Chemical feedback effects on the spatial patterns of the NO x weekend effect : a sensitivity analysis

We examine spatial variations in the weekday– weekend pattern of NO 2 over the Los Angeles metropolitan area using the Ozone Monitoring Instrument (OMI) and then compare the observations to calculations using the WRFChem model. We find that the spatial pattern of the weekday– weekend variations of the NO 2 column in the model is significantly different than observed. A sensitivity study shows that the contrasting spatial pattern of NO 2 on weekdays and weekends is a useful diagnostic of emissions and chemistry. These improvements suggest that constraints from spacebased observations of the processes affecting urban photochemistry (e.g., spatial patterns of emissions, ratios of VOC to NOx emissions, rate constants) are possible at a level of detail not previously described.


Introduction
In the troposphere, NO x (NO+NO 2 ) affects ozone production rates, aerosol formation and the oxidative capacity of the atmosphere.NO x is emitted to the troposphere by fossil-fuel combustion, biomass burning, soil microbial processes, and lightning.During the daytime NO x is removed from the atmosphere by reactions of HO x (OH + HO 2 + RO 2 ) with NO x , primarily by reaction of OH with NO 2 to form nitric acid, RC(O)O 2 with NO 2 to form acylperoxy nitrates, and RO 2 with NO to form alkyl nitrates.As a result, the chemical removal of NO x depends on the concentration of OH and the local mix of volatile organic compound (VOC) precursors to RC(O)O 2 and RO 2 (e.g., Browne and Cohen, 2012).Both OH and RO 2 depend non-linearly on NO x introducing a powerful nonlinear feedback of NO x on its own lifetime (Fig. 1).
Satellite-based observation of the NO 2 column has provided a wealth of information on NO x at local (e.g., Bertram et al., 2005;Boersma et al., 2009;Russell et al., 2012), regional (e.g., Richter et al., 2005;Kim et al., 2009;Hudman et al., 2010), and global scales (e.g., Stavrakou et al., 2008;Lamsal et al., 2011), proving particularly useful for study of seasonal (Jaeglé et al., 2005;van der A et al., 2006), interannual (e.g., Richter et al., 2005;van der A et al., 2008;Russell et al., 2012), and weekday-weekend (e.g., Beirle et al., 2003;Kaynak et al., 2009;Russell et al., 2010) trends in NO 2 column.Typically, the observed columns and their trends are used to assess differences in NO x emissions (e.g., Beirle et al., 2003;Kim et al., 2006).In situ observations (e.g., Thornton et al., 2002;Murhpy et al., 2007) have provided extensive information on the mechanistic details affecting the NO x lifetime on weekdays and weekends, but lack the coverage in space and time of space-based measurements.With a nadir footprint of 13 km × 24 km and daily, global coverage, the Ozone Monitoring Instrument (OMI - Levelt et al., 2006) provides coverage and spatial detail at a scale relevant to boundary layer NO x removal (∼10 km-50 km; e.g., Ryerson et al., 2001;Dillon et al., 2002;Beirle et al., 2011;Valin et al., 2011a, b).Recently, OMI observations have been shown to be sensitive not only to NO x emissions, but also to provide constraints on the NO x lifetime (e.g., Beirle et al., 2011) and its variation with wind speed (e.g., Valin et al. 2013).
Here, we use OMI observations to describe the spatial variation of the NO 2 column over Los Angeles on weekends and  weekdays.We then use a series of WRF-Chem model simulations in a sensitivity study to characterize the information contained in the weekday-weekend patterns.From the range of simulations computed, we identify modifications to the base model that produce improved matches to the spatial pattern OMI observations and use this to guide our understanding of the chemical processes that control the NO 2 column.

