Modeling the uncertainty of several VOC and its impact on simulated VOC and ozone in Houston, Texas

Atmospheric Environment 120 (2015) 404e416

Contents lists available at ScienceDirect

Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv

Modeling the uncertainty of several VOC and its impact on simulated VOC and ozone in Houston, Texas Shuai Pan, Yunsoo Choi*, Anirban Roy, Xiangshang Li, Wonbae Jeon, Amir Hossein Souri Department of Earth and Atmospheric Sciences, University of Houston, 312 Science & Research Building 1, Houston, TX 77204, USA

h i g h l i g h t s
Estimating the uncertainty of VOC in Houston.
Adjusting the VOC emissions using in-situ data.
Investigating the impact of adjusted VOC on ozone.
Validating the adjusted VOC emission using remote sensing.
Finding the peaked ozone in the outflow region.

a r t i c l e i n f o

a b s t r a c t

Article history: Received 1 April 2015 Received in revised form 7 September 2015 Accepted 8 September 2015 Available online 11 September 2015

A WRF-SMOKE-CMAQ modeling system was used to study Volatile Organic Compound (VOC) emissions and their impact on surface VOC and ozone concentrations in southeast Texas during September 2013. The model was evaluated against the ground-level Automated Gas Chromatograph (Auto-GC) measurement data from the Texas Commission on Environmental Quality (TCEQ). The comparisons indicated that the model over-predicted benzene, ethylene, toluene and xylene, while under-predicting isoprene and ethane. The mean biases between simulated and observed values of each VOC species showed clear daytime, nighttime, weekday and weekend variations. Adjusting the VOC emissions using simulated/ observed ratios improved model performance of each VOC species, especially mitigating the mean bias substantially. Simulated monthly mean ozone showed a minor change: a 0.4 ppb or 1.2% increase; while a change of more than 5 ppb was seen in hourly ozone data on high ozone days, this change moved model predictions closer to observations. The CMAQ model run with the adjusted emissions better reproduced the variability in the National Aeronautics and Space Administration (NASA)'s Ozone Monitoring Instrument (OMI) formaldehyde (HCHO) columns. The adjusted model scenario also slightly better reproduced the aircraft HCHO concentrations from NASA's DISCOVER-AQ campaign conducted during the simulation episode period; Correlation, Mean Bias and RMSE improved from 0.34, 1.38 ppb and 2.15 ppb to 0.38, 1.33 ppb and 2.08 ppb respectively. A process analysis conducted for both industrial/ urban and rural areas suggested that chemistry was the main process contributing to ozone production in both areas, while the impact of chemistry was smaller in rural areas than in industrial and urban areas. For both areas, the positive chemistry contribution increased in the sensitivity simulation largely due to the increase in emissions. Nudging VOC emissions to match the observed concentrations shifted the ozone hotspots outside the industrial/urban region and enhanced the peaked ozone in the outflow region with consistent southerly/southeasterly winds during the afternoon time (1e5 pm). This study helps in the understanding of these processes which are critical to constrain high peak ozone values in the outflow regions. The results indicate that formation of ozone in the outflow could complicate attainment status in neighboring counties. © 2015 Elsevier Ltd. All rights reserved.

Keywords: VOC Emission Modeling Ozone Process analysis OMI HCHO Houston

1. Introduction * Corresponding author. E-mail address: [email protected] (Y. Choi). http://dx.doi.org/10.1016/j.atmosenv.2015.09.029 1352-2310/© 2015 Elsevier Ltd. All rights reserved.

Volatile Organic Compounds (VOCs) can be a significant contributor to the criteria pollutant ground-level ozone in urban

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

areas, where ozone concentrations are typically VOC-sensitive (e.g., Choi et al., 2012; Choi, 2014; Choi and Souri, 2015a, 2015b). Examples of VOCs include aromatics such as toluene and unsaturated hydrocarbons such as ethylene which have high ozone forming potential (Carter, 1994). Additionally, VOCs in urban regions often comprise of several air toxic species such as benzene and toluene which have a high potential to cause cancer. Ryerson et al. (2003) mentioned that the emissions of light molecular weight alkenes such as ethylene, propylene, 1,3-butadiene, and butenes are important to explain rapid ozone formation in the Houston area through the Texas Air Quality Study (TexAQS) 2000 aircraft campaign measurements. Vizuete et al. (2008) simulated a series of high industrial point source emissions for the VOC species, npentane, ethylene, propene and o-xylene by using the CAMx model and concluded that ozone concentrations showed the highest sensitivity to o-xylene. The concentrations of the VOC species usually show significant temporal and spatial variations; hence the uncertainty of VOC emissions is one of the major contributions to simulated ozone bias in chemical transport models. Byun et al. (2007) showed ethylene concentrations exhibited significant variability over high ozone episode days as well as pronounced diurnal cycles. The authors indicated that these phenomena were representative of the dilution due to the increased mixing and chemistry. Nam et al. (2006) simulated a series of short-term (1e2-h) release of Highly Reactive Volatile Organic Compounds (HRVOCs) from industrial point sources and indicated that 11 out of 793 emission events produced more than 10 ppb of additional ozone and that 4 out of 793 events produced more than 70 ppb of additional ozone in the industrial area. Model simulations by Webster et al. (2007) using CAMx indicated that the changes in the industrial point source emissions had the potential to cause changes of 10e52 ppb (13e316%) or more in simulated ozone concentrations. Couzo et al. (2012, 2013) analyzed the long term (2000e2009 and 2000e2011, respectively) ground-level ozone measurements and found that in Houston, most non-typical ozone changes were measured at monitors near the ship channel, especially when monitors were downwind of petrochemical facilities. The Houston metropolitan area is characterized by high population density and a large cluster of petrochemical industries. It is the fourth-largest metropolitan area in the United States (US Census Bureau, 2012) and is classified as a nonattainment area for ozone (U.S. EPA's Green Book, 2015). The presence of anthropogenic sources (Czader et al., 2008; Czader and Rappenglück, 2015) such as the petrochemical industry and transportation and biogenic sources (Li et al., 2007) such as vegetation makes the pollutant mix in this area unique in the region. Therefore it is important to understand and quantify the VOC emissions in this region so as to develop appropriate control policies for better air quality and correspondingly attainment status and human health. The modeling study by Kim et al. (2011) indicated that the uncertainties of VOC emissions in the industrial regions affected the surface ozone over Houston during the TexAQS 2006 aircraft campaign. Choi et al. (2012), Choi (2014) and Choi and Souri (2015b) indicated that the urban area of Houston is represented as an extreme NOxsaturated or VOC-sensitive area in the model; one of the reasons for this is the uncertainties of the ozone precursors' emissions in the EPA National Emission Inventory of 2005 or 2011. These studies showed that the high ozone areas in the outflow regions were also influenced by the chemical characteristics of the urban core or industrial regions, where the concentrated ozone precursors may transport to rural areas under proper meteorology. Thus, understanding the chemical conditions of urban or industrial areas and outside of the urban cores in Houston is critical to constrain high ozone values both in source and outflow regions.

