Inferring changes in summertime surface ozone-NO_x-VOC chemistry over U.S. urban areas from two decades of satellite and ground-based observations

Abstract

Urban ozone (O₃) formation can be limited by NO_x, VOCs, or both, complicating the design of effective O₃ abatement plans. A satellite-retrieved ratio of formaldehyde to NO₂ (HCHO/NO₂), developed from theory and modeling, has previously been used to indicate O₃ formation chemistry. Here, we connect this space-based indicator to spatiotemporal variations in O₃ recorded by on-the-ground monitors over major U.S. cities. High-O₃ events vary non-linearly with OMI HCHO and NO₂, and the transition from VOC-limited to NO_x-limited O₃ formation regimes occurs at higher HCHO/NO₂ value (3 to 4) than previously determined from models, with slight inter-city variations. To extend satellite record back to 1996, we develop an approach to harmonizing observations from GOME and SCIAMACHY that accounts for differences in spatial resolution and overpass time. Two-decade (1996 – 2016) multi-satellite HCHO/NO₂ captures the timing and locations of the transition from VOC-limited to NO_x-limited O₃ production regime in major U.S. cities, which aligns with the observed long-term changes in urban-rural gradient of O₃ and the reversal of O₃ weekend effect. Our findings suggest promise for applying space-based HCHO/NO₂ to interpret local O₃ chemistry, particularly with the new-generation satellite instruments that allow evaluations at finer spatial and temporal resolution.

INTRODUCTION

Human exposure to ground-level ozone (O₃) is associated with increased risk of cardiovascular and respiratory diseases, and has been linked to 250,000 O₃-related premature deaths in 2015 globally,¹ and 11,700 deaths over the United States (U.S.).² In the troposphere, O₃ is produced from photochemical reactions involving its precursors: nitrogen oxides (NO_x: NO + NO₂) and volatile organic compounds (VOCs). It is well established that O₃ formation throughout much of the troposphere is largely controlled by the availability of NO_x, but in regions with high NO_x emissions, such as metropolitan areas, O₃ formation can be VOC-limited or in transition between these regimes.^3,4 Identifying the most effective emissions control strategy to lower the O₃ exposure of a densely populated metropolitan area requires knowledge of the local O₃ formation chemistry.

While current satellite-based spectrometers do not retrieve ground-level O₃ abundances, they have provided continuous global observations for two species indicative of O₃ precursors, namely nitrogen dioxide (NO₂) for NO_x,^5–7 and formaldehyde (HCHO) for VOC,^8–13 for over two decades. In theory, the ratio of HCHO to NO₂ (HCHO/NO₂) reflects the relative availability of NO_x and total organic reactivity to hydroxyl radicals.^14,15 We build here upon earlier work proposing this satellite-based HCHO/NO₂ serves as an indicator of O₃ sensitivity to its NO_x versus VOC precursors.^16–19 All of these prior studies use theory as represented in models to link column-based HCHO/NO₂ with surface O₃ sensitivity.^16–19 Models, however, can be biased,²⁰ and airborne measurements suggest large uncertainty in the HCHO/NO₂ threshold values between O₃ production regimes.²¹ Also, modeled and satellite retrieved HCHO and NO₂ often disagree,^19,22,23 and the difference varies by satellite retrievals.^19,24 To overcome these limitations, we derive the threshold values marking transitions in O₃ formation regimes entirely from observations by directly connecting space-based HCHO/NO₂ with ground-based measurements of O₃.

Over the U.S., nationwide anthropogenic NO_x emissions are estimated to have declined by 31% from 1997 to 2016.²⁵ Correspondingly, satellite-retrieved NO₂ tropospheric columns are declining,^7,26,27 although relating NO₂ columns directly to NO_x emissions requires accounting for lifetime changes,²⁸ and accurate partitioning between anthropogenic versus background sources of NO₂.^22,29 Despite the widespread decrease of NO_x emissions, observed O₃ trends are heterogeneous in space and time: decreasing in summer over less urbanized areas, and increasing in winter, night and urban cores, due to the non-linear relationship between O₃ production and NO_x.^30–33As NO_x emissions continue to decline, O₃ formation over VOC-limited urban areas is transitioning towards the NO_x-limited regime,^19,34–36 but the observed long-term O₃ trends may also reflect changes in VOC reactivity,³⁷ as well as meteorology.³⁸ U.S. anthropogenic VOC emissions from vehicles and industry are estimated to have declined by 22% from 1997 to 2016,²⁵ while volatile chemical product emissions may be growing.³⁹ Regionally, summertime U.S. VOC emissions are dominated by biogenic sources, particularly highly reactive isoprene, that vary with meteorology and vegetation density.⁴⁰

A key policy-relevant metric is the turning point between VOC-limited and NO_x-limited O₃ formation regimes. It remains uncertain as to which (and whether) U.S. cities have reached this turning point, and how closely long-term changes in O₃ follow transitions in O₃ production regimes, particularly in light of the strong sensitivity of O₃ to meteorological variability.^41,42 Previous studies used observations of HCHO/NO₂ from single satellite instrument, such as Ozone Monitoring Instrument (OMI), which dates back to 2005^18,19,26 The newly developed, consistently retrieved multi-satellite HCHO and NO₂ products, available from the EU FP7-project Quality Assurance for Essential Climate Variables (QA4ECV),^43–49 offers a new opportunity to extend the record back by a decade to 1996. We first assess if space-based HCHO/NO₂ captures the non-linearity of O₃ chemistry by matching daily OMI observation with ground-based O₃ measurements over polluted areas. We find a robust relationship between space-based HCHO/NO₂ and the O₃ response patterns that is qualitatively similar but quantitatively distinct across cities. Next, we link the long-term changes in the harmonized multi-satellite HCHO/NO₂ to changes in urban-rural O₃ gradients and the O₃ weekend effect from 1996 to 2016. We show that this multi-satellite HCHO/NO₂ complements ground-based networks by providing insights into spatial heterogeneity and long-term evolution of O₃ formation regimes, which could be valuable for future applications over regions lacking dense ground-based monitors.

