Attribution of primary formaldehyde and sulfur dioxide at Texas City during SHARP/formaldehyde and olefins from large industrial releases (FLAIR) using an adjoint chemistry transport model

Authors


Abstract

[1] An adjoint version of the Houston Advanced Research Center (HARC) neighborhood air quality model with 200 m horizontal resolution, coupled offline to the Quick Urban & Industrial Complex (QUIC-URB) fast response urban wind model, was used to perform 4-D variational (4Dvar) inverse modeling of an industrial release of formaldehyde (HCHO) and sulfur dioxide (SO2) in Texas City, Texas during the 2009 Study of Houston Atmospheric Radical Precursors (SHARP). The source attribution was based on real-time observations by the Aerodyne mobile laboratory and a high resolution 3-D digital model of the emitting petrochemical complex and surrounding urban canopy. The inverse model estimate of total primary HCHO emitted during the incident agrees very closely with independent remote sensing estimates based on both Imaging and Multi-Axis Differential Optical Absorption Spectroscopy (DOAS). Whereas a previous analysis of Imaging DOAS data attributed the HCHO release to a Fluidized Catalytic Cracking Unit (FCCU), the HARC model attributed most of the HCHO event emissions to both the FCCU and desulfurization processes. Fugitives contributed significantly to primary HCHO, as did combustion processes, whereas the latter accounted for most SO2 event emissions. The inferred HCHO-to-SO2 molar emission ratio was similar to that computed directly from ambient air measurements during the release. The model-estimated HCHO-to-CO molar emission ratio for combustion units with significant inferred emissions ranged from 2% to somewhat less than 7%, consistent with other observationally-based estimates obtained during SHARP. A model sensitivity study demonstrated that the inclusion of urban morphology has a significant, but not critical, impact on the source attribution.

1 Introduction

[2] Direct emissions of formaldehyde (HCHO) from the combustion of natural gas and chemicals may contribute significantly to ozone formation in Houston and other heavily industrialized areas [Olaguer et al., 2009]. In a related paper, Olaguer [2013] used an adjoint version of the Houston Advanced Research Center (HARC) neighborhood air quality model to demonstrate the plausibility of a significant release of primary HCHO from a large petrochemical flare in the vicinity of the Houston Ship Channel, based on measurements of HCHO and other reactive species during the Second Texas Air Quality Study (TexAQS II). The inverse model results of Olaguer [2013] contradicted the hypothesis of Parrish et al. [2012], who attributed the highest HCHO concentrations observed during TexAQS II (up to 52 ppb) to regional transport of secondary formaldehyde.

[3] For this paper, we investigated a HCHO emission event that occurred in Texas City, Texas during the formaldehyde and olefins from large industrial releases (FLAIR) subexperiment of the 2009 Study of Houston Atmospheric Radical Precursors (SHARP) field campaign. Overviews of SHARP and FLAIR are provided by E. P. Olaguer et al. (Overview of the SHARP campaign: Motivation, design, and major outcomes, submitted to Journal of Geophysical Research Atmospheres, 2013) in this special section, including a description of the various measurement techniques deployed during the FLAIR study.

[4] Whereas the source attribution of Olaguer [2013] was based on ambient air measurements conducted about 8 km away from the source, the FLAIR measurements used as the basis for our study were performed immediately outside the emitting petrochemical facility's fence line. Hence, we will adopt a more refined approach than the simple treatment of meteorology used by Olaguer [2013] by considering the influence of 3-D building morphology on the source attribution wind fields.

[5] The specific release of interest was first noticed in the early afternoon of 13 May 2009 as the Aerodyne mobile laboratory [Kolb et al., 2004; Herndon et al., 2005], equipped with a variety of real-time measuring devices, drove past a petrochemical complex in Texas City that included one of the largest refineries in the U.S. and a chemical plant immediately south of the refinery. At that time, high concentrations of HCHO and SO2 were detected by the mobile lab. Formaldehyde was measured with a Quantum Cascade Laser tunable infrared spectrometer in the mobile lab with 0.6 ppb Limit of Detection (LOD) and 1 s Time Response (TR), while the pulsed fluorescence method (LOD = 2 ppb, TR = 18 s) was used to measure sulfur dioxide. The field study management team then proceeded to direct other measurement platforms to the area, including a portable Imaging Differential Optical Absorption Spectrometer or I-DOAS (O. Pikelnaya et al., Imaging DOAS detection of primary formaldehyde and sulfur dioxide emissions from petrochemical flares, submitted to Journal of Geophysical Research Atmospheres, 2013), and a Piper Aztec aircraft equipped with a Hantzsch fluorescence HCHO monitor (LOD = 50 ppt, TR = 120 s) and a pulsed fluorescence SO2 monitor, among other airborne instruments [Alvarez et al., 2009]. In addition, continuous measurements of area-wide HCHO and SO2 fluxes were performed by two stationary Multi-Axis (MAX) DOAS spectrometers [Stutz et al., 2011], located upwind and downwind of the Texas City industrial core.

