The multispectral satellite approach IASI + GOME2 is designed for observing lowermost tropospheric ozone by synergism of thermal IR atmospheric radiances observed by IASI and UV earth reflectances measured by GOME-2. Both instruments are on board the MetOp satellite series (in orbit since 2006 and expected until 2022), and they both offer global coverage every day (for MetOp-A around 09:30 local time) with a relatively fine ground resolution (12 km diameter pixels spaced by 25 km for IASI at nadir and ground pixels of 80km×40km for GOME-2). As described in detail by Cuesta et al. (2013), IASI + GOME2 jointly fits collocated IR and UV spectra for retrieving a single vertical profile of ozone for each pixel. The horizontal resolution corresponds to that of IASI, using for each pixel the UV measurements from the closest GOME-2 pixel (without averaging). Spectra and Jacobians in the IR and UV are respectively simulated by the KOPRA (Karlsruhe Optimized and Precise Radiative transfer Algorithm; Stiller et al., 2002) and VLIDORT (Vector Linearized Discrete Ordinate Radiative Transfer; Spurr, 2006) radiative transfer codes. The effects of clouds and aerosols are partially taken into account by iteratively adjusting offsets for each of the seven spectral micro-windows (between 980 and 1070 cm-1) used in the IR and effective surface albedos and cloud fractions in the UV (two micro-windows between 290 and 345 nm). Only measurements with cloud fractions below 30 % are used (as determined by the FRESCO algorithm (Fast Retrieval Scheme for Clouds from the Oxygen A-band; Koelemeijer et al., 2001). Ozone profiles are retrieved by a constrained least squares fitting method using a Tikhonov–Phillips-type regularisation (Tikhonov, 1963). Constraint strengths vary with altitude and are optimised for enhancing sensitivity to lowermost tropospheric ozone while keeping acceptable total retrieval errors (in the order of 20 % for the LMT).

Here, we use an updated version of the IASI + GOME2 product, with only minor changes with respect to that of Cuesta et al. (2013). Ozone profiles are retrieved at the vertical grid between the surface and 60 km a.s.l., with steps of 1, 2 and 5 km respectively below 26 kma.s.l., between 26 and 30 kma.s.l. and above 30 kma.s.l.. Three a priori ozone profiles derived from the climatology of McPeters et al. (2007) are used, corresponding to the average over 20–30∘ N, 30–60∘ N and 60–90∘ N, representative of tropical, mid-latitude and polar conditions. These three a priori profiles are used for IASI pixels with tropopause heights (determined by the temperature vertical profile) above 14, between 14 and 9 km, and below 9 km respectively.

IASI + GOME2 products include vertical profiles of ozone, partial columns, averaging kernels (representing sensitivity of the retrieval to the true atmospheric state), error estimations and quality flags. Since 2017, global-scale IASI + GOME2 retrievals have routinely been produced by the French data centre AERIS, and they are publicly available (see https://www.aeris-data.fr and http://cds-espri.ipsl.fr).

In order to analyse the origin and evolution of ozone pollution plumes, the following correlative datasets are used: CO and O3 retrievals from IASI, and NO2 and CH2O observations derived from GOME-2 and OMI. Morning time (around 09:30 local time) datasets from IASI and GOME-2 are derived from the same spectra as those used in synergism by IASI + GOME2. OMI overpass occurs in the early afternoon (near 13:30 local time).

The CO retrievals used in the present paper are derived from IASI radiances using the FORLI algorithm (Fast Optimal Retrievals on Layers for IASI; Hurtmans et al., 2012), from the Université Libre de Bruxelles (ULB) and the Laboratoire Atmosphères, Milieux, Observations Spatiales (LATMOS). This approach uses pre-calculated lookup tables of absorbance cross sections at various pressures and temperatures, and an optimal estimation method for the inverse scheme. The algorithm derives vertical profiles of CO, on a grid of 18 equidistant layers of 1 km of depth from the surface up to 18 km, and a unique layer from 18 to 60 km. Radiative transfer calculations use operational MetOp-A L2 temperature and humidity profiles, and surface emissivity climatologies (Zhou et al., 2011). A priori CO profiles are taken from MOZAIC (Measurements of OZone aboard in-service AIrbus airCraft), ACE-FTS, for higher altitudes (Clerbaux et al., 2005) and the LMDz-INCA global chemistry–transport model (Haugustaine et al., 2004). FORLI provides vertical profiles, total and partial columns of CO derived by profile integrations, averaging kernels, error estimations and quality flags (supplied by AERIS and LATMOS). Comparisons of CO total columns derived from FORLI-IASI showed an agreement better than 7 % and no significant bias with respect to other satellite products (for the Northern Hemisphere; George et al., 2009) and ground-based retrievals from six NDACC stations (Kerzenmacher et al., 2012). A validation of lower (surface–480 hPa) and upper (480–225 hPa) tropospheric columns with respect to MOZAIC measurements found an agreement of respectively 21 and 10 %, and correlations of respectively ∼0.8 and ∼0.7 (De Wachter et al., 2012).

In the present study, we use CO retrievals in the LT integrated from the surface up to 6 kma.s.l. (equivalent to surface–480 hPa), validated by De Wachter et al. (2012) and presenting heights of maximum of sensitivity located at 3–5 km altitude (i.e. at the middle of this partial column; see Sect. 4). LT partial columns are retrieved with 0.83 degree of freedom (DOF, i.e. number of independent pieces of information in the retrieved profile) on average over the region and period studied in the paper. This product provides significant information on CO variability below 3 kma.s.l., as DOF in the LMT are 0.51 on average.

