Changes in US background ozone associated with the 2011 turnaround in Chinese NOx emissions

The chemical data assimilation technique used here is based on a local ensemble transform Kalman filter (LETKF) approach [32] that uses an ensemble forecast to estimate the background error covariance matrix [27, 33]. This approach has been implemented within four different CTMs to construct a Multi-mOdel Multi-cOnstituent Chemical data assimilation (MOMO-Chem) framework [27]. This framework directly accounts for model errors in transport and chemistry and quantifies uncertainties in the data assimilation analysis due to model errors. Parts of the MOMO-Chem system have been used to produce the Tropospheric Chemistry Reanalysis version 1 (TCR-1) and version 2 (TCR-2) products [33, 34].

In this study, we used the GEOS-Chem model within the MOMO-Chem system to evaluate NOx emission changes and their influence on ozone. GEOS-Chem has been widely used to quantify US background ozone [35, 36]. The GEOS-Chem model is driven by assimilated meteorological data from the NASA Global Modeling and Assimilation Office (GMAO) Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) [37]. The adjoint model version 35 [38], which corresponds to version 9 of the forward model, with a horizontal resolution of 2° × 2.5° and 47 vertical levels extending from the surface to 0.1 hPa, was used here as a forward forecast model, without adjoint calculations. The a priori emission data were obtained from the Emission Database for Global Atmospheric Research (EDGAR) version 3 inventory [39] for anthropogenic emissions and from the Global Fire Emissions Database (GFED) version 2 inventory [40] for biomass burning emissions. Emission data for North America were replaced with the 2008 National Emissions Inventory (NEI). Because of strong observational constraints by OMI, the choice of prior emissions had a limited influence on the optimized NOx emissions in most cases [23, 41]. Note that, even the most recent emission inventories have large uncertainties, whereas top-down estimates provide unique information on emission variations and its impacts on ozone [16]. We also use results from the full MOMO-Chem system [27] to provide unique information on the model-related uncertainty of the estimated NOx emissions and of the ozone response to NOx emission perturbations.

Satellite measurements of tropospheric NO₂ columns and of stratospheric ozone profiles over the globe were assimilated in this study. This set-up was designed to explore the impact of changing NOx emissions on surface and tropospheric ozone in a relatively simple yet realistic setting. Assimilation of satellite tropospheric ozone retrievals was not included in this study, except in the context of an observing system experiment (section 3.3), because the measurements lack sensitivity near the Earth's surface and cannot be used in the context of separating out background and local contributions to ozone concentrations. This is in contrast to our previous data assimilation framework, which aimed to provide comprehensive constraints on the entire tropospheric chemistry system through assimilation of satellite measurements of NO₂, CO, SO₂, HNO₃, and ozone to constrain NOx, CO, and SO₂ emissions as well as concentrations of many chemical species, [24, 26, 29, 33, 34]. Nevertheless, the use of the state-of-the-art assimilation, based on flow-dependent error covariance, allows us to provide improved understanding of emission variations. We used the tropospheric NO₂ column retrievals from the QA4ECV version 1.1 Level 2 (L2) product for OMI [42] to constrain NOx emissions, employing a super-observation approach [23] to generate representative data with a horizontal resolution of the forecast model. Version 4.2 ozone L2 products from the Microwave Limb Sounder (MLS) [43] were used to constrain ozone concentrations in the stratosphere. No tropospheric ozone data were assimilated in the standard data assimilation calculations. The data assimilation calculations were conducted for each year from 2006 to 2016 separately, starting from December 1st of the previous year to provide a month spin-up run. In order to measure the impact of the assimilation, model control runs without any assimilation were also conducted.

