Methods and Published by Copernicus Publications Data Systems on behalf of the European Geosciences Data Systems Union.

We analyze the simulation results from a CMAQ model and GOME-2 NO2 retrievals over the United States for August 2009 to estimate the model-simulated biases of NOx concentrations over six geological re- gions (Pacific Coast = PC, Rocky Mountains = RM, Lower Middle = LM, Upper Middle = UM, Southeast = SE, North- east = NE). By comparing GOME-2 NO2 columns to cor- responding CMAQ NO2 columns, we produced satellite- adjusted NOx emission ("GOME2009") and compared baseline emission ("BASE2009") CMAQ simulations with GOME2009 CMAQ runs. We found that the latter exhib- ited decreases of 5.6 %, 12.3 %, 21.3 %, and 15.9 % over the PC, RM, LM, and SE regions, respectively, and increases of +2.3 % and +10.0 % over the UM and NE re- gions. In addition, we found that changes in NOx emis- sions generally mitigate discrepancies between the surface NOx concentrations of baseline CMAQ and those of AQS at EPA AQS stations (mean bias of +19.8 % to 13.7 % over PC, 13.8 % to 36.7 % over RM, +149.7 % to 1.8 % over LM, +22.5 % to 7.8 % over UM,+31.3 % to 7.9 % over SE, and +11.6 % to +0.7 % over NE). The relatively high simulated NOx biases from baseline CMAQ over LM (+149.7 %) are likely the results of over-predictions of sim- ulated NOx emissions, which could shed light on those from global/regional Chemical Transport Models. We also perform more detailed investigations on surface NOx and O3 concentrations in two urban and outflow ar- eas, PC (e.g., Los Angeles, South Pasadena, Anaheim, La Habra and Riverside) and LM (e.g., Houston, Beaumont and Sulphur). From two case studies, we found that the GOME2009 emissions decreased surface NOx concentra- tions significantly in the urban areas of PC (up to 30 ppbv) and in those of LM (up to 10 ppbv) during the daytime and that simulated NOx concentrations from CMAQ with GOME2009 compare well to those of in-situ AQS observa- tions. A significant reduction in NO x concentrations resulted in a comparable increase in surface O3 concentrations in the urban areas of PC (up to 30 ppbv) and the resulting simu- lated O3 concentrations compare well with in-situ surface O3 observations over South Pasadena, Anaheim, and River- side. Over Houston, Beaumont, and Sulphur, large reductions in NOx emissions from CMAQ with GOME2009 coincides with large reduced concentrations of simulated NOx. These concentrations are similar to those of the EPA AQS NOx ob- servations. However, the resulting simulated increase in sur- face O3 at the urban stations in Houston and Sulphur exac- erbated preexisting high O3 over-predictions of the baseline CMAQ. This study implies that simulated low O3 biases in the urban areas of PC are likely caused by simulated high NOx biases, but high O3 biases in the urban areas of LM cannot be explained by simulated high NOx biases over the region. This study also suggests that both in-situ surface NO x and O3 observations should be used simultaneously to re- solve issues pertaining to simulated high/low O3 bias and that remote-sensing data could be used as a constraint for bottom-up emissions. In addition, we also found that day- time O3 reductions over the outflow regions of LM follow- ing large reductions in NOx emissions in the urban areas are significantly larger than they are over outflow regions of PC. These findings provide policymakers in the two regions with

GOME2009 emissions decreased surface NO x concentrations significantly in the urban areas of PC (up to 30 ppbv) and in those of LM (up to 10 ppbv) during the daytime and that simulated NO x concentrations from CMAQ with GOME2009 compare well to those of in-situ AQS observations.A significant reduction in NO x concentrations resulted in a comparable increase in surface O 3 concentrations in the urban areas of PC (up to 30 ppbv) and the resulting simulated O 3 concentrations compare well with in-situ surface O 3 observations over South Pasadena, Anaheim, and Riverside.Over Houston, Beaumont, and Sulphur, large reductions in NO x emissions from CMAQ with GOME2009 coincides with large reduced concentrations of simulated NO x .These concentrations are similar to those of the EPA AQS NO x observations.However, the resulting simulated increase in surface O 3 at the urban stations in Houston and Sulphur exacerbated preexisting high O 3 over-predictions of the baseline CMAQ.This study implies that simulated low O 3 biases in the urban areas of PC are likely caused by simulated high NO x biases, but high O 3 biases in the urban areas of LM cannot be explained by simulated high NO x biases over the region.This study also suggests that both in-situ surface NO x and O 3 observations should be used simultaneously to resolve issues pertaining to simulated high/low O 3 bias and that remote-sensing data could be used as a constraint for bottom-up emissions.In addition, we also found that daytime O 3 reductions over the outflow regions of LM following large reductions in NO x emissions in the urban areas are significantly larger than they are over outflow regions of PC.These findings provide policymakers in the two regions with