For estimating the ratio of enhancements of O3 and CO in the LMT (hereafter referred to as ΔO3/ΔCO) during individual long-range transport events, we use O3 and CO satellite observations (mixing ratios in ppb) after subtracting background levels, as done for analysing airborne in situ data during the PHOBEA I and II experiments over the northeast Pacific (Price et al., 2004). We empirically estimate these background concentrations as the daily average concentration minus the SD over the region of analysis (20–48∘ N, 110–150∘ E). For the event in early May 2009, we derive background levels around ∼46 ppb and ∼126 ppb for the observations of respectively LMT O3 and LT CO. We use the same criteria for deriving ΔO3/ΔCO in the LMT from models (Weather Research and Forecasting with Chemistry (WRF-Chem) and CHASER; see Sects. 2.3 and 4).

We also use single-band IASI-only retrievals of ozone (Eremenko et al., 2008; LISA product in Dufour et al., 2012) in order to analyse the distribution of ozone between 3 and 6 kma.s.l. This single-band IASI product is similar to IASI + GOME2 but only use infrared measurements (both approaches using a Tikhonov–Philips regularisation and the KOPRA radiative transfer code in similar configuration). By comparing IASI-only and IASI + GOME2 ozone retrievals, one may identify ozone plumes located below 3 kma.s.l. and those located between 3 and 6 kma.s.l. (as remarked by Cuesta et al., 2013). Indeed, we expect that high ozone concentrations clearly depicted by IASI + GOME2 and not by IASI are located in the LMT below 3 kma.s.l. Ozone plumes located at 3–6 kma.s.l. are shown by both IASI + GOME2 and IASI. Note that, since the DOF for IASI + GOME2 is lower than 1 in the LMT, multispectral outputs depend as well on ozone concentrations up to 5 or 6 kma.s.l., and they alone do no reveal whether the ozone plumes are located in the LMT or at 3–6 kma.s.l.

Satellite retrievals of NO2 and CH2O from GOME-2 (overpass around 09:30 local time) and OMI (13:30 local time) are used to indicate the availability of ozone precursors. These datasets are provided by TEMIS (Tropospheric Emission Monitoring Internet Service, http://www.temis.nl) and BIRA-IASB (Belgian Institute for Space Aeronomy–Institut royal d'Aéronomie Spatiale de Belgique) for CH2O (http://h2co.aeronomie.be). Retrievals of NO2 from GOME-2 and OMI are derived respectively by the algorithms TM4NO2A version 2.3 (Boersma et al., 2004) and DOMINO version 2.0 (Boersma et al., 2011). These approaches follow three steps: using differential optical absorption spectroscopy (DOAS) to obtain NO2 slant columns from reflectance spectra, separating the stratospheric and tropospheric contribution to the slant column, and converting the tropospheric slant column to a vertical column with the tropospheric air mass factor. Uncertainties for tropospheric NO2 retrievals are estimated as 35–60 % for GOME-2 and 25 % for OMI. Total columns of CH2O are retrieved with the BIRA-IASB algorithm version 14 (De Smedt et al., 2008), also based on DOAS technique and air mass factor estimations. Uncertainties of single CH2O slant column observations typically range from about 10 to 200 % when exceeding the global background (∼4×1015 molcm-2). For reducing random errors, datasets are averaged in regular grids of 1∘×1∘ and 2∘×2∘ respectively for NO2 and CH2O.

In the present study, we use WRF-Chem and CHASER chemistry–transport models for completing the description of the ozone pollution outbreak across East Asia in early May 2009 and for verifying consistency with satellite observations. While WRF-Chem is a regional model (Grell et al., 2005) operating in forecast mode over Asia (Takigawa et al., 2007), CHASER is a global model (Sudo et al., 2002; Sudo and Akimoto, 2007; Sekiya and Sudo, 2014) with coarser spatial resolution but with enhanced accuracy through assimilation of several satellite retrievals of atmospheric pollutants (Miyazaki et al., 2012, 2015) and with higher model top height. These models provide useful insights on the detailed vertical distribution (e.g. surface concentrations) and diurnal evolution of tropospheric O3 and CO over East Asia, in complement to daily satellite observations. The daily evolution of ozone enhancement in the LMT is compared between the models and the satellite data. For this, we estimate background levels of O3 and CO for each dataset using the same criteria (i.e. daily average minus SD over the domain; see Sect. 2.2). We derive concentrations of O3 and CO in the LMT by vertical integration from the surface up to 3 kma.s.l. (without any smoothing by the satellite AVKs). Moreover, we perform sensitivity studies with CHASER (in forecast mode), accounting or not for the stratospheric contribution of ozone in order to identify the tropospheric or stratospheric origin of this pollutant. In the figures of the paper, we show one of the models or both of them according to the following criteria: (i) WRF-Chem describes the structure of plumes of LMT O3 and CO with finer spatial resolution (Sect. 3); (ii) CHASER forecasts with and without stratospheric ozone distinguish tropospheric ozone formed in the troposphere from that originating from the stratosphere (Sects. 3 and 4); (iii) both WRF-Chem and CHASER are used for showing the temporal Lagrangian evolution of O3, CO and NO2 for polluted air masses (Sect. 4); and (iv) CHASER analyses indicate the vertical gradients of ozone between the surface and 2 km altitude with presumably good absolute accuracy provided by assimilation of various observations (Sect. 2).