[6] Figure 1 shows net HCHO and SO2 fluxes to the atmosphere from the area bounded by the vertical scanning planes of the two stationary MAX-DOAS spectrometers around the event of interest, as originally presented by Stutz et al. [2011]. There is a significant correlation between the fluxes of the two chemical species, probably as a result of combustion and desulfurization in petroleum refining operations. Note also the somewhat elevated concentrations of HCHO on the afternoon of 13 May 2009 compared to the afternoon of the preceding day. Based on the stationary MAX-DOAS measurements, Stutz et al. [2011] computed an average net HCHO flux of 45 kg/h from Texas City during FLAIR, of which 22 kg/h was likely to be primary, the latter estimate based on the correlation between HCHO and SO2.

Figure 1.

HCHO and SO2 fluxes measured by the dual MAX-DOAS technique for 3 days during the FLAIR campaign in Texas City, Texas, after Stutz et al. [2011].

[7] Alvarez et al. [2009] analyzed airborne measurements over Texas City that indicated the presence of a petrochemical plume at an altitude of ~400 m on the afternoon of 13 May 2009. For example, at 3:18 P.M. LST, they found an enhancement of about 0.5 ppb in HCHO mixing ratio over a background of around 0.9 ppb, coinciding with concentration peaks of SO2, NOy, and alkenes. Around the same time that the Piper Aztec was flying over Texas City, from approximately 3:00 P.M.–5:00 P.M. LST, the I-DOAS was aimed at a plume coming directly at the instrument from the direction of a Fluidized Catalytic Cracking Unit (FCCU). Based on the I-DOAS absorption data, a simple plume model, and a plume width inferred from simultaneous mobile lab measurements, Stutz et al. [2011] computed a primary HCHO flux of 18.2 + 4.8 kg/h, which they attributed to the unit designated as FCCU 3 in the absence of more refined information about other emission points. This was roughly similar to the estimates of HCHO fluxes from three separate areas within the same complex derived from mobile MAX-DOAS measurements in a study by Johansson et al. [2011].

[8] Whether or not the HCHO release observed by the I-DOAS on 13 May 2009 was the result of a process upset or a routine emission is not clear. The facility in question filed an emission event report to the Texas Commission on Environmental Quality (TCEQ), which stated that a leak from a flange on a propane/propylene (PP) treater associated with FCCU 3 had been discovered at 12:10 A.M. on 14 May 2009. (A PP treater enhances the purity of the FCCU output propylene stream.) A connection between this leak and the late afternoon concentrations of HCHO observed by the MAX-DOAS during the preceding day cannot be established based on Figure 1, as the MAX-DOAS requires sunlight to operate. In any case, HCHO was not among the reported event emissions, although it is listed as a species emitted by FCCU 3 in the reported annual emission inventory for the facility.

[9] The U.S. Environmental Protection Agency (USEPA) [1991] emission factor for formaldehyde released by a refinery FCCU is 2.2 kg HCHO per 1000 lb of feed. The feed reported in USEPA [1991] for the refinery in question was 194,000 lb/day (3667 kg/h), implying HCHO emissions of 17.8 kg/h, which is very close to the estimate derived from the I-DOAS, although much greater than the annual average FCCU 3 emission rate reported by the refinery for 2009, which was 0.02 kg/h (possibly due to control technology). However, FCCU 3 may not have been the only upwind process unit contributing to primary HCHO measured by the I-DOAS, which observed the emission plume head-on rather than from the side.

[10] Our objective for this paper was to perform a rigorous source attribution of primary HCHO emitted from the Texas City petrochemical complex (hereafter referred to as TCPC) on 13 May 2009. To provide an analysis that was independent of remote sensing measurements, we used the 3-D adjoint chemical transport model described in Olaguer [2013] to perform inverse modeling of TCPC emissions based on Aerodyne mobile lab observations and the 4-D variational (4Dvar) data assimilation technique. Unlike Olaguer [2013], however, we assumed realistic 3-D urban morphology to perform the source attribution, using wind fields generated by the QUIC-URB fast response urban wind model [Singh et al., 2008].