MATERIALS AND METHODS

Multi-satellite Observations of O₃ Precursors.

We use 21-year (1996 – 2016) multi-satellite products of tropospheric NO₂ (Ω_NO2) and HCHO (Ω_HCHO) vertical columns developed under the QA4ECV project that retrieves products consistently from three satellite instruments: Global Ozone Monitoring Experiment (GOME), SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) and OMI.^43–49 The nadir resolution is 24 × 13 km² for OMI, 60 × 30 km² for SCIAMACHY and 320 × 40 km² for GOME. The overpass time is around 1:30 PM local time for OMI, 10:00 AM for SCIAMACHY and 10:30 AM for GOME. The a priori vertical profiles used for QA4ECV products are obtained from the same chemical transport model (TM5-MP),⁵⁰ which are better suited for analyzing space-based HCHO/NO₂ than products developed with different prior profiles. The retrieval algorithms are briefly described in the Supplement (S1). We select daily Level-2 observations with: (1) no processing error; (2) less than 10% snow or ice coverage; (3) solar zenith angle less than 80˚ for NO₂, and 70˚ for HCHO; (4) cloud radiance fractions < 0.5. For OMI, we exclude the first and last five rows, which contain large pixels retrieved on the swath edges, and select the rows 5 to 23, which are unaffected by row anomalies throughout the study period.⁵¹ We grid Level-2 swaths by calculating area weighted averages (S2).

Seasonal Harmonization of GOME, SCIAMACHY and OMI.

To study the long-term changes in HCHO, NO₂ and HCHO/NO₂ (Figures 2 to 4), we construct seasonal average Ω_HCHO and Ω_NO2 from the three satellites by calculating the area-weighted averages from 1996 to 2016. The long-term satellite records are based on OMI observations for the years after 2005, the harmonized SCIAMACHY observations for 2002 – 2004, and harmonized GOME observations before 2002. Even with the consistent algorithms for retrieving NO₂ and HCHO under the QA4ECV project, multi-satellite retrievals still need to be harmonized to account for differences in horizontal resolution, overpass time, and any instrumental offsets. We adjust SCIAMACHY and GOME HCHO and NO₂ data with reference to OMI, because OMI has the finest spatial resolution, and the satellites are best able to capture chemical conditions controlling O₃ production during the OMI afternoon overpass, when mixing depths and O₃ production rates are closest to their daily maxima. We first adjust SCIAMACHY Ω_NO2 by decomposing the instrumental differences between SCIAMACHY and OMI into two factors: 1) those associated with different overpass timing or instrumental offsets, which we estimate as the difference in OMI Ω_NO2 and SCIAMACHY Ω_NO2 during the overlap period (2005 – 2011) at a coarse resolution at which we assume the difference is independent of the instrumental resolution ($\overline{{\text{ΔΩ}}_{\text{NO2\_Coarse}}}$, Figure S1); 2) those caused by resolution (RC_NO2, Figure S2), which we estimate as the relative change in OMI Ω_NO2 at a fine-resolution (0.125˚× 0.125˚ ) versus a coarse-resolution (2˚ × 0.5˚) grid that is close to the nadir resolution of GOME (RC_{NO2_OMI}, S3). While previous studies assumed constant resolution correction factors,^27,52 we find that RC_NO2 varies with time, especially over urban areas, and the spatial gradients in Ω_NO2 are larger earlier in the record when Ω_NO2 is higher earlier in the record (Figure S3). Assuming a time-invariant RC_NO2 may thus underestimate the steepness of spatial gradients at high Ω_NO2. We apply the relative temporal variability estimated from RC_{NO2_SCIA} to the long-term summertime average RC_{NO2_OMI} $\left(\overline{R{C}_{\text{NO2\_OMI}}}\right)$. RC_{NO2_OMI} and RC_{NO2_SCIA} correlate well in time (Figure S4), though their absolute values differ. Combining these factors, the adjusted SCIAMACHY Ω_NO2 (Ω_{NO2_adj}) at year yr season m (we focus on summer, June-July-August) and grid cell x is estimated as follows:

$${\text{Ω}}_{\text{NO2\_adj}}\left(x_f,yr,m\right)=\left({\text{Ω}}_{\text{NO2\_coarse}}\left(x_c,yr,m\right)+\overline{{\text{ΔΩ}}_{\text{N02\_Coarse}}}\left(x_c,m\right)\right)\times R{C}_{\text{NO}2}\left(x_f,x_c,yr,m\right)$$ (1)

Figure 2.