For the current application, WRF-Chem resolution is set to 37 vertical layers from the surface up to 100 hPa (∼15 km altitude) and a Lambert conformal conic projection of horizontal pixels of about ∼0.9∘×0.9∘. Emissions from automobiles and other anthropogenic sources are taken from EAgrid 2000 (East Asian Air Pollutant Emissions Grid Inventory; K. Murano, personal communication) and the JCAP (Japan Clean Air Program; Kannari et al., 2007) over Japan for automobiles. Surface emissions over China and North and South Korea are taken from REAS (Regional Emission Inventory in Asia; Ohara et al., 2007) version 1.11, and over Russia from EDGAR (Emission Database for Global Atmospheric Research; Olivier et al., 1996) version 3.2. Biogenic emissions are based on Guenther et al. (1993). The lateral boundaries of chemical species are taken from the global CHASER model every 3 h. The system is driven by meteorological data from the Mesoscale Model (MSM) of the Japan Meteorological Agency (JMA).

CHASER analyses are obtained from an advanced chemical data assimilation system combining satellite observations of several chemical compounds: NO2, O3, CO and HNO3 measurements from OMI, TES, MOPITT and the Microwave Limb Sounder (MLS). Assimilation is performed according to the local ensemble transform Kalman filter technique (Hunt et al., 2007), which simultaneously optimises targeted chemical species, as well as the emissions of O3 precursors (i.e. NOx and CO), while taking their chemical feedbacks into account. Comparisons against independent data show that data assimilation results in substantial improvements. It reduces biases for tropospheric NO2 columns by 40–85 %, for lower-tropospheric CO concentrations in the Northern Hemisphere by 40–90 % and for O3 in the middle and upper troposphere by 30–40 %. Data assimilation also mostly removed the model's negative bias in surface CO concentrations in the Northern Hemisphere. The error reduction for O3 was generally smaller in the lower troposphere than in the middle and upper troposphere because of the reduced sensitivity of the assimilated TES retrievals to lower-tropospheric ozone. The CHASER forecast model includes detailed chemical and transport processes in the troposphere, including 88 chemical and 25 photolytic reactions with 47 chemical species, and has a horizontal resolution of T42 (2.8∘) and 32 vertical levels from the surface to 4 hPa for the present study. Sekiya et al. (2018) have recently developed a high-resolution version of CHASER with 0.56∘ horizontal resolution and demonstrated improved performances over areas with strong local sources with respect to the 2.8∘ resolution version. Nevertheless, the CHASER model with 2.8∘ resolution is capable of properly simulating synoptic ozone patterns.

CHASER is coupled to the Center for Climate System Research/National Institute for Environmental Studies (CCSR/NIES) Atmospheric General Circulation Model (AGCM) version 5.7b. The AGCM fields are nudged toward the National Centers for Environmental Prediction–Department of Energy (NCEP-DOE) Atmospheric Model Intercomparison Project II (AMIP-II) reanalysis (Kanamitsu et al., 2002) at every time step of the AGCM to reproduce past meteorological fields. Anthropogenic, biomass burning and biogenic/soil emissions are respectively based on EDGAR version 3.2 (Olivier et al., 2005), the Global Fire Emissions Data base (GFED) version 2.1 (Randerson et al., 2007) and the Global Emissions Inventory Activity (GEIA) inventory (Guenther et al., 1995). Surface emissions over Asia were obtained from REAS.

For the sensitivity analysis, CHASER simulations of ozone “from the troposphere” (not accounting for stratospheric ozone) are obtained by setting to zero ozone concentrations above 100 hPa, starting from 1 January 2009. Differences between ozone concentrations from full simulations and those not accounting for the stratospheric contribution provide an estimation of the distribution of ozone transported from the stratosphere. For consistency, no data assimilation is performed in either of the two simulations of this sensitivity analysis.

Meteorological conditions leading to production of ozone pollution and transport across East Asia are described in Sect. 3 with ERA-Interim reanalyses (Dee et al., 2011) produced by ECMWF (European Centre for Medium-Range Weather Forecasts). We use meteorological fields (downloaded from http://climserv.ipsl.polytechnique.fr) with global coverage, a horizontal resolution of 0.75∘×0.75∘, 37 pressure levels and a time step of 6 h (interpolated for other hours). Wind, geopotential height and equivalent potential temperature fields describe atmospheric circulation and the locations of synoptic high- and low-pressure systems. Additionally, forecasted atmospheric boundary mixing layer top heights from ERA-I are used in the analysis of the vertical distribution of LMT ozone derived from IASI + GOME2.

Pathways of polluted air masses transported across East Asia are estimated using the Hybrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT) dispersion model (Stein et al., 2015; Rolph et al., 2017; https://ready.arl.noaa.gov). This tool can simultaneously track a total of 12 500 air parcels released at a location and altitude and transported during 24 h according to meteorological fields. For the present analysis, we set the starting altitudes from the surface up to 3 km (i.e. in the LMT, which is the layer observed by IASI + GOME2) and use built-in model reanalysis from NCEP/NCAR (National Center for Atmospheric Research) with a horizontal resolution of 2.5∘×2.5∘ and 18 (29) pressure (sigma) levels. We run the dispersion model at 00:00 UTC (09:00 Japan Local Time (JLT), close to the MetOp satellite overpass) in forward mode for each day of the ozone pollution event, at the mean arrival location of the air parcels travelling on the previous day. The starting point for the pollution event was determined as the southern portion of the North China Plain (NCP), as suggested by model simulations and high concentration of ozone precursors (see Sect. 3.1). For the first day, we determined the region covered by the pollution plume using the dispersion model in backward mode, starting at the mean location of the air parcels on the second day.