2 Methodology

[11] The basic method used for this study was similar to that of Olaguer [2013]. However, we included vertical advection in both the forward and continuous adjoint models, rather than assume a uniform horizontal wind. To suppress numerical instability, we used the simpler Smolarkiewicz [1983] scheme for adjoint vertical advection, while maintaining the Piecewise Parabolic Method for other forward and adjoint advection terms. We also optimized the full vector of species emissions E rather than a single event emission factor, as well as the corresponding vector of species initial conditions C0. As in Olaguer [2013], we maintained the horizontal turbulent diffusion coefficient, Kh, at a constant value of 50 m2/s. The cost function over N time steps was thus defined as follows:

display math(1)

[12] In equation ((1)), the superscript b refers to the prior (or background) estimate of an optimized parameter; Q and B are the error covariances corresponding to their respective optimized parameters in the cost function; and Rk is the measurement error variance associated with the mobile lab observation math formula at time step k.

[13] The optimized values of E and C0 can be obtained from the vector of adjoint values λk as follows:

display math(2)
display math(3)

[14] For this study, we employed a horizontal domain size of 4 km × 4 km with 200 m grid resolution. The vertical domain size was 964 m, with 10 layers and a nonuniform vertical resolution that increased toward the surface (see Table 1). The model time step was set at 20 s.

Table 1. Model Vertical Layer Properties
Layer NumberMidpoint Height (m)Thickness (m)Vertical Diffusivitya (m2/s)
  1. a

    The vertical diffusivity is evaluated at the lower interface of each layer.

18342601.31
260020854.3
3415161163.9
4274121262.7
517086.7278.5
697.358.3209.4
750.235.9113.6
822.619.443.5
98.428.9510.5
101.724.440

[15] A high resolution digital model of the TCPC and the surrounding urban area was developed to facilitate source attribution. Building morphology was represented by 3-D lidar data obtained from the National Geospatial-Intelligence Agency. Annual average emissions of NOx, CO, and reactive VOCs (ethene, propene, 1,3-butadiene, 1-butene, 2-butenes, isobutene, toluene, and xylenes) from each building within the TCPC, as well as corresponding stack heights, were specified based on permit information, official TCEQ emission inventories, and previous remote sensing studies conducted at the complex based on the Differential Absorption Lidar (DIAL) technique [Texas Commission on Environmental Quality, 2010; Randall and Coburn, 2010]. The DIAL studies were used to confirm the spatial coordinates of several process units.

[16] Reported annual mean emissions for NOx, CO, and reactive VOCs were used as first guess emissions for the adjoint model. Reported NOx emissions were assumed to be 90% NO and 10% NO2. The effective release height for several combustion units was set at 30 m above the stack height to account for plume rise.

[17] To perform the HCHO source attribution, we first obtained data from the State of Texas Air Reporting System, specifically the TCEQ's 2009 ozone season day (OSD) HCHO emission inventory for 99 combustion and fugitive sources located in the TCPC. Unlike the annual mean emissions reported by the complex, which ignore primary HCHO emissions except those from FCCU 3, the OSD HCHO emissions were inferred by the TCEQ to facilitate air quality modeling. These emissions were computed using default Environmental Protection Agency speciation profiles for reported unclassified VOCs based on Standard Industrial Classification and Source Classification Code categories. We used the OSD emission inventory to identify the specific TCPC facilities that the TCEQ deemed capable, from an engineering process perspective, of emitting formaldehyde during either fugitive or combustion releases. Rather than use the OSD HCHO emissions as a first guess, however, we assigned a minimal first guess emission of 0.001 short tons per year (TPY) of HCHO to each emission point in the OSD HCHO inventory. This was done to avoid biasing the source attribution for sporadically emitting units such as flares. Moreover, one flare in very close proximity to FCCU 3 was left out, because the I-DOAS observation team did not see any operating flares in their field of view (J. Stutz, personal communication, 2013) and to avoid confusing the source attribution.

[18] To perform the SO2 source attribution, we assigned a minimal first guess emission of 0.001 TPY of SO2 to each of 71 TCPC emission points in the reported annual mean SO2 inventory for 2009. The same flare left out from the HCHO source attribution was left out from the SO2 source attribution.

[19] For the HCHO source attribution, emissions of NOx, CO, HCHO, propene, and toluene from the emission points specified in the OSD HCHO inventory were optimized by the inverse model with chemistry, while emissions from other sources (e.g., storage tanks) were maintained at their first guess values. For the SO2 source attribution, SO2 was treated as an independent passive tracer (due to its relatively long atmospheric lifetime compared to the time interval of interest), so that inferred SO2 emissions did not depend on the emissions of other species. Mobile lab measurements of NOx, CO, HCHO, propene, toluene, and SO2 were used as input to the adjoint model at every time step. To filter out the effects of sporadic, short-duration exhaust plumes intercepted from nearby vehicles, mobile lab measurements were ignored in the HCHO source attribution whenever the measured ambient CO mixing ratio exceeded 800 ppb. A measurement root-mean-square error of 1 ppb was assumed for all species, except for CO, in which case it was 30 ppb. The initial estimate for the emission error variance was 0.02 (g/s)2 for all species other than CO and SO2. For CO and SO2, the initial emission error variances were set at 20 (g/s)2 and 0.2 (g/s)2, respectively.