Introduction
Nitrogen oxides (NO x = NO + NO 2 ) are major O 3 precursors that originate from fossil fuel combustion, lightning, soil, aircraft, and biomass burning.The largest source of NO x over North America is anthropogenic fossil fuel combustion (e.g., Hudman et al., 2007;Choi et al., 2009).Anthropogenic emissions significantly influence the variability of surface NO x concentrations.Several previous studies have shown the proportionality of NO x emissions to satelliteobserved NO 2 column density (e.g., Berlie et al., 2003;Kim et al., 2006Kim et al., , 2009Kim et al., , 2011;;Kaynak et al., 2009;Han et al., 2010;Yoshida et al., 2010;Lamsal et al., 2011;Choi et al., 2012).In particular, Berlie et al. (2003), Kaynak et al. (2009) and Choi et al. (2012) investigated the weekly cycles of the NO x column density using retrieval products from Global Ozone Monitoring Experiments (GOME), SCanning Imaging Absorption spectroMeter for Atmospheric CartograpHY (SCIAMACHY), or GOME2 and found the weekly pattern of the NO 2 column density proportional to that of NO x emissions.
In general, atmospheric scientists obtain daily or weekly patterns of NO x emissions through emissions inventory modeling (e.g., Sparse Matrix Operator Kernel Emissions (SMOKE) modeling (e.g., Houyoux et al., 2000).Although Community Multiscale Air Quality (CMAQ) model users have just begun to use the EPA National Emission Inventory of 2008 (NEI2008), it is still being distributed and tested.Therefore, the National Emission Inventory of 2005 (NEI2005) continues to be used in global and regional CTMs for the simulation of air quality and the impact of meteorological conditions on the chemical environment over the US.The NEI2005 was produced by a bottom-up approach from which a variety of anthropogenic and natural activities were taken into account, and the corresponding emissions efficiency for each activity was estimated (e.g., Hanna et al., 2003).Thus, as previous studies (e.g., Hanna et al., 2003;Napelenok et al., 2008;Kim et al., 2009Kim et al., , 2011;;Han et al., 2010;Choi et al., 2012) have asserted that over some regions of the US, emissions inventory products from the bottom-up approach might exhibit uncertainty reaching a factor of two.Therefore, some other constraints may improve the evaluation/modification of the bottom-up emissions inventory.
Several previous studies pertaining to the NO x emissions inventory have focused on investigating changes in the number of NO 2 columns resulting from either air pollution policy regulations over the eastern, western, and southern US (e.g., Kim et al., 2006Kim et al., , 2009;;2011;Choi et al., 2009Choi et al., , 2012;;Russell et al., 2010) or over China (e.g., Zhao and Wang, 2009;Yang et al., 2011), or the occurrence of extreme weather conditions over coastal urban regions near the Gulf of Mexico (e.g., Yoshida et al., 2010).These studies have shown significant differences between the NO 2 column densities of satellite instruments (e.g., OMI and SCIAMACHY) and WRF-Chem across the western United States (e.g., Kim et al., 2009).In particular, Kim et al. (2009) found that NO x emissions from an updated NEI1998 in WRF-Chem in western urban areas such as Los Angeles were overestimated, resulting in large discrepancies of the simulated NO 2 columns in the areas.Kim et al. (2011) also revealed differences between the NO 2 densities of OMI and those of WRF-Chem with NEI2005 in urban cities over Texas.Brioude et al. (2011) showed differences between the NO y of the model and that of the National Oceanic and Atmospheric Administration (NOAA) and the National Center for Atmospheric Research (NCAR) research aircraft in and around Houston.They claimed that in the Houston Ship Channel (in the eastern part of Houston), either over-predicted NO x emissions were another source of the discrepancies and speculated that surface O 3 over the region could be better simulated if there were fewer NO x emissions.Eder et al. (2009) showed large discrepancies (low or high) in the simulated surface O 3 concentrations from the real-time National Air Quality Forecast Capability (NAQFC) in urban areas of the southern California and Gulf Coast regions of the US.Several other studies focused on investigating causes for simulated surface O 3 biases in the urban areas (e.g., Eder et al., 2009;Zhang et al., 2007;Henderson et al., 2010;Kim et al., 2011).They showed that the uncertainty of the simulated PBL height (e.g., Eder et al., 2009), the emissions inventory of NO x or VOC (e.g., Eder et al., 2009;Kim et al., 2009Kim et al., , 2011)), meteorological uncertainties (e.g., Zhang et al., 2007), or model resolution (e.g., Henderson et al., 2010) introduce simulated O 3 biases.In particular, Kim et al. (2009Kim et al. ( , 2011) ) estimated the uncertainty of the emissions inventory by comparing the NO 2 column densities of the model and remote sensing, but they have not utilized remote-sensing data to derive or adjust the emissions inventory in the model.In some other studies, even with the large uncertainty in remote sensing data, atmospheric scientists have shown the feasibility of utilizing the satellite column density for yielding an accurate NO x emissions inventory by using top-down satellite products for global CTMs (e.g., Martin et al., 2003;Lamsal et al., 2011) and regional CTMs (e.g., Choi et al., 2008;Napelenok et al., 2008;Chai et al., 2009;Zhao and Wang, 2009).
However, as addressed above, most of the previous studies have focused on the evaluation of the NO x emissions inventory by comparing the NO 2 column of the model and remote sensing.Recently, Choi et al. (2012) showed that results from the CMAQ model with satellite-adjusted NO x emissions more accurately captured the weekly cycle of surface NO x over the US for August 2009.In the study, we utilized the adjusted emissions inventory using remote-sensing data and showed that a modified emissions inventory mitigates the discrepancies between the simulated weekly cycle 39  of surface NO x concentrations and the corresponding in-situ surface measurements.Thus, we concluded that estimating the impact of emissions changes on surface NO x and O 3 concentrations is crucial to determining whether a top-down approach can be used for updating/constraining the bottom-up emissions inventory.
The main purpose of this study is not to obtain an accurate emissions inventory or estimate the absolute uncertainty of the emissions inventory, but instead to perform an evaluation of the relative uncertainties of both the NO x emissions inventory and adjusted NO x emissions inventories using remote sensing in the two urban areas that showed large discrepancies between simulated surface O 3 and corresponding observations.As we mentioned above, among these cities, Los Angeles and Houston have been investigated by previous NO 2 remote-sensing studies (e.g., Kim et al., 2009Kim et al., , 2011;;Eder et al., 2009) because of their characteristic as an O 3 nonattainment area and a large discrepancy area of simulated O 3 compared with in-situ measurements.Again, our previous study (Choi et al., 2012) showed how changes in NO x emissions utilizing remote-sensing products mitigate discrepancies between the weekly NO x pattern at EPA AQS measurement stations produced by the model and that produced by observations.In this study, we use the GOME-2adjusted NO x emissions inventory (details regarding on how the emissions inventory was obtained are described in Choi et al., 2012).With the simulation results from both baseline and sensitivity CMAQ with the adjusted emissions inventory for six geological regions -Pacific Coast = PC, Rocky Mountains = RM, Lower Middle = LM, Upper Middle = UM, Southeast = SE, and Northeast = NE) (Fig. 1) -we investigate (1) which geological region produces the largest NO x differences between CMAQ and in-situ surface observations, (2) how satellite-adjusted emissions mitigate the simulated discrepancies of surface NO x concentrations, (3) how the satellite-adjusted emissions affect surface O 3 discrepancies of the model at the stations of two geological regions (PC and LM) of the urban cities, and (4) how changes in the emissions in the urban areas of the geological regions affect the surface O 3 over the outflow regions.