Observations
The Ozone Monitoring Instrument (OMI), a UV/VIS spectrometer developed at KNMI in the Netherlands (Levelt et al., 2006), is mounted on the polar-orbiting, sun-synchronous NASA Aura satellite (Schoerbl et al., 2006).OMI has a 114 • field of view, with 480 detector elements devoted to spatial coverage that are averaged onboard to provide 60 pixels across a 2600 km swath of the Earth, and a 2 second integration period that results in a 13 km pixel dimension in the direction of the spacecraft motion.The resulting spatial footprint is as small as 13 km × 24 km.Aura/OMI orbits the Earth over 14 times a day, providing near global coverage at approximately 1:15 p.m. local solar time.Due to the nonoverlapping orbital pattern, we oversample the native footprint of OMI (13 km × 24 km at nadir) to achieve a spatial resolution of approximately 10 km × 10 km for temporally averaged fields (e.g., Russell et al., 2010).Several different analyses of the spectrum measured by OMI are publically available (Bucsela et al., 2006;Boersma et al., 2007;Zhou et al., 2009;Russell et al., 2011).Here, we use the BErkeley High Resolution (BEHR) retrieval of OMI NO 2 vertical tropospheric columns (available at http://behr.cchem.berkeley.edu;Russell et al., 2011Russell et al., , 2012)).Briefly, BEHR derives tropospheric vertical NO 2 column from the NASA standard tropospheric slant column (Bucsela et al., 2006)  Due to the onshore flow and large coastal emission sources, the May-July, 2005-2007 average weekday NO 2 column increases sharply at the coastline to a regional maximum of 3.9 × 10 16 molecules cm −2 directly over downtown Los Angeles (Fig. 2, top panel).To the east, NO 2 columns decrease gradually through the San Bernardino Valley.To the northwest, in the San Fernando Valley, decreases are sharper than in San Bernardino.On weekends, the average airflow is identical to that on weekdays, but the weekend NO 2 column is markedly lower than the weekday column and the gradients downwind of central Los Angeles are much steeper.Weekend decreases in the NO 2 column are only 25-40 % over coastal and downtown Los Angeles, but are 45-55 % over the San Fernando Valley and an even larger 55-65 % over the San Bernardino Valley.
Large weekend decreases in the NO 2 column and surface concentration have been reported previously using satellite and in situ observations (e.g., Beirle et al., 2003;Murphy et al., 2007;Russell et al., 2010).The changes have been attributed primarily to a weekday-weekend pattern of emissions, with decreases in the US associated with a reduction in freight transport by heavy-duty diesel-powered vehicles (Harley et al., 2005).However, these decreases in emissions result in decreases in NO x concentrations that affect the NO x lifetime through the chemical feedbacks on OH and O 3 (e.g., Valin et al., 2011b;Stephens et al., 2008;Pollack et al., 2012;Pusede and Cohen, 2012 and references therein).Previously, analyses have largely assumed that the changes associated with emissions and chemical feedbacks are uniform across an urban center (Beirle et al., 2003;Brioude et al., 2013).The observation that the NO 2 decreases are not spatially uniform has not, to our knowledge, been previously reported or evaluated in a model.In the following, we simulate the weekdayweekend pattern of NO 2 column using WRF-Chem and compare the results to the observations in an effort to understand which model parameters are important for describing the weekday-weekend patterns in Los Angeles, and by analogy, for describing weekday-weekend patterns of NO 2 in other locations.Our goal is not to produce exact agreement, but rather to illustrate that the spatial pattern of observations contains information that can provide new constraints on our understanding of urban photochemistry and emissions.

WRF-Chem calculations
We simulate the NO 2 column at the OMI overpass time (1 p.m.) using WRF-Chem (Grell et al., 2005) at 4 km horizontal resolution over a 288 km (N-S) by 480 km (E-W) domain, centered at 34 • N, 118 • W. We use 36 vertical layers, 14 of which are in the first 1.5 km.As our base case, we use the Regional Acid Deposition Model 2 (RADM2) chemical mechanism, the 2005 National Emissions Inventory (NEI2005) for anthropogenic emissions and the default online module for biogenic emissions as in Grell et al. (2005).Initial and boundary conditions for meteorology are taken from the North American Regional Reanalysis (NARR) and for chemistry are taken from an idealized profile standard to WRF-Chem.
