Anthropogenic emissions of highly reactive volatile organic compounds in eastern Texas inferred from oversampling of satellite (OMI) measurements of HCHO columns

Satellite observations of formaldehyde (HCHO) columns provide top-down constraints on emissions of highly reactive volatile organic compounds (HRVOCs). This approach has been used previously in the US to estimate isoprene emissions from vegetation, but application to anthropogenic emissions has been stymied by lack of a discernable HCHO signal. Here we show that temporal oversampling of HCHO data from the Ozone Monitoring Instrument (OMI) for 2005–2008 enables detection of urban and industrial plumes in eastern Texas including Houston, Port Arthur, and Dallas/Fort Worth. By spatially integrating the HCHO enhancement in the Houston plume observed by OMI we estimate an anthropogenic HCHO source of 250 ± 140 kmol h−1. This implies that anthropogenic HRVOC emissions in Houston are 4.8 ± 2.7 times higher than reported by the US Environmental Protection Agency inventory, and is consistent with field studies identifying large ethene and propene emissions from petrochemical industrial sources.


Introduction
Anthropogenic highly reactive volatile organic compounds (AHRVOCs) with atmospheric lifetimes of less than a day are important precursors of ozone and organic aerosols in urban air and industrial plumes. Their sources are poorly quantified in emission inventories, as shown by air quality studies in eastern Texas (Ryerson et al 2003, Parrish et al 2012 and in an oil/gas field of northern Colorado (Gilman et al 2013). Satellite column measurements of formaldehyde (HCHO), a high-yield product from atmospheric oxidation of VOCs, have been used to constrain AHRVOC emissions in East Asia (Fu et al 2007) and Nigeria (Marais et al 2014a). However, detection of AHRVOC emissions in the US from satellite HCHO data has been elusive (Martin et al 2004, Millet et al 2008. The highest-resolution data are from the Ozone Monitoring Instrument (OMI), which provides daily global Environmental Research Letters Environ. Res. Lett. 9 (2014) 114004 (7pp) doi: 10.1088/1748-9326/9/11/114004 Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. coverage of HCHO columns by cross-track scanning with 13 × 24 km 2 nadir pixel resolution (Levelt et al 2006). An analysis of OMI urban data in the US by Boeke et al (2011) found only weak HCHO enhancements in the New York and Los Angeles urban cores in summer, and in the Houston urban core in spring and fall.
The difficulty of observing US AHRVOC emissions from space likely reflects their relatively small magnitude. The single-retrieval detection limit for OMI HCHO is 2 × 10 16 molecules cm −2 (Millet et al 2008), which corresponds to ∼4 ppb HCHO in a 2 km deep boundary layer. HCHO concentrations of ∼10 ppb are commonly observed in urban air and industrial plumes (Wert et al 2003, Buzcu Guven and Olaguer et al 2011, Lin et al 2012, Zheng et al 2013 but would be diluted on the scale sampled by the satellite pixels. Temporal averaging of the satellite data greatly improves the detection limit (Boeke et al 2011), though quantifying this improvement is difficult as it depends on the random versus systematic character of the retrieval error. The urban signal can also be masked by large regional emissions of isoprene, the dominant biogenic HRVOC contributing to HCHO (Palmer et al 2003, Martin et al 2004, Boeke et al 2011.
Here we demonstrate that quantitative detection of AHRVOC emissions in eastern Texas can be achieved by oversampling of the OMI HCHO data. 'Oversampling' refers to temporal averaging of the satellite data on a spatial grid finer than the pixel resolution on the instrument. The technique achieves high signal-to-noise ratio at high spatial resolution by sacrificing temporal resolution, i.e., averaging over a long time period. It takes advantage of the spatial offset and changing geometry (from off-track viewing) of the satellite pixels from day to day. Oversampling of OMI data has been applied previously with success to detection of SO 2 and NO 2 from urban and point sources (de Foy et al 2009, Fioletov et al 2011, McLinden et al 2012, Lu et al 2013. We demonstrate here its application to HCHO.

