TES is a nadir-viewing Fourier-transform infrared (FTIR) spectrometer aboard the NASA Aura spacecraft with a high spectral resolution of 0.06 cm⁻¹ and a nadir footprint of 5.3 km × 8.3 km. Here we use Level 1B spectra (V003) for the 1B2 (950–1150 cm⁻¹) and 2A1 (1100–1325 cm⁻¹) bands of the TES instrument [41].

TES retrievals of trace gas profiles are based on an optimal estimation approach (with a priori constraints) that minimizes the differences between the TES Level 1B spectra and a radiative transfer calculation that uses absorption coefficients calculated with the line-by-line radiative transfer model LBLRTM [20,42-45]. Current Level 2 products from TES (V004) include retrieved profiles of CO, O₃, H₂O, HDO, and CH₄. The NH₃ retrieval discussed below will be included in the upcoming V005 of TES products, while the HCOOH retrieval discussed below is a prototype retrieval being developed at AER. The averaging kernel matrix of the retrieval gives the vertical sensitivity of the retrieved profile to the true profile, while the trace of the averaging kernel gives the degrees of freedom for signal (DOFS), which represents the number of independent pieces of information contained in the retrieval [44]. Cloud properties are retrieved by assuming single layer clouds with an effective optical depth that accounts for both cloud absorption and scattering [46]. Due to their small size (count median diameters of ∼0.13 μm [47]), biomass burning aerosols are unlikely to significantly impact radiances in the thermal infrared regions detected by TES, and any impact from larger particles is accounted for by the retrieved effective cloud optical depth.

The TES special observations during ARCTAS-B included nadir observations over eastern Siberia, the North Pacific, and North America for every 0.4° latitude. Biomass burning plumes were identified following the procedure in Alvarado et al. [12], which we briefly outline here. We used maps of Level 3 daily AIRS retrievals of CO at 1° × 1° resolution to identify the transport of CO from major regions of boreal biomass burning. We then used TES Level 2 retrievals of CO (V003) to identify the corresponding plumes that were observed by TES. The CO retrievals for TES special observations between 15 June and 15 July 2008 were filtered for data quality as recommended in the TES Level 2 Data Users Guide [48]. In general, the retrievals had 1 DOFS below 250 hPa with the region of maximum sensitivity in the troposphere near 500 hPa. We defined a plume in the TES special observations as an area where the retrieved CO mixing ratio at 510 hPa exceeded 150 ppb. This threshold ensured that the CO retrievals were significantly different from the a priori values (∼110 ppb). While this procedure will detect thick plumes that are transported between continents [49], it does not detect plumes near the surface (where the sensitivity is low) or very thin or dilute plumes. We then used HYSPLIT back-trajectories [50] to determine if the observed air masses came from boreal biomass burning regions in Siberia (17 plumes) and Canada (5 plumes). The CO and O₃ retrievals for these plumes were analyzed by Alvarado et al. [12]. In this paper, we restrict our analysis to the fresh plumes from Canadian biomass burning.

NH₃ retrievals were performed using TES Level 1B spectra (V003) [41] following the method of Shephard et al. [20], which has been implemented in V005 of the TES Level 2 products. The a priori profiles and covariance matrices for TES NH₃ retrievals are derived from GEOS-Chem model simulations of the 2005 global distribution of NH₃. Figure 1a shows an observed TES brightness temperature spectrum in the region of strong NH₃ absorption for a scan of a fresh Canadian smoke plume. Figures 1b,c show the brightness temperature residuals (observed spectrum minus modeled spectrum) for the unpolluted background NH₃ profile and the retrieved NH₃ profile, respectively, while Figure 1d shows the modeled spectrum of NH₃, calculated as the difference between the modeled spectrum including the retrieved NH₃ profile and the modeled spectrum without any NH₃. We can see a strong residual (−0.8 K) in the background profile spectrum that is substantially reduced following retrieval of NH₃. (The second strong residual feature near 949 cm⁻¹ in panels b and c appears to be due to the ν₇ Q-branch of ethylene (C₂H₄), as is discussed in Section 3.4 below.)