We simulate the average 1 p.m. NO 2 column for 1-14 June 2008 with NEI2005 anthropogenic emissions (weekday) and the same model with an emission rate of 0.625 × E NO x −NEI2005 (weekend), well within the range of values determined previously (∼30-45 % reduction, Harley et al., 2005;Pollack et al., 2012;Oetjen et al., 2013).The first two days are used as spin-up.For the base case, we run WRF-Chem with standard RADM2 chemistry and spatially and temporally uniform weekend emission reductions.We compare results from this base case calculation to weekdayweekend simulations in which we adjust HO x production from ozone photolysis (0.5 × 1.25 × 1.5 × 2 ×), VOC emissions (0.5 × 2 ×), VOC-OH rate constants (0.5 × 2 ×), the NO 2 -OH rate constant (0.5 × 2 ×), the timing of weekend emissions (1 h delay, 2 h delay), and the spatial distribution of weekend emissions (weekend E NO x −San Bernardino of 0.575 × and 0.525 × E NO x −NEI2005 , and E NO x −Los Angeles of 0.640 × and 0.654 ×E NO x −NEI2005 , for domain-average weekend emission of 0.625 × E NO x −NEI2005 ).
The results from these simulations and the OMI observations are summarized in Table 1.We find that the simulated NO 2 columns, regardless of model scenario, are biased low where NO 2 is low.To correct this bias we add a regionally uniform background column of 5 × 10 14 molecules cm −2 to both the weekday and weekend simulations for all model scenarios.A column concentration of 5 × 10 14 molecules cm −2 corresponds to a concentration of 30 ppt if uniformly distributed through the troposphere (0-10 km) and to 200 ppt if restricted to a 1 km boundary layer.This correction does not significantly impact model results where the NO 2 columns are high, but does alter the simulated spatial patterns at the edges of the plume where NO 2 is low (∼20 % effect where NO 2 columns are 2.5 × 10 15 molecules cm −2 ).We test whether a two-week period of simulation is representative of longer time periods by simulating the base case scenario for three months, and find that the magnitude and pattern of decreases simulated in the two-week scenario are comparable to those of the 3-month scenario, indicating their fidelity in representing the meteorological variability of a longer-term simulation.We also test for memory effects by simulating weekday NO 2 column in two-day periods (3-4 June, 4-5 June, . . ., 13-14 June) initialized with the model NO 2 column from 16:00 LST on the previous weekday to test whether simulation of 14 consecu-tive weekend days is biased relative to simulation of two-day weekends within a weekly cycle.We find that the memory effect of weekday emissions on the weekend NO 2 column is small compared to the effects of the other parameters tested.

Model results and analysis
In Fig. 3, we compare the observed weekday-weekend pattern of NO 2 column (top row) to that simulated with the base model (second row).Integrated over the entire basin, there is only a 2 % difference between observation and simulation, and both exhibit similar spatial patterns.Locally, however, there are large differences between the simulated and observed NO 2 columns (up to a factor of two).OMI observes much higher NO 2 values on the south and west edges of the plume, and the model predicts much higher values in the outflow regions, namely the southeast and northwest edges of the plume.This discrepancy indicates several possible model or observational biases.For instance, NO 2 may be transported downwind too quickly in the model, the observations may be biased over the coastline (e.g., solar glint reflectance impacts on the NO 2 retrieval), or OMI with its pixels of 13 km × 24 km may be smearing the spatial pattern that is simulated at a model horizontal resolution of 4 km × 4 km even though our average of the OMI data takes advantage  of oversampling to increase the resolution to approximately 10 km × 10 km (e.g., Russell et al., 2010).The large, localized differences between the observed and simulated NO 2 columns highlight the difficulty of accurately computing the interplay of emissions, chemistry and transport or of accounting for the possibility of locally specific observational biases.However, when investigating the patterns of NO 2 column (e.g., weekday-weekend pattern) within self-consistent model simulations or observational data sets, many of these biases are eliminated.OMI, for instance, measures weekday and weekend NO 2 columns that are subject to the same average meteorological patterns and the same average observational biases, and in our WRF-Chem model setup, the weekday and weekend meteorology is identical.As a result, the agreement of simulated and observed NO 2 trends such as the weekday-weekend pattern (Fig. 3 -right column) is more meaningful than the agreement between observations and simulations for a single time period (Fig. 3 left column).In the base model, simulated NO 2 decreases are too large over the coast (40 % vs. 30 %), are too small over San Bernardino (50 % vs. 60 %), and extend too far inland (150 km vs. 100 km).
