In-situ measurement of pyrolysis and combustion gases from biomass burning using swept wavelength external cavity quantum cascade lasers

Broadband high-speed absorption spectroscopy using swept-wavelength external cavity quantum cascade lasers (ECQCLs) is applied to measure multiple pyrolysis and combustion gases in biomass burning experiments. Two broadly-tunable swept-ECQCL systems were used, with the first tuned over a range of 2089-2262 cm-1 (4.42-4.79 µm) to measure spectra of CO2, H2O, and CO. The second was tuned over a range of 920-1150 cm-1 (8.70-10.9 µm) to measure spectra of ammonia (NH3), ethene (C2H4), and methanol (MeOH). Absorption spectra were measured continuously at a 100 Hz rate throughout the burn process, including inhomogeneous flame regions, and analyzed to determine time-resolved gas concentrations and temperature. The results provide in-situ, dynamic information regarding gas-phase species as they are generated, close to the biomass fuel source.

(PTR-MS) [12], gas chromatography mass-spectrometry (GC-MS) [11], and gas-phase Fouriertransform infrared spectroscopy (FTIR) [7][8][9][10]. Extractive techniques can be extremely sensitive, but even when sampling gases close to the fuel there is a possibility that species may react or adsorb to surfaces before reaching the measurement region, thereby changing the composition or concentrations of gases in the sample. In addition, extractive techniques are fundamentally limited in measurement speed and cannot capture rapid dynamic processes occurring during the burn.
Open-path FTIR (OP-FTIR) has been used extensively to identify and quantify a wide range of trace gases emitted from biomass burning [5,7,8,15]. Use of open-air multi-pass cells to provide measurement path lengths of tens of meters enables detection of gases at ppb-level concentrations at ∼1 Hz rates. Gas-phase molecular species are identifed based on their characteristic absorption spectra arising from ro-vibrational transitions in the infrared spectral regions. Measurement rates of OP-FTIR are usually limited by trade-o˙s in spectral resolution. For biomass burning studies, high spectral resolution ( ∼ 0.5 cm −1 ) is critical for distinguishing broad spectral features of larger NMOCs at low concentrations from the resolved line absorption of more prominent small-molecule species such as CO, CO 2 , H 2 O, etc. [9] but use of these spectral resolutions limits FTIR measurement speeds to ∼1 Hz. OP-FTIR experiments typically measure gases at distances > 1 m away from the fuel source, after the emitted gases are entrained in the ambient air, with corresponding dilution and cooling. Transmission/absorption of fame regions using FTIR is usually not possible due to high optical density of the sample region, strong emission of infrared radiation from the fame, and low spectral radiance of the incoherent sources used for transmission. FTIR emission spectroscopy has been used to measure CO, H 2 O, and temperature in biomass burning fames [16], yet it is diÿcult to obtain quantitative results and measure trace species with this method.
Laser absorption spectroscopy is a proven technique for dynamic measurements in extreme environments such as fames. The high spectral radiance of laser sources enables transmission through optically dense regions and the laser signal of interest can be separated from incoherent emission using spectroscopic or modulation techniques. Tunable diode laser absorption spectroscopy (TDLAS) at visible or near-infrared wavelengths has seen widespread use for measurement of combustion products (H 2 O, CO, CO 2 , etc.) and temperatures in fames [17,18], but is limited to detection of small molecules with resolved ro-vibrational lines due to limitations in laser tuning range. In biomass burning applications, TDLAS has been used to measure molecular species from combustion of wood fuel pellets [4,19] and biomass gasifcation [20].
Quantum cascade lasers (QCLs) have dramatically increased the potential for infrared laser absorption spectroscopy of molecular compounds, allowing access to strong fundamental rovibrational molecular transitions [21,22]. While QCL-based spectroscopy is becoming established for combustion research applications [23][24][25][26], it has seen less use in biomass burning applications. Narrowly-tunable distributed feedback (DFB) QCLs were used for measurement of NH 3 and C 2 H 4 in biomass steam gasifcation [27] and were evaluated for suitability in photo-acoustic detection of species in biomass burning [28].