The NCP is a well-known hotspot of pollutant emissions of worldwide relevance (e.g. Richter et al., 2005). On 2 May 2009, large concentrations of ozone precursors, such as nitrogen dioxide (NO2) and formaldehyde (CH2O), are observed over the NCP. This is suggested by GOME-2 satellite retrievals at 34–37∘ N, 113–117∘ E, marked as a magenta rectangle in Fig. 4a and b. Whereas a dense NO2 plume is mainly formed over the NCP, the highest concentrations of CH2O are mainly located south of it (at 25–36∘ N, 112–120∘ E), reaching the southern part of the NCP. The NO2 and CH2O concentrations observed on 2 May 2009 are approximately a factor of 2 higher than the regional monthly average (also estimated with GOME-2). According to MODIS active fire data over this region, wildfires are negligible in early May 2009 (only very few fire spots are detected; see http://firms.modaps.eosdis.nasa.gov). The short lifetimes of these reactive gases (up to a few hours) prevent the influence of long-range transport. The observed NO2 and CH2O plumes are thus likely associated with local anthropogenic emissions from this densely populated and industrialised region. In the following (Sects. 3 and 4), we particularly focus our analysis on the daily evolution of pollutant concentrations originated from these air masses as they are transported across East Asia. In Figs. 4–12, satellite pixels used to describe the evolution of these polluted air masses are depicted by magenta and red rectangles. These boxes contain valid satellite pixels collocated with at least 5 % of the polluted air parcels trajectories simulated by HYSPLIT.

According to model simulations, ozone concentrations over the NCP are relatively low (below 50 ppb) during the morning of 2 May, both at the surface and within the LMT (up to 3 kma.s.l., shown in Fig. 4c and e for WRF-Chem). Therefore, we do not observe any significant residual ozone plume from the previous day within the LMT. As expected for this time of the day (10:00 JLT), the mixing boundary layer over land is rather shallow (with its top below 1 kma.s.l.; see blue isolines representing the depth of the mixing boundary layer in Fig. 4e). Weather conditions over the region are characterised by very low wind speeds at the lower atmospheric levels associated with marked anticyclonic conditions (see high pressures approaching the NCP from the west at 25–35∘ E, 110–115∘ E in Fig. 4c and d). Due to partial cloud cover, IASI + GOME2 retrievals are not available over this region during this day.

Following up with the typical diurnal cycle, ozone is photochemically produced during the afternoon, until reaching concentrations of ∼90 ppb near the surface in the southern part of the NCP at 16:00 JLT (according to the WRF-Chem model around 34∘ N, 115∘ E; see Fig. 4d). In the afternoon, the mixing boundary layer over this region is deeply developed with its top near ∼2 kma.g.l. (3.5 kma.s.l., Fig. 4f), thus suggesting that the ozone plume freshly formed during the afternoon of this day is well mixed within the LMT.

In the morning of 3 May, anticyclonic conditions prevail over the NCP with a pressure maximum over central China, at the southern outskirts of the NCP (see Fig. 5c). According to the HYSPLIT dispersion model, the ozone plume formed the previous afternoon over southern NCP is transported by weak southerly winds until 35∘ N, 115∘ E (magenta rectangle in Fig. 5a and b). At this location, high concentrations of both LMT ozone (∼90 ppb) and LT carbon monoxide (∼290 ppb) are observed from space respectively by IASI + GOME2 (Fig. 5a) and IASI (Fig. 5b). Collocation of both pollutant plumes over the NCP suggests that these high ozone concentrations are associated with surface anthropogenic emissions, as CO is a tracer for combustion-related emissions.

Meanwhile, another ozone plume is observed over northeastern China (north of 42∘ N and at 115–135∘ E in Fig. 5a). The origin of this plume is likely associated with a low-pressure system (see “L” and concentric isobars north of 44∘ N and at 120–135∘ E in Fig. 5c). Such systems may both entrain ozone from the stratosphere and the UTLS region down to the lower troposphere and also mix pollution-related ozone within the low atmospheric levels, as analysed for other events in the same region by Dufour et al. (2015). Both phenomena may occur in this case. On the one hand, downward transport of ozone from the stratosphere (likely west of 120∘ E and north of 42∘ N) is suggested by enhanced potential vorticity at 300 hPa (a tracer of stratospheric air masses) north of 48∘ N, 130∘ E (Fig. 5d). On the other hand, an anthropogenic contribution of LMT ozone is indicated by the presence of a CO plume observed by IASI east of 122∘ E (and north of 40∘ N, Fig. 5b), likely originating from northeastern Chinese emissions. This pollution plume is observed ahead of a cold front (violet curve in Fig. 5c), associated with the low-pressure system north of 44∘ N. In this region, we expect the formation of a warm conveyor belt. This ascending air stream typically mixes up the air masses near the surface within the low atmospheric levels (e.g. Cooper et al., 2002; Ding et al., 2009; Foret et al., 2014). Such vertical mixing likely contributes to the observation of near-surface pollutants by satellite retrieval sensitive within the LMT. At this location (40–45∘ N, 122–128∘ E), models show ozone concentrations only enhanced up to ∼50 ppb (Fig. 6b for WRF-Chem), near background levels. CHASER clearly simulates the contribution of stratospheric ozone down to the LMT (north of 42∘ N), but shifted to the west (95–105∘ E, not shown). As the paper does not focus on these air masses, a detailed analysis of these differences is beyond the scope of the current paper.