Figure 2a shows the averaging kernel for the same NH₃ retrieval. The retrievals are performed using 14 vertical levels. However, the number of DOFS is generally not greater than one, meaning that it is not possible to obtain information about the shape of the profile from these retrievals. In order to minimize the influence of the a priori constraints on the end result, it is desirable to report one retrieved quantity per DOFS. However, performing a “profile” retrieval offers a major advantage in that us allows the calculation of diagnostics, such as the averaging kernels, that enable characterization of the spatial and temporal variability of the vertical sensitivity of the measurements. Therefore, for this work, the retrievals are performed on 14 levels in order to take advantage of the sensitivity characterization that this enables, then post-processed to calculate a single quantity that better represents the information that is really available in the measurement and that is relatively insensitive to the a priori constraints. Shephard et al. developed a Representative Volume Mixing Ratio (RVMR) metric for NH₃ [20] based on similar techniques used previously for CH₄ [51] and CH₃OH [19]. This RVMR represents a TES sensitivity weighted average value where the influence of the a priori profile is reduced as much as possible. (Note that for NH₃, one of three a priori profiles is selected based on the signal-to-noise ratio in the TES NH₃ band [20]. For the scan shown in Figures 1 and 2, the polluted a priori, shown as a dashed blue line in Figure 2b, was selected.) Figure 2b shows a retrieved profile of NH₃ for a fresh Canadian smoke plume as a solid black line, while the red circle is the RVMR calculated from the profile. The horizontal error bars show the uncertainty in the RVMR due to noise in the TES spectrum, while the vertical bars encompass the region of TES sensitivity to NH₃, defined as the full width at half maximum (FWHM) of the averaging kernel at the pressure level of the RVMR [20]. With the a priori assumption about profile shape that was chosen here the NH₃ RVMR is roughly 20% to 60% of the retrieved surface value for NH₃. The estimated minimum detection level is an RVMR of approximately 0.3 ppb, corresponding to a profile with a surface mixing ratio of about 1–2 ppb [20]. However, it must be noted that this is merely a general estimate of the minimum detection level, which depends strongly on the location of peak NH₃ mixing ratio and the thermal structure of the atmosphere for a given scan [20], and thus retrieved NH₃ RVMRs of less than 0.3 ppb are considered valid and are included in our analysis.

The excess mixing ratio of a trace gas like NH₃ (EMR, ΔNH₃) is defined as the mixing ratio of the gas in the smoke plume minus its mixing ratio in the background. This excess mixing ratio can be normalized using the excess mixing ratio of CO to give the normalized excess mixing ratio (NEMR, ΔNH₃/ΔCO). The emission ratio (ER) is a special case of the NEMR where the measurements are made in fresh smoke near the fire source [3]. The NEMR of a trace gas can be highly variable for reactive gases downwind of fires due to the different rates of deposition and secondary photochemical production and loss for the trace gas and CO. In this paper, we will refer to our derived NEMRs for fresh Canadian smoke observed by TES as emission ratios; however, these emission ratios are inherently convolutions of the initial NEMR and any secondary production and loss processes that have taken place within the smoke plume during plume lofting and transport from the fire source [28]. Furthermore, the lower sensitivity of the TES retrievals near the surface (see Figure 2a) means that TES will preferentially sample smoke from the flaming stages of combustion, as these emissions are more likely to be lofted well above the surface. This lower sensitivity near the surface is due to the physics of nadir thermal infrared sounding: when the surface and the layers of the atmosphere near the surface have similar temperatures (low thermal contrast), we cannot distinguish between radiation emitted by the surface and radiation emitted by the lowest layers of the atmosphere. Similar caution must be used in interpreting emission ratios measured from other platforms: for example, NEMRs of NH₃ measured at the ground are generally much higher than those measured by aircraft, as the aircraft does not sample the emissions from residual smoldering combustion very close to the ground (see Section 3.1 below).