We also performed a three-month calculation in which we assimilated tropospheric ozone profiles from the joint AIRS/OMI version 1 L2 ozone profile product [21], starting on April 1st, 2016. This assimilation is used to evaluate the impact of tropospheric ozone measurements over Asia on our understanding of the global distribution of ozone. The ozone profiles were retrieved using the JPL MUlti-SpEctra, MUlti-SpEcies, Multi-Sensors (MUSES) algorithm that combines AIRS and OMI level 1B (L1B) spectral radiances. The AIRS/OMI ozone profile product, containing both global survey (GS) and regional mapping (RE) mode data, are publicly available via the Aura Validation Data Center website (https://avdc.gsfc.nasa.gov/pub/data/satellite/Aura/TES/AIRS_OMI/O3). The GS mode AIRS/OMI data have been produced with spatial sampling and retrieval characteristics similar to the TES L2 standard data product, which shows good agreement with WOUDC global ozonesonde measurements, with seasonal and global mean biases of 0.9-14.4 ppbv at 750 hPa, 2.2-5.9 ppbv at 510 hPa, and 7.7-2.9 ppbv at 316 hPa [21]. The RE mode data provides more dense spatial coverage for a specific area. RE data over East Asia produced for the 2016 KORUS-AQ field campaign [26] were used in this study.

Data assimilation corrects biases in bottom-up emission inventories and improves understanding of long-term emission changes [16, 24]. As shown in figure 1, the estimated Chinese emissions (108°E–123 °E, 20 °N–44 °N) increased from 2006 to 2012 by about 30% and decreased about 20% from 2013 to 2015, with smaller decreases during 2015 and 2016. Interestingly, there are substantial spatial differences in the estimated positive and negative trends. These variations are attributable to the competing influences of economic development and air quality controls [31], as commonly suggested by previous studies [44–46]. For the Western US (105° W–130 °W, 30 °N–50 °N), the estimated emissions show a rapid reduction of 21% from 2006 to 2010, a slight increase for 2011-2013, and a slow reduction from 2013 through 2016. These emission changes at regional and country scales are broadly consistent with our previous estimates using TCR-1 [24, 47] and TCR-2 [33]. In contrast, even most recent regional inventories, such as an updated NEI inventory over the US and MEIC and CAPPS, still show distinctly different emission patterns from top-down estimates [47, 48], which would provide different ozone changes. The uncertainty of these emissions are estimated to be 21% for East China and 17% for the US based on the multi-model spread of the posteriori emissions using MOMO-Chem [27]. In spite of large differences among the inventories, the choice of prior emissions provided only a limited influence on the posteriori emissions (not shown). This confirms that the OMI measurements provide sufficient constraint on NOx emissions.

Zoom In Zoom Out Reset image size

The ozone analysis from GEOS-Chem has already been validated against independent observations, such as those from ozonesondes, within the MOMO-Chem framework [27]. This evaluation confirmed that data assimilation substantially reduces model errors globally. Here we show an evaluation of the GEOS-Chem ozone data assimilation analysis used in this paper against TES L2 tropospheric ozone retrievals [17] over China and the western US for 2005-2010, when Chinese NOx emissions increased rapidly. The TES retrievals have been extensively validated against ozonesonde observations [49, 50] and provide unprecedented information on global ozone changes for this early time period, including for regions without any ozonesonde observations such as China. Previous work found that the TES retrievals show a large, significant increase in tropospheric ozone at 3-9 km over eastern China (1.08%/year) for the period 2005-2010, and a smaller, insignificant increase over the western US (0.56%/year) [8]. This same work used the TM5 model to show that increases in ozone from China offset about half of the reductions in mid-tropospheric ozone over the Western US associated with US air quality regulations during this period [8]. Changes in surface ozone, however, were not quantified.