Over the NCP (and also south of it), models simulate relatively high ozone concentrations (∼70 ppb) within the LMT at 10:00 JLT (see WRF-Chem in Fig. 6b). As previously mentioned, IASI + GOME2 also retrieves high LMT ozone concentrations over the NCP (around 35∘ N, 115∘ E, Fig. 5a), but also north of it (differing from simulations in the LMT). At the surface, ozone concentrations simulated by WRF-Chem remain rather low (near 40 ppb) at this time of the day (Fig. 6a), probably due to titration during the previous night. This suggests that the high ozone concentrations observed by IASI + GOME2 in the LMT (over the NCP) likely correspond to a plume located within the residual boundary layer and formed during the previous day. Such an ozone plume is simulated by WRF-Chem between 1 and 3 km altitude at 35–38∘ N, while reaching the surface south of 35∘ N (Fig. 6c for WRF-Chem). Both WRF-Chem and collocation with a plume of CO confirm the anthropogenic origin of the ozone plume in the LMT observed by IASI + GOME2 over the NCP. At this continental location, high NO2 concentrations are also both observed by GOME-2 (Fig. 6d) and simulated by the models (not shown).

The origin of this ozone plume may also be estimated from a comparison between CHASER simulations in two configurations: accounting or not accounting for the contribution of ozone from the stratosphere (see more details in Sect. 2.3). This analysis suggests that the stratospheric contribution over the NCP is practically negligible (∼5 ppb) in the LMT (Fig. 6e). Only above the LMT (between 3 and 6 kma.s.l.) is a higher contribution of ozone (∼30 ppb) from a stratospheric filament depicted by CHASER along a front extending from 32∘ N, 110∘ E to 48∘ N, 142∘ E (Fig. 6f). This is also suggested by collocated high values of potential vorticity (PV) at 300 hPa from ERA-I (Fig. 5d). IASI-only retrievals also depict this ozone filament above 3 kma.s.l. (although most pixels are cloudy; not shown).

During the following day (4 May), the anticyclone slowly moves northeastwards and approaches the Yellow Sea, with its pressure maximum at sea level near the coast (36∘ N, 122∘ E, Fig. 7c). According to HYSPLIT, the ozone plume located in the residual boundary layer the previous day is advected northwards by the anticyclonic circulation up to 39∘ N, 117∘ E (see magenta dots in Fig. 7c). At this location over the northern part of the NCP, collocated plumes of LMT ozone and CO are consistently observed respectively by IASI + GOME2 (Fig. 7a) and IASI (Fig. 7b). With respect to the previous day, the observed ozone concentrations remain rather high (∼90 ppb), while CO concentrations start dropping (∼270 ppb). Over northern China (north of 41∘ N), LMT ozone is transported eastwards following a low-pressure system centred at 50∘ N, 140∘ E (not shown). The stratospheric filament depicted by potential vorticity at 300 hPa east of 119∘ E (Fig. 7d) is transported southeastwards, far from the location of the ozone plumes observed in the LMT.

The anticyclone reaches the centre of the Yellow Sea (36∘ N, 122∘ E) on the next day (5 May, Fig. 8c), where it remains for two more days (until 7 May). On 5 May, the CO plume observed by IASI shows an almost identical horizontal structure to the O3 plumes seen by IASI + GOME2, extending across the Yellow Sea coast from the northern part of the NCP until the northern frontier of Korea. The ozone plume originating from the NCP clearly follows the anticyclonic circulation surrounding the Yellow Sea, reaching the northern coast of the Yellow Sea on 5 May (at 41∘ N, 124∘ E; magenta rectangle in Fig. 8a).

As suggested by HYSPLIT and IASI + GOME2, the pollution plume splits into two filaments on 6 May (i.e. two clear pathways are depicted, Fig. 9a and c), one heading south towards the Korean Peninsula as entrained by the anticyclonic circulation around the Yellow Sea and the other one north of it transported by eastwards winds to northeastern China (referred to hereafter as respectively “southern” and “northern” filaments in magenta and dotted red rectangles respectively at 39∘ N, 127∘ E and 43∘ N, 130∘ E; Fig. 9a). The trajectories of the southern and northern pollution plumes (respectively magenta and red in Fig. 9) are obtained by initialising HYSPLIT respectively at the mean arrival location of the trajectories from the previous day and 2∘ northeast from that position. The common geographical origin of the two plumes (before 6 May) is confirmed by HYSPLIT trajectories in backward mode initiated at the location of the two major ozone plumes clearly observed by IASI + GOME2 (e.g. south and north of Japan 2 days later in Fig. 12a).

Pollutants over the Korean Peninsula are carried southwards by relatively strong winds associated with a high- and a low-pressure system respectively to the west and to the east of these plumes (respectively centred at 35∘ N, 120∘ E and 30∘ N, 140∘ E in Fig. 9c). These collocated O3 and CO plumes show a progressive decrease in concentration with respect to the previous days (down to ∼220 ppb and ∼70 ppb respectively for CO and O3, Fig. 9). This reduction in pollutant concentrations may be induced (at least partly) by horizontal dilution during transport away from their sources.