Model and emissions
The CMAQ model, version 4.7.1 (Foley et al., 2010), was applied August 2009 over the CONUS domain using a horizontal resolution of 12 km with 22 vertical layers from the surface to 100 hPa.Meteorological data came from the NOAA National Centers for Environmental Prediction (NCEP) North American Model (NAM), which uses the Weather Research and Forecasting Non-hydrostatic Mesoscale Model (WRF-NMM) (e.g., Eder et al., 2009;Choi et al., 2012).This study used the SMOKE system to process all emissions independent from meteorological conditions (Houyoux et al., 2000) and generated emissions that vary with meteorological conditions in a pre-processor to CMAQ using NMM data fields (e.g., Eder et al., 2009;Choi et al., 2012).We based area and off-road engine emissions for the CMAQ simulations on NEI2005 version 1 and used the electric generating unit (EGU) and non-EGU sources as point sources in the US.Wherever it was possible in the inventory, we substituted continuous emissions monitoring (CEM) data in 2007 for EGU point sources and then updated and projected EGU emissions using emissions projection factors Y. Choi: The impact of satellite-adjusted NO x emissions on simulated NO x and O 3 discrepancies from the Department of Energy 2009 Annual Energy Outlook (AEO) to account for reductions in power plant emissions.Again, the number of point sources has decreased since 2005, and this reduction was determined using CEM data for 2007 and AEO 2009.For mobile sources, we used the EPA Office of Transportation and Air Quality (OTAQ) 2005 on-road emissions inventories and based the number of emissions from wildfires, prescribed burning, and residential wood burning on a multi-year average fire year of 1996-2002(e.g., Choi et al., 2012)).Estimations of biogenic VOC and NO emissions came from the Biogenic Emissions Inventory System (BEIS) version 3 (Houyoux et al., 2000).The baseline emissions and GOME-2-adjusted emissions over the US were 462 and 426 Gg N, respectively, for August 2009.The baseline emissions (referred to as "BASE2009") were obtained from NEI2005, which accounted for reductions in the point sources.The GOME-2-adjusted emissions (referred to as "GOME2009") were obtained from BASE2009 and the GOME-2 and CMAQ NO 2 column ratios.Details relating to the chemistry modules and the chemical boundary conditions for this study were described in the previous study by Choi et al. (2012).

The global ozone monitoring experiment-2 NO 2 column
We used the remotely-sensed NO 2 column density from the Global Ozone Monitoring Experiment-2 (GOME-2) sensor to measure the nadir at 09:30 local time (LT) with footprints of 40 × 80 km 2 , obtained the daily GOME-2 NO 2 column retrievals from http://www.temis.nl/airpollution,and used TM4NO2A version 2.1 for the GOME-2 NO 2 column density.Some data were filtered out with a cloud fraction of > 40 %.Details pertaining to the NO 2 column retrieval products and the reasons for using GOME-2 were provided in the study addressed by the previous studies (Choi et al., 2012 and other references in).

In-situ observed ground-level NO x and O 3
Hourly ground-level NO x and O 3 concentrations (measurement detection limit of 5 ppbv, J. Summers, personal communication from Choi et al., 2008)  geological regions, PC and LM, and used the corresponding in-situ hourly NO x and O 3 data to evaluate each comparison location.

Comparison of the NO 2 columns of CMAQ to those of GOME-2
The monthly mean column retrievals for GOME-2 NO 2 retrievals and the equivalent column-integrated values of NO 2 concentrations were estimated for August 2009.Gorline and Lee (2010) showed that during the three-year period of 2007 to 2009, the month of August 2009 showed the greatest positive O 3 bias of the CMAQ-based National Air Quality Forecast Capability (NAQFC) modeling system.The other setup of CMAQ model showed several over-and under-estimates over CONUS during the summer in a previous study by Choi et al. (2012).The study also showed that a comparison of model-simulated and satellite-observed NO 2 columns exhibited general overestimates in concentrated population areas and underestimates in rural areas (e.g., Choi et al., 2012).In particular, the CMAQ model overestimated the NO 2 column over some urban areas of PC and LM (e.g., Los Angeles, CA, South Pasadena, CA, Anaheim, CA, La Habra, CA, Riverside, CA, Houston, TX, Beaumont, TX, and Sulphur, LA), also found by previous studies (e.g., Martin et al., 2006;Choi et al., 2008Choi et al., , 2009Choi et al., , 2012)).The following subsections will discuss the impact of highly-biased NO x emissions on surface NO x and O 3 concentrations in the urban and outflow areas.

Satellite-adjusted NO x emissions, GOME2009
To filter out the remote region data (with low sensitivity of the satellite sensor), the monthly averaged GOME-2 and CMAQ NO 2 column were estimated and only the regions showing NO 2 column densities > 1 × 10 15 molecules cm −2 were considered to adjust the emission inventories as in our previous study (e.g., Choi et al., 2012).To evaluate the NO x emissions inventory of BASE2009 over six geological regions, we performed additional simulations with GOME2009 for August 2009.Again, this study focused on analyzing the relative uncertainty (instead of absolute uncertainty) of BASE2009 among the six geological regions.The sensitivity simulation, which determined the amount of NO x emissions that increased or decreased according to the ratio, found that NO x emissions decreased by about 7.8 % over the US (from 462 Gg N to 426 Gg N), and changes in the amount of emissions varied in each geological region (e.g., PC = −5.6 %, RM = −12.3%, LM = −21.3%, UM = +2.3%, SE = −15.9%, NE = +10.0%) (Table 1).The large reductions were shown in LM (17.5 Gg N) and SE (14.0 Gg N).The reductions may have been caused by reductions in mobile emissions because the consistent decrease in power plant NO x emissions was accounted for in the baseline emissions, BASE2009, but because of the limited datasets for mobile emission reductions over the contiguous US, changes in mobile sources were not (e.g., Choi et al., 2012).Interestingly, the opposite trend showing an increase in NO 2 emissions appeared over NE (5.4 Gg N) in the satellite emissions, GOME2009.The trend showing an increase in GOME NO 2 columns over the geological region of NE from 1996 to 2002 was found in a previous study by Richter et al. (2005).
Trends showing increased NO x emissions over the region were not well simulated in the emissions modeling.Pickering et al. (2011) found some evidence of an increase in unresolved NO x emissions sources in Pennsylvania, and other neighboring states.The explanation for these results remains unclear.