[20] Figure 2 illustrates the model grid and urban morphology, along with the mobile lab route during the assimilation time period, the I-DOAS location, and the location of FCCU 3 and other important industrial process units in the source attribution. Note that some of the process units are in close proximity and are therefore situated in the same model grid cell.

Figure 2.

Model grid showing urban morphology, approximate mobile lab route on 13 May 2009, and locations of primary formaldehyde sources (see Table 4).

[21] Inverse modeling was conducted for two 1 h periods (2 P.M.–3 P.M. LST and 3 P.M.–4 P.M. LST) on 13 May 2009, the second of which overlapped the I-DOAS measurement period. This second period was further subdivided into twelve 5 min assimilation subwindows, as further explained below. The initial background concentration error covariance matrix was assumed to be diagonal, with error variances of 104 ppb2 for CO, and 10 ppb2 for all other transported species. The transported species initial conditions were optimized for the first hour, but not for the second, as the final condition for the previous 1 h period served as the initial condition at 3 P.M. LST. Thus, only the emissions from the selected emission points were optimized for the 5 min assimilation subwindows.

[22] The meteorological inputs to the model are summarized in Table 2. Hourly-averaged surface air temperature data were obtained from the Texas City Ballpark monitor (CAMS 1022) just outside and to the north of the TCPC. Relative humidity was obtained from the CAMS 1034 station in Galveston, the nearest site for which such measurements were available. The specification of surface pressure and temperature lapse rate was the same as in Olaguer [2013].

Table 2. Model Meteorological Conditions
Local Standard Time (P.M.)Anemometer Level Wind Speed (m/s)Anemometer Level Wind Direction (°)Surface Air Temperature (K)Relative Humidity (%)
2:00–3:005.81168.730170.8
3:00–3:055.96168.830171.9
3:05–3:105.52178.130171.9
3:10–3:155.05187.330171.9
3:15–3:205.93170.230171.9
3:20–3:255.86170.530171.9
3:25–3:306.20165.130171.9
3:30–3:355.10201.930171.9
3:35–3:405.73168.630171.9
3:40–3:456.03157.430171.9
3:45–3:504.78167.130171.9
3:50–3:555.83167.130171.9
3:55–4:005.93158.330171.9

[23] Although high time resolution (2 s) wind measurements were made on board the mobile lab at a height of 3.3 m agl, the presence of wind gusts along the mobile lab route during the first assimilation window (2 P.M.–3 P.M. LST) made it difficult to extrapolate this data to other vertical levels. For this time period, hourly-averaged resultant wind speed and resultant wind direction at anemometer level (10 m agl) were specified from measurements at the Texas City Ballpark monitor. During the second hour (3 P.M.–4 P.M. LST), the mobile lab was stationary and parked close to the I-DOAS. For this time period, 5 min average resultant wind data measured by the mobile lab were extrapolated to anemometer level assuming a power law for wind speed, with an exponent of 0.1. For each assimilation window or subwindow, the anemometer level winds were extrapolated to other height levels using a standard logarithmic profile (as implemented in the QUIC-URB model) and values for roughness length and Monin-Obukhov length of 0.1 m and −100 m, respectively. The resulting wind profiles were fed as background wind input to QUIC-URB, which was then used to compute spatially varying 3-D winds based on the specified urban morphology.

[24] The QUIC-URB grid used to generate the wind fields had a horizontal resolution of 10 m and 41 vertical levels. The QUIC-URB grid origin was positioned so that a lower-resolution, staggered, mass-conserving Arakawa C-grid could be generated for winds and tracers with 10 vertical layers and 200 m horizontal resolution. The higher resolution 3-D wind field was averaged over the appropriate planes to ensure the mass consistency of the HARC model winds.

[25] To compute the vertical diffusion profile, the boundary layer was assumed to be 1 km deep. The turbulence parameterization of Delle Monache et al. [2009], originally developed for an emergency response urban boundary layer model, was employed to better account for the influence of the urban canopy on vertical diffusion. For this purpose, the grid-cell averaged building height and fractional frontal area were set at 7.2 m and 0.034 respectively, while the friction velocity for the time period of interest was set at 1.9 m/s. Values of the vertical diffusivity at the lower interface of each model vertical layer are displayed in Table 1.

