Metropolitan areas with a high number of petrochemical facilities are often struggling to meet current and future air quality standards. The Houston-Galveston area, for example, continues to be in noncompliance with the U.S. federal air quality standard of ozone, despite significant progress in mitigating air pollution. In recent years, the magnitude and role of primary emissions of ozone-forming chemicals, and in particular formaldehyde, from flares in petrochemical facilities have been discussed as a potential factor contributing to ozone formation. However, no direct observations of flare emissions of formaldehyde have thus far been reported. Here we present observations of formaldehyde and sulfur dioxide emissions from petrochemical flares in the Houston-Galveston area during the 2009 Formaldehyde and Olefin from Large Industrial Sources campaign using a new imaging differential optical absorption spectrometer (I-DOAS). Formaldehyde emissions from burning flares were observed directly above the flare stack and ranged from 0.2 to 8.5 kg/h. Unlit flares were found not to emit formaldehyde. SO2 emission rates from a burning acid gas flare ranged between 2 and 4 kg/h. None of the sampled flares coemitted HCHO and SO2. Comparison of the emission fluxes measured by the I-DOAS instrument with those from emission inventories and with fluxes calculated from plumes detected by the long-path DOAS over downtown Houston shows that the flares observed by the I-DOAS were relatively small. While burning flares clearly emit HCHO, a larger observational database is needed to assess the importance of flare emissions for ozone formation.
 Formaldehyde, HCHO, is an important intermediate in the degradation of VOC (volatile organic compounds) in the atmosphere, as well as an important precursor of HOx (OH + HO2) radicals. In many environments the photolysis of HCHO, which leads to the formation of HO2 radicals, is an essential contributor to the HOx budget [e.g., Mao et al., 2010]. Consequently, quantitative knowledge of the primary sources, i.e., emissions, and secondary sources, i.e., chemical formation, of HCHO is crucial to better understand tropospheric chemistry, in general, and the processes leading to urban air pollution, in particular.
 One of the questions that emerged in recent years is that of the magnitude and role of primary formaldehyde emissions from industrial activities, and in particular from flares in petrochemical facilities. Much of this research has focused on the Houston-Galveston area, which is the home of nearly half of the U.S. oil refining capacity (USGS, Circular 1182, http://pubs.usgs.gov/circ/circ1182/pdf/07Houston.pdf). The Houston-Galveston area is regularly in noncompliance with the U.S. federal air quality standard of ozone, which has been linked to the emissions of highly reactive VOC (HRVOC), i.e., predominately ethane and propene, from chemical manufacturing facilities [e.g., Ryerson et al., 2003; Kleinman et al., 2005; Mao et al., 2010]. While the role of these HRVOC in Houston area is well established [e.g., Kleinman et al., 2003; Murphy and Allen, 2005; Vizuete et al., 2008], it is less clear if HCHO is also directly released from these facilities and to what extent these direct formaldehyde emissions contribute to high ozone levels in the Houston area.
 Several studies have used field observations of HCHO and other trace species to argue that directly emitted HCHO plays a significant role in Houston. Friedfeld et al.  reported the ratio of secondary to primary HCHO formation in Houston as 1.7, based on a statistical analysis of HCHO, CO, and O3 observations. Rappenglück et al.  found that based on correlations with CO, peroxyacetylnitrate (PAN), and SO2, 38.5 ± 12.3% of HCHO was due to primary vehicular emissions, 8.9 ± 11.2% was from industrial emissions, 24.1 ± 17.7% was photochemically formed, and the rest could not be classified. Similarly, Buzcu Guven and Olaguer  performed a positive matrix factorization analysis of observations in Houston and found that for a downtown site, on average, 23% of HCHO resulted from traffic emissions and 17% from industrial sources, with the rest stemming from secondary formation due to industrial and biogenic VOC emissions. Interestingly, these factors varied considerably during the day, and the vehicular and industrial contributions were ~10% each during the daytime, when secondary formation clearly dominated. It should be noted, however, that Buzcu Guven and Olaguer  do not specify whether the industrial HCHO sources are primary or if secondary formation took place.
 While these approaches all employed statistical methods based on surface observations, Wert et al.  reached a different conclusion based on aircraft measurements of plumes from the Houston Ship Channel area. They found that ethane, propene, and other VOC emissions account for the majority of HCHO in petrochemical plumes detected during their airborne measurements. The impact of primary formaldehyde emissions, for example from traffic, was found not to be important. Parrish et al.  used these and later airborne HCHO observations of plumes from petrochemical facilities to quantify the contribution of primary emissions (i.e., directly emitted) to the budget of HCHO in Houston to only 4% of the HCHO, while the rest is due to secondary formation. They explain the conclusions of the previous studies by an incorrect interpretation of the correlations between different trace gases, i.e., HCHO with CO, SO2, and PAN, which are all simultaneously impacted by advection, mixing, and dilution. In addition, they argue that secondary HCHO will also correlate with these compounds and that it is thus difficult to separate primary from secondary HCHO.