405

This study used a WRF-SMOKE-CMAQ modeling system to simulate VOC and surface ozone concentrations in southeast Texas for the month of September 2013, during which the NASA DISCOVER-AQ campaign took place in Houston providing an abundance of measurements which can be utilized for model evaluations. Rather than concentrating on the impact on ozone in the industrial area due to the uncertainty of the point source of the VOC emissions (Nam et al., 2006; Webster et al., 2007; Olaguer et al., 2009; Vizuete et al., 2008, 2011), this study focused on the impact on ozone both in the Houston industrial and its outflow regions by the uncertainties of the VOC emissions. The ambient concentrations of various VOCs from the routine surface measurement sites were used, and the model/measurement VOC concentration ratios were used to adjust the VOC emissions. In particular, in this study, the adjusted VOC emissions were indirectly evaluated by comparing model simulated formaldehyde (a proxy for ozone precursor, VOCs) with the NASA's OMI measurements (e.g., Shim et al., 2005; Choi et al., 2010, 2012; Choi, 2014; Duncan et al., 2014; Choi and Souri, 2015b). In TexAQS 2000 studies, TCEQ made its ‘imputed’ emissions by selectively increasing emissions of HRVOC, such as ethylene and propylene, from the regular Texas Emission Inventory (TEI) industrial point sources inside the Houston-Galveston-Brazoria eight counties with the assumption that point sources released the imputed alkene species into the atmosphere (Byun et al., 2007). The purpose of the adjustment of the VOC emissions in this study was to tune VOC concentrations in the model similar to the corresponding observed data and see how the adjustment of VOC emissions impacted the surface VOC and ozone concentrations both in the industrial areas and their outflow regions. 2. Methodology Simulations were performed with the Community Multi-scale Air Quality (CMAQ) model (Byun and Schere, 2006) version 5.0.1 released by the U.S. Environmental Protection Agency (EPA). The model set-up follows Czader et al. (2015). The current analysis is based on the simulations performed with a 4 km grid for the domain covering southeast Texas, with 84 grid cells in the eastwest direction, 66 in the north-south direction, and 27 vertical layers from surface to 100 hPa. Boundary conditions are obtained from the real-time Air Quality Forecasting system at University of Houston (AQF-UH) (http://spock.geosc.uh.edu/) employing a larger 12 km grid covering the United States, southern Canada and northern Mexico. Initial conditions are based on the 4 km AQF-UH predictions from nested southeast domain. Chemistry is simulated with the Carbon Bond 5 (CB-05) chemical mechanism (Yarwood et al., 2005). Emissions were modeled using the Sparse Matrix Operator Kernel Emissions (SMOKE) system (Houyoux et al., 2000) using the US EPA's 2008 National Emission Inventory (henceforth called NEI-2008). Meteorology was simulated with the Weather Research and Forecasting (WRF) model version 3.5 (Skamarock and Klemp, 2008). For this study, the inputs for WRF are National Centers for Environmental Prediction (NCEP) North American Regional Reanalysis (NARR) data provided by the NOAA/OAR/ESRL PSD (Mesinger et al., 2004). Conversion of the WRF output to CMAQ inputs are performed with the MeteorologyeChemistry Interface Processor (MCIP) (Byun and Schere, 2006). Simulations were performed for the month of September 2013. The weather during the month was relatively dry with mostly southerly, easterly or southeasterly winds. From 09/05 to 09/19, there was lack of influence of strong synoptic weather systems. The wind pattern was light northeasterly in the early morning; gradually turning clockwise to southeasterly in the afternoon and evening hours. A cold front passed through in late of 09/20. Rain events

406

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

occurred on 09/02, 09/10, 09/16, 09/19 to 09/21 and 09/28 to 09/30. None of them was heavy. The 09/20 and 09/21 events consisted of widespread light to medium showers. The in-situ measurement data from the TCEQ's Auto-GC system were used for this study. Fig. 1 shows the location of VOC monitoring sites around Houston. The CB-05 mechanism explicitly simulates only six VOC species included in the Auto-GC measurements, namely ethylene, ethane, isoprene, benzene, toluene and xylene. Although other VOC species such as propylene and 1,3butadiene are highly reactive and important to ozone chemistry, they could not be evaluated one-to-one because they were lumped into the CB-05 mechanism in the model. In-situ measurement data for ozone and meteorological variables were drawn from the TCEQ's Continuous Ambient Monitoring Stations (CAMS) system (http://www.tceq.texas.gov/cgi-bin/compliance/monops/daily_ average.pl). The Smithsonian Astrophysical Observatory (SAO) OMI HCHO lez Abad et al., 2015) used in this study had a spatial data (Gonza resolution of 0.25  0.25 and 13:30 local overpass time. The data were retrieved from the NASA Data and Information Services Center (http://disc.sci.gsfc.nasa.gov/Aura/data-holdings/OMI/ omhcho_v003.shtml). This updated OMI product showed improvements in both temporal stability and noises, and was corrected for an apparent increase of background values compared to the previous version. HCHO observations with cloud cover fraction (i.e., O2eO2 OMI cloud product) higher than 0.4, lower than the sensor detection limit, and without good quality flag were filtered out. The CMAQ model was first run using NARR meteorology and the standard NEI-2008. This case will henceforth be referred to as the Base Case (BASE). An evaluation was done using the Auto-GC measurements to understand model-measurement uncertainty of VOC concentrations over Houston. Using these comparisons, a modeled-to-observed ratio of each VOC was calculated for each site. Next, a site-average ratio was calculated for each VOC species. With the assumption that the ambient VOC concentrations are proportional to the amount of those emissions, these ratios were used to adjust the emissions inventory for these species, and the model was re-run using the adjusted emissions. This case will be referred to as the Sensitivity Case (SENS). The time-series of the VOC concentrations was plotted and several statistics such as Correlation (Corr), Index of Agreement (IOA) and Mean Bias (MB) were calculated to understand the impact of the two emissions scenarios. The impact of the uncertainty of the VOC emissions on ozone simulations was assessed. An Integrated Process Rates (IPR)

Fig. 1. Location of the VOC monitoring sites in Houston industrial regions.

analysis using CMAQ process analysis module was conducted to examine the contribution change of individual processes to ozone concentrations. Finally, the impact of the VOC emissions on hotspots outside the industrial/urban regions was assessed. 3. Results 3.1. Evaluation of ethylene, ethane, isoprene, benzene, toluene and xylene for the base case The base-run model performance of the six VOCs was evaluated as shown in Table 1. First column represents each VOC species. For the second column, ALL indicates the averaged value for all the 8 sites. The ideal number of data points (N) for each site equals to 720 (hourly data, 24 h/day times 30 days); the ideal N for ALL is 5760 (720  8). The actual N is usually less due to lack of measurements or valid data. The evaluation metrics are calculated based on hourly data of all the sites for the simulation episode. Compared to correlation, IOA is generally considered to be a more comprehensive measure of how well the concentrations are predicted as it take into account not only scattering of data but also differences in the means and variances (Willmott, 1981). The last column indicates the ratios of modeled values divided by observations (MM/OM), used to adjust model emissions. The MM/OM value corresponding to each site is called Each Site monthly Mean Ratio (ESMR); the overall MM/ OM value at all sites it is called All Sites monthly Mean Ratio (ASMR). For ethylene, the overall correlation of model and Auto-GC measurements for 8 sites is 0.25 which is relatively low. However, the correlation values are relatively acceptable for some specific sites such as Milby Park and Cesar Chavez, 0.60 and 0.55, respectively. The overall IOA for ethylene is 0.42, bigger than correlation. The IOA numbers for Milby Park and Cesar Chavez are 0.75 and 0.71 respectively which means the model values followed the observation at these sites quite closely. MB is equal to the difference between Model Mean (MM) and Observed Mean (OM). The overall OM was 1.7 ppb and the overall base-run MM 2.8 ppb. Hence, the overall MB is 1.1 ppb, a positive value, which means the model is over-predicting ethylene generally. For ethane, the overall IOA is 0.51. The IOAs at each site show very little variability (0.48e0.56) suggesting that model performance for ethane is constrained over the domain. The overall measured ethane mean is 7.8 ppb. The Wallisville Rd site reported the highest OM value of 10.8 ppb. However, it also shows the lowest IOA of 0.48. The MB at all the sites are negative values indicating that ethane is under-predicted by the model. Like ethane, isoprene is also under-predicted by the model. The overall simulated mean (0.10 ppb) is about one third of the observed mean (0.31 ppb). The HRM-3 Haden Rd site has the biggest mean bias (0.58 ppb) due to its significant high observed mean (0.66 ppb); and the modeled value is about only one seventh of the observation (ESMR ¼ 0.14). The order of the underestimation of isoprene simulation in the model varies as we use different meteorological results by changing the input data for WRF (e.g., NARR, NARR with observational objective analysis or North American Mesoscale Forecast System, NAM) (not shown), but all the modeling exercises showed that the model significantly underpredicted the daytime isoprene concentrations in the Houston industrial area: the authors are working to investigate the impact of the uncertainties of the meteorology on the simulated isoprene concentrations. The details are not included here as they are beyond the scope of this study. For aromatics, as shown in Table 1, the overall month-long observed mean of benzene, toluene and xylene are 0.44 ppb, 0.64 ppb and 0.35 ppb respectively. The overall modeled mean of