[26] Inflow boundary conditions for transported species are summarized in Table 3 and were obtained from the CAMS 1034 monitor for NOx and O3, and from automated gas chromatograph (auto-GC) measurements of hydrocarbons at the CAMS 35 monitoring station, which is in an industrialized area near the Houston Ship Channel very similar to Texas City. The inflow boundary condition for SO2 was likewise obtained from the CAMS 35 station. For CO, HCHO, nitrous acid (HONO), and organic nitrate (RNO3), boundary conditions were set equal to typical mixing ratios observed during the 2006 TexAQS II Radical and Aerosol Measurement Project [Lefer et al., 2010] for the same time of day. The relatively clean HCHO and SO2 boundary conditions reflect the location of the model domain at the southern edge of Texas City, in close proximity to Galveston Bay, which makes southerly incoming plumes of primary SO2 or accumulated secondary HCHO unlikely.

Table 3. Model Boundary Conditions
Transported SpeciesInflow Boundary Condition (ppb)
2:00–3:00 P.M. LST3:00–4:00 P.M. LST
Nitric oxide (NO)0.30.4
Nitrogen dioxide (NO2)1.71.5
Ozone (O3)2426
Nitrous acid (HONO)0.10.1
Formaldehyde (HCHO)0.50.5
Carbon monoxide (CO)200200
Ethene (C2H4)0.170.15
Propene (C3H6)0.090.08
1,3-Butadiene (C4H6)0.010.01
1-Butene (BUT1ENE)0.010.01
2-Butene (BUT2ENE)0.020.02
Isobutene (IBUTENE)0.010.01
Isoprene (ISOP)0.320.26
Toluene (TOL)0.040.03
Xylenes (XYL)0.030.05
Organic nitrate (RNO3)11
Sulfur dioxide (SO2)0.60.6

[27] The HARC model lumps together organic species not explicitly represented in the chemical mechanism and assigns them a total OH reactivity. We specified a value of 4 s−1 for this parameter. This was 20% less than assumed by Olaguer [2013] for the Houston Ship Channel region, an area with even more industrial facilities than Texas City. Total OH reactivity in urban areas typically exceeds 10 s−1 [Mao et al., 2010].

3 Results and Discussion

[28] Table 4 presents optimized values of the hourly-averaged emissions for the most important sources. The two major assimilation time periods yielded nearly identical values for total TCPC emissions of HCHO: 22.5 + 0.2 kg/h for 2 P.M.–3 P.M. LST, and 21.9 + 0.1 kg/h for 3 P.M.–4 P.M. LST. (The estimated errors are derived from the optimized emission error variances.) These estimates coincide very nearly with the area-wide primary HCHO flux deduced from the two stationary MAX-DOAS instruments. There was also very little variation in the inferred emissions for the 5 min subwindows from 3 P.M. to 4 P.M. LST. The cost function effectively converged within five iterations, as shown in Figure 3 for the 2 P.M.–3 P.M. LST assimilation window. The maximum decrease in the cost function from its initial value within any of the subsequent 5 min subwindows after five iterations was 0.26%.

Table 4. Facility Emissions Inferred From Inverse Model for the Control Case
Facility2 P.M.–3 P.M. LST3 P.M.–4 P.M. LST
HCHO (kg/h)SO2 (kg/h)HCHO (kg/h)SO2 (kg/h)
Fluidized cat cracker unit 3 fugitives (FCU3-FUGIT)8.2NA7.6NA
Cat feed hydrotreater cooling tower (CFHU-CTWR)2.3NA2.3NA
Residual hydrotreater unit fugitives (RHU-FUGIT)2.1NA2.1NA
Incinerator (SRU-F8CD)1.7681.768
Residual hydrotreater units (RHU)1.4681.469
Cat feed hydrotreater units (CFHU)1.2861.286
Pipestills 3A and 3B (PS3AB)0.96870.9688
Isomerization unit fugitives (ISOM-FUGIT)0.83NA0.83NA
Residual/cat feed flare (TCH-RHU)0.64830.6483
Fluidized cat cracker unit 3 wet gas (CAT3-WGS)0.57NA0.57NA
Total TCPC emissions22.540521.9406
Figure 3.

Convergence of the cost function for the first assimilation window.