In this paper, we apply broadband swept-ECQCL absorption spectroscopy to measure dynamics of combustion and pyrolysis gases emitted from biomass burning. Laboratory-scale wind tunnel experiments were conducted in which a bed of plant material was ignited at one end, and beams from swept-ECQCL systems were propagated perpendicular to the fame front direction, directly above the plant bed. Infrared transmission/absorption spectra were measured continuously at 100 Hz as the fame front propagated along the bed, providing dynamic measurement of gases and temperature adjacent to the fuel source. Two swept-ECQCL systems were used, with one measuring combustion gases CO 2 , H 2 O, and CO, and the other measuring trace species ammonia (NH 3 ), ethene (C 2 H 4 ), and methanol (MeOH). Fitting of measured absorption spectra was used to determine time-resolved concentrations of species and temperature. Temperatures >1000 K were measured as the fame front propagated through the ECQCL beam paths and compared with time-resolved infrared emission intensity. E˙ects of spatial inhomogeneity along the measurement path are observed and found to be important for interpretation of the results. The time-resolved dynamics of multiple species and temperature shows correlation of gases with faming combustion versus smoldering/pyrolysis processes. The results provide in-situ, dynamic information regarding biomass combustion and pyrolysis products as they are generated, in regions close to the fuel source. Figure 1 shows schematic diagrams and photographs of the experimental setup and measurement confguration with the swept-ECQCL systems. The wind tunnel contained the biomass material under study and a fan with fow conditioner at one end provided constant wind velocity along the fuel bed and between the walls of the wind tunnel [38]. The 2 m long x 1 m wide fuel bed consisted of 1 kg dead needles from longleaf pine (Pinus palustris Mill.) interspersed with a grid of living plants (Ilex glabra (L.)). The Ilex plants were grouped into 37 ramekins with 2 plants/ramekin, giving a total of 74 plants. Individual plant heights ranged from 12 to 28 cm and crown areas ranged from 108 to 2477 cm 2 . Pine needle moisture content (dry weight basis) was 8% and plant foliar moisture content was 92%. This range of pine needle moisture content is close to the preferred range used by practitioners in the southern United States [39]. The fre was initiated by applying denatured alcohol (95% ethanol, 5% methanol + isopropanol) to the pine needles at the end of the wind tunnel nearest the fan bank and igniting with a fame. The uniformly distributed pine needles were burned as spreading line fres with wind velocity of ∼1 m/s in a cool/moist atmosphere. The fame front linear spread rate was 0.014 m/s and fame lengths ranged from 0.3 to 0.6 m which are typical for backing fres in southern pine fuel beds [39]. The fame front propagated the length of the wind tunnel in ∼2.5 minutes. Air temperature in the wind tunnel was 12.8 °C and relative humidity was 33%.

Experimental
Two swept-ECQCL systems were used for the measurements. Both were custom-built systems similar to ones described previously [30][31][32][33][34][35][36][37]. We denote the systems by the wavelength band of operation. The frst system operated in the mid-wave infrared (MWIR) and was swept over a tuning range of 2089-2262 cm −1 (4.42-4.79 µm) using a sinusoidal modulation with 100 Hz frequency. The second system operated in the long-wave infrared (LWIR) and was swept over a tuning range of 920-1150 cm −1 (8.70-10.9 µm) using a sinusoidal modulation with 250 Hz frequency.
The beam from each ECQCL was directed through access holes cut into the wind tunnel sides, propagated across the wind tunnel to a mirror, and refected along a path at a slight angle in the horizontal plane. After propagation back through the wind tunnel, a fraction of each return beam was refected from a BaF 2 wedged window and focused onto an infrared photodetector (DET) using a ZnSe lens with 25 mm focal length. The beams were located at a height 36 cm above the base of the plant bed. The MWIR beam was located closer to the start of the plant bed and experienced the fame front frst. The LWIR beam was located 30 cm away from the Beams from the two ECQCL systems were directed across the wind tunnel at a height 36 cm above the plant base to mirrors (M). The mirrors directed the ECQCL beams back to BaF 2 wedged windows (WW) which refected a portion of the beam to a lens (L) and infrared detector (DET). The ECQCL beam paths were separated horizontally by 30 cm and the fame front reached the MWIR beam path before reaching the LWIR beam.
MWIR beam and experienced the fame front second. The total path length accounting for the double-pass confguration inside the wind tunnel for each ECQCL system was 2.64 m, with a 2 m path length directly above the plant bed. The regions of the beam outside the wind tunnel were not included in the measurement path length, as it was assumed the conditions in these regions did not change during the burn.
Both ECQCL systems were operated using a full-depth 500 kHz square wave current modulation to produce a corresponding amplitude modulation (AM) of the output intensity. Use of modulated current in these ECQCLs provides two benefts. It has been previously shown that the modulated current reduces spectral scan noise due to mode-hops [31]. Second, the AM signal on the detector allows separation of the ECQCL light from any background infrared light on the detector using lock-in detection. For operation in ambient conditions, the demodulation eliminates drifts in the detector signal due to undesired background fuctuations. In the measurements performed here, the detector o˙set was used to measure the emission from the high-temperature fame regions. Similar techniques for simultaneous measurement of laser transmission and optical emission have been used with diode lasers for measuring soot in fames [40] and more recently with a swept-ECQCL for characterization of explosive freballs [37].
The signals from the photo-detectors were digitized at a 2 MHz rate. The amplitude of the modulated signals over each 500 kHz cycle provided a signal proportional to the detected ECQCL intensity. The DC o˙set of the modulated signal provided a signal proportional to the broadband infrared emission within the detector feld-of-view. Thus, the measurements provided both the ECQCL transmission and the spectrally-integrated infrared emission at a sampling rate of 500 kHz. The detector used with the MWIR ECQCL (VIGO PVI-4TE-6) has a spectral responsivity from 3-7 µm but atmospheric attenuation limits the range to 3-5 µm. The detector used with the LWIR ECQCL (VIGO PVI-4TE-10.6) has a spectral responsivity from 3-12 µm which spans both the MWIR and LWIR spectral regions.
