To distinguish the type of El Niño, we first use the Oceanic Niño index (ONI) from the Climate Prediction Center (CPC) of the National Oceanic and Atmospheric Administration (NOAA) to filter out El Niño events. The ONI is defined as the 3-month running mean of ERSST.v5 SST anomalies in the Niño 3.4 region (5∘ N–5∘ S, 120–170∘ W), based on centered 30-year base periods updated every 5 years. An El Niño event is defined when the ONI is greater than or equal to 0.5 ∘C for a period of at least five consecutive overlapping seasons. Then we combine two methods, namely the Niño 3 and 4 method in Yeh et al. (2009) and the ENSO Modoki index (EMI) method in Ashok et al. (2007), to distinguish between EP and CP El Niño. When the two methods show consensus results, we define it as a typical EP or CP event.

Ozone abundance in the atmosphere can be measured from space using different remote-sensing techniques. Frequently used tropospheric column ozone datasets include the OMI/MLS carried by AURA and the infrared atmospheric sounding interferometer (IASI) carried by the MetOp satellites. As we focus on lower tropospheric ozone in this study, we chose to use IASI, which can retrieve the ozone from the surface to 6 km. In addition, the IASI is a superior choice considering the spatial coverage, resolution, and data quality. The IASI is a thermal infrared Fourier transform spectrometer onboard the MetOp-A and B satellites. As a spaceborne nadir-viewing instrument, it probes the troposphere using the thermal infrared spectral range, and the atmospheric data are further retrieved by inversion algorithms (Boynard et al., 2009, 2016). The IASI-A and -B instruments have been operationally providing atmospheric products since October 2007 and March 2013, respectively. Ozone monthly gridded data are available on https://cds.climate.copernicus.eu/cdsapp#!/dataset/satellite-ozone-v1?tab=form (last access: 8 November 2021). We use the ozone data from September 2007 to Autumn 2017, mostly from MetOp-A v0001, with substitutes from MetOp-B v0001 for several missing months in 2015.

Meteorological fields for 1980–2017 are obtained from the Goddard Earth Observing System (MERRA-2) database (Bosilovich et al., 2016), which is the current operational met data product from the Global Modeling and Assimilation Office (GMAO). The data are available at http://ftp.as.harvard.edu/gcgrid/data/GEOS_2x2.5/MERRA2/ (last access: 8 November 2021). Meteorological variables used in Sect. 3.2 include surface downwelling solar radiation (SR), relative humidity (RH), total precipitation (TP), temperature (T), sea level pressure (SLP), and wind fields. As for multilevel variables, including RH, T, and winds, we calculate the 0–6 km column averages of these variables to be consistent with column ozone concentration, whereas SR, TP, and SLP are single-level variables.

The GEOS-Chem (GC) chemical transport model (Bey et al., 2001; v12.3.2; http://geos-chem.org (last access: 8 November 2021) is used to explore the EP and CP El Niño-related tropospheric ozone changes. We use the standard chemistry mechanism, which includes both troposphere and stratosphere. The Universal Tropospheric–Stratospheric Chemistry eXtension (UCX) mechanism developed by Eastham et al. (2014) combines tropospheric and stratospheric reactions into a single chemistry mechanism. The model is driven by MERRA-2 meteorological fields with 72 vertical levels and 2 × 2.5∘ horizontal resolution. We first perform a historical run from 1980 to 2017 with anthropogenic emissions fixed at the year 2000, so the difference among different events is only caused by the meteorological fields. A drawback of this setting is that the biomass burning is also fixed at the year 2000; however, the biogenic emission will still change as it interacts with meteorology.