Data and methods
OMI is a UV/Vis nadir solar backscatter spectrometer launched in 2004 on the Aura satellite (Levelt et al 2006). It achieves daily global coverage with an equator crossing time of 13:38 local time. HCHO slant column densities (SCD) along the solar backscatter optical path are fitted in the spectral window 327.5-356.5 nm (Chance et al 2000). We use OMI HCHO Version 2.0 (Collection 3) SCD retrievals for 2005-2008 (http://disc.sci.gsfc.nasa.gov/Aura/data-holdings/ OMI/omhcho_v003.shtml) that (1) pass all fitting and statistical quality checks, (2) have a cloud fraction less than 0.3 and solar zenith angle less than 60°, and (3) are not affected by the 'OMI row anomaly' (www.knmi.nl/omi/research/product/ rowanomaly-background.php). Drift from instrument aging (Marais et al 2012) is removed with a linear temporal regression of background SCD over the North Pacific (130°-125°W, 35°-40°N).
The air mass factor (AMF = SCD/VCD) to convert SCD to vertical column density (VCD, column hereafter) is computed following Palmer et al (2001) with the LIDORT radiative transfer model (Spurr et al 2001). Satellite viewing geometry, cloud fraction, and cloud centroid pressure are from the OMI data. The AMF calculation requires information on HCHO and aerosol vertical distributions, and these are specified locally from the GEOS-Chem model with 0.5°×0.667°horizontal resolution over North America (Zhang et al 2011). Oversampling uses a higher horizontal resolution for the OMI data than 0.5°× 0.667°, but we expect that the error from subgrid variability in the HCHO and aerosol vertical distributions is small relative to other sources of error.
Wintertime observations would avoid biogenic interference on HCHO but we find that OMI HCHO columns are then indistinguishable from noise over the US including over Houston. This does not reflect loss of measurement sensitivity as the mean AMF over Houston in winter (December-February; AMF = 0.99 ± 0.17) is actually higher than in summer (May-August; AMF = 0.78 ± 0.13). The higher AMF in winter is due to longer light path and lower cloud cover, more than compensating for the effects of reduced UV light penetration and shallower planetary boundary layer (PBL). We attribute the lack of detectable HCHO in winter to low OH concentrations, delaying the oxidation of AHRVOCs to HCHO and thus smearing the HCHO signal. Average 12:00-15:00 (local time) OH concentrations in the GEOS-Chem model over Houston are a factor of 5 lower in December-February than in May-August. Mean surface wind speeds are 30% higher in December-February than May-August (www.wunderground.com/), also contributing to the smearing. We limit our attention here to May-August HCHO columns.
We oversample the OMI HCHO data over eastern Texas (99.5°-92.5°W, 28°-34°N) for May-August 2005-2008 by averaging individual pixels onto a 0.02°× 0.02°(∼2 × 2 km 2 ) grid. In this procedure, the column measurement for a given pixel is assumed to apply to a circle defined by the center point of the pixel and an 'averaging radius' of 24 km. Previous OMI oversampling studies have used the same strategy. Fioletov et al (2011) chose an averaging radius of 12 km for oversampling OMI SO 2 pixels, and McLinden et al (2012) used 20 km and 24 km for OMI NO 2 and SO 2 pixels, respectively. We find that a 12 km or 20 km averaging radius for HCHO pixels leads to excessive noise. Our oversampling approach leads to ∼800 OMI measurements being averaged in each 0.02°× 0.02°grid square. Here AHRVOCs are defined as having atmospheric lifetimes of less than 1 day against oxidation by OH. Table 1 lists the main NEI05 AHRVOCs emitted in the Houston plume area defined in figure 1 (gray line). Ethene and propene are the most important HCHO precursors, as also observed in the TexAQS campaigns (Wert et al 2003, Parrish et al 2012. Wert et al (2003) found from speciated VOC samples that 78% of the HCHO production potential was from terminal alkenes including 30% from ethene, 22% from propene, 14% from isoprene, and 12% from other alkenes.