Each ECQCL system was swept repeatedly over its tuning range using a sinusoidal modulation and measurements were performed continuously throughout the entire burn process. After detection and demodulation, the ECQCL wavelength scans were subdivided into blocks separated by half the scan period (the same wavelength region is covered twice during each sinusoidal scan period), yielding scans at 200 Hz (MWIR ECQCL) and 500 Hz (LWIR ECQCL). Relative wavelength calibration was performed using data acquired before the burn by measuring transmission through a solid silicon etalon with 0.416 mm length, inserted temporarily into each beam path. The absolute wavelength calibration was determined by comparing a measured gas spectrum to a known reference absorption spectrum. For the MWIR ECQCL, a measured spectrum of CO + CO 2 acquired during the burn was used, and for the LWIR ECQCL a measured spectrum of Freon-152a (1,1-difuoroethane) from an air duster sprayed into the beam path was used.
After wavelength calibration, the ECQCL scans were further averaged to a common measurement rate of 100 Hz, corresponding to a 10 ms interval between scans. This averaging time was selected based on a balance between tracking the fuctuations in spectra during the turbulent fame region and reducing the number of spectra for ftting. The ECQCL intensity for each scan was converted to absorbance units by normalizing to a background portion of the data recorded before ignition of the fre: A i (ν) = −ln[I i (ν)/I 0 (ν)]. Here, A i (ν) is the base-e absorbance of the i-th scan, I i (ν) is the measured ECQCL intensity for the i-th scan, and I 0 (ν) is the intensity of the background scan. The background I 0 (ν) was taken as the average of 100 scans (1s total duration) from the beginning of the measurement run and is shown in Fig. 2(a) for the MWIR ECQCL and in Fig. 2(b) for the LWIR ECQCL. The absorbance spectra calculated in this manner show the changes in absorbance relative to the ambient conditions during the background. Thus, the species concentrations determined from the absorbance spectra refect the di˙erence relative to the starting ambient conditions, directly providing the excess mixing ratio (EMR) for each analyzed species [2].

Spectral analysis
After averaging of scans, the data sets consisted of 36,000 absorbance spectra for the MWIR ECQCL and 38,000 absorbance spectra for the LWIR ECQCL, corresponding to a measurement time of ∼6 minutes with a time interval of 10 ms between spectra. Large temporal variations in temperature were present during the measurement period, especially as the fame front propagated through the measurement paths, and the e˙ects of temperature on the absorbance spectra must be accounted for in the spectral ftting to obtain meaningful results. More critically, the measured absorbance spectra are path-integrated and may contain contributions from spatial regions with highly variable temperature and species concentrations along the measurement path, especially when localized fame fronts intersect the beam path. Inverse modeling of measured spectra to determine spatial variations of parameters along the measurement path presents a considerable challenge and often requires additional measurement data (e.g. high-speed imaging or point measurements) along with radiative transport considerations [23], and is beyond the scope of the e˙orts presented here. Instead, the spectral ftting model used here assumes uniform conditions along the measurement path. Despite the limitations inherent to ftting path-integrated spectra in this manner, the spectra may still be analyzed to identify species, estimate chemical concentrations, estimate temperatures, and especially to study the temporal dynamics of these parameters. For the results presented in this manuscript, it is to be understood that the reported concentrations and temperatures do not imply uniform conditions along the measurement path. Furthermore, results from analysis of the path-integrated spectra will have a high uncertainty in absolute concentrations and temperatures for times at which large spatial gradients exist.
The absorbance spectra from each ECQCL system were analyzed in two stages. The frst stage used a weighted nonlinear least squares (WNLS) ft on a spectral subset to estimate the temperature based on strong variations with temperature for the relative absorption line areas of combustion gases (CO 2 , CO, H 2 O). In contrast to traditional TDLAS analysis where it is common to use ratios between two lines to determine temperature [25], the ECQCL spectra were analyzed using broadband fts to multiple lines (10-100's). In this case, variations in peak areas and band profles provide an estimate of temperature averaged over many transitions [37]. The second stage used a linear weighted least squares (WLS) ft to determine the concentrations of various species, using the estimated temperature as a parameter. Breaking the analysis into two stages was found to give better spectral fts (lower residuals) than attempting a single WNLS ft to the entire spectrum and signifcantly decreased the computation time for analysis. In addition, the WLS fts provided better convergence in regions with low concentrations where the WNLS ft parameters were indeterminate. In both cases, the fts were weighted by I 0 (ν) 2 to account for variations in detected intensity [41] and points with absorbance >5 were excluded from the fts.