On 7 May, the southern pollution plume over Korea elongates southwards (magenta rectangle at 37∘ N, 128∘ E in Fig. 10a and b), with an apparent decrease in CO concentrations (∼200 ppb). We remark that the O3 plumes over Korea (magenta rectangle) and the northeastern Chinese coast (dotted red rectangle) are probably located below 3 kma.s.l. as they are only clearly shown by IASI + GOME2 (Fig. 10a) and not by the IASI-only retrieval (Fig. 10e).

Over the area of the northern pollution filament (over the northeastern Chinese coast), enhancements of O3 and CO concentrations are shown respectively by IASI + GOME2 and IASI near 42∘ N, 125–132∘ E (dotted red rectangles and dots in Fig. 10a and b). This pollutant plume is observed ahead of the strong southward winds of a cold front (seen north of 42∘ N between 115 and 125∘ E, violet curve in Fig. 10b), probably transporting freshly emitted pollution (as suggested by moderately enhanced NO2 concentrations at 43–45∘ N, 126–132∘ E up to 6×1015 molcm-2 in Fig. 10c). These pollutants may originate from the densely populated agglomeration around the megacity of Harbin (43–46∘ N, 125–127∘ E). This enhancement of pollutant concentrations along the cold front is also clearly simulated by WRF-Chem (Fig. 10d for CO), although located slightly west of the plumes depicted by the satellite retrievals (Fig. 10b). To account for this difference in the location of the plumes, the dotted red square is shifted by 4∘ to the west in Fig. 10d with respect to the other panels. This freshly emitted or produced pollution likely mixes with the aged pollution air masses originating from the NCP. Moreover, we expect the formation of a warm conveyor belt ahead of that cold front (in violet in Fig. 10b), which typically mixes up air masses in the low atmospheric levels (as also remarked over this region on 3 May 2009, Fig. 5). Vertical mixing of freshly emitted pollutants within the LMT may also explain the enhancements of O3 and CO observed by the satellite approaches.

Another ozone plume is observed south of the Yellow Sea on 6–7 May by IASI + GOME2 and IASI retrievals respectively in the LMT and 3–6 kma.s.l. (Figs. 9a, 10a, e). The origin of this plume is probably related to downward transport from the stratosphere, as suggested by a collocated PV filament at 300 hPa (green contours in Fig. 10e). CHASER simulations suggest that this stratospheric ozone filament does not reach the LMT (Fig. 10f) and does not affect ozone concentrations of the tracked pollution plumes over Korea and the northeastern Chinese coast (rectangles).

The high-pressure system moves southwards to the East China Sea on 8 May and then eastwards until reaching the Pacific Ocean on 9 May (centred respectively at 28∘ N, 122∘ E in Fig. 11b and 27∘ N, 130∘ E in Fig. 12b). The southern polluted air masses coming from Korea (magenta rectangle) are entrained by the southwards circulation on an axis around 130∘ E between the high- and low-pressure systems (Fig. 11a and b). They reach southern Japan (the island of Kyushu) on 8 May (31∘ N, 130∘ E in Fig. 11a) and the Pacific on 9 May (27∘ N, 130∘ E in Fig. 12a). As only captured by IASI + GOME2 (Fig. 11a) and not by IASI (Fig. 11c), the moderately high O3 concentrations over the island of Kyushu are probably located below 3 kma.s.l. Ozone and carbon monoxide concentrations at the location depicted by HYSPLIT (magenta rectangles) are rather close to the background levels (particularly for CO).

The northern O3 plume coming from the northeastern Chinese coast is transported over the Sea of Japan by strong anticlockwise winds around the low-pressure system east of Japan on 8 May (Fig. 11a) until reaching the Pacific southeast of Japan on 9 May (Fig. 12a). This ozone plume is likely located in the LMT, as clearly depicted by IASI + GOME2 and not by IASI (red rectangles respectively in Fig. 11a and c). According to CHASER and low PV at 300 hPa, we do not expect a significant contribution of stratospheric ozone reaching the LMT over the Sea of Japan on 8 May (Fig. 11e and f) or over central Japan on 9 May (Fig. 12e and f). Therefore, these ozone plumes in the LMT are likely associated with photochemical production during transport from precursors emitted over land (from northeastern China on 7 May and Japan on 8 May). Simulations from WRF-Chem also suggest significant photochemical production of LMT ozone during transport from northeastern China to central Japan and the Pacific (see Sect. 4).

Downward transport of stratospheric ozone occurs southeast of Japan on 8–9 May, as suggested by the location of the PV filament at 300 hPa (Figs. 11e and 12e), which travels eastwards (located south of the Yellow Sea on 7 May). This ozone plume originating from the stratosphere reaches the lower troposphere above 3 kma.s.l. (as observed by both IASI + GOME2 and IASI in Figs. 11a and c, and 12a and c) but not the LMT (indicated by CHASER in Figs. 11f and 12f). This suggests distinct origins and vertical locations for the two elongated ozone plumes southeast of Japan observed by IASI + GOME2 on 9 May (27–40∘ N, 132–141∘ E in Fig. 12a). The one closer to Japan is associated with photochemical production, and the other one with stratospheric transport (indicated by green contours of PV in Fig. 12e), respectively located in the LMT and at 3–6 kma.s.l.