The transient ozone simulation is further validated against tropospheric ozone within the same altitude range retrieved by IASI. Because IASI only retrieves column ozone concentration between 0 and 6 km, our comparison and analysis also focus on 0–6 km integrated column ozone concentration, referred to as lower tropospheric ozone (LTO) hereafter. This focus on column ozone concentration can also reduce the impact of mismatch in anthropogenic emission between IASI and GC, which mainly influence the near-surface ozone concentration. As satellite observation starts in October 2007, to ensure comparability, we select the 2015–2016 and 2009–2010 events to represent EP and CP El Niño, respectively. A 10-year (September 2007–August 2017) seasonal average is used as the climatological state. The missing month of IASI data in September 2007 is filled as NaN in our calculation. As we focus on the ozone changes, the bias induced by the mismatch of anthropogenic emissions is further mitigated by subtracting the climatological state. Therefore, we expect the ozone changes in ENSO years to show similar patterns during the ENSO years between GC simulation and IASI observation. Figure S1 in the Supplement shows the seasonal mean SST anomalies for the two periods selected, which correspond well to EP (2015–2016) and CP (2009–2010) El Niño patterns. The comparison results are shown in Fig. 1 and discussed in Sect. 3.1.

To further distinguish the ozone changes between EP and CP El Niño, we also perform three composite model simulations driven by the composite meteorological fields of the four seasons of (1) the three most typical EP events (1982–1983, 1997–1998, 2015–2016), (2) the four most typical CP events (1994–1995, 2002–2003, 2004–2005, 2009–2010), and (3) a 30 year averaged climatology (September 1985–August 2015). Figure S2 in the Supplement shows the composites of seasonal mean SST anomalies, which well corresponded to EP and CP El Niño. To save computing resources and time, we calculate the seasonal mean and archive it in daily data files; each season was run for 10 d with the same seasonal-averaged meteorological fields every day. These three simulations started on the same day from a previous transient run to save time for spin-up. In this set of composite simulations, the difference between the result of simulations 1 and 3 (simulations 2 and 3) can represent the ozone changes driven by EP (CP) meteorological changes.

Moreover, to explain the physical and chemical drivers of the ozone changes, we analyze the composite meteorological fields to check the ENSO-related meteorological changes. We also diagnose the 0–6 km ozone budget changes of different model processes and quantify the absolute contribution of each process. These budget diagnoses are calculated by taking the difference in 0–6 km vertically integrated column ozone mass before and after major GEOS-Chem simulation components, including chemistry, transport, mixing, and convection, at each time step.

An ENSO event usually develops in autumn (September–October–November, SON0), reaches its peak in winter (December–January–February, DJF0-1), and decays in the following spring (March-April-May, MAM1) and summer (June–July–August, JJA1) (Xu et al., 2017). We denote the ENSO developing year as year 0 and the following year as year 1. We first compare the climatology state for ozone (Fig. S3 in the Supplement) between observation and simulation. Model performance is comparable to that in previous modeling works (Dang et al., 2021; Lu et al., 2019; Ni et al., 2018). The bias mainly comes from the resolution, chemical mechanism, microphysics processes, and site representativeness (Sun et al., 2019; Young et al., 2018). Then we examine the change in satellite-retrieved and simulated 0–6 km column ozone during the 2015–2016 EP and 2009–2010 CP events relative to the climatology state (Fig. 1) to validate the model response to ENSO-related signals.

The EP El Niño generally exerts negative effects on LTO in China in both observation and simulation, except for a dipole mode change over southern China during pre-EP autumn and post-EP summer. Satellite-retrieved LTO shows an increase in the south and a decrease in the north in autumn, whereas this dipole mode is obscure in the simulation. In winter and spring, both the satellite-retrieved and simulated LTO exhibit coherent decreases throughout China, but the intensity in the model is much smaller. In summer, the observation still shows declines over most regions except for a slight increase over the southeast coastal area and southwestern China. The simulation shows a similar pattern but with much stronger positive signals over southern China. In contrast, in CP El Niño, there are more prominent LTO increases, such as over southern China in autumn, northeastern China in spring, and northern China in summer. In autumn, the satellite observation and simulation both exhibit a dipole mode change in the north and south, with LTO decrease over northern and increase over southern China. In winter, the observed and simulated LTO both show a reverse change with slightly positive and negative signals. The LTO changes in spring and summer are consistent between observation and simulation.