Results and discussion
Biogenic isoprene makes a large background contribution to HCHO over eastern Texas, as seen in figure 1 and previously noted by Martin et al (2004). Isoprene emissions in MEGAN v2.1 are particularly high over forested Northeast Texas, explaining the high OMI HCHO columns there. Some distinction between biogenic and anthropogenic contributions to OMI HCHO can be made on the basis of correlation with surface air temperature. Isoprene emission increases exponentially with temperature (Guenther et al 2006), and this dependence is apparent in regional HCHO satellite data over the Southeast US (Palmer et al 2006). Figure 2 shows the relationships of OMI HCHO with surface air temperature over Northeast Texas and the Houston core for May-September 2006-2008. The data over Northeast Texas show a strong exponential relationship with temperature (R 2 = 0.64) with an argument of 0.11 K −1 , consistent with that expected for isoprene emission (Guenther et al 2006, Palmer et al 2006. By contrast, OMI HCHO columns over the Houston core show no significant relationship with temperature (R 2 = 0.03), supporting the dominant anthropogenic influence. The lack of correlation in the Houston data partly reflects a cluster of three points in figure 2 with T > 300 K and HCHO column <2 × 10 15 molecules cm −2 , but even excluding these points the relationship with temperature yields only R 2 = 0.34. Some correlation of HCHO with temperature would be expected even over Houston due to (1) the regional HCHO background contributed by isoprene (Wert et al 2003), (2) the temperature dependence of AHRVOCs oxidation, and (3) the association of high temperature with stagnation. . In our case, the AHRVOC enhancement is on top of a large regional background (figure 1). We constrain AHRVOC emissions for the Houston plume area by integrating the OMI HCHO enhancement over the Houston plume as the difference between the observed HCHO column (Ω) and the regional background (Ω 0 ) contributed by biogenic and long-lived anthropogenic emissions. From the HCHO lifetime (τ HCHO ) we deduce the corresponding HCHO source S per unit time as HCHO 0 where the integral is over the area A = 1.9 × 10 4 km 2 of the plume as defined in figure 1. This represents the total emission of AHRVOCs within the plume area weighted by their prompt yield of HCHO. We can convert this quantity to a total AHRVOC emission (E) by applying independent estimates of the fraction f i of the total emission contributed by species i and the corresponding HCHO yield Y i (table 1): Here we estimate the regional background (Ω 0 ) as the HCHO column downwind of the discernible Houston plume, corresponding roughly to the green color in figure 1 (8-10 × 10 15 molecules cm −2 ). Figure 3 shows HCHO columns averaged across the prevailing wind as a function of the distance from Houston city center. The plume decays to a regional background value at about 110 km downwind of the city center. For a mean wind speed of 3.3 m s −1 (www. wunderground.com/) this corresponds to an aging time of 9.3 h, long relative to the lifetimes of ethene and propene (table 1). From the downwind asymptote of the plume we   derive a best estimate for the regional background of Ω 0 = 9.6 ± 0.3 × 10 15 molecules cm −2 (figure 3). The actual uncertainty in the regional background is likely larger considering the fine-scale heterogeneity in isoprene emissions (Gulden, Yang Z L 2006). We therefore adopt Ω 0 = 9.6 ± 0.5×10 15 molecules cm −2 as a more conservative estimate of the uncertainty. The uncertainty in the Houston plume enhancement (Ω Ω − 0 ) is largely defined by the uncertainty in Ω 0 , considering that any systematic errors in the retrieval would likely be canceled in computing the difference between Ω and Ω 0 . We thus obtain a total HCHO column enhancement integrated over the plume of 400 ± 180 kmol (red area in figure 3).
Loss of HCHO is by photolysis and oxidation by OH. From table 1, we estimate τ HCHO to be 1.6 h. Photolysis accounts for half of total HCHO loss and is relatively well constrained. Loss by reaction with OH is not as well constrained because of uncertainty in OH concentrations. Here we adopted a mean OH concentration of 1.1 × 10 7 molecules cm −3 at 09:30-13:30 local time from measurements in downtown Houston (Mao et al 2009) and estimate the overall uncertainty in τ HCHO to be 30%. The resulting anthropogenic HCHO source S over the area of the plume is 250 ± 140 kmol HCHO h −1 , propagating in quadrature our estimated uncertainty in τ HCHO .