For the southern filament, a sustained increase of the ratio ΔO3/ΔCO is clearly derived during the whole ozone outbreak across East Asia from satellite observations (Fig. 13a). It evolves from ∼0.25 over the NCP on 3 May to ∼0.46 on 8–9 May over the Pacific and southern Japan (in magenta in Fig. 13a). As dilution of the air masses affects equally the concentrations of both pollutants, a monotonous enhancement of ΔO3/ΔCO is clear evidence of the production of O3 during transport. The evolution of this ratio from 0.25 over the main source regions to 0.46 after long-range transport estimated here with satellite retrievals is consistent with other estimations from airborne in situ measurements ranging from 0.2 to 0.5 (from flights at 2–3 km altitude) for other events of trans-Pacific long-range transport of industrial/urban pollution and in absence of stratospheric intrusions (Price et al., 2004). Overall consistency is also found with model estimates of 0.3 for typical air masses downwind from Asian pollution sources in springtime (Mauzerall et al., 2000) and the same value from in situ ground-based measurements over the United States in summer (Chin et al., 1994). These IASI + GOME2 O3 and IASI CO values are fairly higher than those estimated in the lower troposphere with IASI-only O3 retrievals (0.16–0.28 for the column below 6 kma.s.l.) for an East Asian pollution event in May 2008 (Dufour et al., 2015) and lower than retrievals in the free troposphere (400–700 hPa) from OMI and AIRS (∼0.6) over Tokyo (Kim et al., 2013).

Assuming that most CO is emitted at the beginning of the event (for the southern filament, magenta curve in Fig. 13a), the evolution of the ΔO3/ΔCO values suggests a production of O3 during transport of ∼60 % after the first 3 days (mainly over northern China and Korea) and of ∼84 % during the whole event (6 days) with respect to that over the NCP. In this case, the greatest growth of ΔO3/ΔCO occurs on 3–6 May when the air masses are transported over the most industrialised areas (the NCP and northern China) with the greatest emissions of ozone precursors as NOx (as shown in Fig. 10c for NO2). In the following days, a slower growth with almost constant ΔO3/ΔCO occurs over the Korean Peninsula (on 7 May), southern Japan (8 May) and the Pacific (9 May), far from the main sources of O3 precursors over China. Less ozone production over this oceanic region is consistent with low availability of NO2, indicated by satellite observations and both models (Fig. 14j), and a regime of NOx-limited photochemical production of O3, as observed over Fukue Island (Kanaya et al., 2016). This behaviour is also simulated by WRF-Chem and CHASER, showing ozone diurnal cycles with greater ozone production in the afternoon of the first 3 days of the event and significantly less thereafter (see hourly outputs of the models in Fig. 14a and c). Simulated diurnal cycles of ozone also reveal the strong nocturnal reduction in ozone concentrations (down to 30–40 ppb for WRF-Chem and 40–60 ppb for CHASER, dotted light colour curves in Fig. 14a and c), particularly significant over China (3–6 May), probably associated with nitrogen monoxide (NO) titration over the continent, and less pronounced thereafter, near or over the ocean. Moreover, WRF-Chem clearly suggests a reduction in CO concentrations (also observed by IASI in the lower troposphere), particularly significant after 6 May (Fig. 14b) and likely linked to atmospheric dilution (horizontal and/or vertical). Sinks of CO are not expected to be significant during a period of 3 days.

The ΔO3/ΔCO ratios derived from the CHASER and WRF-Chem models in the LMT follow a similar relative evolution to that from satellite retrievals, with a minimum at the beginning of the event and a relative monotonous increase by the end (Fig. 13b and c). This is particularly observed for WRF-Chem and until 6 May for CHASER. In absolute values, the ratio ΔO3/ΔCO derived from the satellite measurements is higher than that from models. At the beginning of the event (3–5 May), satellite estimates of the ratio are 0.1 to 0.15 higher than those from models. After 6 May, satellite and WRF-Chem ratios are closer (with differences between 0.05 and 0.1). Differences between the models and with respect to satellite-derived ΔO3/ΔCO ratios are likely associated with photochemical schemes in the models, model resolutions, precursors availability, the location of the plumes etc. The non-steady enhancement of ΔO3/ΔCO for CHASER after 6 May could be partly associated with significantly less availability of NO2 in the LMT, as compared to WRF-Chem (a factor of 3 higher, Fig. 14i) and a less marked reduction of CO concentrations in the LMT (while it clearly decreases for WRF-Chem, Fig. 14b and d).

Figure 14g and h show that the steady increase of ΔO3/ΔCO observed for the southern pollution plume does not seem to be linked to changes in sensitivities of the satellite retrievals. This is described in terms of the degrees of freedom and the altitude of maximum sensitivity of the retrieved atmospheric columns, which respectively quantify the amount of information provided by the satellite retrieval and the altitude it comes from. Neither of these two variables for either O3 or CO reflect such a steady variation, greater during the first 3 days and nearly flat afterwards, as that observed for ΔO3/ΔCO. The LMT O3 retrieval sensitivity peaks between 2.5 and 3 kma.s.l. for most of the days (and near 4 kma.s.l. over oceanic cold waters on 9 May), with degrees of freedom fluctuating from 0.2 to 0.3 (for the LMT, Fig. 14g, and around 5.5 to the O3 total column, not shown). The CO lower tropospheric column is retrieved with 0.8 to 1 DOF with a peak of sensitivity from 3.5 to 5 km altitude (Fig. 14g and h).

