. The time series of Comb-N along with CO₂ and with MCE for Fire 037 (ponderosa pine) are plotted in Fig. 10. As expected they show an excellent correlation of Comb-N with CO₂ (R²=0.942), since all the species are flaming compounds, but nonlinear correlation of Comb-N with MCE (R²=0.363), since the latter factors in a smoldering compound (CO), similar to the NO_x∕N_r vs. MCE plot for Fire 047 in Fig. 7. The excellent correlation of Comb-N with CO₂ is a broadly applicable result; the R² parameters for all the fires shown in Fig. S5 had an average of 0.898 and ranged from 0.806 to 0.966. As a consequence, we can conclude that CO₂ would be the best tracer for Comb-N in many western US ecosystems where conifers predominate, provided ambient CO₂ backgrounds can be properly accounted for as described by Yokelson et al. (2013a).

Figure 11Details of the PMF factors for Fire 037 (ponderosa pine). Panel (a) shows the total N_r signal (magenta) and the Comb-N factor (black); panel (b) shows the HT-N factor (green) and HT- VOC factor (blue), and panel (c) shows the LT-N factor (red) and LT-VOC factor (orange). The insets (d) and (e) show the correlation of the two HT factors and the correlation between the two LT factors, respectively.

Download

Our Comb-N factor did not correspond to the high-temperature VOC factor (HT-VOC) found by Sekimoto et al. (2018) in their pyrolysis study, because our broader study includes flaming combustion, which produces NO_x and HONO, and almost none of the compounds classified as VOCs survive flaming conditions. However, the HT-N and HT-VOC factors are both linked to pyrolysis and were well correlated for many fires. An example of this is shown in Fig. 11 for Fire 037, a sample that was broadly representative of ponderosa pine (i.e., canopy and litter). This result can be rationalized by the fact that while HT-VOC factors have large contributions from many more compounds than the N compounds measured here, they also have large contributions (>85 %) from HCN, HNCO, and HONO (in other words > 85 % of HCN, HNCO, and HONO are found in the HT-VOC factor). Since the HT-N factors are also heavily weighted by HCN and HNCO, it is reassuring that both of these PMF analyses have independently identified these species as important contributors to the HT fire regime. The R² correlation coefficients between the HT-N and HT-VOC factors for the coniferous fires shown in Fig. S5 averaged 0.866 and ranged from 0.419 to 0.959. As a consequence of this correlation, we can conclude that HCN is the best marker for the HT-N and HT-VOC factors in most western US wildfires, since HCN is essentially inert on the timescales of fire plumes (Li et al., 2000). It should be noted that other nitriles, particularly acetonitrile, also show up in the HT-N factor, and acetonitrile has also been used as a tracer of biomass combustion. However, it has been shown that this acetonitrile signal can be obscured in urban or industrial areas by solvent usage or can be quite small in woodstove emissions due to low N in the fuel (Coggon et al., 2016).

The correlations of LT-N and LT-VOC factors were not particularly high for most of the coniferous fires shown in Fig. S5. The average R² was 0.427 with a range of between 0.072 and 0.827. The reasons for this lack of correlation are not clear, as NH₃, amines, and amides appear predominantly in both LT factors, and the absolute concentrations of NH₃ are usually quite high in these fires relative to VOCs (Sekimoto et al., 2018). However, the LT-VOC factor includes many more compounds with a variety of functional groups not found in the LT-N factor, so it appears that the VOC and N compounds have sufficiently different pyrolysis chemistry that the LT factors do not show much correlation. We conclude that NH₃ (and particle ${\mathrm{NH}}_{\mathrm{4}}^{+}$) will be the best marker for the LT-N factor in western US coniferous wildfires, but the LT-VOC chemistry might not be captured reliably by this marker.

3.7 PMF analysis of chaparral fuels

Chaparral is an important ecosystem of concern in wildfires that occur in central and southern California and other areas of the southwestern US. The emissions from burning chaparral fuels (manzanita and chamise) collected at two sites in California were also analyzed as a group and yielded three separate factors in a fashion similar to the coniferous fuels (see Fig. S7 for the PMF timeline). As with the coniferous fuels, there was essentially no change in the three-factor solution with F_peak, so F_peak=0 was used, and the Q∕Q_exp was 3.8. The chaparral factors had slightly different composition (Fig. S8); the combustion factor was mostly NO, with small amounts of HNCO, HONO, and NH₃; the high-temperature factor was dominated by NO₂ and included HONO, HCN, and HNCO; and the low-temperature factor was mostly NH₃ with a slight amount of NO contributing. The NVOC species were found in both the medium- and low-temperature factors.

There was less similarity between the Comb-N factor and CO₂ emissions for chaparral fuels compared to those found for coniferous fuels, with an average correlation coefficient (R²) of 0.689, with a range from 0.244 and 0.950. As a result, there may not be a simple conserved tracer for the combustion factor of these fuel types; however total odd nitrogen (NO_y) which is NO_x and all the compounds that are produced from NO_x in the troposphere, may be useful as it is a reasonably conserved tracer in the absence of wet or dry deposition of particles. Correlation coefficients between the HT-N and HT-VOC factors were on average R²=0.551, with a range 0.047–0.911. The correlations between LT-N and LT-VOC factors were in the same range for chaparral fuels as for coniferous, with average R²=0.447, range 0.028–0.827.

There were some fuels that do not sustain flaming combustion well, specifically duff, Yak dung, and Indonesian peat. These fires exhibited little or no NO emission commensurate with minimal flaming combustion. Instead the emissions were mostly the pyrolysis products NH₃, (0.22–0.53 N_r fraction), and HCN (up to 0.32 N_r fraction for peat). It was also apparent that these fires also had unaccounted-for N_r, close to or just over 0.30 (Table S1). The distribution of N_r compounds in the one peat fire that we measured (Fire 055) is in line with those reported for fires measured in situ, which showed relatively high EFs for HCN and NH₃ (Stockwell et al., 2016b, 2015).

3.8 Application to real-world fires

The application of our N_r emissions results to real-world fires will depend somewhat on the nature of the information available on a particular fire or fire complex. As a good starting point, or in the absence of detailed N and C analyses of fuels, a N_r∕C ratio of 0.37 % appears to capture most of the fires studied in this work. The N_r could be apportioned according to the results summarized in Table 3. Adjustments to those fractions can be made by scaling slightly by average MCE, with the knowledge that intermediate species (HT-N pyrolysis species) such as HCN and HNCO do not scale in the simple manner that NH₃ and NO_x do. If measurements of marker compounds are available then CO₂, HCN, and the sum ${\mathrm{NH}}_{\mathrm{3}}+{\mathrm{NH}}_{\mathrm{4}}^{+}$ can be used for the combustion, high-temperature pyrolysis, and low-temperature pyrolysis factors, respectively.