Combining this result with the data on f i and Y i in table 1, we deduce a total AHRVOC emission E for the HGB of 190 ± 100 kmol h −1 , which can be compared to the NEI05 estimate of 39 kmol h −1 from table 1. The OMI observations thus suggest that AHRVOC emissions in the Houston plume area are underestimated by a factor of 4.8 ± 2.7 in the NEI05 inventory for 2005-2008. This is consistent with previous studies pointing to a large underestimate of alkene emissions in the HGB (e.g., Wert et al 2003, de Gouw et al 2009, Parrish et al 2009, Mellqvist et al 2010. Our estimate of the HCHO source in the Houston plume is consistent with the estimate of 240 ± 90 kmol HCHO h −1 from Parrish et al (2012) computed using an improved alkene inventory (Kim et al 2011) with updated ethene and propene emission factors from petrochemical facilities (Mellqvist et al 2010).
Previous analyses of HCHO data in Houston have reached contradictory conclusions on whether most of the HCHO is primary, i.e., directly emitted (Rappengluck et al 2010, Buzcu Guven and Olaguer 2011, Olaguer 2013, Johansson et al 2014 or secondary, i.e., produced within the plume from alkene oxidation (Friedfeld et al 2002, Wert et al 2003, Parrish et al 2012, Zhang et al 2013. The distinction is important because primary HCHO would accumulate at night and photolyze in early morning, providing a source of radicals to initiate ozone formation. Our inability to detect the Houston urban plume in winter from the OMI data (see above) argues against a major primary source of HCHO. We attempted to constrain the speciation of AHRVOCs by using the shape of the OMI plume in figure 3 and a simple constant-wind model, as primary HCHO would decay closer to the core. We were unsuccessful, partly because of the complexity arising from primary emissions at night and in early morning when the HCHO lifetime is long.
We see from figures 1 and 3 that the anthropogenic HCHO enhancements in the Houston plume and elsewhere are on top of a larger regional background (Ω 0 ). This background is dominantly from biogenic isoprene. For the Houston plume area in figure 3, the 24 h average isoprene emission calculated by MEGAN in May-August is 95 kmol h −1 . Assuming a HCHO molar yield of 2.3 from isoprene oxidation (Millet et al 2006), this yields a HCHO production rate of 220 kmol h −1 , comparable in magnitude to the anthropogenic source.
Observations from the TexAQS aircraft campaigns in 2000 and 2006 documented a decrease of AHRVOC emissions from the HGB over that period. Gilman et al (2009) reported 56% and 51% decreases in ethene and propene median concentrations, respectively; Washenfelder et al (2010) found emission decreases in the Houston ship channel of 41% for ethene and 8% for propene from 1999 to 2006. Presently the useful OMI HCHO record is limited to 2005-2008; data after 2008 are too noisy for trend analysis because of the row anomaly. De Smedt et al (2010) used data from the GOME and SCIAMACHY satellite instruments to infer global 1997-2009 trends in HCHO, but we find that the pixel resolution of these instruments is too coarse for detection of the Houston plume. Post-2008 OMI data are expected to be corrected in a future product (González Abad et al 2014), which will then allow analysis of AHRVOC trends as well as examination of AHRVOC emissions associated with the large increase in oil/gas exploration across the US over the past five years.

Conclusions
We have shown that multi-year oversampling of summertime OMI HCHO satellite data enables detection of HCHO enhancements from large US urban/industrial sources of anthropogenic highly reactive volatile organic compounds (AHRVOCs). The enhancement in the Houston urban plume is sufficiently extensive to allow quantitative interpretation in terms of AHRVOC emissions. Our resulting estimate of AHRVOC emissions for Houston is 4.8 ± 2.7 times higher than the US EPA inventory and consistent with previous field estimates that identified large ethene and propene emissions from the petrochemical industry. The lack of detectable OMI HCHO enhancements in winter suggests that anthropogenic HCHO is mainly produced by photochemical oxidation of alkenes rather than directly emitted.