The impact of GOME2009 on surface NO x over six geological regions
As we addressed above, the explanations for the large differences between the NO 2 columns of CMAQ and those of GOME-2 remain unclear.However, if we assume that GOME-2 involves an additional constraint on the emissions inventory, we could utilize the GOME-2 NO 2 columns to produce the GOME-2-adjusted emissions inventory, GOME2009.Particularly over LM and PC, the pronounced large reductions in simulated NO x concentrations (in absolute amount) from the baseline to sensitivity CMAQ with GOME2009 (10.7 ppbv over LM and 4.7 ppbv over PC) (Table 2) suggest that NO x emissions from BASE2009 at the EPA AQS stations of geological regions LM and PC are likely to be high.The reduction in simulated NO x concentrations from the baseline CMAQ over SE was estimated to be 2.4 ppb at the EPA AQS stations, which is smaller than the reduction at the other stations over the LM and PC regions.The following sections will provide details regarding the impact of large reductions in NO x emissions on NO x and O 3 concentrations in the urban areas of the LM and PC regions.We developed the new emissions inventory, GOME2009 and then evaluated the results from the CMAQ with GOME2009 by comparing them with the results of other in-situ surface observations in the urban areas and their outflow regions.
Considering all the uncertainties of the chemistry and transport in the model and remote sensing observations addressed by previous studies (e.g., Richter et al., 2005;Lamsal et al., 2008;Kim et al., 2011), the relatively high simulated NO x biases from baseline CMAQ in urban areas over LM are likely the result of over-predictions of simulated NO x emissions.Furthermore, the mean NO x concentrations at AQS stations over PC from high NO x emissions in urban areas such as Los Angeles are the largest (13.8 ppbv) (Table 2).Interestingly, whereas the baseline CMAQ over-predicted NO 2 columns compared to GOME-2 NO 2 columns in the urban areas in southern California, the model under-predicted NO 2 columns over neighboring rural regions.The explanations for the contrasting trends in the urban and rural areas are not clear, but by enhancing emissions in rural areas and reducing those in urban areas, we can at least produce similar chemical environments in terms of the amount of surface NO x concentrations.By doing so, we can examine how changes in the chemical environments (by modifying NO x emissions) impact surface NO x and O 3 concentrations in the urban areas and their neighboring outflow regions of PC and LM.

The impact of GOME2009 on NO x and O 3 concentrations over PC
As addressed in the previous section, the EPA AQS observations sites over PC yielded relatively higher NO x concentrations than those of other geological regions (Table 2).
In addition, the large discrepancies between simulated NO x concentrations of baseline CMAQ simulation and those of AQS observations occurred in the urban areas of PC (Table 3), of which Los Angeles is a representative.To examine how large reductions in NO x emissions in urban areas (e.g., 1: Los Angeles, 2: South Pasadena, 3: Anaheim, 4: La Habra, and 5: Riverside, CA) (Fig. 3) affect surface NO x and O 3 , we estimated the impact of large changes in emissions on surface concentrations in or near the urban cities during the daytime (13:00-17:00 LT) (Fig. 3).Large reductions in emissions (> 3.0 mol s −1 ) in the urban areas resulted in large   reductions in NO x concentrations in the cities (> 15.0 ppbv) (Fig. 3).In addition, several increases in NO x emissions in the eastern parts of the urban cities in the satellite emission, GOME2009 (see the negative values on the left panel of Fig. 3) also appear, but the impact of changes in emissions on NO x concentrations are not evident (see the right panel of Fig. 3), likely the result of the eastward transport of significantly reduced NO x air in urban cities (Fig. 3).In conclusion, the large changes in surface NO x emissions mitigated the large discrepancies in CMAQ simulated NO x concentrations (see the circles on the right panel of Fig. 3), compared to corresponding EPA AQS observations (Fig. 3).
To evaluate how the GOME2009 emissions inventory affects the NO x concentrations at five station grids (including the discrepancies of > 20 ppbv NO x concentrations during the daytime (13:00-17:00 LT), compared to the corresponding EPA AQS observations), we estimate three sets of NO x concentrations from CMAQ simulations with BASE2009, CMAQ simulations with GOME2009, EPA AQS observations (Table 3).Estimates of the mean values of the surface NO x concentrations at the AQS sites were 34.7, 24.7, 20.3, 21.5, and 23.1 ppbv at the five station grids (1: Los Angeles, 2: South Pasadena, 3: Anaheim, 4: La Habra, and 5: Riverside) (Table 3).Estimated surface NO x concentrations of the baseline CMAQ model were 59.6, 52.2, 43.4, 38.5, and 51.8 ppbv at the grids.The corresponding estimates from CMAQ with GOME2009 were 29.8, 25.2, 23.1, 18.7, and 31.2 ppbv, which were similar to those of EPA AQS observations.
Because of the complexity of O 3 production, changes in surface O 3 following changes in NO x emissions are more difficult to understand.Large reductions in NO x emissions in urban cities such as Los Angeles result in large increases in daytime O 3 (13:00-17:00 LT) in the areas and the downwind from the urban cities is caused by westerly see breezes during the daytime in the summer.As Los Angeles is a typical NO xsaturated regime area, the large reductions in NO x emissions resulted in large increases in surface O 3 concentrations in the urban city and its neighboring areas (see the right panel of Fig. 5).Interestingly, the CMAQ with BASE2009 underpredicted surface O 3 in and near Los Angeles and South Pasadena and their outflow areas, including Riverside (up to 30 ppbv) (see the circles of the right panel of Fig. 5).Thus, the large increases in simulated surface O 3 following significant reductions in NO x emissions in Los Angeles, Pasadena, and Anaheim resulted in trends of pre-existing simulated under-prediction to those of over-prediction in the areas (see the right panel of Fig. 5).
Large reductions in NO x emissions resulted in reductions in NO x concentrations and increases in O 3 concentrations in the urban cities (Figs. 4 and 5), which is a typical trend shown in NO x -saturated regime.For example, the baseline CMAQ model under-predicted surface O 3 concentrations by −22.7, −27.1, −23.4,−13.1 and −31.3 % at the five station grids, and the large reductions in NO x emissions increased surface O 3 concentrations, which resulted in overestimation of the surface O 3 predictions of +37.6, +31.3, +19.5, +38.1, and +9.6 % (Table 4).In other words, the large reduction in NO x emissions introduced the over-prediction of surface O 3 concentrations in CMAQ with GOME2009 (Fig. 6).
We also investigated how O 3 concentrations vary during the daytime (13:00-17:00 LT) from baseline CMAQ to CMAQ with GOME2009.Similarly, during the daytime, the CMAQ model under-predicted surface O 3 concentrations by −12.9, −29.0, −24.8, −20.9, and −27.2 % at the five station grids, and large reductions in NO x emissions significantly increased surface O 3 concentrations; thus, the under-prediction patterns became over-prediction patterns of +25.4,+12.0, +11.5, +19.6, and +9.2 % at the five grids (see the parentheses in Table 4).During the daytime, except for the station in Los Angeles, the large biases of surface O 3 at the stations in South Pasadena, Anaheim, La Habra, and Riverside decreased as a result of reductions in NO x emissions.The large NO x reductions mitigated the large discrepancies between the NO x concentrations of the baseline CMAQ and those of corresponding AQS observations and enhanced simulated surface O 3 concentrations, but the increases in O 3 concentrations were extreme, likely stemming from the high sensitivity of surface O 3 to changes in NO x emissions.Interestingly, the low biases of baseline CMAQ during the daytime (13:00-17:00 LT) decreased in a more similar manner than those of the baseline CMAQ during the whole day, even with the large increase in nighttime O 3 caused by a reduction in surface NO x emissions (Fig. 6).Explanations for this phenomenon remain unclear, probably because the CMAQ model represents the urban areas as extreme NO x -saturated regime (instead of normal NO x -saturated areas or mixed areas, shown in Fig. 3 of the previous study by Choi et al., 2012), likely due to the overestimated NO x emissions.
Several previous studies have also suggested that the uncertainty in the emissions of O 3 precursors such as NO x is closely associated with large low O 3 biases over Southern California (e.g., Eder et al., 2009;Kim et al., 2009).Kim et al. (2009) showed that two satellite (OMI and  They suggested that these differences were caused by overestimated NO x emissions in the areas from an updated NEI1999 in their study.Again, in this study, unlike in previous studies, we reveal that satellite-adjusted emissions mitigated simulated NO x and daytime O 3 discrepancies compared to the corresponding observations.