[29] The HCHO source attribution did not implicate the FCCU 3 as the only facility responsible for the observed HCHO emission event. Rather, it also implicated the nearby hydrotreaters and other process units associated with desulfurization activities. If we exclude the isomerization unit to the west (and therefore not upwind) of the I-DOAS location (see Figure 2), the two assimilation time periods yielded nearly identical values for the HCHO emissions from the remaining units in Table 4: 19.1 kg/h for 2 P.M.–3 P.M. LST, and 18.5 kg/h for 3 P.M.–4 P.M. LST. These estimates are in precise agreement with the local HCHO flux deduced from the I-DOAS. The HCHO emissions attributed to the Residual/Cat Feed Flare are within the range of 0.3–2 kg/h ascribed to routine operating flares by O. Pikelnaya et al. (submitted manuscript, 2013) based on I-DOAS measurements during SHARP/FLAIR.

[30] Whereas both fugitives and combustion sources were substantially implicated by the HCHO source attribution, combustion sources were found to be mainly responsible for the SO2 event emissions. Most of the combustion emissions of SO2 are due to desulfurization activities, including hydrotreatment and associated flaring, with significant contributions from distillation (pipestill emissions) and incineration.

[31] The inferred emissions in Table 4 can be compared to the corresponding values from the OSD HCHO and annual mean SO2 inventories listed in Table 5. For the incinerator and flare, the inferred HCHO emissions are about three orders of magnitude larger than the inventory values, as are the inferred SO2 emissions from the catalytic feed hydrotreater. On the other hand, the inferred HCHO emissions from the cooling tower, hydrotreaters, and FCCU wet gas, and the inferred SO2 emissions from residual hydroteatment, are only one to two orders of magnitude higher than the values in Table 5. Inferred pipestill and incinerator emissions of SO2 are higher by about a factor of five, while the estimates of fugitive emissions of HCHO from catalytic cracking, hydrotreatment, and isomerization agree roughly within a factor of three. The best agreement occurs for pipestill emissions of HCHO and flare emissions of SO2. Olaguer [2013] has already noted that HCHO emissions from flares are routinely undercounted. The large emissions of both HCHO and SO2 attributed to the hydrotreater units, however, may indicate an accidental release rather than a systematic underestimate in the inventory.

Table 5. Facility Emissions From the TCEQ OSD HCHO Inventory and the TCPC-Reported Annual Mean SO2 Inventory
FacilityHCHO (kg/h)SO2 (kg/h)
Fluidized cat cracker unit 3 fugitives (FCU3-FUGIT)3.2NA
Cat feed hydrotreater cooling tower (CFHU-CTWR)0.016NA
Residual hydrotreater unit fugitives (RHU-FUGIT)0.66NA
Incinerator (SRU-F8CD)0.0007615
Residual hydrotreater units (RHU)0.0740.16
Cat feed hydrotreater units (CFHU)0.0200.069
Pipestills 3A and 3B (PS3AB)0.5015
Isomerization unit fugitives (ISOM-FUGIT)0.26NA
Residual/cat feed flare (TCH-RHU)0.0007652
Fluidized cat cracker unit 3 wet gas (CAT3-WGS)0.041NA

[32] To further evaluate the inferred emissions, we computed HCHO:CO and HCHO:SO2 molar ratios for the major combustion sources, as well as for total TCPC emissions including fugitive releases of HCHO. These ratios are presented in Table 6. Typical HCHO:CO ratios inferred from near-source measurements of industrial plumes either during SHARP/FLAIR [Alvarez et al., 2009; Stutz et al., 2011; Wood et al., 2012] or the more recent TCEQ Flare Study [Allen and Torres, 2011; Knighton et al., 2012] range from about 2% to as high as 10%. The model-estimated HCHO-to-CO molar emission ratio for combustion units with significant inferred emissions ranges from 2% to somewhat less than 7%. This is consistent with the independent estimates noted above. The inferred HCHO:SO2 molar ratio is 12% for total TCPC emissions. Aerodyne mobile lab data collected for all transects in Texas City during the FLAIR campaign suggest a typical molar ratio of 12%, with a range of 7% to 16% [Stutz et al., 2011]. Once again, our inferred molar ratio is consistent with these estimates.