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

407

Table 1 Evaluation of the six VOC species for the base case simulation. Species

Site name

N

Corr

IOA

RMSE

MAE

MB

OM

MM

MM/OM

Ethylene

ALL Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez ALL Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez ALL Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez ALL Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez ALL Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez ALL Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez

4776 526 512 576 641 625 614 641 641 4776 526 512 576 641 625 614 641 641 4352 526 512 576 217 625 614 641 641 4766 526 511 576 641 625 605 641 641 4766 526 511 576 641 625 605 641 641 4766 526 511 576 641 625 605 641 641

0.25 0.41 0.47 0.13 0.60 0.53 0.30 0.28 0.55 0.40 0.52 0.38 0.43 0.45 0.33 0.32 0.44 0.46 0.23 0.65 0.32 0.07 0.34 0.21 0.55 0.33 0.35 0.18 0.21 0.27 0.29 0.56 0.29 0.52 0.10 0.59 0.32 0.35 0.35 0.27 0.51 0.31 0.60 0.31 0.50 0.34 0.34 0.62 0.12 0.52 0.30 0.63 0.40 0.60

0.42 0.49 0.62 0.25 0.75 0.65 0.52 0.35 0.71 0.51 0.55 0.52 0.56 0.49 0.49 0.48 0.56 0.51 0.44 0.58 0.49 0.45 0.52 0.49 0.66 0.54 0.53 0.34 0.41 0.44 0.31 0.73 0.48 0.59 0.24 0.66 0.46 0.34 0.53 0.32 0.59 0.41 0.52 0.40 0.60 0.51 0.51 0.61 0.29 0.57 0.41 0.48 0.57 0.75

4.9 3.2 2.5 8.5 1.8 1.8 3.6 8.8 1.5 9.2 6.6 5.6 7.7 9.1 7.1 15.1 9.4 8.6 0.4 0.3 0.3 0.9 0.2 0.3 0.3 0.2 0.3 1.5 0.9 0.8 1.7 0.5 0.8 0.5 3.3 0.5 1.5 1.2 1.0 2.5 1.3 1.4 0.9 1.8 1.5 0.7 0.5 0.4 1.3 0.8 0.8 0.5 0.7 0.5

2.2 2.1 1.3 3.4 1.2 1.1 2.0 5.3 0.9 5.1 4.4 3.2 5.2 5.6 4.7 7.8 5.2 4.4 0.2 0.3 0.2 0.6 0.2 0.2 0.2 0.1 0.2 0.5 0.6 0.2 0.9 0.3 0.4 0.3 1.4 0.3 0.8 0.8 0.5 1.6 0.8 0.9 0.5 1.1 0.6 0.4 0.3 0.3 0.7 0.5 0.5 0.3 0.4 0.3

1.14 1.26 0.46 1.31 0.49 0.51 0.06 4.59 0.39 4.32 4.18 1.97 4.45 5.39 4.24 6.46 3.64 3.81 0.20 0.25 0.14 0.58 0.13 0.18 0.11 0.03 0.19 0.17 0.23 0.02 0.76 0.00 0.10 0.21 0.45 0.20 0.57 0.54 0.17 1.32 0.63 0.63 0.46 0.87 0.05 0.26 0.09 0.22 0.31 0.44 0.37 0.26 0.27 0.10

1.67 1.72 1.14 2.68 1.36 1.28 2.11 1.99 1.10 7.83 7.24 5.23 8.67 8.18 6.87 10.82 8.03 7.16 0.31 0.36 0.22 0.66 0.18 0.24 0.27 0.16 0.29 0.44 0.51 0.27 0.43 0.39 0.48 0.23 0.79 0.43 0.64 0.65 0.53 0.68 0.76 0.72 0.31 0.46 1.01 0.35 0.42 0.20 0.48 0.37 0.42 0.13 0.30 0.44

2.81 2.98 1.60 3.99 1.85 1.79 2.05 6.58 1.49 3.52 3.06 3.26 4.22 2.79 2.63 4.37 4.39 3.35 0.10 0.11 0.08 0.09 0.05 0.07 0.17 0.12 0.10 0.61 0.74 0.29 1.19 0.39 0.38 0.43 1.24 0.23 1.22 1.19 0.70 2.00 1.38 1.34 0.77 1.33 0.96 0.61 0.51 0.42 0.80 0.81 0.79 0.39 0.57 0.54

1.68 1.73 1.40 1.49 1.36 1.40 0.97 3.31 1.35 0.45 0.42 0.62 0.49 0.34 0.38 0.40 0.55 0.47 0.32 0.31 0.36 0.14 0.28 0.29 0.63 0.75 0.34 1.39 1.45 1.07 2.77 1.00 0.79 1.87 1.57 0.53 1.91 1.83 1.32 2.94 1.82 1.86 2.48 2.89 0.95 1.74 1.21 2.10 1.67 2.19 1.88 3.00 1.90 1.23

Ethane

Isoprene

Benzene

Toluene

Xylene

Notation: N e Number of data points; Corr e Correlation; IOA e Index of Agreement; RMSE e Root Mean Square Error; MAE e Mean Absolute Error; MB e Mean Bias; OM e Observed Mean; MM e (Base case) Model Mean; MM/OM e (Base case) Model Mean divided by Observed Mean. Units for RMSE/MAE/MB/OM/MM: ppb.

the 3 species are all larger than observed mean which indicates model generally over-predicts benzene (ASMR ¼ 1.39), toluene (ASMR ¼ 1.91) and xylene (ASMR ¼ 1.74). The Lynchburg Ferry site showed the highest observed benzene numbers (0.79 ppb); Cesar Chavez the highest observed toluene (1 ppb) and HRM-3 Haden Rd the highest xylene measurements (0.48 ppb), which suggested different distribution patterns between VOC species and spatial variations for each VOC. Lynchburg Ferry has lowest benzene IOA (IOA ¼ 0.24), and HRM-3 Haden Rd has lowest toluene IOA (IOA ¼ 0.32) and lowest xylene IOA (IOA ¼ 0.29). When we calculated the month-long observed mean of VOC species, we did not exclude the occasional spikes of relatively high or sometimes extremely high values, which might be caused by episodic flare

VOC emissions (e.g., Murphy and Allen, 2005; Nam et al., 2006; Webster et al., 2007; Vizuete et al., 2008). Higher monthly observed mean values might indicate more spikes happened at this site which could be one possible reason behind the lower IOA. Another reason of the low IOA could be the uncertainty in speciation profiles used to construct the emissions inventory. Overall, the model over-predicted benzene, ethylene, toluene and xylene and under-predicts ethane and isoprene. The biases between simulated and observed VOCs were relatively large (up to 200%). Similar magnitudes were also seen from other air quality studies (e.g., Byun et al., 2007; Li et al., 2007; Kim et al., 2011). These biases could be attributed to uncertainty in the emission inventories, primary radicals' concentrations in the model and PBL

408

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

height dynamics modeling (Vizuete et al., 2008; Olaguer et al., 2009; Czader et al., 2012). Table S1 in the Supplement document showed monthly mean values of observation, base case simulation and biases at daytime, nighttime, weekday and weekend. Isoprene is mainly from biogenic sources and the daytime observed and simulated values were significant larger than those in nighttime and varied little from weekday to weekend. The other VOCs generally showed larger values in nighttime than daytime and increased biases in nighttime which may be attributed to the impact of the change in the simulated PBL height and vertical transport, in addition to their emission inventory uncertainty; although one-month simulation may not be enough to describe the weekday/weekend patterns, both the observed and simulated VOC concentrations decreased from weekday to weekend, similar to the results by Blanchard et al. (2008) that reported a moderate reduction (19%) in ambient VOC concentrations on weekends over large cities in U.S. over the period of 1998e2003.