Maps of summertime average: (a) satellite-based Ω_NO2, (b) Ω_HCHO and (c) HCHO/NO₂ for seven cities (New York, Los Angeles, Chicago, Washington DC, Pittsburgh, Atlanta and Houston) in 1996 – 2000 and 2013 – 2016. The white area in (c) indicates HCHO/NO₂ above 6. The numbers show the mean and the range of Ω_NO2, Ω_HCHO, and HCHO/NO₂ for each core-based statistical area (CBSA, outlined in black). The red star shows the location with highest Ω_NO2 in the CBSA. The red circles in the bottom two rows label the locations of three AQS sites where the highest O₃ occurred in the region, and the color represents the summertime mean O₃ (color bar inset in bottom right panel). Maps for 2001 – 2004, 2005 – 2008, and 2009 – 2012 are shown in Figure S11.

Figure 4.

(a) Satellite-based summertime average HCHO/NO₂ in seven cities during five periods. (b)Weekday-to-weekend difference in average 1–2 pm summertime O₃ (weekend ΔO₃, mean O₃ Saturday-Sunday minus mean O₃ Tuesday-Friday) within each city at: (b) high temperature (> median summer average temperature 1–2 pm) observed at AQS sites during five periods. Satellite-based HCHO/NO₂ is sampled over ground-based AQS O₃ sites. The error bars represent year-to-year variability in a given period. (c) Scatter plot between summertime average satellite-based HCHO/NO₂ and the weekend ΔO₃ with colors representing different cities and symbols representing different periods. The blue line is the fitted linear regression line with the 95% confidence interval shaded.

where $\overline{{\text{ΔΩ}}_{\text{NO2\_Coarse}}}\left(x_c,m\right)$ is the difference between OMI Ω_NO2 and SCIAMACHY Ω_NO2 at coarse resolution averaged during the overlap period (n years):

$$\overline{{\text{ΔΩ}}_{\text{N02\_Coarse}}}\left(x_c,m\right)=\frac{1}{n}{\displaystyle {\sum }_{yr=2005}^{yr=2011}\left({\text{Ω}}_{\text{NO2\_OMI\_coarse}}\left(x_c,yr,m\right)-{\text{Ω}}_{\text{N02\_SCIA\_coarse}}\left(x_c,yr,m\right)\right)}$$ (2)

$R{C}_{\text{NO}2}\left(x_f,x_c,yr,m\right)$ ; is the resolution correction factor where x_f is the grid cell at fine resolution, and x_c is the coarse grid cell where x_f falls.

$$R{C}_{\text{NO}2}\left(x_f,x_c,yr,m\right)=\overline{R{C}_{\text{NO}{2}_{-}\text{OMI}}}\left(x_f,x_c,m\right)\times \frac{R{C}_{\text{NO}{2}_{-}\text{SCIA}}\left(x_f,x_c,yr,m\right)}{R{C}_{\text{NO}{2}_{-}\text{SCIA}}\left(x_f,x_c,m\right)}$$ (3)

$$\overline{R{C}_{\text{NO}{2}_{-}\text{OMI}}}\left(x_f,x_c,m\right)=\frac{1}{n}{\displaystyle {\sum }_{yr=2005}^{yr=2016}\frac{{\text{Ω}}_{\text{NO}{2}_{-}\text{OMI\_fine}}\left(x_f,yr,m\right)}{{\text{Ω}}_{\text{NO}{2}_{-}\text{OMI\_coarse}}\left(x_c,yr,m\right)}}$$ (4)

$$R{C}_{\text{NO2\_SCIA}}\left(x_f,x_c,yr,m\right)=\frac{{\text{Ω}}_{\text{N02\_SCIA\_fine}}\left(x_f,yr,m\right)}{{\text{Ω}}_{\text{N02\_SCIA\_coarse}}\left(x_c,yr,m\right)}$$ (5)

$$\overline{R{C}_{\text{NO2\_SCIA}}}\left(x_f,x_c,m\right)=\frac{1}{n}{\displaystyle {\sum }_{y=2002}^{y=2012}R}{C}_{\text{NO2\_SCIA}}\left(x_f,x_c,yr,m\right)$$ (6)

To harmonize GOME Ω_NO2, we apply the same correction factors that we applied to SCIAMACHY except that the temporal variability in RC_NO2 is driven by the variability in RC_NO2 of GOME. We do not adjust for any systematic differences between GOME and SCIAMACHY at coarse resolution, because the overpass time is close, and the overlap period (August 2002 to June 2003) does not cover an entire summer.

We similarly decompose the instrumental differences in Ω_HCHO to differences caused by resolution (RC_{HCHO_OMI}) versus overpass time (ΔΩ_{HCHO_Coarse}). We find that RC_{HCHO_OMI} is much smaller than ΔΩ_{HCHO_Coarse}, and the spatial pattern of RC_{HCHO_OMI} tends to be noisy (Figure S5). We find little resolution dependence of the difference between OMI Ω_HCHO and SCIAMACHY Ω_HCHO, likely due to widespread summertime isoprene emissions, the dominant summertime precursor to HCHO over the U.S., as well as HCHO produced during oxidation of longer-lived VOCs.⁵¹ Therefore, we do not apply a resolution correction to SCIAMACHY Ω_HCHO or GOME Ω_HCHO. We calculate the climatology of the systematic difference $\left(\overline{{\text{ΔΩ}}_{\text{HCHO}}}\right)$ between OMI Ω_HCHO and SCIAMACHY Ω_HCHO at 0.25˚× 0.25˚ resolution, and adjust Ω_HCHO (Ω_{HCHO_adj}) by applying these differences to the original SCIAMACHY and GOME Ω_HCHO (Ω_{HCHO_Ori}) for the years without OMI observations:

$${\text{Ω}}_{\text{HCHO\_adj}}(x,yr,m)={\text{Ω}}_{\text{HCHO\_Ori}}(x,yr,m)+\overline{{\text{ΔΩ}}_{\text{HCHO}}}(x,m)$$ (7)

$$\overline{{\text{ΔΩ}}_{\text{HCHO}}}(x,m)=\frac{1}{n}{\displaystyle {\sum }_{yr=2005}^{yr=2011}\left({\text{Ω}}_{\text{HCHO\_OMI}}(x,yr,m)-{\text{Ω}}_{\text{HCHO\_SCIA}}(x,yr,m)\right)}$$ (8)