Table 6. Molar Emission Ratios Inferred From Inverse Model
Facility2 P.M.–3 P.M. LST3 P.M.–4 P.M. LST
HCHO/COHCHO/SO2HCHO/COHCHO/SO2
Incinerator (SRU-F8CD)0.0340.0530.0340.053
Residual hydrotreater units (RHU)0.0580.0440.0580.043
Cat feed hydrotreater units (CFHU)0.0610.0300.0590.030
Pipestills 3A and 3B (PS3AB)0.0690.0240.0540.023
Residual/cat feed flare (TCH-RHU)0.0330.0160.0330.016
Fluidized cat cracker unit 3 wet gas (CAT3-WGS)0.019NA0.018NA
Total TCPC emissions0.0790.120.0760.12

[33] The performance of the HARC model can be gauged not only from the reasonableness of the inferred emissions but also from the associated concentration fields. Figure 4 illustrates the optimized initial conditions and predicted final concentration for ambient HCHO near the surface (at the level of the mobile lab observations) for 2 P.M.–4 P.M. LST, while Figure 5 shows the corresponding concentration fields for SO2. The little resemblance between the optimized initial conditions and the final concentration fields demonstrates that a sporadic emission event has occurred within the 2 h assimilation period, consistent with the temporal variation in the net fluxes detected by the dual stationary MAX-DOAS instruments. The final concentration field of HCHO shows two distinct plumes. The larger plume to the east is associated with catalytic cracking and desulfurization emissions, while the smaller plume to the west is associated with fugitive emissions from the isomerization unit. The final SO2 concentration field shows an amalgamated plume from multiple combustion sources.

Figure 4.

HCHO mixing ratio isopleths near the surface at (top) 2 P.M. LST (contour spacing = 0.1 ppb) and (bottom) 4 P.M. LST (contour spacing = 2 ppb) for 13 May 2009.

Figure 5.

SO2 mixing ratio isopleths near the surface at (top) 2 P.M. LST (contour spacing = 0.05 ppb) and (bottom) 4 P.M. LST (contour spacing = 20 ppb) for 13 May 2009.

[34] Figure 6 compares mobile lab measurements of ambient HCHO with corresponding HARC model predictions along the mobile lab route, while Figure 7 presents the same comparison for SO2. The HARC model reasonably simulates temporal variations in ambient HCHO and SO2 for both 1 h assimilation windows, although it does not capture the highest concentration peaks, a shortcoming that may be explained by emissions transience, chaotic turbulence, or overly efficient vertical mixing by the model. The simulation errors are larger during the second assimilation time period, when the mobile lab was stationary.

Figure 6.

Model-predicted (solid line), observed 20 s average (dashed line), and observed 5 min average (dotted line) HCHO mixing ratio from (left) 2 P.M. to 3 P.M. LST and (right) 3 P.M. to 4 P.M. LST along mobile lab route on 13 May 2009.

Figure 7.

Model-predicted (solid line), observed 20 s average (dashed line), and observed 5 min average (dotted line) SO2 mixing ratio from (left) 2 P.M. to 3 P.M. LST and (right) 3 P.M. to 4 P.M. LST along mobile lab route on 13 May 2009.

[35] Figure 8 compares Aztec aircraft measurements of HCHO and SO2 at 304 to 463 m agl with corresponding model predictions aloft for time segments during which aircraft data were available. The aircraft-observed peaks are surprisingly well represented by the model, especially for SO2. For HCHO, the model peak amplitudes are generally within about a factor of two of the observations. The HCHO peak at around 3:48 P.M. LST, however, is precisely simulated. Note that we did not use the aircraft data to drive the adjoint model.

Figure 8.

Model-predicted (solid line) and aircraft-observed (dashed line) 20 s average mixing ratios of (left) HCHO and (right) SO2 after 3 P.M. LST on 13 May 2009. The time is displayed in minutes after 3 P.M. The model predictions are shown only for time segments during which aircraft data was available.

[36] Table 7 displays performance statistics for the simulation of the ambient concentrations of the two compounds of interest. For SO2 near the surface, we have considered only those cases for which the observed concentration is greater than 1 ppb. The model performance improves significantly for longer averaging time (5 min versus 20 s), as one would expect. Given that we are simulating a highly transient emission event for which available information is otherwise poor, the simulation error metrics are quite reasonable. To further improve the simulation, higher temporal and spatial resolution or weaker vertical diffusion may be required.

Table 7. Ambient Concentration Simulation Performance Statisticsa
MetricNear Surface 2 P.M.–3 P.M. LSTNear Surface 3 P.M.–4 P.M. LSTAloft 3 P.M.–4 P.M. LST
HCHOSO2HCHOSO2HCHOSO2
  1. a

    The SO2 performance statistics near the surface are for cases when the observed mixing ratio exceeded 1 ppb.