For the MWIR region, the temperature was estimated by performing a WNLS ft on the absorbance spectra from 2150-2262 cm −1 , which contains CO and CO 2 lines/bands sensitive to temperature. A detailed description of the WNLS procedure in the MWIR was provided in [37]. A 2 nd order polynomial ft to the baseline was subtracted before ftting. Line parameters for CO and CO 2 were obtained from HITEMP [42] and standard isotopic abundances listed in HITRAN [43] were used. To reduce computation time, lines with peak areas <0.001× the maximum peak area in the spectral region of interest were excluded from the simulation, with the peak areas calculated at 1000 K. Absorbance peaks were modeled using Lorentzian lineshapes to approximate the instrumental linewidth of the ECQCL measurement. All peaks of each species were modeled using the same linewidth and all species were assumed to be present at the same temperature. For spectra with low CO/CO 2 concentrations the temperature could not be determined with confdence. Therefore, a threshold was set whereby only temperatures determined from spectra with CO 2 column density > 2 σ CO2 were retained, where σ CO2 is the standard deviation of CO 2 concentrations before ignition of the fame (refecting the measurement noise). The temperature was set to a default value of 298 K for spectra failing to meet this condition. After thresholding, the temperatures were smoothed using an adjacent averaging flter with 1 s width to use in the subsequent WLS ftting routine.
The WNLS ftting to estimate temperature for the LWIR data was similar to the MWIR data, but used the spectral region 1090-1150 cm −1 , which contains multiple H 2 O lines sensitive to temperature. A 7 th order polynomial ft to the baseline was subtracted before ftting. The H 2 O spectrum was modeled using line parameters from HITRAN [43] with standard isotopic abundances and all lines were kept (no signifcant di˙erence was found versus using HITEMP parameters). The threshold for determining temperature was computed similarly to the MWIR data but using the H 2 O concentration. The low absorption cross-section of the H 2 O lines in the LWIR spectral region resulted in more spectra falling below the threshold, and the uncertainty in temperature is thus higher for the LWIR data than for the MWIR data. Figure 3 shows three examples of WNLS fts to H 2 O in the LWIR spectra, from which the strong temperature dependence of the H 2 O lines is apparent. After the temperatures were estimated from the WNLS fts, a subsequent WLS ftting was performed on the absorbance spectra over the full spectral range. For the MWIR spectra, the ftting library consisted of CO, CO 2 , and H 2 O. For each experimental absorbance spectrum, a library absorption spectrum for each species was modeled using HITEMP parameters at the temperature determined previously from the WNLS ft. For the WLS fts, the modeled spectra used Lorentzian peak shapes with 0.2 cm −1 full width at half-maximum (FWHM), determined from the average best-ft widths found during the WNLS ft routine; this represents the e˙ective instrumental linewidth of the absorbance spectra measured with the MWIR ECQCL.
For the LWIR spectra, the ftting library consisted of CO 2 , H 2 O, NH 3 , C 2 H 4 , MeOH, and EtOH. Library absorption spectra for CO 2 , H 2 O, and NH 3 were modeled using HITRAN parameters at the temperature estimated previously from the WNLS ft. The modeled spectra used Lorentzian peak shapes with 0.25 cm −1 FWHM, the e˙ective instrumental linewidth of the absorbance spectra measured with the LWIR ECQCL. Absorbance spectra for C 2 H 4 , MeOH, and EtOH were taken from the Northwest Infrared (NWIR) spectral library [44]. For consistency throughout the full burn, NWIR library spectra at a temperature of 298 K were used and increased uncertainty in the absolute concentrations is thus expected for ftting spectra at higher temperatures. However, by examining the ft residuals it was found that the NWIR library spectra provided good fts to most measured experimental spectra, even for temperatures above 298 K. Attempts to model C 2 H 4 and MeOH spectra using HITRAN parameters resulted in higher ft residuals with signifcantly longer computation times due to the high number of spectral lines; therefore, the NWIR library spectra were used instead.
For the WLS analysis of the MWIR (LWIR) spectra, the spectral baseline was ft as follows. Principal component analysis (PCA) was performed on a set of 60(80) absorbance spectra acquired before the fre was ignited, and the frst 4(8) principal component vectors were included in the ftting library to account for systematic drifts in the absorbance baseline due to variations in the MWIR (LWIR) ECQCL scans, respectively [31]. In addition, a 7 th order polynomial was used to account for random fuctuations in absorbance occurring over time scales shorter than the scan time, which were most prevalent during high turbulence when measuring through the fame region.