3.2. Evaluation of ethylene, ethane, isoprene, benzene, toluene and xylene for the sensitivity case The monthly mean ratios of base case simulation to observation for ALL sites (ASMR) and each site (ESMR) for each species were obtained using the MM/OM ratios in Table 1. These ratios were applied to adjust the SMOKE output emission files. Even though there is some variability of the ratios for each species, we used the specific ratios for measurement grid cells and site-wide averaged ratios for other grid cells due to the limitation of the spatial coverage of the measurement data. For the model grid cells hosting each Auto-GC measurement site, each VOC emission value was divided by the ESMR corresponding to the site; while for other grid cells, VOC emission values were divided by the ASMR. This method

was applied to all the six available VOC species. The distribution of difference between adjusting VOC emissions and baseline emissions were shown in Fig. 2. A slight decrease of monthly mean of sum of six VOC emissions in urban and industrial areas (Fig. 2(a)) was caused by the reduction of ethylene and aromatics, emissions of which were elevated in industrial area and urban core (Fig. 2(d) and Fig. S1). A relatively large amount of emission increase in northern Houston was caused by increase of isoprene and ethane emissions which were high in this region (Fig. 2(b) (c)). A substantial amount of HCHO can be produced by oxidation of HRVOCs such as isoprene and ethylene (Abbot et al., 2003; Couzo et al., 2013). In order to evaluate the amount of the emission increase, we compared the OMI HCHO column and the simulated HCHO columns from CMAQ runs with two different emissions (i.e., BASE and SENS), as shown in Fig. 3. Since each satellite-measured HCHO concentration bears large uncertainty (e.g., Duncan et al., 2014), the averaged HCHO columns over the land part of the simulation domain was considered. Overall, the results indicated that the simulated HCHO columns using adjusted emission were closer to OMI HCHO values in term of RMSE (22%) and MAE (10%) metrics in respect to BASE emission, which indicated the relatively significant emission increase was reasonable (also can be indicated by comparing the spatial distribution of HCHO columns (Fig. S2)). The other time series comparisons of HCHO mixing ratios between two simulations and aircraft measurements (Fig. S3) indicated that compared with the HCHO from the baseline simulation, the HCHO peaks in SENS simulation increased and matched with measurements slightly better; R, Mean Bias and RMSE improved from 0.34, 1.38 ppb and 2.15 ppb to 0.38, 1.33 ppb and 2.08 ppb respectively. The sensitivity case model performance of the six VOC species was evaluated as shown in Table 2. Comparison of Table 2 with Table 1 for ethylene, ethane and isoprene indicates that the

Fig. 2. The difference between monthly mean surface adjusting emissions (SENS) and baseline emissions (BASE) for September 2013 for (a) sum of six VOC species; (b) Isoprene; (c) Ethane and (d) Ethylene (Note the different color-bar scales for Ethylene). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

409

Fig. 3. Time series of averaged HCHO column between OMI measurements, simulated base case (BASE) and sensitivity case (SENS). Only HCHO columns over the land part of research domain were considered.

sensitivity-run overall model means became quite closer to observed means than the base-run model values. The overall MBs of the 3 species decrease from 1.1 ppb, 4.3 ppb and 0.20 ppb to 0.16 ppb, 0.38 ppb and 0.03 ppb respectively. The overall IOA for the 3 species all increased. Among the 8 sites, the IOA for ethylene, ethane and isoprene at Channelview site all showed significant improvement, changing from 0.49, 0.55 and 0.58 to 0.64, 0.72 and 0.78 respectively. These changes could be partly because at this site the base-run ESMR of the 3 species applied to adjust emissions were all relatively closer to the corresponding ASMR. The results indicate that the practice to adjust the emissions of these three VOCs (ethylene, ethane and isoprene) improved the simulation results. Comparison of Table 2 with Table 1 indicates that the overall MB of benzene, toluene, and xylene changed from 0.17 ppb, 0.57 ppb and 0.26 ppb to 0.01 ppb, 0.01 ppb and 0 ppb respectively. The trends indicate a significant reduction. The overall correlations for the three aromatics improved slightly. The overall IOA for toluene and xylene increased from 0.46 to 0.55 and from 0.51 to 0.53 respectively; while the overall IOA for benzene decreased from 0.34 to 0.31. Also, this sensitivity test produced a better comparison of the three aromatics with corresponding observations. The distributions of hourly ratios of 8 sites averaged model simulated values to Auto-GC measurements were plotted in Fig. 4. The comparison of SENS/OBS to the BASE/OBS showed that the distribution ranges of emission-reduced species along with y-axis became significantly smaller. Also, the medians became closer to 1 for all the six VOCs. Fig. 5 shows several reasonably good time series comparisons between measured and model simulated VOC mixing ratios. For example, the BASE toluene mixing ratios at the Clinton site generally followed the daily variations of the Auto-GC measurements, but the magnitude was significantly higher. After adjusting emissions in the sensitivity case, the SENS toluene concentrations decreased significantly, approaching the observation. To summarize, after the emissions of benzene, ethylene, toluene and xylene were decreased, and the emissions of ethane and isoprene increased, the ratio of model to observation value for VOC species showed that the sensitivity-run overall model means became quite closer to observed means, although among some individual observation sites or during some specific hours, this ratio still revealed high discrepancy.

3.3. Change of simulated ozone by using adjusted VOC emissions Once the surface concentrations of the six VOCs were adjusted, the impacts of the changed VOC concentrations on ozone were quantified. Ozone statistics were calculated for 52 sites located in the domain and two simulations, and listed in Table 3. Only the 52

sites averaged (ALL), six industrial sites which have both VOC and ozone measurements, and several suburban and rural sites were listed. In terms of the monthly mean values, BASE simulation were over-predicted for ALL and most of the sites, and the SENS simulated mean ozone became larger, a 0.4 ppb or 1.2% increase over the domain. The increase between SENS and BASE is relatively bigger at rural sites than industrial sites (i.e. MB D value is bigger). The correlation and IOA didn't change much. Fig. 6 plots the time series of ozone mixing ratios between CAMS measurements and two simulations over the month. The HRM-3 Haden Road site was chosen as a representative industrial site as it is surrounded by other stations and near the center of ship channel; the UH WG Jones Forest site (whose location was shown in Fig. 2(b), also inside high isoprene emission enhancement area) in northern Houston was chosen as the representative rural site. At HRM-3 Haden Road site, from September 1st to 18th, the model simulated values generally captured the observation trends well, except during September 4th-8th and 15th-18th when the model over-predicted to some extent. Additionally, during September 19th and 20th, the model over-predicted significantly due to low measurements values caused by precipitation. The error could be attributed to the high uncertainties in the simulations of cloud fraction and other meteorological variables (e.g. Pour-Biazar et al., 2007). On September 25th, a high exceedance of ozone was observed during the afternoon time when the model significantly under-predicted. The high ozone bias may be partly due to the uncertainty of the simulated wind patterns such as recirculation and/or stagnation (e.g., Rappenglück et al., 2008). After that, both two model cases and observation started deviating significantly but still correlated with each other to some extent. Simulated ozone at the UH WG Jones Forest site had similar over-/under-prediction patterns over the month, except the high observed ozone peak which appeared on the 26th, when the high-peaked ozone formed in the industrial region on the previous day was transported by southeasterly air flows and past this site. Eder et al. (2009) evaluated the performance of National Oceanic and Atmospheric Administration (NOAA) National Air Quality Forecast Capability (NAQFC) on 8-h max ozone prediction during the summer of 2007, and reported that much of model over-prediction was associated with observed concentrations less than 50 ppb and model was more likely to under-predicted when concentrations exceeded 70 ppb. It is interesting that the statement is also suited to describe the hourly ozone comparisons shown from our simulations. Comparisons for the same episode but with zoomed in y-axis are plotted in Fig. S4, so the difference between SENS and BASE could be clearly seen. Generally, at both sites the SENS values had expected increases at early afternoon; and increase at rural site was bigger than at industrial site. It can be noticed that the sensitivity-run peaked