Root-mean-square error (20 s averages)2.0 ppb13 ppb2.8 ppb29 ppb0.7 ppb2.1 ppb
Mean normalized bias (20 s averages)7.5%36.8%87.7%8.13%−67%45.5%
Mean normalized error (20 s averages)66.1%90.2%123%80.5%71%128%
Root-mean-square error (5 min averages)1.3 ppb5.9 ppb1.3 ppb19 ppbNANA
Mean normalized bias (5 min averages)−28.2%−9.5%13.6%−31.7%NANA
Mean normalized error (5 min averages)36.5%36.5%41.5%43.7%NANA

[37] Lastly, we conducted a sensitivity experiment to determine how important the urban morphology was to the source attribution. For this experiment, we set the horizontal components of the wind speed to their horizontal average values at each level and neglected the vertical wind. The resulting wind speed profile is illustrated in Figure 9. Note that the urban canopy drag results in a weaker average horizontal flow than indicated by the anemometer level wind values used as input to the model. The lack of horizontal wind variation in the sensitivity experiment, however, eliminates the channeling effects of the urban canopy.

Figure 9.

Wind speed profile for 2 P.M.–3 P.M. LST in sensitivity experiment.

[38] Table 8 presents the results of the model sensitivity experiment. The inferred total TCPC emissions of HCHO have been reduced by roughly 14% as a result of ignoring the urban morphology, while the corresponding total SO2 emissions have increased by less than 1%. The inferred HCHO emissions from the FCCU have been reduced only by about 12%. From this, we conclude that the inclusion of urban morphology has a significant, but not critical, impact on the source attribution.

Table 8. Facility Emissions Inferred From Inverse Model for the Sensitivity Experiment
Facility2 P.M.–3 P.M. LST3 P.M.–4 P.M. LST
HCHO (kg/h)SO2 (kg/h)HCHO (kg/h)SO2 (kg/h)
Fluidized cat cracker unit 3 fugitives (FCU3-FUGIT)7.2NA6.6NA
Cat feed hydrotreater cooling tower (CFHU-CTWR)2.2NA2.2NA
Residual hydrotreater unit fugitives (RHU-FUGIT)1.9NA1.9NA
Incinerator (SRU-F8CD)1.5681.568
Residual hydrotreater units (RHU)1.3681.369
Cat feed hydrotreater units (CFHU)1.0881.088
Pipestills 3A and 3B (PS3AB)0.88860.8887
Isomerization unit fugitives (ISOM-FUGIT)0.59NA0.59NA
Residual/cat feed flare (TCH-RHU)0.57810.5781
Fluidized cat cracker unit 3 wet gas (CAT3-WGS)0.52NA0.52NA
Total TCPC emissions19.440118.8403

[39] While our approach to neighborhood scale source attribution represents a considerable advance in methodology, there remain some important sources of uncertainty that should be addressed in future studies. For example, we were not able to incorporate aerosol lidar data that would have helped to constrain the model vertical mixing within the planetary boundary layer. Second, some highly reactive volatile organic compounds, such as ethene, that were not included in our analysis may have contributed to secondary formation of formaldehyde near combustion sources. Advances in real-time detection of trace gases, such as the use of new reagents (e.g., Kr+) in chemical ionization-mass spectrometry to expand the range of measured species, may help to increase the accuracy of source attribution using inverse modeling techniques.

4 Conclusion

[40] Based on the application of the adjoint version of the HARC neighborhood air quality model, we conclude that industrial activities in Texas City, Texas emit a significant amount of primary formaldehyde. Although some of this is from flaring, other combustion and fugitive emission sources can make significant contributions to the primary formaldehyde flux. This includes desulfurization and possibly other processes associated with catalytic cracking of refinery feed. The particular emission event observed on 13 May 2009 may have been the result of such activities.

[41] Our study demonstrates the value of real-time mobile measurements as a complement to remote sensing in deducing the spatial and temporal variations of industrial emissions. When paired with accurate inverse modeling of reactive chemical species, such measurements offer an advantage over short-range lidar or infrared remote sensing techniques in that they can be conducted in stand-off mode, i.e., outside industrial fence lines.

[42] Source attribution using a combination of real-time mobile measurements and adjoint modeling on neighborhood scales can best be improved by more accurate characterization of local wind flow. Portable wind lidars that can scan horizontally on neighborhood scales are available from commercial vendors. Wind measurements from such instruments can be used to provide model-assimilated wind fields based on sophisticated computational fluid dynamics models and high-resolution urban morphology data. Such advanced treatments will further improve our ability to accurately quantify industrial emission events, which are not adequately resolved by current emission inventories used in atmospheric science or regulatory applications.

Acknowledgments

[43] We would like to acknowledge Bernhard Rappenglück for providing access to the University of Houston's Piper Aztec data set for the FLAIR campaign and Jochen Stutz of the University of California at Los Angeles for his helpful comments and suggestions. This work was supported by the U.S. Department of the Interior, Fish and Wildlife Service, Coastal Impact Assistance Program through Harris County, Texas.

Ancillary