The output from the WLS ftting routine is a set of column densities versus time for each analyzed species. To facilitate comparison with other biomass burning studies, we report the results as concentration mixing ratios (ppm v ) relative to air at 298 K by dividing the column densities by the 2.64 m measurement path. For times at which conditions are uniform along the measurement path, the reported concentrations represent average values. However, for times with variations in temperature or concentrations along the measurement path, the reported concentrations and temperatures should not be considered as average values and have a high uncertainty due to the ftting of path-integrated spectra. Nevertheless, the reported values do provide valuable information on relative changes in temperature and concentrations, as shown below. To indicate that reported concentrations and temperatures result from ftting of path-integrated spectra, we denote the reported parameters with a tilde -e.g. C ft or T ft . Additional details on the WNLS ftting routine can be found in Phillips et al. [37] and additional details on the use of PCA in the WLS ftting routine can be found in Phillips et al. [31]. Details on the methods used to reduce e˙ects of turbulence on broadband absorption spectra as measured with the swept-ECQCL, which include scanning at rates above the dominant turbulence time scales and inclusion of the PCA to ft the baseline, are reported in [36]. After the fame front propagates through the beam paths, the emission signals return to near-zero values. Figure 4(b) shows the temperatures estimated from the WNLS fts to the MWIR and LWIR absorption spectra. As expected, the temperature profles versus time show a strong correlation with the measured infrared emission. The temperatures show a more gradual rise and fall and a broader profle in time than the emission signals, likely due to heated gases propagating into the beam paths before/after observing the strong infrared emission from the fame front itself. The ft temperatures (averaged to 1 s) range from ∼300-1100 K, which are reasonable values observed for biomass burning [2], although local temperatures within fames will be higher. In addition, the temperature ranges estimated from the absorption spectra are similar for the MWIR and LWIR channels. Figure 4(c) shows the ft concentration of H 2 O, and Fig. 4(d) shows the ft concentration of CO 2 determined from the WLS ft for both MWIR and LWIR channels. The H 2 O and CO 2 ft concentrations for both channels are highest during the fame front region, but both are detected throughout the burn. The exact shapes of the temporal profles compared between MWIR and LWIR channels are di˙erent due to the di˙erent spatial positions of the beam and di˙erent local conditions of the burn; however, the ranges of CO 2 and H 2 O ft concentrations determined from the analysis are similar for both.

Results
The data can be divided into three general regions depending on timing relative to the fame front passing the ECQCL beams. The frst region consists of times before the fame front reaches the beam paths. During this time period, infrared emission signals are low, and the temperatures are near ambient levels. Gases emitted by the burning material propagates downwind and into the ECQCL beam paths. The second region consists of the time period when the fame front passes through the ECQCL beam path and strong emission signals are detected, along with gases present at elevated temperatures. The high thermal and spatial gradients from the fames cause large signal fuctuations and increased noise via beam steering and turbulence. There may also be gases emitted upwind from previously burned regions which are detected during this time period. The third region consists of times after the fame front has passed the ECQCL beam path, resulting in low infrared emission signals with temperatures slightly above ambient. Gases are emitted from previously burned regions, which continue smoldering after the fame passes. We present the results of each region in turn below. We frst consider the initial time period before the fame front reaches the ECQCL beam paths. Figure 5(a) shows the concentrations versus time for H 2 O, CO 2 , and CO determined from fts to the MWIR spectrum, during these initial stages of the burn. Before ignition at 10:40:00, the ft concentrations of the H 2 O, CO 2 , and CO are low with fuctuations due to measurement noise. A small rise in all gases is seen before ignition, likely due to increased human activity in the wind tunnel during this time. After ignition, the ft concentrations of H 2 O, CO 2 , and CO all increase above the noise levels, and exhibit correlated fuctuations between the species. The results show detection of the plume of gaseous combustion products generated from the burning plants at the start of the bed, which propagate downwind through the tunnel and intercept the MWIR beam path.
The modifed combustion eÿciency (MCE) is defned as: where [CO] and [CO 2 ] are the detected EMRs of the gases [45]. The MCE provides a measure of the combustion process from faming (high MCE) versus smoldering (low MCE). The lower panel in Fig. 5(a) shows the MCE during the times shortly after ignition. The initial burning of the plant material is dominated by faming combustion and the gases detected downwind refect this behavior with a high MCE > 0.98. The LWIR results are shown in Fig. 5(b) for the initial time period before the fame front reaches the beam path. The H 2 O and CO 2 signals do not show a strong increase, due to low absorption cross-sections in the LWIR (especially at temperatures near ambient). NH 3 and C 2 H 4 are not detected above the noise levels. A plume of MeOH and EtOH vapor is detected starting just before the fame is ignited, originating from the denatured alcohol used to start the fame. The strong temporal correlation between MeOH and EtOH serves as a benchmark of the ability of the swept-ECQCL to measure chemical mixtures and distinguish actual fuctuations from measurement noise. Figure 6(a) shows an absorption spectrum measured by the MWIR ECQCL before arrival of the fame front. Measured and best-ft spectra are overlaid to