The impact of GOME2009 on NO x and O 3 concentrations over LM
Large discrepancies in the simulated NO x concentrations were shown in the urban areas of LM compared to the corresponding in-situ observations (see the circles on the right panel of Fig. 7).To examine how the large reductions in NO x emissions at the stations in the urban areas (e.g., Houston and Beaumont, TX and Sulphur, LA) (Fig. 7) affect surface NO x and O 3 , we investigated the impact of large changes in emissions on surface concentrations in or near the three urban cities over LM (Fig. 7).During the daytime (13:00-17:00 LT), the large reductions in emissions (> 2.0 mol s     in the urban areas resulted in reductions in NO x concentrations in the cities (> 8.0 ppbv) (Fig. 7).In addition, some increases in NO x emissions over the western parts of Houston resulted in increases in NO x concentrations (> 2.0 ppbv) during the daytime (Fig. 7).
To investigate how satellite-adjusted emissions, GOME2009, affect surface NO x concentrations in the    model, we examine four station grids for the urban areas (Fig. 8).We chose the four grids because they exhibited large differences between the baseline CMAQ simulated and AQS observed NO x concentrations of > 10 ppbv during the daytime (13:00-17:00 LT) (see the right panel of Fig. 7).For more details, we compared the surface NO x concentrations from baseline CMAQ and CMAQ with GOME2009 to the corresponding in-situ observations at four different station grids (Fig. 8).Our analysis showed that the baseline CMAQ model over-predicted surface NO x concentrations by +288.0,+464.2, +683.1, and +402.5 % at the four station grids (Table 5 and Fig. 8).Interestingly, from the baseline CMAQ, the diurnal differences in the surface NO x concentrations in the urban areas of LM were significantly larger than those in the urban areas of PC.The baseline CMAQ significantly overestimated the surface NO x , particularly during the nighttime in the urban areas of LM, partly because of the underestimated PBL heights over the Gulf Coast areas, as Eder et al. (2009) previously found.
The large reductions in NO x emissions led to significant reductions in NO x concentrations and thus mitigated the discrepancies between simulated NO x concentrations of baseline CMAQ and corresponding AQS observations, and those of CMAQ with GOME2009 became as +15.8, +65.5, +172.7, and +46.8 % (Table 5 and Fig  Table 5.The number, the mean, and the standard deviation of the EPA AQS NO x observations (in ppbv), CMAQ NO x simulations with the baseline emissions, BASE2009 and CMAQ NO x simulations with GOME-2 adjusted emissions, GOME2009 at the four different EPA AQS measurement sites (see the right panel of Fig. 7, 1: Houston A, 2: Houston B, 3: Beaumont, and 4: Sulphur).
CMAQ with BASE2009 (O 3 ) CMAQ with GOME2009 (O 3 ) EPA AQS (O 3 ) at the four urban areas, but the overestimates of the baseline CMAQ model significantly decreased (Fig. 8).The large reductions in NO x emissions in the urban areas of LM region generally resulted in large reductions in daytime O 3 (13:00-17:00 LT) over regions downwind from the urban cities (e.g., over forested regions represented by an extreme NO x -sensitive regime) because of the distinctive southerly or southeasterly sea breezes during the daytime in the summer (see the right panel of Fig. 9).In another words, in the forested areas over the region downwind from Houston and Beaumont, TX and Sulphur, LA (e.g., Sam Houston National Forest and Devy Crockell National Forest), monthlyaveraged daytime O 3 concentrations significantly decreased by > 6 ppbv as a result of reductions in NO x emissions in the urban cities (see the right panel of Fig. 9).During the summertime, dominant sea breezes from the ocean blow into the area following a large reduction in NO x emissions in the central cities during the daytime, resulting in large O 3 reductions over the NO x -sensitive regime (Fig. 9).
Large reductions in NO x emissions resulted in several increases in surface O 3 concentrations in the core cities (e.g., Houston and Sulphur) (Table 6).For example, during the daytime (13:00-17:00 LT), the baseline CMAQ originally over-predicted surface O 3 (about 8 ppbv) in Houston and (about 2 ppb) Sulphur (see the parentheses of Table 6).Thus, the large increases in simulated surface O 3 following the significant reductions in NO x emissions in central Houston and Sulphur exacerbated pre-existing simulated over-prediction trends in the areas (Table 6).
This study showed that overestimates of simulated surface O 3 in the urban centers or industrial areas were relatively smaller than those of the simulated O 3 in the other areas of Houston (see the left panel of Fig. 9) likely the result of the limitation of the CMAQ model, that is, its inability to capture temporary high-peak O 3 phenomena in central or industrial areas.The monthly-averaged surface O 3 concentration plot does not fully represent specific-surface O 3 peaks near the Houston Ship Channel as Daum et al. (2003) and Xiao et al. (2010) showed in their studies.Thus, the ratio of VOC/NO x as a proxy of the chemical environment needs further investigation.
We examined how the GOME2009 emissions impact surface O 3 concentrations at selected station grids, as we did in the NO x study.At the four selected grids, the baseline  −19.4, −36.6, and −18.6 % (see the blue colored line of Fig. 10).The large reductions in NO x emissions (Fig. 7) generally increased surface O 3 concentrations and the underprediction trends of CMAQ became weaker or changed to overprediction trends.For example, at the four station grids, the estimates of the biases of the CMAQ model with GOME2009 were +55.5, +18.5, −9.1 and +38.1 % (see the red colored line of Fig. 10).