 A complicating aspect of this disagreement is the spatial scale on which the impact of direct HCHO is considered. Based on the analysis of Parrish et al. , the contribution of direct emissions of HCHO to the overall budgets is not significant. However, this conclusion may not hold when considering scales of a few kilometers or less downwind of the primary HCHO source. Directly emitted HCHO can play the role of an early HOx radical precursor in the plumes, jump-starting the oxidation of VOC and NOx in the plumes and therefore accelerating the associated formation of ozone. This effect may not be captured by urban airshed models which typically run at grid sizes of 4 km or larger. The analysis of Parrish et al.  would also not be affected by this effect, as this study only considers the conversion of all HRVOC to HCHO and not the speed of conversion, or the speed of ozone formation. A recent high-resolution modeling study by Olaguer  shows that in a plume 8 km downwind from a petrochemical flare, the presence of primary HCHO increased the amount of ozone formed in the core of the plume by up to 10 ppb. This increase is associated with the HOx radical enhancement provided by primary HCHO photolysis in the plume. The impact of directly emitted HCHO thus seems to be constrained to the immediate vicinity of petrochemical flares, i.e., within ~10 km, and decreases farther downwind. Other modeling studies by Al-Fadhli et al.  and Pavlovic et al.  also concluded that HRVOC emissions from flares with less than 98%–99% combustion efficiency have the greatest impact on ozone formation in a few kilometers vicinity of a flare. This could explain the result by Parrish et al. , which were predominately sampled farther downwind.
 Despite this discussion of the role of primary emissions of HCHO, one piece of crucial evidence—the unequivocal identification of direct HCHO emissions from industrial facilities—is still missing. Here we present the methodology as well as field observations of direct HCHO and SO2 emissions from a small selection of flares in Houston area during the “Formaldehyde and Olefin from Large Industrial Sources” (FLAIR) campaign in the spring of 2009 by imaging differential optical absorption spectroscopy (I-DOAS). In recent years, the I-DOAS method has been used for visualization of NO2 plumes from power plants [Lohberger et al., 2004; Heue et al., 2008; Lee et al., 2009] and BrO plumes from volcanoes [Bobrowski et al., 2006]. It is ideally suited for remote observations of individual pollution sources, such as flares and smoke stacks of petrochemical facilities.
2 I-DOAS Method Description
2.1 Instrumental and Measurements Setup
 The UCLA imaging DOAS instrument was designed to be portable, allowing for fast setup and takedown, and to operate for extended periods of time from a portable power source while still maintaining spectroscopic characteristics necessary to detect target species such as HCHO. The instrument consists of an Acton SP-2150i imaging spectrometer coupled to a Princeton Instruments PIXIS camera equipped with the 1024 × 256 pixels back-illuminated E2V CCD30-11 CCD array (pixel size 25 µm × 25 µm) mounted in the focal plane of the spectrometer. For all observations presented here, a grating of 1200 groves/mm was used, covering the wavelength range between 290 and 407 nm. The spectral resolution of the instrument is 0.6 nm. A shutter is mounted directly behind the entrance slit and connected to the camera controller to ensure accurate exposure times. The CCD array is cooled to −70°C during the measurements in order to eliminate detector dark current. Five rows of the CCD array are binned into one spectrum, providing measurement of 50 separate spectra per exposure. Under nominal atmospheric conditions, exposure time for one measurement of all 50 spectra was between 50 and 200 ms.
 A 45° elliptical mirror collects scattered sunlight from the direction of the plume, with most of the light coming from behind the plume. The light collected by the mirror is imaged onto the entrance slit of the spectrometer by way of a Hoya 340 UV band-pass filter, a 100 mm focal length quartz lens, and a turning mirror (Figure 1a). Translational stages are used to align the telescope and scanner assembly in the laboratory before deployment. The light on the slit represents a field of view of ~7° vertical by 0.2° horizontal. Each of the 50 vertical spectra thus represents a part of the plume 0.14° vertical by 0.2° horizontal. The scanning mirror is rotated in the horizontal (azimuth) by a small stepper motor with a minimum step width of ~0.11°. The azimuth scanner is able to cover an overall angle of ~160°, although smaller scanning intervals are typically used to characterize most plumes.
 The spectrometer-detector combination, together with the telescope-scanner assembly, is mounted in a rigid aluminum frame with approximate dimensions of 40 × 30 × 30 cm and ~15 kg weight, which can be tilted with a 0.1° precision. The I-DOAS is equipped with an aiming scope and digital camera to further improve the aiming capability of the system. The instrument is controlled by a laptop that controls the detector, spectrometer, and scanner using the DOASIS software package (Institute of Environmental Physics, Heidelberg University, Germany, https://doasis.iup.uni-heidelberg.de/bugtracker/projects/doasis/). In the field, the I-DOAS is powered by a portable power system based on three commercial lead-acid batteries and a DC-AC pure sine wave converter.