show their quantitative agreement, and di˙erences are shown by the residual spectrum in the lower panel. Sharp spikes in the ft residuals near strong absorption peaks result from small o˙sets in line position or linewidth between the measured and modeled spectra as are commonly observed when detecting narrow spectral lines with the swept-ECQCL [30,35,37]. The CO 2 absorption cross-sections within this spectral window are relatively weak, arising from the P branch of the minor isotopologue 13 CO 2 , and higher energy rotational lines from the P branch of 12 CO 2 . Figure 6(b) shows a LWIR absorption spectrum of the MeOH and EtOH vapor plume resulting from the denatured alcohol used to start the fame.  shows a steady decrease as the fame front approaches the MWIR beam path and decreases at a faster rate after the fame front passes. Large fuctuations in the ft parameters are observed as the beam passes through the highly dynamic and spatially inhomogeneous fame region, with corresponding large uncertainty in absolute concentrations and temperature. Increased turbulence is expected due to large refractive index gradients present in and near the fames. The high spectral acquisition rate of the swept-ECQCL helps to mitigate the e˙ects of turbulence on the measured absorption spectra and reduces random noise in the spectral ft parameters, as has been previously shown [34]. The remaining fuctuations in the ft parameters thus largely refect actual changing conditions in time and not solely noise from turbulence, as will be addressed in the Discussion section. Figure 7(b) shows the time dependence of ft concentrations for species detected as the fame front passes the LWIR ECQCL beam. Elevated levels of H 2 O and CO 2 are detected, with similar magnitudes to those detected by the MWIR ECQCL. An increase in NH 3 is detected, which peaks later in time than H 2 O and CO 2 . This observation suggests that the NH 3 is emitted more strongly from smoldering sections, including the previously burned material upwind of the beam path [7,10]. Elevated levels of C 2 H 4 and MeOH are also detected as the fame front propagates through the beam path. The ft temperatures estimated from the LWIR spectra are of similar magnitude to the MWIR spectra when averaged to 10 Hz, peaking at values near 1000 K. The short-term extremes in temperature at 100 Hz are higher for the MWIR than the LWIR measurements, which may refect the use of spectra from di˙erent species to estimate the temperature, but are still within expected ranges for biomass burning fames [2]. Figure 8 shows MWIR absorption spectra measured in the fame region. A wide range of physical conditions are spanned during the fame front region, and example spectra are plotted for two times at which the instantaneous ft temperature was low [ Fig. 8(a), T ft =656 K] and high [ Fig. 8(b), T ft =1170 K]. The most notable feature compared to Fig. 6(a) is the increase in strength and spectral density of the hot lines of 12 CO 2 due to the elevated temperatures. Additional higher energy lines from CO are also observed due to the increased temperatures. The ft residuals increase in the fame front region, partly due to increased measurement noise from turbulence but more notably from defciencies in the spectral ftting model. In particular, the ft residuals in Fig. 8(b) suggest additional CO present at a temperature lower than the best-ft temperature of 1170 K (which is determined primarily by the ft to the CO 2 spectrum). As addressed in the Discussion section, this observation is consistent with spatial inhomogeneity in temperature and species along the measurement path, which is not accounted for in the spectral ft. Figure 9(a) shows an example LWIR spectrum measured as the fame front enters the beam path, showing increased ft concentrations of H 2 O and CO 2 at elevated temperature. Figure 9(b) shows a spectrum obtained later in the fame region, detailing the presence of NH 3 lines and H 2 O lines, also at elevated temperature. For both spectra, the ft residuals are signifcantly higher than in the pre-fame region due to increased turbulence at the elevated temperatures. But it is also apparent that not all the spectral lines are ft accurately, partly due to limitations of the ftting model. In particular, Fig. 9(b) shows a number of spectral lines not accounted for in the ftting model which could be due to additional species or variations in line-strengths due to deviations from the average (best-ft) temperature. Nevertheless, it is clear visually from the agreement between line positions and band shapes that the identifed species (H 2 O, CO 2 , NH 3 ) are present in the spectra shown in Fig. 9.  with the MWIR ECQCL after the fame front has passed the beam. The H 2 O, CO 2 , and CO ft concentrations were observed to decrease initially after the fame front passed the beam, but then increased again from 10:43-10:44. The MCE is poorly determined when the CO 2 ft concentration is near zero, but after the rise in concentration the MCE remains nearly constant at ∼0.7 which indicates smoldering. Figure 10 Figure 11(a) shows an example MWIR absorption spectrum from the time period after the fame front has passed with the relative heights of the CO and CO 2 absorption features reversed from the pre-fame region, and also indicative of smoldering. Figure 11(b) shows a LWIR spectrum obtained during this post-fame smoldering region in which spectral features of NH 3 , C 2 H 4 , and MeOH are clearly identifed.
The results show detection of multiple species, all of which change in time during the burn process. The presence of each chemical species was confrmed by visual examination of measured absorption spectra and comparison with identifying features in the modeled spectra, and examples  were shown for each species. However, spectra were also identifed which exhibited poor spectral fts arising from a source other than random noise. As discussed in the next section, we attribute these poor fts to a limitation of the spectral ftting model, which does not account for spatial inhomogeneity along the measurement path. We also show that spatial and temporal variations must be considered when interpreting the measurement results.