As shown in figure 2, the model control run without any data assimilation generally underestimates ozone in the free tropopsphere over China (108 °E–123 °E, 20 °N–44 °N) by up to 10 ppb, except during spring. The data assimilation removes most of the model negative bias, with a mean bias reduction of 37% (from −3.1 ppb to −1.9 ppb) and an increase in temporal correlation from 0.88 to 0.96. The interannual changes are also well reproduced by the reanalysis, which shows increasing trends from 2005 to 2010 of +0.33 ppb/year (p > 0.05 from the Mann-Kendall test. p-values greater than a significance level of 0.05 fail to reject the null hypothesis of no trends), consistent within the uncertainties with the +0.44 ppb/year increase (p > 0.05) in the TES observations. The model control simulation, in contrast, has only a + 0.11 ppb/year increase (p > 0.05). Over the western US (105 °W–130 °W, 30 °N–50 °N), the model reproduces the observed variations in the free troposphere relatively well. However, it underestimates the springtime peak by up to 6 ppb in some years. The data assimilation reduces these model negative biases and decreases the root mean square error (RMSE) from 4.5 ppb to 2.6 ppb.

Zoom In Zoom Out Reset image size

Free tropospheric ozone variations are largely controlled by NOx emissions, including the recent rapid changes during the COVID-19 pandemic [51–53], and by variations in transport of ozone from the stratosphere to the troposphere [8, 54]. Since tropospheric ozone observations were not assimilated, the improved agreement with independent TES measurements demonstrates that the assimilation provides a reasonable representation of NOx emissions and of stratospheric ozone. Nevertheless, there are many other factors affecting ozone variations, including precursor emissions other than NOx. While we focus on NOx in this paper, these influences need to be explored to fully address the causal mechanisms of observed ozone changes.

3.2.1. Spatial distributions

To measure the impact of changing NOx emissions on ozone for each year during 2006-2016, two sets of model simulations were conducted using the optimized emissions obtained from the data assimilation runs. In baseline simulations, the optimized emissions for each year were used. In sensitivity calculations, the emissions at 2006 levels were used for 2007-2011, and the emissions at 2011 levels were used for 2012-2016. Differences between the baseline and sensitivity model simulations illustrate global ozone anomalies due to changing emissions. While we have assimilation results for 2005 (c.f. Figure 2), we start with 2006 in this analysis in order to examine the same number of years for both increasing and decreasing emissions (5 years each).

First, we investigated the impacts of global NOx emissions changes. Variations in global NOx emissions relative to 2006 led to differences in monthly mean ozone concentrations of up to 18 ppb at the surface and up to 20 ppb in the lower troposphere (750 hPa), with mostly negative values during 2007-2009 and mixed positive and negative values in 2010 and 2011 (figure S1 available online at stacks.iop.org/ERC/4/045003/mmedia). Large surface ozone anomalies were found over polluted areas. At 750 hPa, the anomalies were propagated across the northern extratropics, including remote areas, through atmospheric transport. These changes in ozone reflect changes in both natural and anthropogenic sources of NOx across the globe, including biomass burning emissions, and confirm the substantial impact of NOx emissions on global background ozone levels. In addition, year-to-year ozone variations can also be driven by changes in meteorological or climate conditions [55], such as wind, temperature and humidity. They control ozone production efficiency and transport patterns, including inflows from the stratosphere [54].