The systematic difference is mainly attributed to the diel cycle in HCHO.⁵⁴ As we adjust the morning retrieval of HCHO with respect to the afternoon retrieval, upward adjustment is expected due to the diel cycle in temperature, which controls biogenic VOC emissions, and in OH, which controls HCHO production from its parent VOCs (Figure S5).^40,55

Connecting Satellite HCHO/NO₂ with Ground-based O₃ observations.

We use observations of hourly O₃ from the U.S. Air Quality System (AQS) from 1996 to 2016. We first aggregate daily OMI data (used in Figure 1) by sampling the gridded daily OMI Ω_HCHO and Ω_NO2 coincident with ground-based observations of O₃ at 0.125˚×0.125˚ resolution. Retrievals from SCIAMACHY and GOME are not used for daily analysis because harmonization at the daily time scale is unrealistic. We average hourly O₃ measurements at 1 PM and 2 PM local time to match the OMI overpass time. We first select 1,221 O₃ monitors located in polluted regions, defined as summertime 2005–2016 average OMI Ω_NO2 > 1.5×10¹⁵ molecules/cm². OMI retrieved Ω_NO2 and Ω_HCHO are sampled daily over AQS O₃ sites for the warm season (May to October) from 2005 to 2016, yielding over 700,000 paired observations, and we calculate the probability of O₃ exceeding 70 ppb from this dataset. Next, we focus on seven metropolitan areas to evaluate the satellite-based HCHO/NO₂ and study the long-term evolution of O₃ production regimes from 1996 to 2016, as the resolution of the harmonized satellite products (~ 10 km) is more suitable for studying cities spanning larger areas. We first select Los Angeles, New York, Chicago, the three most populous cities in the U.S.A. We then include four additional cities: Washington DC, Pittsburgh, Atlanta and Houston, where long-term ground-based observations of O₃ and NO_x are available, and which also cover different U.S. climate regions. To assess if satellite HCHO/NO₂ captures the long-term changes in O₃ production regimes, we include ground-based measurements of O₃ from 1996 to 2016 in each of the seven cities and their surrounding rural areas, from which we analyze the changes in urban-rural O₃ gradient, and the weekday-to-weekend differences defined as weekend (Saturday-Sunday) O₃ - weekday (Tuesday-Friday) O₃.

Figure 1.

(a) Probability of O₃ exceeding 70 ppbv (high-O₃ probability) as a function of OMI Ω_NO2 and Ω_HCHO. All ground-based hourly O₃ observations (averaged at 1 PM and 2 PM local time) in the warm season (May to October) from 2005 to 2016 are first aggregated based on corresponding daily OMI Ω_NO2 and Ω_HCHO (interval: 0.5 ´ 10¹⁵ molecules/cm²). We only include sites over polluted regions (defined as long-term average OMI Ω_NO2 > 1.5×10¹⁵ molecules/cm²). The probability is the number of observations with O₃ higher than 70 ppbv divided by the total number of observations at given OMI Ω_NO2 and Ω_HCHO. The black lines delineate OMI HCHO/NO₂ values of 2 and 4. (b) Probability of O₃ exceeding 70 ppbv as a function of OMI HCHO/NO₂ for all selected sites (black) and seven cities individually. High-O₃ probability is calculated by first matching hourly O₃ observations with daily OMI HCHO/NO₂, dividing these paired observations to 100 (200 for black dots) bins based on OMI HCHO/NO₂, and then calculating the high-O₃ probability (y axis) for each OMI HCHO/NO₂ bin (x axis, labeled as a dot). The solid lines are fitted third order polynomial curves, and the shading indicates 95% confidence intervals. The vertical lines indicate the maximum of the fitted curve (labeled in the legend), and the vertical shading represents the range over the top 10% of the fitted curve (regime transition). The uncertainty is two standard deviation (2σ or 95% confidence interval) of the derived peaks using statistical bootstrapping by iteratively running the model on 50 randomly selected subsets of 30 data pairs.

RESULTS AND DISCUSSION

Nonlinear O₃ Chemistry Captured by Satellite-based HCHO/NO₂.