2.2 Data Analysis
 The spectral retrieval of the I-DOAS measurements considers each of the 50 rows as an independent spectrometer; therefore, each row had its own wavelength-to-pixel conversion function and instrument function, which were determined by measuring mercury atomic emission lines. At the beginning of each instrument setup, a mercury spectrum is recorded, and a set of reference trace gas absorption spectra is convoluted using an instrument function determined with the measured Hg line at 334.4 nm for each of the 50 rows. The references for the absorption cross sections used in the analysis are listed in Table 1. The spectral retrieval was performed using a combination of linear and nonlinear least squares fit, as described in Stutz and Platt . Table 2 presents the wavelength intervals used for detection of different trace gases from the I-DOAS observations. For HCHO a small interval between 329 and 334.6 nm was excluded from the fit to minimize interference from solar Fraunhofer bands and other spectral interferences in this region. In addition to the trace gas references, an atmospheric reference was included in the fit. For measurements presented here, this atmospheric reference was obtained by performing a single measurement in an azimuth direction upwind of the targeted flare (where the concentration of the target pollutant was expected to be low). This spectrum also contains absorptions of the background trace gas levels. Consequently, the atmospheric background trace gas concentration upwind of the flare is subtracted spectroscopically, considering that the radiative transfer conditions upwind and downwind of the flare are similar (see section 2.3).
Table 1. Trace Gas References Used for I-DOAS Analysis
 Simulated Ring spectra of the solar reference scan, along with the linear and quadratic expansion of the Ring spectra [Vountas et al., 1998; Langford et al., 2007], were fitted together with a polynomial after tenfold triangular smoothing low-pass filtering. To allow for uncertainties in the grating position caused by temporal drift of the spectrometer during a measurement, spectral shift of trace gases during evaluation was allowed. During the evaluation, all trace gases were linked in shift and squeeze to each other, and spectral shift typically did not exceed 1 pixel. Similarly, the spectral shifts of the Ring and Fraunhofer spectra were linked to each other. The error of the measurement is calculated by multiplying the statistical error of the fit by a factor of 3 according to Stutz and Platt .
 Systematic errors of our observations are dominated by errors associated with the reference absorption cross sections used in the spectral retrievals. Reported errors for reference absorption cross sections used for spectral evaluation are listed in Table 1.
 The results of the I-DOAS retrieval are differential slant column densities (SCDs), i.e., path-averaged trace gas concentrations, relative to the reference spectrum. Figure 2 shows an example of HCHO retrieval for one of the I-DOAS measurements on 12 May 2009 at 16:27 UTC. The HCHO differential slant column density was found to be 3.4 ± 0.15 × 1016 molecules/cm2.
2.3 Radiative Transfer Considerations
 Trace gas SCDs derived from scattered sunlight observations depend on the conditions of radiative transfer at the time of the measurement. For the observations of industrial flares, however, radiative transfer modeling can be omitted. This is due to the fact that these observations were performed from distances of only few hundred meters away. In addition, the observed flares did not emit any particulate matter as operators regulate the flares to avoid “smoke” (as required by the Federal Regulations 40CFR §60.18 [Title 40 Code of Federal Regulations]), making radiative transfer effects inside the plume itself negligible. Therefore, a geometric approximation based on the I-DOAS viewing direction and distance to the observed object can be used to estimate emission fluxes.
 In order to verify this assumption, we examined O4 SCD values resulting from the spectral evaluation of the I-DOAS spectra. If our assumption were inaccurate and radiative transfer inside the plume played a role, radiative transfer effects would emerge as variations of O4 SCDs within the plume. Upon investigation of O4 SCD images, we found no changes in O4 SCD that can be attributed to the observed plumes from flares.
 Also, for most of our observations, radiative transfer conditions along the entire light path of the reference measurement and any other measurement in a scan were similar, i.e., clear-sky conditions and only few days had some scattered clouds. O4 SCD and UV intensity images collected on days with scattered clouds show some small variations due to clouds; however, these variations are due to differences occurring far from the instrument, before the light encountered the flare plume, and therefore have no effect on retrieved SCDs.
2.4 I-DOAS Viewing Direction Calibration
 In order to produce accurate flux calculations, the viewing direction of the instrument has to be determined with high precision. At the time of the measurement, the instrument coordinates were recorded by GPS and the azimuthal viewing direction was determined using a compass and the aiming scope/digital camera. The elevation angle of the instrument was determined with a high-precision tiltmeter at the time of setup. The I-DOAS elevation viewing angle was determined with an accuracy of 0.1°. For a flare in a 400 m distance, this translates to an uncertainty of 0.7 m for the flare height determination. Using Google Earth and the instrument coordinates and azimuthal viewing direction, the distance between the instrument and the flare under observation was determined. The accuracy of this determination was determined to be 10 m based on previous experience comparing Google Maps distances with GPS coordinates and comparisons with maps. In addition, the I-DOAS viewing direction was directly verified by comparing the digital photograph of the area to the UV intensity image of the I-DOAS scan. This viewing direction verification is illustrated in Figure 3, showing the photograph and UV intensity image of the I-DOAS scans over the acid gas flare in Texas City on 8 May 2009. The tip of the acid gas flare and three power lines in the view of the I-DOAS show as dark shadows in the UV intensity image.