During the daytime (13:00-17:00 LT), estimates of the biases of the baseline CMAQ were +21.0, +17.9, −3.0, and +6.9 %, and emissions of the biases of the CMAQ with GOME2009 changed to +39.1, +24.0, +3.2, and +20.7 % at the four station grids in the urban areas (see the parentheses of Table 6).The baseline CMAQ originally overestimated or slightly underestimated surface O 3 concentrations in the urban areas during the daytime, and the large reduction in the surface NO x emissions resulted in the large increase in simulated O 3 concentrations in the areas.Thus, the increased surface O 3 concentrations from the model with GOME2009 increased trends of overestimation of the baseline CMAQ (Table 6).
Consistent with the results of a previous study by Choi et al. (2012), the results of this study show that CMAQ recognizes the central urban areas of LM as extreme NO xsaturated regime areas (by the overestimates of NO x emissions).The high-biased NO x concentrations in the urban cores might enhance surface O 3 more over the regions downwind from the urban cities (particularly in the forecasted areas in northeastern Houston) (Fig. 9).Because of the opposing chemical characteristics of these two areas regarding O 3 sensitivity (e.g., urban cores: NO x -saturated area and forests: NO x -sensitive area) to changes in NO x emissions/concentrations, designing an efficient O 3 control strategy for both areas poses a challenge.Whereas Li et al. (2007) showed the impact of the reduction in biogenic VOC emissions in forested areas in northeastern Houston on surface O 3 in the urban core of Houston, this study showed the impact of the reduction in surface NO x emissions in urban cores on surface O 3 over the outflow regions from Houston (Fig. 9).
The explanation for preexisting high O 3 biases (> 6 ppbv) in CMAQ in Houston (see the circles on the right panel of Fig. 9) remains unclear.However, the over-prediction of NO x emissions in Houston, shown in Fig. 7, is likely a major cause of the over-predicted NO x concentrations, and the large NO x emissions reductions significantly mitigate the discrepancies of the baseline CMAQ compared to the corresponding AQS observations (see the right panel of Fig. 7), but the overestimated NO x emissions cannot solely explain the overprediction of the surface O 3 in the area.The large NO x reductions in GOME2009 mitigated the high O 3 biases over the outflow region of and around the urban cities, but they exacerbated the O 3 high biases of the baseline CMAQ in the core of Houston and Sulphur (Table 6).In Beaumont, the large NO x reductions increased the simulated surface O 3 concentrations, which resulted in mitigating low O 3 biases of the baseline CMAQ in urban areas such as Los Angeles.
From comparisons of WRF-Chem model results and aircraft and remote-sensing measurements, Kim et al. (2011) concluded that the Houston NO x emissions in NEI2005 were overestimated and the Houston VOC emissions in NEI2005 were underestimated.They hypothesized that less NO x and more VOC emissions in the Houston industrial area might have introduced better O 3 modeling performance, but they still found large deficiencies in model-simulated O 3 even when they accounted for these two factors and suggested the need for future study that would clarify O 3 high bias issue.The results of this study showed that significant reductions in NO x emissions (more than those found by Kim et al., 2011) mitigate the discrepancies of the simulated NO x concentrations both in the urban and outflow areas from the baseline CMAQ, but they do not mitigate the discrepancies of the simulated O 3 in the urban core areas.The results from this study also showed that the large reductions in NO x emissions in urban cities lead to large reductions in simulated O 3 over the outflow regions around the urban cities, which would be critical for mitigating the simulated O 3 discrepancies over the outflow regions.
To evaluate the baseline emissions inventory, BASE2009, we analyzed simulation results from CMAQ over six geological regions over the CONUS.We obtained NO x emissions inventory, GOME2009, from BASE2009 over the CONUS using ratios of CMAQ to GOME-2 NO 2 column density.We found large reductions in NO x concentrations in CMAQ with GOME2009, at EPA AQS stations over LM, SE and PC (i.e., +149.7 to −1.8 % for LM and +19.8 to −13.7 % for PC), which resulted from large reductions in NO x emissions over the regions.The GOME2009 significantly reduced the biases of the NO x concentrations in the urban areas of the LM and PC regions of CMAQ simulations and AQS observations, resulting in the mitigation of the simulated discrepancies of baseline CMAQ.
The results of this study indicated that NO x emissions from BASE2009 in the urban areas of LM (e.g., Houston, Beaumont, and Sulphur) and PC regions (e.g., Los Angeles, South Pasadena, Anaheim, La Habra, and Riverside) are abnormally high, indicating large relative uncertainty of BASE2009 NO x emissions.The analysis of simulation results from global/regional CTMs and climate models must account for these high NO x biases in the urban areas of LM and PC.For example, significant reductions in NO x emissions led to large reductions in model simulations of surface NO x at AQS stations in the urban areas of LM and PC.In particular, in the urban areas of LM, the largest discrepancies in simulated surface NO x concentrations are significantly mitigated in the areas, compared to corresponding observations from AQS.These results suggest that remote-sensing data could be another useful constraint of the bottom-up emissions inventory.Furthermore, changes in NO x emissions in the urban areas of LM and PC in GOME2009 compared to those in BASE2009 are useful for not only updating the emissions inventory but also mitigating the discrepancies in the chemical environment such as NO x concentrations.