We first evaluate if satellite-based HCHO/NO₂ can capture the well-established non-linearities in O₃ chemistry. Pusede et al.⁵⁶ proposed a conceptual framework that uses the observed O₃ exceedance probability to interpret the non-linear dependence of O₃ production on precursor emissions. This framework assumes stagnant meteorology so that measured O₃ is sensitive to its local chemical production, and the local changes in chemical or depositional loss are insignificant on average. We follow this approach by calculating the probability that surface O₃ exceeds 70 ppbv (high-O₃ probability) at OMI overpass, given the OMI Ω_NO2 and Ω_HCHO (Figure 1a). Figure 1a, derived solely from observations, resembles O₃ isopleths that are typically generated with analytical models.^4,57 Consistent with O₃ isopleths, three regimes can be roughly identified from Figure 1a: (1) high Ω_NO2 and low Ω_HCHO, where high O₃ events become more likely at lower NO_x, indicating NO_x-saturated (or VOC-limited) chemistry; (2) low Ω_NO2 and relatively high Ω_HCHO, where the probability of high O₃ events increases with Ω_NO2, indicating NO_x-limited chemistry; (3) high Ω_NO2 and high Ω_HCHO, where the probability of high-O₃ events peaks, and increases with both Ω_NO2 and Ω_HCHO. While Figure 1a resembles this overall O₃-NO_x-VOC chemistry, the high O₃ probabilities span a broad range, with an uncertain, blurry transition between NO_x-limited and VOC-limited regimes. The lack of sharp transitions between O₃ production regimes in Figure 1a likely reflects the influence from other factors such as varying meteorology, chemical and depositional loss of O₃, noisy satellite retrievals, the spatial mismatch between the area satellite observations and the point measurements of surface O₃, and in some cases, small sample size that lacks statistical power to calculate high-O₃ probability. Despite these uncertainties, Figure 1a qualitatively illustrates the non-linear relationship between the occurrence probability of high-O₃ events and the HCHO and NO₂ proxies for precursor VOC and NO_x, respectively.

Having established this qualitative approach, we next derive quantitative relationships by calculating high-O₃ probabilities at given OMI HCHO/NO₂ and examining their statistical relationship across different U.S. cities. We investigate three possible empirical relationships by applying moving average, second-order polynomial and third-order polynomial models to observations over seven U.S. cities (Figure S6). The third-order polynomial model is used to derive the maximum high-O₃ probability (the peak of the curve in Figure 1b), because it best fits data, with the smallest uncertainty (estimated with statistical bootstrapping, Figure S6) and higher correlation coefficient (R) than second-order model. Assuming that the peak of the curve marks the transition from VOC-limited to NO_x-limited regime,⁵⁶ we define the transitional regime as the range of HCHO/NO₂ spanning the top 10% of the high-O₃ probability distribution.

Aggregating over all available observations used in Figure 1a, we find that the high-O₃ probability peaks at HCHO/NO₂ = 3.6, with the transitional regime ranging from 3.2 to 4.1, hereafter as [3.2, 4.1]. Evaluating the relationship for the seven cities individually, we find robust non-linear relationships between the high-O₃ probability and OMI HCHO/NO₂, despite differences in the overall high-O₃ probability, which reflect other factors such as emissions, meteorology, and transport. The HCHO/NO₂ marking the regime transition varies slightly among these cities, which is highest for LA (4.5 [4.1, 5.0]), and lowest for Houston (3.0 [2.6, 3.5]). We evaluate the uncertainty in the derived peaks using statistical bootstrapping by iteratively applying the model to 50 randomly selected subsets of the data. We define the uncertainty as two standard deviation (2σ or 95% confidence interval) the derived maxima. The uncertainty is generally within 2 except for Atlanta (2σ = 2.8) and Houston (2σ = 2.4), where the fitted curve is relatively flat. Separating the observations into two periods (before and after 2009), the derived thresholds are slightly higher in the later period, which may reflect more high HCHO/NO₂ values in the recent period, which drive the curve to move towards a higher turning point, but the uncertainty also increases as we halve the number of observations (Figure S7).

The HCHO/NO₂ thresholds derived in Figure 1b are higher than previously reported model-based values,^16,17,19 implying that at the same HCHO/NO₂, our observation-based approach suggests O₃ production is more VOC-limited. The difference originates from the distinct approaches used to link HCHO/NO₂ with O₃ production regimes. Previous modeling studies derive the threshold by simulating the response of surface O₃ to an overall reduction in NO_x or NMVOC emissions with coarse resolution models, which best capture regional as opposed to local O₃-NO_x-VOC sensitivity.^16,19 Our thresholds derived with in situ observations should be more indicative of the local O₃ chemistry, including the effect of NO_x titration over urban areas. Schroeder et al.²¹ also found VOC-limited chemistry occurring at high HCHO/NO₂ (1.3 ~ 5.0) in their analysis of column HCHO/NO₂ from aircraft measurements.

Declining NO₂ Over Time.