In order to use the TES retrievals of NH₃ and CO to calculate the emission ratio of NH₃, we first calculated the RVMR for NH₃ following the procedure of Shephard et al. [20]. Since the DOFS for the CO retrieval are generally higher than for NH₃, in order to obtain a comparable metric we transform the TES CO retrieval using the same grid and weightings as were used to generate the NH₃ RVMR to obtain a pseudo-RVMR for CO. Figure 2c shows the retrieved CO profile for a fresh Canadian smoke plume as a solid black line while the pseudo-RVMR for CO is shown in green. The emission ratio of NH₃ was then calculated as the slope of a least squares linear regression of the NH₃ RVMR and the CO pseudo-RVMR.

We have developed a prototype retrieval for formic acid (HCOOH) from TES Level 1B spectra. The retrieval approach is similar to that used for NH₃. The spectroscopic parameters for HCOOH were taken from the HITRAN 2008 database, which substantially improved the estimates of the strengths of HCOOH lines by removing interference from the formic acid dimer [52]. All other spectroscopic parameters were taken from v1.4 of the TES spectroscopic line parameters [53]. The a priori constraint and covariance matrix were compiled from GEOS-Chem model simulations of the 2004 global background mixing ratios of HCOOH; however, these background profiles may be biased low, as GEOS-Chem tends to underestimate HCOOH in the northern mid-latitudes [28]. For HCOOH, the initial guess profile is the same as the a priori constraint vector. LBLRTM was run using TES Level 2 (V004) retrievals of temperature, average cloud optical depth, emissivity, reflectivity, H₂O, CO, O₃, and CH₄. Profiles of CO₂ and N₂O were taken from the TES Level 2 supplemental data files. The top panel of Figure 3 shows the observed TES brightness temperature spectrum between 1090–1130 cm⁻¹ for a scan of fresh smoke from a Canadian fire with the HCOOH retrieval microwindows shown in red. Note that only some sections of the HCOOH band have been retained in order to remove interfering lines from other species, principally water vapor. The second panel shows the residuals (observed spectrum minus modeled spectrum) for the background HCOOH profile. We see a strong (−2 K) residual in the HCOOH Q-branch (∼1105 cm⁻¹), which is removed after retrieval of HCOOH as seen in the third panel of Figure 3.

An RVMR for HCOOH and a corresponding CO pseudo-RVMR were calculated following the procedure of Shephard et al. for NH₃ [20], as illustrated in Figure 4. The RVMR for HCOOH is lower than the retrieved value at the level of maximum sensitivity to HCOOH, poissibly because the TES sensitivity to HCOOH covers a wide pressure range. The emission ratio of HCOOH was calculated as the slope of the least squares linear regression of the HCOOH RVMR and the CO pseudo-RVMR.

In order to use TES to detect PAN in boreal smoke plumes, we calculated the differences (residuals) between the TES Level 1B spectra (V003) and a forward run of LBLRTM using v1.4 of the TES spectroscopic line parameters for the region of strong absorption by PAN (1140–1180 cm⁻¹). As for the retrievals of HCOOH, the model was run using TES Level 2 (V004) retrievals of temperature, emissivity, reflectivity, average cloud optical depth, H₂O, CO, O₃, and CH₄ and profiles of CO₂ and N₂O were taken from the TES Level 2 supplemental data files. Preliminary model runs using the spectrally resolved effective cloud optical depth retrieved by TES led to unphysical slopes in the residuals versus wavenumber between 1100 and 1200 cm⁻¹. We removed these slopes by setting cloud optical depth in this spectral region to the average effective cloud optical depth included in the TES Level 2 products; however, the spectral signature of PAN was detectable regardless of which TES cloud product was used. The absorption cross section of PAN was taken from HITRAN 2008 [52].

A similar procedure was used to detect C₂H₄ in TES spectra, with the addition that the surface emissivity, surface temperature, and NH₃ profile were taken from the final results of the NH₃ retrieval described in Section 2.2 above. The line parameters for C₂H₄ were taken from v1.4 of the TES spectroscopic line parameters.