410

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

Table 2 Evaluation of the six VOC species for the sensitivity case simulation. Species

Site name

N

Corr

IOA

RMSE

MAE

MB

OM

MM

MM/OM

MM/OM D

Ethylene

All Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez All Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez All Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez All Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez All Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez All Channelview Deer Park2 HRM-3 Haden Rd Milby Park Clinton Wallisville Rd Lynchburg Ferry Cesar Chavez

4776 526 512 576 641 625 614 641 641 4776 526 512 576 641 625 614 641 641 4352 526 512 576 217 625 614 641 641 4766 526 511 576 641 625 605 641 641 4766 526 511 576 641 625 605 641 641 4766 526 511 576 641 625 605 641 641

0.28 0.41 0.48 0.13 0.59 0.52 0.3 0.3 0.55 0.4 0.53 0.37 0.43 0.46 0.33 0.32 0.43 0.47 0.28 0.64 0.3 0.11 0.28 0.21 0.57 0.36 0.28 0.19 0.21 0.28 0.27 0.55 0.29 0.52 0.11 0.6 0.37 0.35 0.37 0.26 0.5 0.33 0.6 0.34 0.51 0.36 0.34 0.61 0.12 0.52 0.3 0.64 0.41 0.61

0.44 0.64 0.65 0.22 0.72 0.71 0.45 0.53 0.68 0.6 0.72 0.57 0.6 0.64 0.55 0.53 0.62 0.66 0.5 0.78 0.49 0.44 0.39 0.42 0.61 0.46 0.47 0.31 0.43 0.42 0.41 0.7 0.44 0.68 0.2 0.6 0.55 0.54 0.56 0.46 0.68 0.56 0.75 0.54 0.48 0.53 0.55 0.76 0.26 0.7 0.53 0.73 0.59 0.69

3.4 1.8 1.7 7.3 1.5 1.3 3.4 3.8 1.3 9.8 5.9 7.1 9 8.1 7.1 16.7 11.4 8.3 0.5 0.3 0.3 0.8 0.4 0.4 0.5 0.4 0.4 1.3 0.7 0.7 1 0.5 0.8 0.3 3 0.5 0.9 0.6 0.7 1.2 0.7 0.8 0.4 0.9 1.5 0.5 0.4 0.2 1.1 0.4 0.5 0.2 0.5 0.5

1.4 1.2 0.9 2.3 0.9 0.8 1.7 2.3 0.7 5.2 3.8 4.4 5.1 4.8 4 8.5 6.3 4.6 0.3 0.2 0.2 0.6 0.2 0.2 0.3 0.2 0.3 0.4 0.4 0.2 0.6 0.3 0.4 0.2 1.1 0.3 0.5 0.4 0.4 0.7 0.5 0.5 0.2 0.5 0.6 0.3 0.2 0.1 0.4 0.3 0.3 0.1 0.3 0.2

0.16 0.11 0.21 0.34 0.16 0.12 0.8 0.59 0.17 0.38 0.71 1.71 0.2 2.24 1.29 1.45 1.18 0.07 0.03 0.01 0.05 0.28 0.02 0.01 0.2 0.15 0.06 0.01 0.03 0.05 0.29 0.06 0.15 0.09 0.04 0.22 0.01 0.03 0.12 0.2 0.01 0.03 0.07 0.13 0.36 0 0.09 0.03 0.01 0.07 0.01 0.07 0.02 0.09

1.67 1.72 1.14 2.68 1.36 1.28 2.11 1.99 1.1 7.83 7.24 5.23 8.67 8.18 6.87 10.82 8.03 7.16 0.31 0.36 0.22 0.66 0.18 0.24 0.27 0.16 0.29 0.44 0.51 0.27 0.43 0.39 0.48 0.23 0.79 0.43 0.64 0.65 0.53 0.68 0.76 0.72 0.31 0.46 1.01 0.35 0.42 0.2 0.48 0.37 0.42 0.13 0.3 0.44

1.51 1.61 0.93 2.34 1.2 1.16 1.31 2.59 0.93 7.45 6.52 6.94 8.87 5.94 5.58 9.37 9.21 7.09 0.34 0.37 0.27 0.38 0.2 0.25 0.47 0.31 0.36 0.44 0.54 0.22 0.72 0.33 0.33 0.32 0.83 0.2 0.64 0.62 0.41 0.88 0.77 0.75 0.39 0.59 0.65 0.34 0.32 0.23 0.47 0.44 0.43 0.2 0.31 0.34

0.90 0.94 0.82 0.87 0.88 0.91 0.62 1.30 0.85 0.95 0.90 1.33 1.02 0.73 0.81 0.87 1.15 0.99 1.10 1.03 1.23 0.58 1.11 1.04 1.74 1.94 1.24 1.00 1.06 0.81 1.67 0.85 0.69 1.39 1.05 0.47 1.00 0.95 0.77 1.29 1.01 1.04 1.26 1.28 0.64 0.97 0.76 1.15 0.98 1.19 1.02 1.54 1.03 0.77

0.78 0.79 0.58 0.62 0.48 0.49 0.35 2.01 0.50 0.50 0.48 0.71 0.53 0.39 0.43 0.47 0.60 0.52 0.78 0.72 0.87 0.44 0.83 0.75 1.11 1.19 0.90 0.39 0.39 0.26 1.10 0.15 0.10 0.48 0.52 0.06 0.91 0.88 0.55 1.65 0.81 0.82 1.22 1.61 0.31 0.77 0.45 0.95 0.69 1.00 0.86 1.46 0.87 0.46

Ethane

Isoprene

Benzene

Toluene

Xylene

Notation: MM e (Sensitivity case) Model Mean; MM/OM e (Sensitivity case) Model Mean divided by Observed Mean; MM/OM D e ratios Difference between Sensitivity case and Base case; other statistical parameters are same as Table 1.

ozone has about 5 ppb increase at Haden Road site on the 25th of September. However, this was not enough to mitigate the high bias. Hence adjusting the model VOC emissions using month-long averaged ratios may improve VOC predictions, but it might not be an efficient approach to correct high-peaked ozone. In order to estimate the changes in the contribution of atmospheric processes, especially chemistry process to ozone concentrations, a process analysis (PA) was conducted for both BASE and SENS simulations at two areas in the domain as shown in Fig. 2(a), the blue box includes industrial and most part of urban area, and red box covers one portion of rural area (outflowed from the ship channel area due to the southerly winds). The vertical extents of the two areas followed the hourly change in PBL. So

each PA volume was 52 km  40 km  PBL (13 easting model grid cells  10 northing model grid cells, including vertical layers from surface up to PBL). Fig. 7 shows the PA results over the industrial and urban areas for the simulation episode. From both BASE and SENS, the chemistry reaction is the main process contributing to ozone production at daytime (e.g., Kimura et al., 2008; Henderson et al., 2010, 2011), especially from morning to early afternoon. Other positive contributions come from vertical transport (VDIF þ ZADV) from layers above PBL. The results also indicate that horizontal advection is the main sink of ozone. Hence for industrial and urban areas, the normal routine at daytime is chemical reaction producing ozone along which is vertically transported ozone, horizontally outflow to other areas. Compared