Figure 2a shows summertime average Ω_NO2 over seven metropolitan areas in 1996 – 2000 versus 2013 – 2016 produced from the harmonized multi-satellite data. NO₂ is concentrated over urban areas and near combustion sources. Applying the resolution corrections to GOME NO₂ reveals spatial gradients not detected directly with the coarse resolution of GOME (Figure S8). We find the largest urban-rural gradients in NYC and LA, where Ω_NO2 varies by a factor of ten within their core-based statistical areas (CBSA, outlined in Figure 2). Satellite observations show large decreases in Ω_NO2 over the past two decades, consistent with previous studies (Figure 2a).^7,52,58 The mean Ω_NO2 in each CBSA has decreased by 40% (Atlanta) to 56% (LA) in 2013 – 2016 relative to 1996 – 2000. We use ground-based measurements of NO_x to evaluate the long-term changes of satellite-based Ω_NO2, since our approach assumes Ω_NO2 is a good indicator of ground-level NO_x. Satellite-based Ω_NO2 captures the decrease of ground-level NO_x over LA, Chicago, and Washington to within 5%, but underestimates the decrease over NYC, Pittsburgh, and Houston, while overestimates the decreases in Atlanta (Figure S9). Both satellite-based Ω_NO2 and ground-level NO_x show the largest decline before 2004 over Pittsburgh, associated with emission controls on coal-fired power plants.^59,60 Satellite-based Ω_NO2 does not show decreases over NYC and Houston before 2000, but ground-based NO_x suggests large decreases (Figure S10). This discrepancy is likely due to the coarse resolution of GOME; while we have corrected the spatial patterns of GOME Ω_NO2, the total Ω_NO2 may still be biased low, due to the contributions from the nearby ocean where NO₂ is low. Satellite-based Ω_NO2 does capture the large decreases between 2005 and 2012 in NYC and Houston (Figure S10). Over LA, Chicago, Washington and Atlanta, both satellite and ground-based observations suggest the largest reductions occurred between 2005 to 2012 (Figure S10), when emission controls on power plants and stricter vehicle emission standards were implemented.^26,61 The substantial decrease in 2008 – 2010 may also reflect the economic recession.^7,61 In the most recent period (2013 – 2016), satellite data show flattening trends in Ω_NO2 in all seven cities (Figure S10), possibly related to a slowdown of NO_x emission reductions,²⁹ changes in NO_x lifetime,²⁸ and the relatively larger influence of upper tropospheric NO₂ as anthropogenic contributions declines.²²

Heterogeneous Trends of HCHO.

Figure 2b compares summertime multi-satellite Ω_HCHO in 1996 – 2000 versus 2013 – 2016. The spatial patterns of HCHO over the U.S. are largely driven by variations in biogenic VOCs, especially isoprene, which is mainly emitted from broadleaf trees, and is most abundant in the southeastern USA.⁵³ As expected, the mean Ω_HCHO is highest over the southeastern city of Atlanta, followed by Washington and NYC. Ω_HCHO shows strong inter-annual variability (Figure S12), driven by inter-annual variability of meteorology, temperature in particular.^55,62 Over urban areas, satellite-based Ω_HCHO decreased by 7% in LA, 4% in NYC, 3% in Pittsburgh, 4% in Atlanta and 3% in Houston in 2013 – 2016 relative to 1996 – 2000 (Figure S9), consistent with the widespread reduction of anthropogenic VOC emissions.²⁵ Over surrounding rural areas, satellite-based Ω_HCHO decreased near LA, Washington, Atlanta, and Houston, but increased near New York, Chicago, and Pittsburgh. These changes in Ω_HCHO correspond to estimated long-term changes in isoprene emissions (Figure S13), which have previously been shown to be related to changes in vegetation coverage.⁶³ In addition, NO_x reductions could lead to lower the HCHO yield from isoprene oxidation,^64,65 but the available observations are insufficient to conclusively determine the changes in HCHO yield. Overall, long-term changes in HCHO are driven by several factors,⁶⁶ such as anthropogenic and biogenic emissions, OH abundance, and HCHO yield dependence on NO_x, which warrant further investigation as more measurements become more available.⁶⁷ Most relevant to our study is that the overall changes in HCHO are much smaller than the NO₂ changes over the last two decades (Figure 2).

Spatial Expansion of NO_x-limited Regime Over Time.

As Ω_NO2 decreased over time, while changes in Ω_HCHO were relatively small, satellite-based HCHO/NO₂ increased from 1996 – 2000 to 2013 – 2016, indicating a shrinking extent of NO_x-saturated O₃ formation in urban areas (Figure 2c). Using the thresholds derived from Figure 1b to identify the O₃ production regimes, NO_x-saturated chemistry existed during summer in all cities during 1996 – 2000, with the largest areal extent in Pittsburgh. By 2013 – 2016, NO_x-saturated chemistry only occurred in the center of LA, Chicago and NYC. The spatial expansion of the NO_x-limited regime suggests that NO_x emission reductions are more effective today at reducing O₃ pollution, as confirmed from prior modeling^35,36,68 and ground-based observational studies.^34,56 In recent years, as Ω_NO2 remains at low levels, Ω_HCHO plays a more important role in determining the spatial and temporal variability in HCHO/NO₂. For example, the mean Ω_HCHO over LA is 8.2 × 10¹⁵ molecules/cm² in 2010, but increases to 15.2 × 10¹⁵ molecules/cm² in 2011, leading to the mean O₃ formation regime to shift from NO_x-saturated to NO_x-limited (Figure S14). Also, Atlanta and Pittsburgh show similar Ω_NO2 in 2013 – 2016, but Ω_HCHO is 50% higher in Atlanta, leading to 76% higher HCHO/NO₂ and thus more NO_x-limited chemistry in Atlanta, consistent with the well-understood regional differences in summertime O₃ sensitivity.^69,70

LA, NYC and Chicago are the three cities where we find strong urban-rural gradients in HCHO/NO₂, where O₃ production transitions from NO_x-saturated at city centers towards a NO_x-limited regime over rural areas in both periods. Figure 3 shows summertime average satellite-based NO₂, HCHO and HCHO/NO₂ as a function of the distance to the city center during 1996 – 2000 and 2013 – 2016 over these three cities. Satellite observations detect large urban-rural gradients of NO₂ in LA and NYC with 20 × 10¹³ molecules/cm²/km in 1996 – 2000, which decrease to 8 × 10¹³ molecules/cm²/km in 2013 to 2016. The urban-rural gradient has decreased from 11 × 10¹³ molecules/cm²/km to 3 × 10¹³ molecules/cm²/km in Chicago. We find a small enhancement of Ω_HCHO in urban areas over NYC and LA of 2 to 3 × 10¹³ molecules/cm²/km, and negligible urban-rural difference of Ω_HCHO in Chicago. The urban-rural gradient of OMI HCHO/NO₂ is therefore mainly driven by the variations in NO₂. Using the regime thresholds we estimated, we infer the regime transition occurred at 110 to 130 km away from the city center in LA, 80 to 120 km in NYC, and 120 to 130 km in Chicago in 1996 – 2000. By 2013 – 2016, the locations of regime transition have moved closer to the city centers: 50 to 70 km for LA, 40 to 60 km for NYC and 30 to 60 km for Chicago (Figure 3).