Figure 5 shows a map of the retrieved NH₃ RVMR values within smoke plumes from Canadian biomass burning in the summer of 2008. We have filtered the retrievals for quality following Shephard et al. [20]. First, NH₃ retrievals are rejected if the DOFS are less than 0.1 or if the average cloud optical depth is greater than 1. Second, retrievals with DOFS less than 0.5 are rejected if the thermal contrast between the surface and the atmospheric layer nearest the surface is between −7 K and 10 K.

Figure 6 shows the NH₃ RVMR versus the CO pseudo-RVMR for all of the retrievals shown on the map in Figure 5. The most elevated amounts of CO and NH₃ come from a fresh smoke plume from a single Canadian fire observed by TES between 52.5–54.1°N and 90.5–91.8°W at 19:03 UTC on 1 July 2008 (TES Run #7656). Due to the high retrieved values for CO (>500 ppb), this plume likely contains concentrated fresh smoke. Also included are retrievals of NH₃ within a more dilute smoke plume observed between 55.5–59.8°N and 88.8–91.4°W at 18:51 UTC on 17 June 2008 (TES Run #7472).

We calculate the NH₃ emission ratio (ΔNH₃/ΔCO) for these fires as 1.0% ± 0.5%. However, there is a large amount of scatter in this plot, which leads to a low correlation between CO and NH₃ (r² = 0.30). The scatter is likely due to the combination of: (1) variations in plume heights, which change the relative sensitivity of TES to CO and NH₃ in the smoke plume; (2) variations in the age of the smoke plumes, as older plumes could have lost more NH₃ to chemical reaction or deposition, and (3) variations in the initial emissions of ammonia from the fires, which are in turn related to differences in the relative fraction of flaming versus smoldering combustion as well as variations in fuel nitrogen content.

Table 1 compares our derived emission ratio for NH₃ to previous satellite, aircraft, and ground measurements of ammonia emission ratios for boreal biomass burning. Our value is close to the value of 1.3% derived from IASI retrievals in Siberian smoke over East Mongolia [9]. The results from the aircraft and ground studies are substantially different with ground studies showing values of ΔNH₃/ΔCO that are a factor of 5 higher than the aircraft or satellite studies. The higher ground values are likely due to emissions of NH₃ from residual smoldering combustion very close to the ground [3] where aircraft cannot sample and satellites have little sensitivity to NH₃ (see Figure 2). Thus, both aircraft and satellites are likely to underestimate the true emissions of NH₃ from biomass burning fires unless the emission factors from aircraft and satellite studies are carefully combined with ground observations, such as in the review of Akagi et al. [3]. In addition, given the discrepancy between aircraft and ground observations, nadir-viewing satellite observations of NH₃ are most appropriately compared with aircraft observations, as satellites will not be very sensitive to the ground-level residual smoldering emissions detected in ground studies. Our NH₃ emission factor of 1.0% ± 0.5% is similar to the values of 1.3% ± 0.7% and 1.7% ± 0.9% reported for Alaskan forest fires in the aircraft studies of Nance et al. [54] and Goode et al. [55], respectively, but is substantially higher than the value of 0.16% ± 0.14% reported by Radke et al. for boreal forests [56]. Furthermore, our value lies within the uncertainty range of the average value of 1.0% ± 0.7% (boreal forests, aircraft studies only) reported in the review of Akagi et al. [3], as well as the value of 2.6% ± 2.2% (all extratropical forests, all study types) in the review of Andreae and Merlet [57].