Discussion
The swept-ECQCL measurements provide dynamic information on physical and chemical conditions over both fast and slow time scales at locations close to the fuel source. To provide detail on the fast temporal fuctuations, Fig. 12(a)-MWIR and Fig. 12(b)-LWIR show the highspeed temporal variations in ft concentrations, ft temperature, and emission intensity for 5 s time periods in the fame period, plotted at 10 ms temporal resolution. Figure 12 shows that the ft concentrations of CO 2 and H 2 O vary by orders of magnitude, and ft temperatures vary by 100's of K over sub-second time scales when measuring through the fames, and the 100 Hz spectral rate used for the ECQCL measurement and analysis tracks these variations without discontinuities. For both the MWIR and LWIR measurements, the strong temporal correlation between H 2 O, CO 2 , emission intensity, and ft temperature indicates the H 2 O and CO 2 are measured in the high-temperature fame conditions. The correlated behavior provides additional evidence that the fuctuations are due to actual source variations in space and time and not solely due to measurement noise or turbulence. The results in Fig. 7 and Fig. 9 appear to indicate the presence of NH 3 , C 2 H 4 , and MeOH at temperatures well above their fash points or autoignition temperatures. The temperature was determined from spectral fts primarily to H 2 O and CO 2 , which are produced from combustion in the faming regions and thus exist initially at high temperatures. While the spectral ft assumes a set of species at a uniform temperature, in reality each species is present over a distribution of temperatures along the path-integrated measurement. Thus, the high temperature determined from fts to the H 2 O and CO 2 spectra does not preclude the existence of localized cooler regions along the beam path, in which fammable gases may exist. Since the fame front is inhomogeneous and does not cross the beam path all at once, localized pockets of gases from faming, unburned, and smoldering material may all exist simultaneously along the measurement path. It is also possible that pyrolysis species are released near the burning fame region, or that they are emitted upwind of the measurement from smoldering regions and propagate downwind into the beam path.
The e˙ect of spatial inhomogeneity can also explain the ft residuals in Fig. 8(b), which strongly suggest the presence of CO at multiple temperatures. This observation is consistent with CO being emitted from cooler smoldering regions, while CO 2 is emitted from higher temperature faming regions. Figure 12(a) shows that the CO ft concentration exhibits lower fuctuations and less correlation with CO 2 , H 2 O, and temperature, providing additional evidence that the CO may be measured in lower temperature regions along the beam path. These observations are consistent with the CO being emitted from smoldering regions upwind of the fame front and propagating downwind into the measurement path. Figure 12(b) shows similar uncorrelated behavior of NH 3 with the other species, indicating the NH 3 may be measured in a di˙erent spatial location, and likely at a lower temperature, than the H 2 O and CO 2 .
Spatial inhomogeneity along the measurement path also helps interpret the MCE results. Despite measurement through a strongly faming region, the MCE in Fig. 7(a) is signifcantly lower than would be expected for these conditions. Due to the wind tunnel confguration, the measurement may detect gases from all points upwind of the beam path. Therefore, even when the beam is passing through the fame front region, there may be contributions to the detected gas composition from upwind smoldering regions which the fame front has already passed. In particular, the detected CO likely originates largely from smoldering regions upwind of the beam path. The steady decrease in MCE during the burn process is consistent with this mechanism, as the fraction of smoldering material upwind of the measurement path increases during the fame front propagation. The measured MCE in the post-fame region of ∼0.7 is also lower than the value ∼0.8 typically observed from smoldering in biomass burning studies [5,10,15]. In addition to the mechanism discussed above, the lower observed MCE may result from the in-situ measurement at close proximity to the smoldering fuel source which reduces the likelihood of reactive CO decreasing in concentration during propagation to a remote measurement or through sampling e˙ects.
The emission ratio (ER) was calculated as the ratio of time-integrated EMRs between a given species and CO. Table 1 shows calculated ERs when the time-integral was performed over the entire burn duration (excluding the initial plume of denatured alcohol vapor), and also when the integral was performed over a subset of data after the fame front passes. For NH 3 , C 2 H 4 , and MeOH, the ERs calculated from the ECQCL are in excellent agreement with prior published results [11,12]. Species detected using the LWIR ECQCL with high confdence were H 2 O, CO 2 , NH 3 , C 2 H 4 , and MeOH, verifed by visual examination of the absorbance spectra and identifcation of distinctive spectral features. For these species, the absorption cross-section in the LWIR ECQCL tuning range multiplied by the concentration is relatively high, resulting in peak absorbance values ∼ 10× the absorbance spectrum baseline noise. Other potential pyrolysis species (EtOH, formaldehyde, SO 2 , furan, acetic acid, formic acid, propene, acetaldehyde, acrolein, methyl vinyl ketone, and 2,3-butadione) were considered by adding them to the WLS analysis library; however, these species all have predicted peak absorbances approximately equal to the absorbance spectrum noise, based on the product of infrared absorption cross-section [44] and the ranges of concentrations expected to be present in biomass burning experiments [11]. The output results from the WLS algorithm indicated possible increases in EtOH, propene, acetic acid, and formic acid in the post-fame region, but the results were ultimately inconclusive because no distinctive spectral features could be reliably and consistently identifed in the experimental spectra.