In general, the LTO changes range from -1 to 1 DU (Fig. S4 in the Supplement), accounting for 5 %–10 % of the 0–25 DU mean range. The spatial patterns of the simulated and observed LTO changes agree well, despite an overall underestimation by the model. This underestimation can be explained by the fixed biomass-burning emission in the simulation that weakens the sensitivity of tropospheric ozone to ENSO, as this leads to milder changes in ozone precursors such as carbon monoxide. The underestimation in spring and summer is the most significant in high-latitude areas, such as northeastern China, for both EP and CP events. This deviation probably represents the interferences of other high-latitude climate variabilities. Another reason is that the model underestimates the average ozone concentration at high latitudes in winter and spring (Fig. S3), which leads to less ozone transport from polar regions to northern China in the model. The IASI-retrieved data exhibit high ozone concentration in the Arctic during winter and spring (Fig. S3f, g); this phenomenon is also shown in previous studies (Cooper et al., 2014). However, the GEOS-Chem simulation did not capture the high values in polar regions. A possible explanation for this underestimation is that the Brewer–Dobson circulation (BDC) may be insufficiently represented in the model. The BDC consists of an upward transport branch across the tropopause in the tropics and has a strong poleward and downward circulation branch in the winter hemisphere (Hu et al., 2017), which contributes to the high LTO concentration in polar regions through the stratosphere–troposphere exchange. Another potential reason for the underestimation is due to the imprecise halogen chemistry in GEOS-Chem. Wang et al. (2021) point out that the halogen chemistry can worsen the underestimation of tropospheric ozone in the Northern Hemisphere by halogen-catalyzed loss. Thus, the ozone transport from polar regions to northern China can be much less in the model. The overall consistency between simulated and observed LTO changes gives us the confidence to use the model for composite analysis, as the satellite record only covers limited El Niño events.

To include more El Niño events and check the response of ozone to meteorological fields, we further use the composite meteorological fields of three EP events and four CP events to drive the GEOS-Chem model. Figure 2 shows the LTO changes in China during different seasons of the EP and CP El Niño. The patterns agree well with the composite results from historical simulations (Fig. S7 in the Supplement) but show stronger changing magnitudes due to the more direct response of ozone to meteorological changes. It is seen that LTO decreases over most regions in both EP and CP types in the range of 5 %–10% (2 %–5% in the composite of historical run), whereas only some regional increases are seen in pre-El Niño autumn and post-El Niño summer. During winter and spring, LTO decreases consistently, reaching ∼ 10 % for western and northern China. The changes associated with CP El Niño are more extensive, spatially uniform, and stronger than EP. For summer, however, EP appears to correspond to a more substantial LTO decrease, especially for the northern and southwestern parts. The region exhibiting the most LTO change differences between EP and CP events is southern China. The differences between EP and CP patterns will be further examined in the next section. It appears that the seasonal alternation of LTO changes in southern China may represent the extension of the remarkable ozone changes over the tropical regions. During the EP (CP) El Niño developing, sustaining, and first decaying periods, there are significant dipolar (tripolar) modes of ozone changes over the tropical Pacific area (Fig. S5 in the Supplement), which is consistent with the result of previous studies (Chandra et al., 1998; Oman et al., 2013). These ozone changing patterns correspond well with solar radiation changes (Fig. S6 in the Supplement) since they can modulate the photolysis rates and biogenic emissions.

Because El Niño is generally associated with decreased tropospheric ozone concentration, we also briefly examine the LTO changes during the negative phase, i.e., La Niña events (Fig. S7). In contrast to El Niño, La Niña tends to be associated with extensive LTO increases by ∼2 %–5 %, especially over northern China, indicating an adverse impact on the already severe tropospheric ozone pollution in this region. An increase in ozone concentration during the post-La Niña spring has also been reported by Wie et al. (2021). However, because El Niño teleconnections are typically stronger and better established, we still focus on El Niño in this study.