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

411

Fig. 4. Box and Whisker plots of the hourly ratios of 8 sites averaged model simulated values to Auto-GC measurements (OBS). (a) BASE to OBS; (b) same as (a), with zoomed in yaxis; (c) SENS to OBS; (d) same as (c), with zoomed in y-axis. Box tops, middles and bottoms give the 75th, 50th and 25th percentile values.

with the PA for the BASE simulation, the PA for SENS simulation showed that the positive chemistry contribution slightly increased, even the total VOC emissions decreases at part of this area as shown in Fig. 2(a). The reason might be that isoprene has relatively high ozone forming potential (Carter, 1994). Two additional sensitivity simulations were performed with PA turned on: (1) reducing only ethylene emissions (SENS-rETH); (2) increasing only isoprene emissions (SENS-iISOP). The PA results of the two new cases were shown in Fig. S5; compare with Fig. 7, SENS-rETH is similar to BASE, while SENS-iISOP is similar to SENS, which indicated that the simulated ozone increase at industrial and urban areas might largely contribute by the increase in isoprene emissions. Fig. S6 shows PA results at rural area for BASE and SENS simulations. The results indicate that chemistry is still a significant process contributing to ozone production in rural areas, while the magnitude is smaller than that in industrial and urban areas. The positive chemistry contribution increases in SENS in the rural area due to emission adjustment, similar to industrial and urban areas. Interestingly, adjustment of VOC emissions/concentrations did not only affect the ozone concentrations in the industrial measurement sites, it also (sometimes) had a definitive influence

outside the industrial/urban regions. As shown in Fig. 8, ozone concentrations in the SENS case increased in several areas outside the industrial area compare to the BASE case. This increase might be due to the net increase of total VOC emissions in SENS case, while the feature of ozone increase does not coincide with that of VOC emissions. This discrepancy might be caused by complexity and non-linearity of ozone formation. The three notable ozone enhancement areas (shown as red, in the web version) in Fig. 8(b) shows general agreement with three noticeable NOx concentration areas shown in Fig. 8(d). Especially, the ozone concentration near north part of study domain did not increase much, despite of significantly increased amount of VOC emissions. This result reveals that the effect of adjusted VOC emissions on ozone formation strongly depends on the background distribution of NOx concentration. The slight difference between three ozone-increased and NOx-rich areas is due to transport of NOx by sea breeze from the south-east direction. In fact, the ozone difference cores appeared right north-west side of NOx concentration cores. Meanwhile, the feature of ozone increase, which is shown in Fig. 8(b), generally agrees with the feature of HO2 increase and NOx decrease shown in Fig. 8(f) and (e), respectively. The increased VOC emissions in SENS case induced the increased HO2 production rate by reacting with

412

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

Fig. 5. Time series of several VOC mixing ratios between Auto-GC measurements (OBS, black dot), simulated base case (BASE, green dash line) and sensitivity case (SENS, red line). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

more OH radical. Consequently, the increased HO2 reacted with more NOx and the increased conversion rate of NOx to NO2 led the decrease in NOx concentration and increase in ozone concentration.

The ozone simulations at several measurement sites in the peaked ozone enhancement area were shown in Fig. S7 in Supplementary document. In summary, the increase of ozone concentration by

Table 3 Ozone statistics between observation, simulated base and sensitivity cases. Site name

N

B Corr

B IOA

B MB

OM

BM

SM

Corr D

IOA D

MB D

ALL Channelview Deer Park2 HRM-3 Haden Rd Clinton Wallisville Rd Lynchburg Ferry UH WG Jones Forest Conroe Relocated Northwest Harris Co.

33,308 656 656 713 708 714 707 652 714 704

0.73 0.74 0.76 0.77 0.70 0.78 0.74 0.72 0.77 0.80

0.80 0.81 0.84 0.85 0.81 0.87 0.84 0.84 0.87 0.85

8.3 8.1 5.9 5.8 5.1 3.3 5.3 0.7 3.0 7.8

24.4 21.7 25.5 23.3 22.3 26.1 23.8 29.3 28.2 22.6

32.7 29.8 31.5 29.0 27.3 29.4 29.1 28.5 31.2 30.5

33.2 30.2 31.9 29.5 27.8 29.8 29.5 29.5 32.0 31.3

0 0.01 0 0.01 0.01 0 0 0 0 0

0 0 0 0.01 0.01 0 0.01 0 0 0

0.4 0.4 0.4 0.4 0.5 0.4 0.3 0.9 0.8 0.9

Notation: N e Number of data points; B Corr, B IOA, B MB e Base case Correlation, Index of Agreement, Mean Bias; OM e Observed Mean; BM e Base case model Mean; SM e Sensitivity case model Mean; Corr D, IOA D, MB D e Differences of Correlation, Index of Agreement, Mean Bias between Sensitivity case and Base case; Units for B MB/OM/BM/ SM/MB D: ppb.

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

413

Fig. 6. Time series of ozone mixing ratio between CAMS measurements (OBS, black dot), simulated base case (BASE, blue line) and sensitivity case (SENS, red line) at industrial and rural sites. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 7. Contributions of atmospheric processes to ozone concentrations from surface to PBL height at industrial and urban area during Sept. 2013 for BASE and SENS simulations (HADV e horizontal advection, ZADV e vertical advection, HDIF e horizontal diffusion, VDIF e vertical diffusion, DDEP e dry deposition, CLDS e cloud process, CHEM e chemistry reaction).

414

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

Fig. 8. (a) BASE simulated monthly mean of ozone mixing ratio; (b) the difference of monthly mean of ozone mixing ratio between two simulations (SENS e BASE); (c) SENS e BASE monthly mean of sum of six VOC species emissions; (d) BASE simulated monthly mean of NOx mixing ratio; (e) SENS e BASE monthly mean of NOx mixing ratio; (f) SENS e BASE monthly mean of HO2 mixing ratio. Note: the monthly mean values were calculated only for the daytime (13:00e17:00 LT) of September of 2013.

adjusted VOC emissions in SENS depended on complicated factors like the concentration of NOx, VOC and radicals and their interactions. 4. Summary and discussion A WRF-SMOKE-CMAQ modeling system was used to study the uncertainty of the simulated VOC and surface ozone concentrations in southeast Texas during September 2013. The model simulation results were evaluated against the Auto-GC measurement data from the Texas Commission on Environmental Quality. The evaluation statistics indicated that the baseline model over-predicted benzene, ethylene, toluene and xylene, and under-predicted isoprene and ethane. Uncertainty in meteorological fields and speciation profiles could be contributors towards the modelmeasurement biases. The mean biases between simulated and observed values of each VOC species showed clear daytime, nighttime, weekday and weekend variations. Adjusting the VOC emissions using simulated-to-observed ratios improved model

performance of each VOC species, especially mitigating the mean bias substantially. Simulated monthly mean ozone did not change significantly, about 0.4 ppb or 1.2% increase; while a change of more than 5 ppb was seen in hourly ozone data on high ozone days which moved model closer to observations. The results indicated that the simulated formaldehyde columns using adjusted emissions were closer to OMI remote-sensing values with respect to BASE scenario, which indicated the relatively significant emission increase was reasonable. Additionally, the other time series comparisons of formaldehyde mixing ratios between two simulations and aircraft measurements indicated the HCHO peaks in SENS simulation increased and matched with measurements slightly better vis-a-vis the baseline case. A process analysis conducted to both industrial/urban areas and rural area whose vertical extents followed the hourly change in PBL suggested that chemistry reaction is the main process contributing to ozone production in both areas; while the magnitude of chemistry reaction is smaller in rural area than in industrial and urban areas. For both areas, the positive chemistry contribution increased