Figure 7 shows a map of the retrieved HCOOH RVMR values within smoke plumes from boreal biomass burning over Canada. We have removed retrievals with less than 0.5 DOFS for HCOOH, which also eliminated retrievals with an HCOOH RVMR of less than 0.5 ppbv. The smoke plumes included in the analysis are the same as described above for NH₃. Figure 8 shows the HCOOH RVMR versus the CO pseudo-RVMR for all of the retrievals shown on the map in Figure 7. We calculate the emission ratio ΔHCOOH/ΔCO for these fires as 0.31% ± 0.21%. As we saw for NH₃, there is a large amount of scatter in this plot that leads to a low correlation between CO and HCOOH (r² = 0.41). As discussed above, this scatter is likely due to the combination of variations in (1) plume heights, (2) plume ages, and (3) initial smoke emissions of HCOOH. The large pressure range of TES sensitivity to HCOOH may also contribute to the low correlation of HCOOH with CO. It is worth noting that the ACE-FTS observations of HCOOH within boreal biomass burning plumes showed a much higher r² value (0.86), likely due to the fact that the higher vertical resolution of limb retrievals means that ACE is better able to separate the different vertical layers of the plume which are averaged together by nadir-viewing sounders like TES and IASI [31].

Table 2 compares our derived emission ratio for HCOOH to previous satellite, aircraft, and ground measurements of formic acid emission ratios for boreal biomass burning. Our HCOOH emission ratio is slightly lower than the ratio of 0.38% ± 0.06% derived from ACE-FTS observations [31], possibly due to the fact that a limb-sounder like ACE is more sensitive to plumes that reach high altitudes; this type of plume is more common over more flaming fires since fire power output peaks during flaming combustion [4]. Table 2 shows that our value is similar to the emission ratios for HCOOH derived from aircraft and ground studies, with the exception of the value of 3.7% ± 2.0% reported for Canadian forests by Lefer et al. [60], which is about a factor of 10 higher than the average value for boreal biomass burning in the review of Akagi et al. [3]. However, Lefer et al. note that the elevated enhancement ratios of HCOOH they observed (0.82% to 6.2%) are likely due to HCOOH production within the plumes in the first couple hours after emission, and do not necessarily reflect the initial emissions. When the Lefer et al. study is removed, the Andreae and Merlet average becomes 0.85% ± 0.53%, with the relatively higher value of HCOOH likely due to the inclusion of midlatitude fires as well as boreal fires in the average [57].

The solid black line of Figure 9a shows the brightness temperature residuals (data minus model) for the scan of fresh smoke from a Canadian fire at 53.32°N and 90.84°W.

The two lobes of PAN absorption are clearly visible on either side of the water line at ∼1165 cm⁻¹. The dotted red line shows the residuals when a hypothetical PAN profile with a peak concentration of 1.9 ppb at 560 hPa is added to the forward model, and Figure 9b shows the modeled PAN spectrum as a solid black line. Adding the PAN profile substantially reduces the mean residuals, but this peak concentration of PAN in the assumed profile is about a factor of 2 higher than the concentration of PAN observed 30–60 min downwind of the Lake McKay fire in Saskatchewan, Canada during ARCTAS-B [12]. We have also detected PAN in a scan of aged smoke from Siberian biomass burning observed near Kamchatka (not shown). These two scans shared several common features, including relatively high mixing ratios of CO (mixing ratio > 250 ppbv at 510 hPa) and low average cloud optical depth (<0.1). This low cloud optical depth may be a necessary but insufficient criterion for detecting PAN in biomass burning smoke plumes with TES.

As mentioned in Section 2.2 above, there is a second strong residual feature in Figure 1c near 949 cm⁻¹ that appears to be due to absorption by C₂H₄.

Figure 10b shows this residual feature, which partially overlaps with the strong CO₂ line [39] that is visible in the TES spectrum shown in Figure 10a. Figure 10c shows that the addition of a hypothetical C₂H₄ profile to the model (with a peak concentration of 1.9 ppb at the surface) removes this feature. This peak C₂H₄ mixing ratio is approximately the 85th percentile of C₂H₄ observations observed downwind of the Lake McKay fire in Saskatchewan, Canada during ARCTAS-B [12]. While the scan in Figure 10 did have a low cloud optical depth, this criterion is likely to be less important for detecting C₂H₄ within fresh biomass burning plumes using TES, as the residual feature of C₂H₄ is both stronger and sharper than the PAN feature. However, the short lifetime of C₂H₄ may make it difficult to detect in smoke that is several hours old.