The remainder of our analysis focuses on the influence of Chinese NOx emission changes on ozone. The turnover in Chinese emissions during the satellite period provides a unique opportunity to evaluate the linearity of the ozone response to both increases and decreases in NOx. As shown in figures 3 and 4, the increase in Chinese NOx emissions from 2006-2011 had substantial influence on ozone across the NH extratropics, including the Arctic region. The ozone response varies strongly with region and season. In winter over highly polluted areas, ozone near the surface was decreased due to the NOx titration effect under high NOx conditions (figure not shown). This was most obvious over northern China. During other seasons and in the free troposphere, there was a broad increase in ozone due to increased Chinese NOx emissions. In boreal spring (figure 3), the ozone anomaly was strongly propagated into the Pacific and North America via westerly winds at NH mid latitudes. The cross-Pacific transport was stronger in spring than in other seasons (not shown). The increase in Chinese NOx emissions increased monthly mean ozone at 750 hPa by up to 1.8 ppb in East Asia and up to 0.4 ppb over the Western US in 2011 relative to 2006. Meanwhile, the impacts of Chinese emission variations and differences in meteorology reveal strong year-to-year changes in ozone, such as weaker anomalies over the Pacific and North America in 2009 than in 2008 and 2010. The ozone anomaly over Canada and the Arctic region is largest in 2010, with up to a 0.4 ppb increase.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Similarly, the reduction in Chinese NOx emissions from 2012 onward decreases ozone, with maximum reductions in 2015 over many regions, e.g., up to a 0.7 ppb reduction over the Northern Pacific and 0.4 ppb reduction over the Western US at 750 hPa. The monthly mean changes in ozone are typically a + 0.2–0.5 ppb and −0.15–0.25 ppb, respectively, over these regions for the 2013-2017 period. The interannual variability of the simulated ozone concentration driven by meteorology is typically 1-3 ppb over Northern Pacific and 0.5–15.5 ppb over the Western US (not shown), which is larger than the evaluated influence of Chinese emissions. Nevertheless, the systematic ozone anomaly seen throughout the 2013-2017 period provides evidence of a robust influence of the Chinese NOx emissions. An exception is northern China, where ozone near the surface increases, again reflecting NOx titration effect. The general spatial patterns at 750 hPa, with the largest anomalies extending from southern China to the mid-latitude Pacific Ocean and North America, are common for the positive anomalies during 2007–2011 and for the negative anomalies during 2012-2016. These results suggest that variations in NOx emissions in Asia affect background ozone levels and explain part of the observed variability over many remote regions and the Western US. Peroxyacetyl nitrate (PAN) is a long-lived reservoir species for NOx and can be transported long distances before decomposing. Figure S2 shows large anomalies in PAN in the free troposphere associated with changes in Chinese NOx emissions, highlighting the role of long-range transport of ozone precursors in remote impacts of international NOx emission changes on US background ozone.

Again, because changes in emissions other than NOx, such as those of VOCs and CO, were not considered, the actual ozone responses to Chinese anthropogenic emission changes could be broader and more complex [56]. Satellite measurements of formaldehyde (CH₂O) from OMI and GOME-2 have been used to evaluate VOC emission changes in previous studies, focusing on biogenic isoprene emissions [57–59]. The VOC emission analysis can be extended using more recent satellite CH₂O products, such as from TROPOMI [60] and OMPS, including anthropogenic VOC sources [61]. Nevertheless, evaluation of anthropogenic VOC emissions remains largely uncertain for many regions of the world [16]. Using model sensitivity calculations, changes in VOC emissions by 20% led to local monthly mean ozone concentration changes by 3 ppb over eastern China at the surface [52]. Meanwhile, MOPITT CO measurements have been used to infer −2% per year decrease in primary CO sources in East Asia from 2005–2016, mainly attributing to source reductions in China [62]. These changes in VOC and CO emissions would amplify the impact of Chinese emission influences on the US background ozone. Further studies would be required to evaluate decadal changes in global anthropogenic VOC emissions and its impacts on ozone.