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

in sensitivity simulation largely due to the emission increase. The impact of adjusting VOC emission on ozone formation largely depends on the background distribution and concentrations of NOx, VOC and radicals and their interactions. Understanding the impact of the difference in the amount of VOC emissions on the high-peaked ozone in the outflow regions is critical to prepare for an efficient air pollution control policy in the Houston metropolitan area. The results suggest that regional transport of these high ozone concentrations could influence nonattainment status in neighboring counties. Additionally, the possible reduction of the NOx emissions in the Houston urban/industrial region from EPA's National Emissions Inventory 2011 (visa-vis NEI-2008) might affect the ozone chemistry relating to the changes in the VOC emissions in the source areas. Czader and Rappenglück (2015) addressed that NEI and TEI from TCEQ for TexAQS 2006 are quite different in terms of VOC emission rates and spatial and temporal patterns. In another word, it would be interesting to check how the changes of emissions from NEI-2008 to NEI-2011 or TEI affect the emissions/concentrations of VOCs, NOx, and ozone chemistry. Also, the uncertainties of other reactive VOC species besides these six VOC species in the Houston area may need to be investigated. Further, it may be that the inaccurate simulation of cloud fraction cover and wind patterns (e.g., sea breeze) by the meteorological model attributed to the discrepancy in the ozone productivity. Hence a potential future direction from this work could be to improve the simulated cloud fraction and wind patterns using remote sensing data. Acknowledgments This work is partially supported by Air Quality Research Program (AQRP) 14-014. Authors acknowledge the free use of the TCEQ Auto-GC VOC measurements and CAMS system ozone measurements, the Smithsonian Astrophysical Observatory OMI and NASA DISCOVER-AQ aircraft campaign HCHO measurements data. Thanks Beata H. Czader for her helpful instructions on the WRF-SMOKECMAQ modeling system setup. Also thanks are extended to Bernhard Rappenglück, Barry Lefer and Robert Talbot at University of Houston, for their useful comments to this study. Finally, we appreciate the anonymous reviewers who helped improve quality of this paper greatly. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.atmosenv.2015.09.029. References Abbot, D.S., Palmer, P.I., Martin, R.V., Chance, K.V., Jacob, D.J., Guenther, A., 2003. Seasonal and interannual variability of North American isoprene emissions as determined by formaldehyde column measurements from space. Geophys. Res. Lett. 30 (17), 1886,. http://dx.doi.org/10.1029/2003GL017336. Byun, D.W., Kim, S.-T., Kim, S.-B., 2007. Evaluation of air quality models for the simulation of a high ozone episode in the Houston metropolitan area. Atmos. Environ. 41, 837e853. Blanchard, C.L., Tanenbaum, S., Lawson, D.R., 2008. Differences between weekday and weekend air pollutant levels in Atlanta; Baltimore; Chicago; Dallas-Fort Worth; Denver; Houston; New York; Phoenix; Washington, DC; and surrounding areas. J. Air Waste Manag. Assoc. 58 (12), 1598e1615. http:// dx.doi.org/10.3155/1047-3289.58.12.1598. Byun, D., Schere, K.L., 2006. Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) Modeling System. Appl. Mech. Rev. 59, 51e77. Carter, W.P.L., 1994. Development of ozone reactivity scales for volatile organic compounds. J. Air Waste Manag. Assoc. 44, 881e899. Choi, Y., Souri, A., 2015b. Chemical condition and surface ozone in large cities of Texas during the last decade: observational evidence from OMI, CAMS, and model analysis. accepted in Remote Sens. Environ.. http://dx.doi.org/10.1016/ j.rse.2015.06.026.

415

Choi, Y., Souri, A., 2015a. Seasonal behavior and long-term trends of tropospheric ozone, its precursors and chemical conditions over Iran: a view from space. Atmos. Environ. 106, 232e240. Choi, Y., 2014. The impact of satellite-adjusted NOx emissions on simulated NOx and O3 discrepancies in the urban and outflow areas of the Pacific and Lower Middle US. Atmos. Chem. Phys. 14, 675e690. http://dx.doi.org/10.5194/acp-14-6752014. Choi, Y., Kim, H., Tong, D., Lee, P., 2012. Summertime weekly cycles of observed and modeled NOx and O3 concentrations as a function of satellite-derived ozone production sensitivity and land use types over the Continental United States. Atmos. Chem. Phys. 12, 6291e6307. http://dx.doi.org/10.5194/acp-12-62912012. Choi, Y., Osterman, G., Eldering, A., Wang, Y., Edgerton, E., 2010. Understanding the contributions of anthropogenic and biogenic sources to CO enhancements and outflow observed over North America and the western Atlantic Ocean by TES and MOPITT. Atmos. Environ. 44, 2033e2042. http://dx.doi.org/ 10.1016/j.atmosenv.2010.01.029. Couzo, E., Jeffries, H.E., Vizuete, W., 2013. Houston's rapid ozone increases: preconditions and geographic origins. Environ. Chem. 10, 260e268. Couzo, E., Olatosi, A., Jeffries, H.E., Vizuete, W., 2012. Assessment of a regulatory model performance relative to large spatial inhomogeneity in observed ozone in Houston, TX. J. Air Waste Manag. Assoc. 62, 696e706. Czader, B.H., Choi, Y., Li, X., Alvera, S., Lefer, B., 2015. Impact of updated traffic emissions on HONO mixing ratios simulated for urban site in Houston, Texas. Atmos. Chem. Phys. 15, 1253e1263. http://dx.doi.org/10.5194/acp-15-12532015. Czader, B.H., Byun, D.W., Kim, S.-T., Carter, W.P.L., 2008. A study of VOC reactivity in the Houston-Galveston air mixture utilizing an extended version of SAPRC-99 chemical mechanism. Atmos. Environ. 42, 5733e5742. Czader, B.H., Rappenglück, B., 2015. Modeling of 1, 3- butadiene in urban and industrial areas. Atmos. Environ. 102, 30e42. http://dx.doi.org/10.1016/ j.atmosenv.2014.11.039. Czader, B.H., Rappenglück, B., Percell, P., Byun, D., Ngan, F., Kim, S., 2012. Modeling nitrous acid and its impact on ozone and hydroxyl radical during the Texas Air Quality Study 2006. Atmos. Chem. Phys. 12, 6,939e6,951. http://dx.doi.org/ 10.5194/acp-12-6939-2012. Duncan, B.N., Prados, A.I., Lamsal, L.N., Liu, Y., Streets, D.G., Gupta, P., Hilsenrath, E., Kahn, R.A., Nielsen, J.E., Beyersdorf, A.J., Burton, S.P., Fiore, A.M., Fishman, J., Henze, D.K., Hostetler, C.A., Krotkov, N.A., Lee, P., Lin, M., Pawson, S., Pfister, G., Pickering, K.E., Pierce, R.B., Yoshida, Y., Ziemba, L.D., 2014. Satellite data of atmospheric pollution for U.S. air quality applications: examples of applications, summary of data end-user resources, answers to FAQs, and common mistakes to avoid. Atmos. Environ. 94, 647e662. http://dx.doi.org/10.1016/j.atmosenv. 2014.05.061. Eder, B., Kang, D., Mathur, R., Pleim, J., Yu, S., Otte, T., Pouliot, G., 2009. A performance evaluation of the National Air Quality Forecast Capability for the summer of 2007. Atmos. Environ. 43, 2312e2320. http://dx.doi.org/10.1016/ j.atmosenv.2009.01.033. lez Abad, G., Liu, X., Chance, K., Wang, H., Kurosu, T.P., Suleiman, R., 2015. Gonza Updated smithsonian astrophysical observatory ozone monitoring instrument (SAO OMI) formaldehyde retrieval. Atmos. Meas. Tech. 8, 19e32. http:// dx.doi.org/10.5194/amt-8-19-2015. Henderson, B.H., Kimura, Y., McDonald-Buller, E., Allen, D.T., Vizuete, W., 2011. Comparison of Lagrangian process analysis tools for Eulerian air quality models. Atmos. Environ. 45 (29), 5200e5211. Henderson, B.H., Vizuete, W., Jeffries, H., Kim, B., 2010. The influence of model resolution on industrial releases and ozone formation. J. Air Waste Manag. Assoc. 60 (9), 1105e1117. Houyoux, M., Vukovich, J., Brandmeyer, J., 2000. Sparse Matrix Kernel Emissions Modeling System: SMOKE User Manual. MCNC-North Carolina Supercomputing Center. Available at: http://www.cmascenter.org/. Kim, S.-W., McKeen, S.A., Frost, G.J., Lee, S.-H., Trainer, M., Richter, A., Angevine, W.M., Atlas, E., Bianco, L., Boersma, K.F., Brioude, J., Burrows, J.P., de Gouw, J., Fried, A., Gleason, J., Hilboll, A., Mellqvist, J., Peischl, J., Richter, D., Rivera, C., Ryerson, T., te Lintel Hekkert, S., Walega, J., Warneke, C., Weibring, P., Williams, E., 2011. Evaluations of NOx and highly reactive VOC emission inventories in Texas and their implications for ozone plume simulations during the Texas Air Quality Study 2006. Atmos. Chem. Phys. 11, 11361e11386. http:// dx.doi.org/10.5194/acp-11-11361-2011. Kimura, Y., McDonald-Buller, E., Vizuete, W., Jeffries, H.E., Allen, D.T., 2008. Application of a Lagrangian process analysis Tool to Characterize ozone formation in Southeast Texas. Atmos. Environ. 42, 5743e5759. Li, G., Zhang, R., Fan, J., Tie, X., 2007. Impacts of biogenic emissions on photochemical ozone production in Houston, Texas. J. Geophys. Res. 112, D10309. http://dx.doi.org/10.1029/2006JD007924, 2007. Mesinger, F., et al., 2004. North American Regional Reanalysis: a long-term, consistent, high-resolution climate dataset for the North American domain, as a major improvement upon the earlier global reanalysis datasets in both resolution and accuracy. submitted to BAMS. Available at: http://www.esrl.noaa. gov/psd/. Murphy, C.F., Allen, D.T., 2005. Hydrocarbon emissions from industrial release events in the Houston-Galveston area and their impact on ozone formation. Atmos. Environ. 39, 3785e3798. http://dx.doi.org/10.1016/j.atmosenv.2005.02.051. Nam, J., Kimura, Y., Vizuete, W., Murphy, C., Allen, D.T., 2006. Modeling the impact of emission events on ozone formation in Houston, Texas. Atmos. Environ. 40,