2.5 Flux Calculation
 Emission strengths from individual point source were calculated using the combination of the I-DOAS observations and meteorological data, using the approach described below. I-DOAS observations of individual sources normally were set up such that the instrument's line of sight (main azimuth viewing angle) was approximately perpendicular to the direction of the wind and therefore to the direction to which pollutants emitted from the flare would travel (see Figure 4).
 By summing up the 50 trace gas SCDs in one measurement of the I-DOAS at a given azimuth viewing angle, an integrated trace gas SCD in a vertical “slice” through the plume was obtained (equation (1)):
 Taking into account the distance between the instrument and the flare, wind direction and speed, and the optical setup of the instrument, a trace gas flux was then calculated for each observed azimuth angle using the following equation (equation (2)):
where γ is the I-DOAS projection factor that relates the dimensions of the image in a distance to its projection onto the spectrometer slit (, where 6 mm is the slit height of the spectrometer, 50 is the number of rows in the detector, and 100 mm is the focal length of the telescope lens), D is the distance between the I-DOAS and the flare (cm), Vw is the wind speed (cm/s), and β is the angle between the I-DOAS line of sight and the direction of the wind.
 In order to reduce uncertainties and the variability of the emissions, we performed flux calculations for several neighboring azimuth viewing angles to obtain an averaged flux value. For all of the calculations presented here, ∑ SCDj is an average value of 8 to 10 azimuthal steps downwind from the flare. To account for the possible “background” amount of the respective trace gas, the vertically integrated trace gas SCD from the part of the image that is upwind of the source was determined. This background value was then subtracted from the downwind ∑ SCDj. The background vertically integrated SCD is also averaged over the 10 azimuthal steps. The final emission fluxes reported here are thus calculated using the following equation (equation (3)):
 The error of the trace gas fluxes was calculated using statistical error propagation of errors from the least squares fit of atmospheric spectra to retrieve trace gas SCDs, errors in determination of azimuth and distance between I-DOAS and observed flare, and uncertainties in wind speed and wind direction. The error for the SCDs was calculated by multiplying the statistical error of the spectral fit by 3 in accordance with Stutz and Platt . This error is the smallest contribution to the uncertainty of the flux calculation. Distance between the I-DOAS and flare/stack under observation was determined using the Google Earth ruler tool, and the uncertainty of this measurement was set to 10 m; while Google does not provide information on the accuracy of this tool, online users report it to be 10–15 m. The uncertainty for the I-DOAS azimuth viewing direction was determined to be 5°. Wind speed information used for the flux determination was estimated from the meteorological data available for the region at the time of measurements using a logarithmic wind profile relationship [Oke, 1987] (see section 2.6). Wind speed represents the largest uncertainty in the flux calculations.
2.6 Winds at Flare Altitude
 As presented in section 2.5, accurate wind speed information is essential for determination of emissions from flares via I-DOAS measurements. The petrochemical flares sampled by the I-DOAS during the FLAIR 2009 campaign were at altitudes between 40 and 80 m above ground level. Determination of the wind speed in this lowest region of the boundary layer (lowest 100 m) presents challenges. Most meteorological data are collected at the altitude of 10 m above the ground (World Meteorological Organization standard anemometer height). However, within the lowest 100 m of the atmosphere, wind speeds can vary significantly [Oke, 1987]. This change in wind speed has to be considered for our flux calculations.
 In order to determine the wind speed at the altitude of the observed flare, we used a logarithmic wind profile, a semiempirical relationship used to describe wind profiles within the lowest 100 m of the boundary layer [Oke, 1987]. We estimate the wind speed at altitude z2 from the wind speed measurements at altitude z1 of the boundary layer using the following equation:
where and are wind directions at altitudes z1 and z2, respectively, z0 is the roughness length, and ψ is a stability term. All I-DOAS observations presented in this manuscript were performed on sunny days, during afternoon hours, when the boundary layer is fully developed and the atmosphere is mostly neutral, and therefore, ψ can be assumed to be zero. Oke  reports for urban environments a roughness length from 0.5 m for urban areas moderately covered by low buildings at relative separation of three to seven obstacle heights and no high trees, to 2.0 m for densely built-up city centers with a mixture of low and high buildings. Industrial areas where I-DOAS observations were performed represent somewhat of a mixture between these two urban extremes, and therefore, a roughness length of 0.75 m was used for calculations.
 This calculation introduces uncertainties in the wind speeds. For example, the uncertainty of ~1.5 m/s in the 6 m/s wind speed observations at 10 m altitude translates to an uncertainty of ~2.5 m/s for the wind speed at 40 m altitude. In addition, one needs to consider the uncertainty in the surface roughness of 0.25 m, which translates to an uncertainty of approximately 1 m/s in the wind speed at 40 m. The assumption of neutral conditions versus the possibility of a more unstable atmosphere also introduces an uncertainty of ~1 m/s. A quadratic propagation of these uncertainties results in an uncertainty of ~ 2.8 m/s or approximately 30%.