3.2.2. Impacts on Western US background ozone

Surface ozone over the US west coast is heavy influenced by long-range transport from China [8, 63]. Our model simulations constrained by satellite measurements show that the ozone changes over Asia associated with Chinese NOx emissions are propagated across the Pacific to the Western US. While the magnitude of the ozone anomalies decay significantly during transport, they are nevertheless non-negligible in the US at daily scales. As seen in figure 5, daily surface ozone anomalies over Los Angeles associated with Chinese NOx emissions are about 3.8 ppb higher in 2012 than in 2006 at 750 hPa and 0.7 ppb higher at the surface, with strong day-to-day and seasonal variations. The ozone anomaly is largest from spring through early summer at both levels. The seasonality is driven by changes in transport and ozone production efficiency [64]. In total, over the Western US, the number of exceedances of the U.S. national ozone standard (70 ppb for daily maximum 8-hour concentrations: MDA8) increased by about 6% from 2006 to 2012 due to the Chinese NOx emission changes (figure S3). Because we stored 6-hourly mean ozone concentration outputs only and used them for this analysis, actual exceedance cases due to the Chinese NOx emission changes could be biased in some locations. Using 2-hourly model output, we confirmed that the use of 6-hourly averages, instead of MDA8, tend to slightly underestimate daytime maximum ozone concentrations over the Western US and overestimate it over central and eastern US, by up to 1 ppb in monthly means and 1.5 ppb in daily concentrations over polluted areas with high ozone levels. Differences in daily concentrations are as large as 5 ppb over remote areas like the oceans. A more accurate evaluation of the impact on the US standard needs to be obtained using our results as boundary conditions to a regional model that better represents the detailed spatiotemporal evolution of ozone at the relevant scales, as similarly demonstrated by previous studies [63, 65].

Zoom In Zoom Out Reset image size

We also conducted model simulations in which we fully removed anthropogenic NOx emissions in China. The surface ozone anomaly due to all Chinese anthropogenic NOx emissions is about 7 ppb at 750 hPa and 4 ppb at the surface over Los Angeles (figure 5). Chinese anthropogenic NOx emissions are associated with approximately a 97% increase in the number of exceedances of the US national ozone standard (figure S2). This demonstrates the important contributions of human activity in remote countries to local air quality and the resulting human health impacts, not only in the Western US, but in any location downwind of significant ozone sources. In the western US, the remote influence is greatest at high elevation locations, consistent with previous findings [13]. The Chinese emission impact was also large over Mexico (up to 4 ppb), which increased the occurrence of the exceedances of the Mexican national ozone standard (also 70 ppb) by 6 days.

Meanwhile, the large difference in the ozone response per unit NOx emission in the two simulations (about 30% emission reductions in the upper left plot in figure 5 and 100% emission reductions in the upper right) and in the number of Western US ozone exceedances (6% versus 97%) confirms the non-linearity of ozone response, the degree of which also differs between the surface and free troposphere. This nonlinearity suggests that realistically-achievable decreases in remote emissions using current technologies will have at most a modest effect on local ozone. Anthropogenic emissions from countries other than China, such as India and Southeast Asian countries, generally had smaller influences on US background ozone than China (results not shown). Nevertheless, we find that the rapid increase in Indian emissions from 2006 to 2016 substantially increased background ozone levels over remote regions at low latitudes (within ± 30°). Biomass burning emissions from tropical regions also had substantial influences on background ozone levels and the tropospheric ozone burden [66]. Better understanding of these complex effects on background ozone over the US and other countries is needed to further improve air quality control strategies.

We emphasize that the ozone response to changing emissions can be strongly model dependent. Model sensitivities are the primary drivers of differences in chemistry-climate estimates of quantities such as ozone radiative forcing [67, 68]. Our previous work has shown that the sensitivity of ozone to NOx emissions varies by a factor of 2 among the four different chemical transport models in the MOMO-Chem system, revealing fundamental differences in the representation of fast chemical and dynamical processes [27]. Thus 200% can be regarded as the possible uncertainty range in the estimated ozone response we show here. In another study assessing background ozone from multiple models and other approaches, the uncertainty in total US background ozone was estimated to be about 10 ppb for seasonal mean values and higher for individual days [13]. Future work will focus on better quantifying model-based uncertainties in US background ozone estimates. In particular, the use of a suite of modeling systems in a consistent manner would provide a more systematic assessment, as suggested by previous study [13]. Meanwhile, the ozone response showed large spatial variations. Thus, monitoring detailed spatial patterns of both ozone and its precursors regionally or globally may provide important information in understanding background ozone changes.