416

S. Pan et al. / Atmospheric Environment 120 (2015) 404e416

5329e5341. Olaguer, E.P., Rappenglueck, B., Lefer, B., Stutz, J., Dibb, J., Griffin, R., Brune, B., Shauck, M., Buhrg, M., Jeffries, H., Vizuete, W., Pinto, J., 2009. Deciphering the role of radical sources during the second Texas air quality study. J. Air Waste Manag. Assoc. 59 (11), 1,258e1,277. Pour-Biazar, A., McNider, R.T., Roselle, S.J., Suggs, R., Jedlovec, G., Byun, D.W., Kim, S., Lin, C.J., Ho, T.C., Haines, S., Dornblaser, B., Cameron, R., 2007. Correcting photolysis rates on the basis of satellite observed clouds. J. Geophys. Res. 112, D10302. http://dx.doi.org/10.1029/2006JD007422. Ryerson, T.B., Trainer, M., Angevine, W.M., Brock, C.A., Dissly, R.W., Fehsenfeld, F.C., Frost, G.J., Goldan, P.D., Holloway, J.S., Hubler, G., Jakoubek, R.O., Kuster, W.C., Neuman, J.A., Nicks Jr., D.K., Parrish, D.D., Roberts, J.M., Sueper, D.T., Atlas, E.L., Donnelly, S.G., Flocke, F., Fried, A., Potter, W.T., Schauffler, S., Stroud, V., Weinheimer, A.J., Wert, B.P., Wiedinmyer, C., Alvarez, R.J., Banta, R.M., Darby, L.S., Senff, C.J., 2003. Effect of petrochemical industrial emissions of reactive alkenes and NOx on tropospheric ozone formation in Houston, Texas. J. Geophys. Res. 108 (D8), 4249. http://dx.doi.org/10.1029/2002JD003070, 2003. Rappenglück, B., Perna, R., Zhong, S., Morris, G.A., 2008. An analysis of the vertical structure of the atmosphere and upper-level meteorology and their impact on surface ozone levels in Houston, Texas. J. Geophys. Res. 113, D17315. http:// dx.doi.org/10.1029/2007JD009745. Shim, C., Wang, Y., Choi, Y., Palmer, P.I., Abbot, D.S., Chance, K., 2005. Constraining global isoprene emissions with Global Ozone Monitoring Experiment (GOME) formaldehyde column measurements. J. Geophys. Res. 110, D24301. Skamarock, W.C., Klemp, J.B., 2008. A time-split non-hydrostatic atmospheric

model for weather research and forecasting applications. J. Comput. Phys. 227, 3465e3485. U.S. Census Bureau, 2012. In: Annual Estimates of the Population of Metropolitan and Micropolitan Statistical Areas: April 1, 2010 to July 1, 2011 (CBSAEST2011e01). U. S. Census Bureau. U.S. EPA Green Book, 2015. 8eHr Ozone (2008) Nonattainment State/Area/County Report, as of January 30, 2015. Available at: http://www.epa.gov/airquality/ greenbook/hncs.html#TEXAS. Vizuete, W., Kim, B.-U., Jeffries, H., Kimura, Y., Allen, D.T., Kioumourzoglou, M.-A., Biton, L., Henderson, B., 2008. Modeling ozone formation from industrial emission events in Houston. Tex. Atmos. Environ. 42, 7641e7650. http:// dx.doi.org/10.1016/j.atmosenv.2008.05.063. Vizuete, W., Jeffries, H., Tesche, T.W., Olaguer, E.P., Couzo, E., 2011. Issues with ozone attainment methodology for Houston, TX. J. Air Waste Manag. Assoc. 61 (3), 238e253. Webster, M., Nam, J., Kimura, Y., Jeffries, H., Vizuete, W., Allen, D.T., 2007. The effect of variability in industrial emissions on ozone formation in Houston, Texas. Atmos. Environ. 41, 9580e9593. http://dx.doi.org/10.1016/ j.atmosenv.2007.08.052. Willmott, C.J., 1981. On the validation of models. Phys. Geogr. 2, 184e194. Yarwood, G., Rao, S., Yocke, M., Whitten, G.Z., 2005. Updates to the Carbon Bond Chemical Mechanism: CB05. Final Report to the US EPA, RT-0400675, December 8, 2005. Available at: http://www.camx.com/publ/pdfs/CB05_Final_Report_ 120805.pdf.