Figure 3.

Satellite-based summertime Ω_NO2 (blue dots), Ω_HCHO (green dots), HCHO/NO₂ (red dots) and summertime average O₃ (bars) as a function of distance to the city center during 1996 – 2000 and 2013 – 2016 for three cities: (a) Los Angeles, (b) New York, and (c) Chicago. City center is defined as the grid cell with highest summertime Ω_NO2 within this region (labeled as red stars in Figure 2), which we find do not change over time in these cities (Figures 2 and S11). The curves shown in the top row are a polynomial fit (third order for Ω_NO2 and HCHO/NO₂, second order for Ω_HCHO) curves. The gray area indicates the regime transitions for HCHO/NO₂, which is derived for each city individually as shown in Figure 1b. Summertime average O₃ is calculated from hourly AQS observations at OMI overpass time (averaged at 1 PM and 2 PM local time). AQS O₃ sites are grouped by distance to the city center at 20 km intervals.

Observed Response of Ground-level O₃ to Regime Transitions.

Theoretically, O₃ production regime transitions should correspond to the conditions at which O₃ formation is most efficient.⁵⁷ As the regime transition moves closer to populated city centers, peak O₃ production efficiency is expected to move towards the city center. We hypothesize that we should observe the highest O₃ concentration where the transitional regime occurs, assuming that local changes in meteorology, chemical and depositional loss do not contribute strongly to the observed summertime mean urban-to-rural O₃ gradients. We find that the ground-based sites measuring the highest summertime mean O₃ in each region move towards the city centers over time, except for Atlanta and Houston, where the highest O₃ is found near the city center in both periods (Figure 2c). We aggregate ground-based O₃ sites based on their distance to the city center for LA, NYC and Chicago (Figure 3), where the VOC-limited regime still existed in 2013 – 2016. As expected, peak O₃ has moved towards the city center from 1996 – 2000 to 2013 – 2016 in LA and Chicago: from ~100 km to ~60 km in LA, from 120 km to 20 km in Chicago. The locations of peak O₃ are largely consistent with the locations of the regime transition identified by the satellite-based HCHO/NO₂. In NYC, we find that O₃ peaks ~140 km away in 1996 – 2000, which is consistent with the regime transition inferred from satellite-based HCHO/NO₂. In 2013 – 2016, O₃ shows peaks at 100 and 160 km, however, which may be due to the non-circular nature of city shape and possibly confounding role of nearby ocean. If we only consider the small region within 100 km, O₃ peaks at 40 km away from the city center, more consistent with the regime transition inferred from satellite-based HCHO/NO₂.

Regionally, surface O₃ in summer has decreased over the past two decades over the USA, especially over the eastern USA.^30,33,71,72 As expected, summertime mean O₃ is smaller in 2013 – 2016 than 1996 – 2000 over the three megacities, but the reduction is larger over rural areas where O₃ formation falls in the NO_x-limited regime (Figure 3). The faster decline in O₃ over rural areas than urban areas has previously been demonstrated.³³ In NYC and Chicago, we find an increase in O₃ at the city center where O₃ formation is NO_x-saturated. In the NO_x-saturated regime, NO_x emission reductions decrease NO_x titration, which increases O₃ directly, and also increases OH available for VOC oxidation and subsequent O₃ production. The spatial difference between maximum and minimum O₃ narrows from 13 ppbv in 1996 – 2000 to 7 ppbv in 2013 – 2016 in NYC, and 10 ppbv to 2 ppbv in Chicago. In LA, O₃ decreases in urban areas, which we attribute to decrease in anthropogenic VOC emissions.⁷³ The largest O₃ decreases occur in the transitional regimes in LA, where reductions in both anthropogenic VOCs and NO_x lower O₃.

Reversal of the O₃ Weekend Effect.