The weekday-weekend pattern of the NO 2 column depends both on patterns of NO x removal and patterns of NO x emissions.The pattern of removal depends on the relationship of OH and RO 2 concentrations to NO x (Fig. 1).In Los Angeles, where NO x concentrations are high (i.e., to the right side of the NO x -OH curve), a decrease in NO x emissions on the weekends results in higher concentrations of OH and RO 2 .As a result, NO x is removed faster, and the reduction in NO 2 columns on the weekend is larger than the reduction in NO x emissions alone.The spatial map of these nonlinear effects reflects the timescale for NO x removal (Fig. 3 right column).If the chemical removal of NO x is slow, the nonlinear decreases will be small over the source where NO x concentrations are high, but will persist and accumulate far downwind where the concentrations eventually decrease to a value that corresponds to a maximum OH concentration and a minimum in the NO x chemical lifetime (Fig. 1).If the chemical removal of NO x is rapid near the source, the nonlinear feedbacks will be large in that location.The discrepancy between the observations and the base simulation (Fig. 3 first row vs. second row) suggests that the model falls into the first category while the observations fall into the second category, an indication that the simulated OH and RO 2 concentrations are too low (i.e., the model is too far to the right of the NO 2 -OH curve presented in Fig. 1).
To test this hypothesis, we evaluate the effects of parameters that influence the concentration of OH and RO 2 , and thus affect the timescale of NO x removal.We double the rate of ozone photolysis to increase the concentration of OH and RO 2 (i.e., shift the entire NO 2 -OH curve in Fig. 1 upward).Separately, we double the anthropogenic emissions of VOC to increase the concentration of RO 2 throughout the model domain.
Figure 3 (third row) shows the simulated weekdayweekend pattern of the NO 2 column when HO x production is increased.Due to the increase in HO x , the rate of NO x removal increases, and both weekday and weekend NO 2 columns are smaller than those simulated in the base model.As a result of higher OH and RO 2 , the timescale of NO x losses is shortened compared to the timescale of transport, and, as hypothesized, the weekend decrease is larger near the source and does not extend as far downwind relative to the base case.Here, we adjust HO x production using ozone photolysis, but other sources of HO x (or Cl), such as HONO, ClNO 2 , or CH 2 O photolysis may play a role.
Figure 3 (fourth row) shows the simulated weekdayweekend pattern of the NO 2 column when VOC emissions are doubled.VOC both removes OH and leads to OH formation through its feedback on HO x sources (e.g., O 3 , CH 2 O).At low NO x concentrations, the role of VOC as a sink of OH dominates, and OH decreases for any increase in VOC.At high NO x , however, the role of VOC as a sink of OH is small, and the concentration of OH increases with increases in VOC.As a result, RO 2 increases everywhere for an increase in VOC emissions, and OH increases where NO x is high (NO 2 column > 7.5 × 10 15 molecule cm −2 ) but decreases where NO x is low (NO 2 column < 5 × 10 15 molecule cm −2 ).Due to higher OH and RO 2 concentrations over Los Angeles, the timescale of NO 2 removal is shortened when VOC emissions are increased relative to the base model, and the agreement of the simulated weekday-weekend pattern with observations improves.Here, we alter VOC concentrations by altering anthropogenic VOC emissions, but biogenic emissions may also be a significant source of VOC in Los Angeles.
The comparison of modeled and observed weekdayweekend patterns of the NO 2 column suggest that HO x chemistry in Los Angeles is NO x -saturated (Fig. 1), but also indicate that the Los Angeles atmosphere is not as far to the right of the NO x -OH-RO 2 relationship as the model simulation.We find that agreement between model and observation is improved when we increase HO x production or VOC emissions in the model, but the improvement is similar for both changes (Fig. 3 -third row vs. fourth row).Despite their similarities, there are significant differences between these two adjustments, particularly differences in the relative roles of RO 2 and OH as NO x sinks (i.e., PAN and RONO 2 vs. HNO 3 ).PAN, for example, is thermally unstable and is much more likely to re-release NO x downwind than is HNO 3 , which is more likely to deposit before undergoing photolysis or reaction with OH.Distinguishing the roles of HNO 3 and PAN chemistry may require more information, such as higher spatial resolution satellite-based measurements at the plume edges where the differences between the two simulations seem the largest, or time-resolved measurements as will become available from geostationary orbit (e.g., GEO-CAPE; Fishman et al., 2008, TEMPO;Chance et al., 2013).