To clarify the mechanism associated with different LTO changes of the two types of El Niño, we further examine the changes in meteorological variables, including SR, RH, TP, T, SLP, and wind fields during EP (Fig. 3) and CP events (Fig. 4). The leading two variables (SR and RH) impact the local production, and the circulation changes represented by SLP and winds control the regional transport. Although wet scavenging of ozone by TP is negligible because ozone is insoluble in water, TP is closely related to SR and RH; it is also the primary variable examined to identify ENSO teleconnections. We thus also include TP in the comparison. In addition, we calculate the budget changes corresponding to the EP and CP events from GEOS-Chem simulations. The simulated ozone concentration is mainly determined by four processes, namely chemistry, transport, mixing, and convection. Since each process can contribute to ozone either positively or negatively, in Fig. 5 we calculate the absolute value of the column integrated ozone budget in each grid box and then calculate the mean value of the chosen domain (24.0–42.0∘ N, 100.0–117.5∘ E; purple box in Fig. 6a) to better quantify the impact of each process. Because chemistry and transport are the two dominant processes accounting for more than 70 % of the ozone changes in all conditions, we focus our following discussions on these two processes. Figure 6 shows the spatial distribution of ozone budgets corresponding to the chemistry and transport processes from the simulation driven by composite meteorological fields.

In the autumn before El Niño, LTO changes for EP type show a general decrease in China (Fig. 2a), especially in the southeastern part. The EP El Niño is always accompanied by an anomalous anticyclone in the Philippine sea (Fig. 3q), which produces strong southwesterly wind anomalies that transport moisture from the ocean, resulting in increased TP and RH but decreased SR over southeastern China (Fig. 3i, e, a). These changes are unfavorable for ozone production but efficient for ozone removal, thus leading to a chemical loss of LTO over southern China (Fig. 6a). Some regional increase over southwestern China has been observed and likely is due to the positive contribution of transport (Fig. 6e) from India as indicated by the westerly wind anomalies (Fig. 3q). During the CP event, there is a moderate dipole mode change (Fig. 2e), with decreases in northern China and increases in the southern part. In contrast to EP, an anomalous cyclone appears over the Philippine Sea, leading to northwesterly wind anomalies over southern China that produce a dry condition with increased SR (Fig. 4i, e, a). The slight decrease in LTO over northern China is likely attributed to the decreased chemical production (Fig. 6i) associated with negative temperature anomalies (Fig. 4m), although the signal is not statistically significant. The opposite atmospheric circulation patterns over the Philippine Sea during EP and CP events are responses to the different SST anomaly regions under these two conditions, as shown by Wang and Wang (2013) using simple atmospheric model experiments.

In winter, when the Pacific SST anomalies reach their maxima, EP and CP El Niño are both associated with increased TP and RH, and decreased SR, over southern China (Figs. 3b, f, j and 4b, f, j). These similar changes are due to the moisture transport induced by western North Pacific anomalous anticyclones (WNPAC) that occur in both EP and CP El Niño, while EP exhibits greater meteorological changes than CP due to the much stronger anomalous anticyclone (Figs. 3r and 4r). WNPAC is a critical system that links El Niño and East Asia climate change, and its formation and maintenance mechanisms are discussed thoroughly in Li et al. (2017). WNPAC is initiated and maintained by local atmosphere–ocean interaction (Wang et al., 2000) and the moist enthalpy advection/Rossby wave modulation (Wu et al., 2017a, b). Although the meteorological variables change in the same direction, the EP- and CP-related LTO changes in winter are still opposite over southern China (Fig. 2b, f), where the El Niño teleconnection signal is the most prominent (Wang et al., 2020). Budget analysis reveals that this phenomenon is due to the varying contribution of different model processes. Consistent with the increased RH and decreased SR, the contributions of chemical processes are both negative over this region during EP and CP (Fig. 6b, j). The southwestern wind anomalies (Figs. 3r and 4r) bring not only water vapor from the ocean but also ozone from India and the China–Indochina Peninsula to southern China, contributing to LTO concentration there. During EP, the chemical loss (Fig. 6b) suppressed the positive transport (Fig. 6f) due to the severe change of SR and RH over southern China (Fig. 3b, f). However, for CP conditions, the chemical loss (Fig. 6j) due to the increased RH is much weaker and is offset, or even exceeded, by transport (Fig. 6n). This is also consistent with the much larger absolute contribution of transport than chemistry for CP (Fig. 5f).