 The wind direction can also potentially shift from the altitude of the observation to the flare height. However, analysis of wind profiles in the region and also theoretical calculations based on an idealized Ekman spiral indicate that this shift is no more than 5°, within the uncertainty of ±5° assumed for the observed wind direction.
 In the following, we will discuss three specific examples of observations performed during the experiment. The results from all observations will be discussed in section 4.
3.1 HCHO Emissions From a Flare of a Chemical Manufacturing Facility Along the Houston Ship Channel
 On 29 May 2009 between 20:00 and 21:30 UTC, formaldehyde emissions were detected from a flare in a chemical manufacturing plant in the Houston Ship Channel. During the observations, the I-DOAS instrument was positioned 400 m southeast from the burning flare and performed a series of measurements over the flare starting at an azimuth of 280° downwind of the flare and continuing to an azimuth of 295° upwind of the flare.
 Three sets of scans were performed. All three sets showed clear enhancements of HCHO relative to the background, originating directly from the tip of the flare. Figure 5 presents one of the three sets of measurements. Figures 5a and 5b show HCHO SCD and SCD error images, respectively, and Figure 5c shows vertically integrated HCHO SCDs (see section 2.5) for this scan. Horizontal axes of Figures 5a and 5b represent a number of individual measurements in the scan, and vertical axes on the left hand side are representative of the row of the CCD detector, altitude determined from the observation geometry is indicated by the vertical axis on the right hand side of the panel a. Based on the HCHO SCD images and photographs taken during measurements, the tip of the observed flare was determined to be at approximately pixel 21 in the vertical, which corresponds to 50 m altitude and pixel 22 in the horizontal, and clear HCHO enhancement observed at these coordinates (see Figure 5a) indicates direct HCHO emissions by the flare. Figure 5 (b) shows the error for HCHO SCD retrieval; this image is uniform within and outside of the HCHO plume, showing the quality of spectral retrieval. No SO2 emissions above the detection limit of 1.3 kg/h were observed from this flare.
 Figure 5c shows vertically integrated HCHO SCDs calculated using data in Figure 5a. As expected, vertically integrated HCHO SCDs are low upwind of the flare (−20–0 m away from the flare on the left of the image), and they increase rapidly and stay relatively constant downwind of the tip of the flare (0–15 m right of the flare). Beyond that, vertically integrated HCHO SCD values begin to drop because the plume is being dispersed in the atmosphere and the I-DOAS instrument no longer captured the entire plume. The following parameters were used for the HCHO flux calculation. The distance between the I-DOAS and the flare was determined to be 400 ± 10 m. Wind speed and direction at the altitude of the flare were determined using a logarithmic wind profile using 4 ± 1 m/s and 180 ± 5° at 10 m altitude as measured by the Texas Commission for Environmental Quality (TCEQ) Lynchburg Ferry meteorological station in the Houston Ship Channel. Based on the HCHO SCD images, vertically integrated HCHO SCDs averaged between horizontal pixels 22 and 30 were used for flux calculations (green crosses in Figure 5c). Averages of vertically integrated SCD between pixels 2 and 10 were used for background values of HCHO (red crosses in Figure 5c). The HCHO flux determined for this scan was 8.50 ± 2.75 kg/h.
3.2 HCHO Emissions From the Flare of a Chemical Manufacturing Facility in Mont Belvieu
 During the FLAIR campaign, on 19 May 2009, a unique opportunity presented itself to observe a flare from a chemical plant in the Mont Belvieu area before and after it was ignited. This flare was identified by TCEQ personnel as releasing uncombusted VOC, based on images from a handheld infrared camera. The Aerodyne mobile laboratory performed in situ measurements downwind of the flare before it was ignited and attributed elevated concentrations of propane and lower levels of butene, benzene, and toluene to this flare [Wood et al., 2012]. HCHO was not enhanced in the flare plume [Wood et al., 2012]. I-DOAS observations of the unlit flare were performed, followed by observations of this flare after it was ignited. Figures 6 (a) and 7 (a) show HCHO SCD images for two scans over the unlit and ignited flares, respectively. While the scans over the flare were performed between the same angles azimuthally, they were performed at two different elevation viewing angles, and therefore, the vertical axes of the two images are different. For both images however, flare tip is located at approximately 44 m altitude and pixel 33 in the horizontal. For the unlit flare, no significant enhancement of HCHO over the background is observed. However, for the burning flare, there is a clear HCHO enhancement starting at the tip of the flare. No SO2 emissions were detected from either unignited or lit flare.
 Observations of this flare lead to the conclusion that HCHO emissions from petrochemical flares are the result of incomplete combustion processes inside the flare. In fact, almost all burning flares we observed in Houston during the FLAIR campaign emitted HCHO, while no HCHO emissions were observed from unignited flares.