The weekday-weekend pattern of NO 2 columns also depends on weekday-weekend patterns of emissions.The timing of emissions shifts later in the day on weekends (Harley et al., 2005).The delay of emissions on weekends relative to weekdays is expected from a shift in light-duty gasoline vehicle activity later in the day.Weekend emission reductions also vary regionally, and likely locally, due to differing weekday-weekend trends in industry and heavy-duty diesel truck activity relative to passenger vehicle activity (McDonald et al., 2012).
Figure 3 (fifth row) shows the weekday-weekend pattern of NO 2 simulated when we delay the timing of all weekend emissions, both VOC and NO x , by two hours.When weekend emissions are delayed, the 1 p.m. NO 2 column decreases because the total emissions in the hours prior to the satellite overpass decrease.The timing of emissions affects the magnitude of the observed decreases but does not significantly affect the spatial pattern.
Figure 3 (sixth row) shows the weekday-weekend pattern of NO 2 calculated with a different weekend emission reduction over San Bernardino (0.525 × E NO x −NEI2005 ) than over Los Angeles (0.654 × E NO x −NEI2005 ), where the total weekend emission reduction is no different than the base case (0.625 × E NO x −NEI2005 ).Not surprisingly, the contrast between the East basin (San Bernardino) and the West basin (Los Angeles) is larger when the regional weekday-weekend emission pattern is altered, a modest effect, but one that is consistent with observations that show larger decreases over San Bernardino and smaller decreases over the Los Angeles Basin.This pattern might be used in conjunction with detailed vehicle activity assessments to understand whether the changes are consistent with a differential weekend decrease of heavy-duty diesel truck traffic in the two regions.
In the set of simulations discussed above, we show that the NO 2 column and more specifically weekday-weekend patterns of NO 2 column depend strongly on HO x production, VOC, and the timing and spatial distribution of weekend emissions.While no single modification provides an unbiased representation of the observed weekday-weekend pattern over Los Angeles, our comparison of the simulated and observed patterns provides a framework for future work aimed at optimally reproducing the weekday-weekend pattern of NO 2 column over Los Angeles or any location.The launch of geosynchronous UV/VIS instruments such as TEMPO (planned for 2018 - Chance et al., 2013) will provide hourly daytime measurements and sampling every 2.5 km × 4.5 km at the surface and a significant increase in the information available to constrain the processes affecting the pattern of weekend NO 2 decreases and the NO x lifetime.

Conclusions
The high spatial resolution of OMI (13 km × 24 km) captures a phenomenal level of detail over the Los Angeles metropolitan area (200 km × 150 km).Comparisons of these observations to a high-resolution model demonstrate that this spatial detail provides information on local patterns of NO x transport, emissions and chemistry.The observations show that weekend decreases in NO 2 column are smallest over central Los Angeles (25-40 %) where NO 2 columns are high, and are largest downwind of Los Angeles over the San .
Using WRF-Chem, we simulate weekday and weekend NO 2 column using a range of model parameters.In the model, the weekend NO 2 column decreases by more than the weekend emissions reduction, a nonlinear feedback that results from the increase in OH concentration and the corresponding decrease in NO x lifetime that occur when NO x concentrations are reduced.We show that simulated pattern of weekend NO 2 decreases is sensitive to HO x production, VOC emissions, and the spatiotemporal distribution of NO x emissions.The magnitude of the simulated weekdayweekend pattern of column NO 2 is in better agreement with observations when HO x production is increased, when VOC emissions are increased, or when weekend emissions are delayed or shifted spatially.This is a unique conclusion indicating that satellite measurements of NO 2 can simultaneously teach us about weekday-weekend trends in NO x emissions and the coupled weekday-weekend variation of the NO x lifetime.