In general, while the large reductions in NO x emissions in GOME2009 mitigated the discrepancies in the simulated O 3 concentrations in the urban areas of PC region (e.g., South Pasadena, La Habra and Riverside) and the LM region (e.g., Beaumont), the large reductions in emissions exacerbated over-predictions of pre-existing biased surface O 3 in urban core areas in Houston and Sulphur of LM in the baseline CMAQ resulting from the increase in surface O 3 from reductions in NO x emissions.However, the large reductions in NO x emissions mitigate the high O 3 biases over outflow region or around the main core of Houston (see the right panel of Fig. 9).Again, reductions in the over-predictions of pre-existing simulated large O 3 in the urban core cities of LM in baseline CMAQ did not occur as a result of reductions in the high over-predictions of NO x emissions except in Beaumont; thus, to explain simulated high O 3 bias over the southern US from CTMs, shown in previous studies (e.g., McKeen et al., 2009;Kim et al., 2011), we need to divide urban areas into urban core cities and near or outflow region of the urban core and test two areas separately.In addition to NO x concentrations, the emissions of VOCs over the LM and PC regions require further investigation.In future research, we will introduce another remote-sensing product such as HCHO columns as a proxy for VOC measurements.In addition, we will divide high O 3 -biased regions into chemical regimes, which might prove useful for determining the cause of high O 3 biases.
As Kim et al. (2011) stated, if we have more VOC emissions and fewer NO x emissions in the industrial areas of Houston, we might simulate similar chemical environments in the areas.However, they also showed that simulated O 3 concentrations still exhibit large discrepancies compared to in-situ observed O 3 concentrations.In a future study, we will first modify the chemical environment (both NO x and VOC) in the areas, which might alter the chemical characteristics of an extreme NO x -saturated regime region to those of a less NO x -saturated regime region or a mixed regime region in industrial and urban areas.For example, as we addressed above, as a proxy for VOC emissions, we will also evaluate simulated HCHO column densities in the model and compare them to remote-sensing HCHO columns.After performing this evaluation, we can evaluate the ratio of the simulated HCHO/NO 2 columns (as a proxy for the chemical environment, VOC/NO x ) to those of corresponding remote-sensing observations.Of course, one challenge of this task is how we consider the uncertainty of remote sensing, such as that stemming from the products of HCHO columns.
This study also found that during the summertime, the significant reduction of NO x emissions in urban core cities resulted in an increase in surface O 3 in the urban areas (e.g., Los Angeles, South Pasadena, Anaheim, La Habra, Riverside, Houston, Beaumont, and Sulphur), but resulted in large reductions in surface O 3 in forested areas or outflow regions over northeastern or northern regions downwind from urban cities (e.g., Houston and Beaumont).These large reductions are not clearly shown in outflow regions of urban cities in southern California (e.g., near Los Angeles) likely because of lower VOC concentrations in the outflow regions.In other words, while the NO x -saturated regime becomes NO x -sensitive regime as the location moves from urban cities of Houston and Beaumont of LM to their outflow rural or forested regions, the characteristics of the NO xsaturated regime of the urban area of Los Angeles remain stable, as do those of the outflow region from the urban city because of limited VOC sources.In a future study, we plan to investigate how quickly the chemical regime from urban areas to outflow areas changes, the purpose of which is to design an optimal O 3 pollution control strategy for each urban area.
The direct satellite-adjusting method in this study gave general success in mitigating the discrepancies of modelsimulated surface NO x concentrations compared with in-situ measurements, but further research is needed to addresses some of remaining issues.First, the assumption that remotesensing NO 2 columns are closer to actual true values compared with model-simulated NO 2 columns was not perfectly met by the results.Thus, ideally, in order to get accurate emission inventories, we need to estimate uncertainties of remote-sensing NO 2 column and model simulated NO 2 column/NO x emission inventories and use the uncertainties for the application of data assimilation approach (e.g., Napelenok et al., 2008;Chai et al., 2009;Zhao and Wang, 2009).Second, emissions were adjusted using morning time satellite NO 2 column data (e.g., GOME-2) and the resulting emission inventory could miss its diurnal cycle.In a following study, we will adjust the diurnal cycles of emissions using two different remote-sensing data from GOME-2 (morning time) and OMI (afternoon).Some other uncertainties regarding the use of NO 2 columns as a proxy for NO x concentrations/emissions over the surface were described in detail in the previous study (e.g., Choi et al., 2012).
More interestingly, the high simulated NO x biases are still shown in the comparison of the NO x concentrations from CMAQ including NEI2008 from Air Quality Forecasting system at UH (AQF-UH) and the corresponding observations from the CAMS sites over southeastern Texas for the DISCOVER-AQ Houston campaign (September of 2013) (Appendix A), but they are not shown to be significant as much as in those of CMAQ including the modified NEI2005 in this study.The detailed study needs to be followed to examine how high biases of NO x emissions found in this study are changed in the modeling study with NEI2008 using same resolution and same time simulations.