 HCHO emissions rates for this flare estimated by the method described earlier are shown in Figures 6 (bottom) and 7 (bottom). For the unlit flare (Figure 6), HCHO emissions are below 0.02 kg/h, while for the burning flare, emissions were determined to 0.22 ± 0.13 kg/h.
3.3 SO2 Emissions From a Flare of a Refinery in Texas City
 On 8 May 2009, the I-DOAS instrument sampled air upwind and downwind of an acid gas flare located on the southeast side of the Texas City Industrial Complex. Between 17:53 and 18:45 UTC, three consecutive sets of scans over the flare were performed. Figure 8 shows the SCD image of SO2 observed from the flare between 18:29 and 18:45 UTC. All three scans revealed SO2 emissions, ranging from 3.99 ± 1.39 to 2.0 ± 0.6 kg/h, or from 95.8 to 48 kg/d, respectively. No HCHO emissions above the detection limit of 0.28 kg/h were observed. For the SO2 flux calculation shown in Figure 8, a distance of 200 m between the I-DOAS and the flare was determined using Google Earth; the wind speed at the altitude of the flare was calculated using logarithmic wind profile and meteorological data collected at the altitude of 10 m by the meteorological station at the Texas City courthouse (6.25 m/s and the wind direction 180°). Pixels 160–170 in the horizontal were used for background subtraction, and pixels 140–150 were used for the flux calculations. The SO2 flux farther from the stack decreases, as part of the plume moves beyond the view of the I-DOAS instrument (similarly to the HCHO flux in Figure 5).
 In the 2009 ozone season emissions inventory provided to us by the Texas Commission for Environmental Quality (TCEQ), the acid gas flare daily emissions are reported at 0.17247 metric tons/day, corresponding to average hourly emissions of 7.2 kg/h-a value that is similar to our observations for this flare. It is also interesting to note that since acid gas flares are common for any refining facility, in some states, e.g., California, daily SO2 emissions from acid gas flares are posted by states' regulatory agencies. For example, the Bay Area Air Quality Management District reports daily emissions from the acid gas flare of Valero Benicia on their website (http://hank.baaqmd.gov/enf/flares/). Examination of reported daily emissions values over the past 9 years revealed that SO2 emissions from that particular acid gas flare are highly variable. On average, only for a few days of each month, daily SO2 emissions are above zero and vary from low of 2.9 kg/d to high of 612 kg/d, with emissions of a few kg/h being more common.
 The previous section clearly showed the direct emission of both HCHO and SO2 from certain flares. Here we want to further discuss these results with regard to the question of the coemission of HCHO and SO2, as well as the question whether the observed emission strengths have an impact for the air quality of the greater Houston area. Table 3 summarizes all successful measurements performed by the I-DOAS instrument during FLAIR. While these results only provide a snapshot of the many different types of flares in petrochemical facilities, a number of conclusions can be drawn. First and foremost, our observations clearly show that burning flares directly emit formaldehyde. HCHO is observed within 1–2 m of the tip of the flare stack and thus in or at the edge of the flame. This together with the fact that the few unignited flares that were observed did not emit HCHO leads to the conclusion that HCHO is a product of the feed gas combustion of the flare, rather than the feed gas itself. It is well known that HCHO is one of the first intermediate products of combustion of methane and many other hydrocarbons [Gardiner, 2000]. Consequently, unless 100% flare destruction and removal efficiency is achieved, formaldehyde emissions from VOC-burning flares can be expected to be a common phenomenon. The TCEQ 2010 Flare Study conducted at the John Zink facility [Allen and Torres, 2011], where flare experiments were designed to simulate various flare operations of industrial facilities, found that under many conditions flares had lower than expected destruction and removal efficiency leading to emissions of a variety of products of incomplete combustion, including HCHO [Herndon et al., 2012; Torres et al., 2012]. For example, this study found large propene emissions from the flare operating at low destruction efficiencies (DRE) of ~73%–88%. During this study, unburned vent gases and CO contributed to a large fraction of the emissions; however, small, but measurable, emissions of HCHO also occurred [Herndon et al., 2012; Torres et al., 2012]. Torres et al.  found that industrial flares operating at low-flow conditions often had DRE well below 99% (DRE value prescribed by the industry for flare emission calculation), resulting in the release of CO, ethylene, formaldehyde, acetylene, and acetaldehyde.