In spring, LTO decreases extensively over the entire northern China under both EP and CP conditions (Fig. 2c, g), coherent with the large-scale reduction in SR and increase in RH (Figs. 3c, g and 4c, g). WNPAC persists under EP conditions according to the moist enthalpy advection mechanism (Wu et al., 2017a), whereas it nearly disappears in CP (Feng et al., 2011). In EP conditions, with the slight westward shift of the anticyclone center from winter to spring, the wind anomalies also shift from southwesterly to southerly, bringing more moisture and further enhancing TP in higher latitudes where RH increases and SR decreases coherently. Although these changes are generally unfavorable to the local ozone production, the chemistry process still contributes positively in eastern China (Fig. 6c). We attribute this pattern to the large-scale increase in temperature related to the warm southerly wind anomalies (Fig. 3o, s). As the climate warms from winter to spring, the role of temperature becomes increasingly important and may compensate or even exceed the impact of SR reduction. On the other hand, as the southerlies blow low ozone air from the ocean, the severe negative transport (Fig. 6g) dominates the overall ozone decrease. In CP, regional transport is weaker due to the unremarkable change in circulation patterns over the western North Pacific compared with the EP condition; thus, the absolute contributions of transport and chemistry are comparable to those for CP (Fig. 5g).

The situation for the post-El Niño summer is more complicated as El Niño teleconnections substantially involve air–sea interactions and interbasin teleconnections (Feng et al., 2011; Kug et al., 2009). Ozone changes for the EP conditions show a decrease over central and northern China and a band-like ozone increase over southeastern China (Fig. 2d). Although the chemical production (Fig. 6d) increases with the slight SR increase and RH decrease (Fig. 3d, h) over China's eastern coastal line, the transport process (Fig. 6h) controlled by southwestern wind anomalies dominates the ozone decline over the Yangtze River basin and increases over the southeastern coastal line. The circulation anomalies manifest as a tripolar pattern with an anomalous anticyclone (AAC) over the southern China Sea and an anomalous cyclone circulation (ACC) over Japan (Fig. 3t). This pattern appears to be induced by the Indian Ocean capacitor (IOC) effect, which indicates the Indian Ocean memory of ENSO influence (Chen et al., 2012; Xie et al., 2009; Yang et al., 2007). Since the convection is suppressed in the anomalous anticyclone, the drier condition corresponds well to the positive LTO changes over the Philippine Sea (Fig. S5d). This positive signal extends to southeastern China's coastal areas due to the transport by the southwest wind anomalies. During CP, ozone decreases coherently over most of China (Fig. 2h). As no significant Indian Ocean warming appears (Fig. S2h), the summer climate is influenced more by the western Pacific warm pool. The negative SST anomalies in the central–east Pacific imply an upcoming La Niña. According to a previous study, the western tropical Pacific warm pool spreads eastward near the surface as El Niño builds (Johnson and Birnbaum, 2017). As La Niña generally shows opposite characteristics to El Niño, the western tropical Pacific warm pool under the former condition will shrink. Associated with the SST drop, SLP increases over the northwestern Pacific (Fig. 4t), resulting in an enhanced western Pacific subtropical high (WPSH), a typical feature of CP El Niño (Chen et al., 2019). Controlled more by the local Pacific than the Indian Ocean, the SLP anomaly center shifts eastward during CP El Niño compared with the anomalous anticyclone during EP El Niño, and the positive LTO anomalies also move eastward accordingly (Fig. S5h). Regional transport (Fig. 6p) by the southwest wind anomalies surrounding the positive SLP center (Fig. 4t) exerts a consistent negative contribution to LTO in southern China (Fig. 2h; Jiang et al., 2021). In sum, the post-El Niño summer LTO change is dominated by the IOC effect for EP and WPSH enhancement for CP.