Further work should focus on the evolution of weekdayweekend NO 2 patterns over time.Its seasonal evolution, for instance, can teach us about chemical feedbacks, as seasonal variations of weekday-weekend emission trends are expected to be small.The interannual evolution, on the other hand, would provide an additional set of constraints on emission trends and the NO x lifetime.
relationship of OH (red) and RO 2 (÷20; blue) concentrations to the concentration NO x ionship derived in Murphy et al. under conditions typical of a polluted summertime urban The lifetime of NO x with respect to reaction of NO 2 with OH is indicated on the right axis.

Fig. 1 .
Fig.1.The steady-state relationship of OH (red) and RO 2 (÷ 20; blue) concentrations to the concentration NO x using the relationship derived inMurphy et al. (2007) under conditions typical of a polluted summertime urban environment.The lifetime of NO x with respect to reaction of NO 2 with OH is indicated on the right axis.

Fig. 2 .
Fig. 2. May-July, 2005-2007 average BEHR NO 2 column (molecules cm −2 ) for weekdays (top panel) and weekends (second panel), and the percent weekend decrease (third panel).MODIS RGB image of Southern California (fourth panel) with the average North American Regional Reanalysis winds for the region (33.7 • -34.0 • N; 117.7 • -118.4 • W) reported in the bottom-left corner, and rectangles marking the San Fernando (NW), Los Angeles (SW) and San Bernardino (E) basins.

Fig. 3 .
Fig. 3.Weekday NO 2 column (molecules cm −2 ), weekend NO 2 column (molecules cm −2 ), and percent change of NO 2 column observed by OMI for May-July 2005-2007 (top row), and simulated with WRF-Chem for 3-14 June 2008 at 13:00 LST using a uniform 37.5 % reduction in NO x emissions with standard RADM2 chemistry (second row), a 2 × increase of J O 3 (third row), a 2 × increase of anthropogenic E VOC (fourth row), a two-hour delay of anthropogenic weekend E NO x and E VOC relative to weekdays (fifth row), and spatially distinct weekend E NO x reductions (sixth row; weekend E NO x −San Bernardino 0.525 × E NO x −NEI2005 , and E NO x −Los Angeles 0.654 × E NO x −NEI2005 , such that the domain-average weekend E NO x is 0.625 × E NO x −NEI2005 ).www.atmos-chem-phys.net/14/1/2014/Atmos.Chem.Phys., 14, 1-9, 2014 • -34.1 • N, 117.8 • -118.4 • W), the San Fernando Valley to the northwest (34.1 • -34.3 • N, 118.2 • -118.8 • W), and the San Bernardino Valley to the east (33.8 • -34.2 • N, 117.0 • -117.8 • W).During the summer, the typical daytime meteorology consists of onshore flow entering the Los Angeles Basin at the coast flowing out through the San Bernardino Valley to the east and the San Fernando Valley to the northwest.

Table 1 .
Table1reports the percentage reduction in NO 2 column over the entire Los Angeles plume (defined as the area over which OMI observes weekday NO 2 columns exceeding 2.5 × 10 15 molecules cm −2 within the domain presented: Summary of observed and simulated weekday-weekend patterns of NO 2 column in Los Angeles.Bold-face entries denote results from simulations shown in Fig.3. www.atmos-chem-phys.net/14/1/2014/Atmos.Chem.Phys., 14, 1-9, 2014 a The Los Angeles plume is determined as the region over which the observed weekday NO 2 column exceeds 2.5 × 10 15 molecules cm −2 within the domain presented (33.3 • N-34.8 • N, 116.25 • W-119.25 • W). b Both weekend E NOx and E VOC are delayed by 2 h.c E NOx−S.Bernardino is 0.575 × 0.525 × E NOx−NEI2005 , and E NOx−Los Angeles is 0.640 × 0.654 ×E NOx−NEI2005 , respectively, for domain-average weekend emission of 0.625 × E NOx−NEI2005 .33.3 • N-34.8 • N, 116.25 • W-119.25 • W) and the fraction of the Los Angeles plume with weekend decreases in NO 2 column that exceed 35 %, 45 %, and 55 %.