The decrease of urban NO_x emissions associated with road traffic on weekends provides an observation-based natural test for investigating O₃ sensitivity to NO_x emissions;^74,75 over urban areas where O₃ formation is NO_x-saturated, reduction of NO_x emissions on weekends increases in O₃ (referred to as the O₃ weekend effect). Figure 4 shows the mean satellite-based HCHO/NO₂ sampled over long-term O₃ sites for the seven selected cities in five periods, and the corresponding in situ observed weekday-to-weekend difference in average summertime O₃ (weekend ΔO₃) within each metropolitan area. Here we only select days with high temperature (> median summertime average), as they are generally associated with high pressure, clearer skies and slower winds, conditions suitable for efficient O₃ production.^37,57 As O₃ production becomes more sensitive to NO_x, the weekend ΔO₃ lessens and even reverses in some cities. The extent of the NO_x-saturated regime is largest in LA, as suggested by the lowest average satellite HCHO/NO₂ (Figure 4a). The O₃ weekend effect in LA persists from 1996 to 2016, but is smallest in the most recent period. During 1996 – 2000, we find a positive weekend ΔO₃ in 18 (11 with p < 0.1) out of 20 sites along southern California (Figure S15), but only 11 out of 18 sites (5 with p < 0.1) during 2013 – 2016. The shrinking O₃ weekend effect after 2000 in LA is reported in previous studies.^76,77 Chicago has the second lowest HCHO/NO₂, and the weekend ΔO₃ changes from positive to negative in 2009 – 2012. Over Chicago, the O₃ weekend effect is strongest during 2001 – 2004, when 32 out of 34 sites show positive weekend ΔO₃, and diminishes to 10 out of 23 sites during 2013 – 2016 (Figure S15). The reversal of O₃ weekend effect occurs earlier in 2001 – 2004 over NYC, Pittsburgh, and Washington, though satellite HCHO/NO₂ does not change much compared with 1996 – 2000. In Houston, we find the reversal of weekend ΔO₃ occurs around 2009 – 2012. In Houston, 17 out of 24 sites show positive weekend ΔO₃ on weekends during 2001 – 2004, but they all changed sign during 2013 – 2016 (Figure S15). In Atlanta, where O₃ formation is most NO_x-limited based on our metric, O₃ concentration remains lower on weekends than weekdays at high temperature, but a reversal of the O₃ weekend effect does occur at moderate temperature during 2005 – 2008 (Figure S16).

The observed long-term changes in the O₃ weekend effect are overall consistent with the increasing sensitivity to NO_x, as suggested by the increasing satellite-based HCHO/NO₂ (Figure 4a). We find that satellite-based HCHO/NO₂ and weekend ΔO₃ is moderately correlated (R = −0.57, p < 0.001, Figure 4c). The regression line intercepts 0 at HCHO/NO₂ = 3.4, which is close to the regime transition derived in Figure 1b. Using this satellite-based indicator to quantitatively predict the occurrence of O₃ weekend effect in any particular city for a given time period, however, is subject to uncertainties. The definition of the O₃ weekend effect we invoke here assumes that the only difference in O₃ is directly attributable to changes in NO_x emission. The observed O₃ differences, however, may also be influenced by variability in meteorology.^42,78 The early reversal of the O₃ weekend effect in 2001 – 2004 over northeastern cities (NYC, Washington and Pittsburgh) is better explained by the overall colder temperature on weekends than weekdays over these three cities (Figure S17). Pierce et al.⁴² suggest the long-term trend in the O₃ weekend effect over the northeast USA is strongly influenced by the inter-annual variability in meteorology. We find larger fluctuations of the weekend ΔO₃ at moderate temperatures in most cities except for LA (Figure S16), which may be related to meteorological conditions that act to weaken urban-to-rural gradients through regional-scale O₃ transport that dilutes the signal of local urban O₃ production.

Limitations and Future Directions.

Our study is the first attempt to directly connect satellite-based HCHO/NO₂ with ground-based O₃ observations. We show that space-based HCHO/NO₂ captures the nonlinearities of O₃-NO_x-VOC chemistry, and the detected spatial expansion of the NO_x-limited regime is supported by ground-based observations. However, using satellite HCHO/NO₂ to quantitatively diagnose the effectiveness of emission controls is subject to following uncertainties that warrant further investigation. First, theoretical studies that relate indicator ratio to O₃-NO_x-VOC sensitivity show variations among different locations, which are subject to uncertainties of deposition and interactions with aerosol.^14,79 Second, satellite instruments measure the vertically integrated column density, and inhomogeneities in vertical distributions degrade the ability of satellite-based column HCHO/NO₂ to identify the near-surface O₃ sensitivity.^19,21 Third, we use an empirical observation-based approach to derive the thresholds making the transitions between chemical regimes, which are likely to be affected by not only biases in the satellite retrieval algorithms,¹⁹ but also by sampling size and biases of both ground-based and space-based observations. Fourth, the extent to which satellite-based Ω_HCHO relates to local surface organic reactivity is unclear. Satellite-based Ω_HCHO shows small decreasing trends over urban areas, and are mostly insensitive to observed decreases in anthropogenic VOCs,^56,73 partially due to relatively small HCHO yields from some classes of anthropogenic VOCs (e.g. alkanes).⁸⁰ As HCHO is a weaker UV-Visible absorber than NO₂, satellite retrieval of HCHO is more prone to errors,⁴⁶ which may limit its ability to detect HCHO from local sources of anthropogenic VOCs. We find small enhancements of satellite Ω_HCHO over urban areas, but the magnitudes of the enhancement are insensitive to resolution, suggesting satellite Ω_HCHO is more indicative of the regional VOC reactivity, which is mainly influenced by biogenic isoprene emissions across much of the U.S.A. in summer.⁸¹ Finally, although the retrieval uncertainty associated with different instruments has largely been reduced in the QA4ECV products,⁴⁹ our applications of satellite-based HCHO/NO₂ are nonetheless limited to long-term averages or data aggregations sufficiently large sample sizes to reduce retrieval noise. It is challenging to use current satellite retrievals to observe short-term variability and detailed spatial patterns within urban cores. The new generation of satellites, including the newly launched TROPOMI aboard Sentinel-5P, and the upcoming geostationary satellite instruments such as TEMPO will offer an unprecedented view to characterize the near-surface O₃ chemistry at finer spatial and temporal scales.^82,83