Measurement sensitivity and accuracy
Measurement sensitivity expressed as noise-equivalent concentration (NEC) was determined via a standard Allan-Werle analysis [46]. A set of 7,000 spectra acquired over a 70 s time period before the fame was ignited was analyzed to determine the chemical concentrations, which are expected to be zero (relative to background levels). Figure 13 shows the results of the Allan-Werle analysis. All species show excellent averaging behavior indicating good instrument stability and lack of measurement drift over time scales up to 10 s. The Allan deviation results were obtained for stable conditions with low turbulence and near ambient temperatures, and sensitivity will be degraded when measuring through the fame region due to increased turbulence. It is important to note that the rapid fuctuations in species concentrations and temperatures limit the ability to improve the signal-to-noise ratio (SNR) via averaging, since in these cases averaging will reduce the signal levels as well as the noise. In practice, acquiring spectra at a faster rate than the time scales of turbulence and source fuctuations provides the most fexibility for averaging to best match the time scales of the physical system [34,36,37]. Measurement accuracy is diÿcult to quantify without additional reference measurements under known and stable conditions. At temperatures near ambient levels, corresponding to regions before or after the fame front, the agreement of measured spectra with reference data (NWIR or HITRAN) is high and accuracy is primarily limited by uncertainty in species distribution along the measured path. For measurements through the fame front regions, the accuracy is degraded as evidenced by the increased spectral ft residuals. The increased measurement noise due to turbulence partly explains the increased residuals; however, there are also clearly identifable and systematic discrepancies between the modeled spectra and experimental data. These discrepancies may be due to limited applicability of database spectra at elevated temperature, additional absorbing species in the measurement not included in the selected ftting library, or spatial inhomogeneity along the measurement path. Additional work is needed to validate the measurement and spectral analysis in controlled high-temperature conditions, and to quantify the accuracy and precision of the results.
The path-integrated measurement also limits the accuracy due to the assumption of uniform conditions along the beam path for the modeled spectrum. For higher accuracy, the absorption spectrum should be modeled as a composite mixture of species with varying temperatures and concentrations along the measured beam path using additional spatially-resolved experimental data or assumptions about the distributions of species and temperatures. For example, the spectral fts shown in Fig. 8(b) could be improved by assuming di˙erent temperatures for CO and CO 2 , or by adding spectra from multiple populations of CO, with each at a di˙erent temperature.
The measurements performed here involved a relatively short path length of 2.64 m and a high measurement speed to capture the rapid changes in species concentrations during the burn. Sensitivity could be increased by using a multi-pass confguration across the fame region or by sampling gases generated from the burn followed by o˜ine analysis [9]. In the latter case, by measuring the gases at lower temperatures and in less turbulent conditions it will be possible to detect species at much lower concentrations. For example, using a 50-100 m path length multi-pass cell [31] with 10 s averaging would provide predicted detection limits for many species at low-to sub-ppb concentrations. However, using an extractive or sampling method would reduce the ability to measure transient or reactive species existing only near the fame or smoldering regions. In practice, in-situ and extractive measurements provide complementary information for understanding the burn process and both are useful measurement confgurations.

Summary and conclusions
The measurements reported here used swept-ECQCL spectroscopy to measure gases close to the fuel and fame source in laboratory large-scale biomass burning experiments. A combination of two swept-ECQCL systems provided sensitive detection of CO 2 , H 2 O, and CO in the MWIR spectral region, and NH 3 , C 2 H 4 , MeOH in the LWIR spectral region. Species were measured throughout a dynamic burn experiment and showed the evolution of temperature and gas concentrations over fast and slow time scales. The 100 Hz measurement speed was able to track the rapid and large variations in gas concentrations and temperatures, especially when measuring through turbulent fame regions. Temporal correlations between gas species, temperature, and emission intensity on sub-second time scales indicate the presence of localized high temperature regions dominated by combustion gases CO 2 and H 2 O. Uncorrelated variations in CO and NH 3 indicate that they exist in cooler localized regions along the measurement path and may be generated from upwind regions of previously burned material. The results are consistent with observed variations in MCE during the burn process.
The swept-ECQCL measurements demonstrate a new tool for studying biomass burning dynamics in-situ and on faster time scales than have been previously studied. The high spectral radiance of the ECQCL permits measurement through high-temperature fame regions, which is needed to access spatial regions close to the fuel source where transient species may be present. However, path-integrated measurements through inhomogeneous fame fronts leads to high uncertainty in determining absolute concentrations and temperature at these times, and spatial inhomogeneity must be considered when interpreting the results of the measurement and analysis. Modifying spectral analysis algorithms to account for non-uniform temperatures and species distributions is expected to improve accuracy. The preliminary results presented here focused primarily on gases with highest concentrations and infrared cross-sections, with NH 3 , C 2 H 4 , and MeOH being identifed unambiguously. The broad tuning range of each swept-ECQCL provides the ability for multi-species detection, and the experiments show an example of using multiple ECQCL systems simultaneously to access di˙erent wavelength regions. Although the tuning range of each swept-ECQCL is not as large as FTIR-based instruments, the combination of high measurement speed, high spectral resolution, and high spectral radiance in the swept-ECQCL provides a valuable and complementary capability for infrared absorption spectroscopy. With increased measurement path lengths, the sensitivity of the ECQCL measurements is expected to improve, enabling time-resolved in-situ measurement of a wider range of trace gases emitted from biomass burning.

Funding
Strategic Environmental Research and Development Program (RC-2640).