Table 3. Summary of I-DOAS Observations of HCHO and SO2 Emissions From Flares During FLAIR 2009
HCHO and SO2 Fluxes (kg/h)
Texas City, acid gas flare
17:53 UTC—F(SO2) = 3.99 ± 1.39
18:14 UTC—F(SO2) = 2.99 ± 0.99
18:29 UTC—F(SO2) = 2.80 ± 1.03
18:50 UTC—F(SO2) = 2 ± 0.6
F(HCHO) < 0.28
Texas City, flare of the chemical plant
F(HCHO) = 0.67 ± 0.24
F(HCHO) = 0.91 ± 0.38
F(SO2) < 1.2
Mont Belvieu, chemical plant
11:45 UTC—F(HCHO) < 0.02 (unlit)
13:20 UTC—F(HCHO) = 0.22 ± 0.13 (burning)
F(SO2) < 0.22
Houston Ship Channel area, burning flare of the chemical facility
20:54 UTC—F(HCHO) = 6.88 ± 3.81
20:54 UTC—F(HCHO) = 8.50 ± 2.75
F(SO2) < 0.5
 HCHO emissions from burning flares during the I-DOAS deployments in Houston area in 2009 were found to range from 0.22 to 8.50 kg/h (Table 3). It should be noted here that the observed flares were small and none of them were operated in an emergency release, which is known to produce the largest VOC releases [Allen et al., 2004]. Table 3 also lists the SO2 emissions observed in the HCHO-emitting flares and the acid gas flare, for which we identified emissions of SO2. The acid gas flare SO2 emissions were in the range of 2.0–4.0 kg/h. The emissions were quite variable between individual scans. A test of the temporal variations in emissions determined over several 15 min intervals (not shown here) confirms the apparent variability in the emissions shown in Figure 8. This flare is listed in the TCEQ 2009 ozone season emission inventories with average hourly emissions of 7.2 kg/h. This value is very close to the emissions from this flare observed by the I-DOAS on 8 May 2009.
 The SO2 fluxes in the HCHO-emitting flares presented above were below the respective for the presence limits for these fluxes of 1.2, 0.22, and 1.5 kg/h. Despite the two higher detection limits, which would still allow for the presence of considerable SO2 emissions, we conclude that our sample of flares does not coemit HCHO and SO2. This contradicts some previous observations of the correlation of HCHO and SO2 in plumes observed near downtown Houston (see also below). For example, direct HCHO emissions from flaring and other activities associated with the petrochemical industry have been estimated based on the assumption that there is a clear correlation of the emissions of these two species [e.g., Rappenglück et al., 2010]. Our observations cannot confirm this assumption. The most likely reason for the observed correlation of HCHO and SO2 in the plumes observed near downtown Houston is the presence of two independent but colocated sources of HCHO and SO2 in the Houston Ship Channel, an area which is known for its very high density of petrochemical industry. We want to emphasize that with our limited number of observations, we cannot completely exclude the possibility that certain flares do coemit both species.
 While it is difficult to assess the significance of our observations for ozone chemistry in Houston, we can compare previous observations of transient HCHO and SO2 plumes in downtown Houston. These nocturnal plumes, which were previously mentioned, but not further discussed, in Olaguer et al. , were observed by UCLA's long-path differential optical absorption spectrometer (LP-DOAS) during the TexAQS Radical Measurement Program (TRAMP) air quality measurement campaign in the summer/fall 2006 from 15 August through 30 September 2006. Details on the LP-DOAS instrument and observations during the TRAMP campaign can be found in Stutz et al. [2010a, 2010b]. In short, the LP-DOAS measured in open air between the University of Houston's Moody Tower, 70 m above the ground, and downtown Houston 4–5 km away. Three retroreflectors were mounted at three altitudes above the ground, and measurements of horizontally averaged HCHO and SO2 mixing ratios were performed in three vertical intervals: lower (20–70 m), middle (70–130 m), and upper (130–300 m).
 During TRAMP, six nighttime events with elevated HCHO and SO2 were observed lasting for periods between 2 and 5 h. In all but one, instances of elevated HCHO and SO2 levels coincided. During all these events, wind speeds, measured at 80 m altitude on Moody Tower, were low (between 0 and 5 m/s) and the observed wind direction indicated that the air masses passed over the Houston Ship Channel a few hours earlier. Wind directions during plume events were approximately perpendicular to the LP-DOAS light path. Figure 9 presents an example of such a nighttime enhanced HCHO and SO2 event. At about midnight CST on 3 September, HCHO and SO2 levels in the upper height interval began to increase, remaining elevated for the following 5 h of the night. To maintain the observed vertical mixing ratio, profiles of HCHO and SO2 emissions must have occurred after the formation of the nocturnal boundary layer, i.e., after daytime vertical mixing had ceased. For comparison, Figure 9 shows the profile of NO2, which is elevated near the surface due to its ground sources. The average wind speed was ~2 ± 0.3 m/s. The Houston Ship Channel area is only approximately 6–7 km, or 0.8–2 h air travel time, to the west/northwest of the University of Houston. We hypothesize that the observed HCHO and SO2 enhancements aloft were due to hot plumes emitted from flares or smoke stacks, which were lofted above the nocturnal boundary layer and transported aloft to our site. Since photochemistry, and thus efficient HCHO formation, halts during the night, and observed HCHO enhancements aloft are decoupled from possible surface sources, it is reasonable to assume that the observed HCHO and SO2 enhancements are a result of direct emissions, likely from a burning flare(s) in the Houston Ship Channel.