Our previous studies have shown that the simultaneous assimilation of ozone and its precursors is the most powerful way to constrain the vertical distribution of ozone for the entire troposphere within East Asia [26] and for the globe [23]. Using the regional (RE for East Asia: 20°N-50°N, 80°E-130°E with denser spatial sampling) multi-spectral AIRS/OMI tropospheric ozone profile retrievals [21] generated for the KORUS-AQ campaign studies [26, 69], we demonstrate the impact of dense satellite measurements of tropospheric ozone profiles over Asia on ozone within the broader Northern Hemisphere during April-May 2016. The purpose here is to explore the value of such measurements in isolation using OSEs rather than to provide a best estimate of the impact of Asian ozone on remote locations. While utilizing the optimized NOx emissions and stratospheric ozone constrained by satellite measurements (c.f., sections 3.1 and 3.2), our OSEs are designed to provide insight into the ability of satellite measurements of ozone over Asia to inform global atmospheric composition in a realistic data assimilation setting. Assimilating multiple satellite tropospheric ozone measurements simultaneously from the current satellite observing system would provide a more quantitative assessment of the value of individual measurements. Nevertheless, the presence of systematic biases among measurements [70] complicates such an evaluation, and thus this was not considered in our stand alone OSEs. In the future, similar observational information on the vertical distribution of tropospheric ozone could be obtained from geostationary satellites such as the Korean Geostationary Environment Monitoring Spectrometer (GEMS) if combined with a high spatial sampling, high accuracy thermal IR instrument in geostationary or even polar orbit (for once per day profiles at the overpass time).

As shown in figure 6, large negative data assimilation increments in free tropospheric ozone appear over Asia early on in the AIRS/OMI data assimilation cycle. The magnitude and structure of the data assimilation increments reflect complex factors such as the model-observation differences (departure), averaging kernel profile, retrieval uncertainty, and sampling pattern of the assimilated measurements. The positive model bias relative to the AIRS/OMI data, suggested by the negative data assimilation increments (figure 6), is opposite to the negative model bias relative to TES seen in figure 2. This reflects the large impacts of optimizing NOx emissions and stratospheric ozone concentrations in the OSEs, together with errors in the model initial condition used in the OSEs, especially for the first few assimilation cycles. In addition, any systematic biases between TES and AIRS/OMI will be reflected as disparate model biases. We focus here on the spatial propagation of the information provided by the assimilation of the AIRS/OMI measurements.

Zoom In Zoom Out Reset image size

The increments are mainly limited to the free troposphere because of the maximum vertical sensitivity of AIRS/OMI to this region. The ozone anomalies, however, are carried into the lower troposphere through vertical atmospheric transport, decaying as they near the surface. After a few days, the adjustments made over East Asia begin to spread into a larger area. After 40 days, we find large negative ozone anomalies across the Northern Hemisphere, with up to 13 ppb reductions in the continental US, relative to the control model simulation without any assimilation. The maximum ozone anomaly in the Western US occurs during April-May 2016 and is about 20 ppbv at 500 hPa and 3 ppb at the surface. Over Los Angeles, it reaches 9 ppb at 500 hPa but is greatly reduced at 750 hPa and at the surface (figure S4). These results demonstrate the value of direct satellite-based ozone measurements over Asia for informing the global distribution of ozone given model biases associated with chemical production of ozone and difficulty in reproducing free tropospheric ozone over Asia [71].

However, the changes in background ozone levels associated with ozone assimilation are strongly dependent on model performance, including errors in emissions of many precursors and the impacts of stratospheric ozone. Ozone assimilation would have a smaller impact in a simulation that better represented 2016 conditions of the complex atmospheric system. Meanwhile, more comprehensive OSEs, for instance, using a combination of other satellite measurements of ozone, such as those from current polar orbiting satellites, as well as surface observations of ozone and its precursors, would provide further insights into the relative value of the Asian measurements in the current observing system. This type of OSE needs to be done in a separate study.