Figure 1 .
Figure 1.Map of the six geological regions of the US, Pacific Coast (PC), Rocky Mountains

Fig. 1 .
Fig. 1.Map of the six geological regions of the US, Pacific Coast (PC), Rocky Mountains (RM), Lower Middle (LM), Upper Middle (UM), South East (SE), and Northeast (NE) for the performance evaluation (different colors represent six geological regions and letters locate EPA AQS stations of NO x measurements).

Figure 2 .
Figure 2. Surface NO x concentrations at EPA AQS stations (black crosses), corresponding CMAQ simulations with the baseline emissions, BASE2009 (blue), and CMAQ simulations including GOME-2-adjusted NO x emissions, GOME2009 (red) over six geological regions (see Figure 1, PC: Pacific Coast, RM: Rocky Mountain, LM: Low Middle, UM: Upper Middle, SE: South East, and NE: North East) for August 2009.

Figure 3 .
Figure 3.The difference between the surface NO x emissions of the baseline emissions BASE2009 and GOME-2-adjusted emissions GOME2009 (left panel, the difference is estimated only when monthly NO 2 column averages are > 10 15 molecules cm -2 from GOME-2 and CMAQ over the continent); the differences between the surface NO x concentrations of the baseline CMAQ with BASE2009 and CMAQ with GOME2009 (right panel); and the differences between

Fig. 3 .
Fig.3.The difference between the surface NO x emissions of the baseline emissions BASE2009 and GOME-2-adjusted emissions GOME2009 (left panel, the difference is estimated only when monthly NO 2 column averages are > 10 15 molecules cm −2 from GOME-2 and CMAQ over the continent); the differences between the surface NO x concentrations of the baseline CMAQ with BASE2009 and CMAQ with GOME2009 (right panel); and the differences between the baseline CMAQ with BASE2009 and EPA AQS observations (circles on the right panel) for the daytime (13:00-17:00 LT) of August 2009.Among the circled stations, the five marked stations (1: Los Angeles, 2: South Pasadena, 3: Anaheim, 4: La Habra, and 5: Riverside) include large discrepancies in > daytime 20 ppbv NO x (baseline CMAQ simulations -EPA AQS observations).

Figure 5 .
Figure 5. Surface O 3 concentrations from the baseline CMAQ with baseline emissions, BASE2009 (left panel), and EPA AQS measurements (circles on the left panel), the difference

Fig. 4 .
Fig. 4. Surface O 3 concentrations from the baseline CMAQ with baseline emissions, BASE2009 (left panel), and EPA AQS measurements (circles on the left panel), the difference between the surface O 3 of the baseline CMAQ with BASE2009 and CMAQ with GOME-2-adjusted NO x emissions GOME2009 (right panel) and the difference between the surface O 3 of the baseline CMAQ with BASE2009 and EPA AQS (circles on the right panel, baseline CMAQ simulations -EPA AQS observations) for the daytime (13:00-17:00 LT) in August 2009.

Figure 7 .
Figure 7.The differences between surface NO x emissions of baseline emissions, BASE2009 and GOME-2-adjusted emissions, GOME2009 (left panel, the difference is estimated only when monthly NO 2 column averages are > 10 15 molecules cm -2 from GOME-2 and CMAQ over the continent); the differences between the surface NO x concentrations of the baseline CMAQ with BASE2009 and CMAQ with GOME2009 (right panel); and the differences between surface NO x concentrations of the baseline CMAQ with BASE2009 and EPA AQS observations (circles on

Fig. 7 .
Fig. 7.The differences between surface NO x emissions of baseline emissions, BASE2009 and GOME-2-adjusted emissions, GOME2009 (left panel, the difference is estimated only when monthly NO 2 column averages are > 10 15 molecules cm −2 from GOME-2 and CMAQ over the continent); the differences between the surface NO x concentrations of the baseline CMAQ with BASE2009 and CMAQ with GOME2009 (right panel); and the differences between surface NO x concentrations of the baseline CMAQ with BASE2009 and EPA AQS observations (circles on the right panel) for the daytime (13:00-17:00 LT) of August of 2009.Among the circled stations, the four marked stations (1: Houston A, 2: Houston B, 3: Beaumont, and 4: Sulphur) include the large discrepancies of > daytime 10 ppbv NO x (baseline CMAQ simulations -EPA AQS observations).

Figure 8 .
Figure 8. Surface NO x concentrations at EPA AQS stations (black crosses), corresponding baseline CMAQ simulations with baseline emissions, BASE2009 (blue), and CMAQ simulations with GOME-2-adjusted NO x emissions, GOME2009 (red) at the four station grids (see the right panel of Figure 7, 1: Houston A, 2: Houston B, 3: Beaumont, and 4: Sulphur) in August 2009.

Figure 9 .
Figure 9.The differences between the O 3 concentrations from a baseline CMAQ with baseline emissions, BASE2009 (left panel) and EPA AQS measurements (circles on the left panel); the

Fig. 9 .
Fig. 9.The differences between the O 3 concentrations from a baseline CMAQ with baseline emissions, BASE2009 (left panel) and EPA AQS measurements (circles on the left panel); the differences between surface O 3 of the baseline CMAQ with BASE2009 and CMAQ with GOME-2-adjusted NO x emissions, GOME2009 (right panel); and the differences between surface O 3 of the baseline CMAQ with BASE2009 and EPA AQS (circles on the right panel, baseline CMAQ simulations -EPA AQS observations) for the daytime (13:00-17:00 LT) of August of 2009.

Table 2 .
The total number, the mean, and the standard deviation of the EPA AQS NO x observations (in ppbv), CMAQ NO x simulations with the base emissions, BASE2009 and CMAQ NO x simulations with the GOME-2 adjusted emissions, GOME2009 at the EPA AQS NO x measurement sites over six geological regions of the US (PC: Pacific Coast, RM: Rocky Mountain, LM: Lower Middle, UM: Upper Middle, SE: Southeast, and NE: Northeast).

Table 3 .
The total number, the mean, and the standard deviation of the EPA AQS NO x observations (in ppbv), CMAQ NO x simulations with the baseline emissions, BASE2009 and CMAQ NO x simulations with GOME-2 adjusted emissions, GOME2009 at the five EPA AQS NO x measurement sites (see the right panel of Fig.3, 1: Los Angeles, 2: South Pasadena, 3: Anaheim, 4: La Habra, and 5: Riverside).CMAQ with BASE2009 (O 3 ) CMAQ with GOME2009 (O 3 ) EPA AQS (O 3 )

Table 6 .
The number, the mean, and the standard deviation of the EPA AQS O 3 observations, CMAQ O 3 simulations with the baseline emissions, BASE2009 and CMAQ O 3 simulations with GOME-2 adjusted emissions, GOME2009 at the four EPA AQS O 3 measurement sites (see the right panel of Fig.7, 1: Houston A, 2: Houston B, 3: Beaumont, and 4: Sulphur).The parentheses indicate the corresponding data during the daytime (13:00-17:00 LT).