 In order to estimate the source strength for this plume, we assume that the plume was entirely in the LP-DOAS light path in the vertical and horizontal. Since neither of these two assumptions can be verified, our calculation is a lower limit of the trace gas flux through the plane formed from our three light paths. We can estimate the emission strength for this HCHO and SO2 source using equation (5):
where [A]enh represents the concentration of species i (HCHO or SO2) in excess of background derived for each LP-DOAS height interval, li is the length of the respective LP-DOAS light path through each interval, hi is a vertical extend of the altitude interval, and Vw is the average wind speed observed at Moody Tower, which has an uncertainty of ~0.3 m/s.
 We estimate a minimum HCHO flux through our light paths of ~15.4 kg/h. The SO2 emission flux in this plume was determined to be at least 194 kg/h. The uncertainty of these fluxes is dominated by the uncertainty in the wind speed, direction, and profile. We estimate this error to be ~30%. The HCHO flux is about 6–60 times larger than the fluxes observed by the I-DOAS system. Two possible explanations for this larger number are possible. The plume was caused by a multitude of smaller flares, such as those we observed with the I-DOAS. Alternatively, a large flare, such as those used in emergencies, could be responsible for the observed plumes.
 For example, reported SO2 emission rates for emergency releases from the acid gas flare at the Valero Benicia Refinery, California, are normally on the order of few hundred kg/h, occasionally reaching as much as 600 kg/h (as of 8 September 2004, see http://hank.baaqmd.gov/enf/flares/index_2004.htm).
 Table 4 summarizes HCHO and SO2 fluxes calculated for all nocturnal HCHO and SO2 plumes observed by the LP-DOAS over downtown Houston in 2006. Values for HCHO and SO2 fluxes as well as SO2/HCHO ratios in the observed plumes are highly variable. For example, the HCHO flux during the night of 23 August 2006 is comparable to the one observed by the I-DOAS from the flare observed on 29 May 2009 (see Figure 5). Interestingly, the ratio of SO2/HCHO in the downtown plume is approximately 12.6, while the upper limit ratios from the HCHO-emitting flares in Table 4 are below 2. This seems to confirm the conclusion that many of the flares do not coemit HCHO and SO2.
Table 4. HCHO and SO2 Fluxes Calculated From Nocturnal Plumes Observed Over Downtown Houston in Fall 2006
Date and Time (CST)
08/22/06 23:04–08/23/06 00:40
09/02/06 22:30–09/03/06 05:45
 Previous studies of air pollution in the greater Houston area have suggested that direct emission of formaldehyde from burning flares may accelerate the formation of ozone [Rappenglück et al., 2010; Buzcu Guven and Olaguer, 2011]. This interpretation was supported by observations of plumes of elevated HCHO in downtown Houston, which were often found to correlate with SO2 [Olaguer, 2011]. However, other studies have argued that the direct emission of HCHO does not significantly contribute to ozone pollution in Houston [Parrish et al., 2012]. To shed light on this open issue, I-DOAS observations of flares in the Houston-Galveston area were performed. While only a small selection of flares was investigated, the following conclusions can be drawn from our observations:
 Burning flares directly emitted HCHO at rates of 0.2–8.5 kg/h, while unlit flares were found not to emit formaldehyde. HCHO is thus a combustion product of the flare, and it can be expected that many burning flares will emit varying amounts of HCHO.
 SO2 emission rates from a burning acid gas flare ranged between 2 and 4 kg/h. However, many of the investigated flares did not emit SO2.
 None of the sampled flares coemitted HCHO and SO2. Observations of correlated plumes of HCHO and SO2 are thus most likely dominated by meteorological factor and/or colocated, but different sources of these two species. The approach of correlating HCHO with SO2 in situ concentrations to determine emission rates therefore has to be performed with care.
 Comparison with lower limit HCHO and SO2 fluxes determined near downtown Houston using LP-DOAS indicates the releases of these two compounds by many colocated smaller flares such as those observed by the I-DOAS or by a few larger flares, such as those used in emergencies.
 While our observations clearly show that flares emit formaldehyde at rates of several kg/h, the number of flares sampled in our study was too small to provide an overall estimate of the contribution of this source to HCHO and ozone formation in Houston. More systematic observations are needed in order to create a database of HCHO and other trace gas emissions from flares, based on the type of flare, operational conditions, volume and composition of feed, etc. As we have demonstrated here, the I-DOAS method is ideally suited for this task.
 The authors would like to thank Dejian Fu, now at the Jet Propulsion Laboratory, Pasadena, CA, for his invaluable help with the setup and I-DOAS measurements at the beginning of the FLAIR campaign in the spring of 2009. We would also like to thank Evan Couso from the University of North Carolina at Chapel Hill for sifting through emission inventories in order to identify reported emissions from flares sampled with the I-DOAS. The preparation of the manuscript is based on work supported by the state of Texas through the Houston Advanced Research Center and the Air Quality Research Program administered by the University of Texas at Austin by means of grants from the Texas Commission on Environmental Quality.