Figure 14e, f and i show evidence of the negligible influence of stratospheric ozone on the evolution of ΔO3/ΔCO in the LMT for the polluted air masses tracked on 3–9 May. Ozone amounts within the LMT originating from the troposphere are a factor of ∼12 greater than the contribution from the stratosphere, according to CHASER simulations (accounting or not accounting for stratospheric contributions, Fig. 14e). This is consistent with meteorological tracers of stratospheric air masses, such as the potential vorticity at 500 hPa (Fig. 14i). At this pressure level, no particular enhancement of potential vorticity is clearly remarked in correlation with the days of high concentration of ozone in the LMT. On the other hand, the ozone contribution of stratospheric downward transport in the upper troposphere (from 6 to 12 kma.s.l.) fluctuates significantly during the whole event (Fig. 14f). Potential vorticity on 3 May 2009 is high only at 300 hPa in consistency with an ozone enhancement in the upper troposphere (Fig. 14e), but not below (see potential vorticity at 500 hPa in Fig. 14i). As a quality check, we remark that similar concentrations of ozone in the LMT and the upper troposphere are retrieved by IASI + GOME2 and simulated by CHASER (adding in contributions from the troposphere and stratosphere) averaged over the whole event (differences of 13 ppb at most).

For the northern pollution plume, satellite-derived ΔO3/ΔCO ratios show an increase on 6 May (curve red in Fig. 13a) with respect to the previous days, as remarked for the southern plume. On 7 May, the eastern plume air masses exhibit lower ΔO3/ΔCO ratios of ∼0.25, probably due to mixing with freshly emitted pollutants from the northern China megacities (suggested by CO observations on 7 May, Fig. 10b, and NO2 concentrations from WRF-Chem and CHASER in Fig. 14j). This value of ΔO3/ΔCO is practically the same as the one observed on 3 May over large pollution sources from NCP. From 7 to 9 May, the ΔO3/ΔCO ratio (in red) rises up monotonically from ∼0.25 to ∼0.4, thus suggesting photochemical production during transport (as remarked for the days following emission of ozone precursors over the NCP). This evolution in terms of ΔO3/ΔCO ratios corresponds to an ozone production of about ∼60 % with respect to that on 7 May, within 2 days. The relative evolution of satellite-derived ΔO3/ΔCO ratios is consistent with WRF-Chem simulations (blue curve in Fig. 13b), which also show a relative increase from 5 to 6 May and then lower values on 7 May (with an additional pollution plume) that rise up monotonically until 9 May. The CHASER model shows an enhancement from 7 to 8 May, but it drops on 9 May (Fig. 13c). The latter might be linked to low availability of NO2 in the LMT in CHASER simulations (a factor of 3 lower than for WRF-Chem, Fig. 14j) and eventually a difficulty to represent such small-scale pollution plumes with the coarser resolution of this global model.

During this period, air masses are transported from northeastern China to the Sea of Japan and then over central Japan, finally reaching the Pacific. Ozone precursors might originate from northeastern Chinese and central Japanese megacities. Both WRF-Chem and CHASER simulations suggest a relatively higher availability of NO2 in the LMT (although 3 times higher for WRF-Chem) for the northern pollution filament (dotted curves with respectively blue ovals and green stars in Fig. 14j) than for the southern plume (light blue and light green in Fig. 14j). This is consistent with the greater growth of ΔO3/ΔCO (and therefore ozone production) from 7 to 9 May for the northern pollution plume with respect to that at the south, as estimated with satellite retrievals (Fig. 13a). WRF-Chem and CHASER simulations also suggest the occurrence of ozone production during transport after 7 May by a succession of marked diurnal cycles of ozone with greater amounts in the afternoon (Fig. 14a and c). As compared to the period before 6 May, ozone diurnal cycles simulated by both models exhibit smaller amplitudes, which are likely associated with less nighttime titration over non-continental areas. The reduction of this ozone reservoir may also enhance the growth of ΔO3/ΔCO during transport.

As for the southern plume, stratospheric contribution of ozone down to the LMT at the location of the northern pollution filament seems negligible according to CHASER simulations (Fig. 14e) and low values of potential vorticity (Fig. 14i). Furthermore, satellite-derived ΔO3/ΔCO ratios may be affected by changes in sensitivity for the CO IASI retrievals, which peaks at the middle troposphere on 7–8 May, instead of the lower troposphere (Fig. 14h). According to sensitivity analyses, these uncertainties induce under- or overestimations for ΔO3/ΔCO that remain below ±11 % for changes of 1 and 3 km at the heights of maximum sensitivity for respectively O3 and CO retrievals. These estimations are obtained using typical vertical profiles of O3 and CO for a pollution plume (from WRF-Chem) smoothed with averaging kernels of the satellite retrievals and taking into account the concomitant change in the heights of maximum sensitivity for O3 and CO retrievals, as in both cases they depend on the difference between surface and air temperatures. These uncertainties are significantly lower than changes observed for ΔO3/ΔCO from satellite retrievals (up to 84 % during the whole event). Therefore, conclusions drawn on the occurrence and quantification of photochemical ozone production in this period are not significantly affected by changes in satellite retrieval sensitivities.