A global three-dimensional chemistry transport model, GEOS-Chem (version v9-02) (Bey et al., 2001a, http://geos-chem.org), is employed to simulate the global ozone distributions and the source–receptor relationship between Africa and Asia. GEOS-Chem includes a detailed description of tropospheric O3–NOx–hydrocarbon–aerosol. It is driven by assimilated meteorological data from the GEOS at NASA Global Modeling and Assimilation Office (GMAO).

In this study, the simulations are driven by GEOS-4 meteorology at a 4∘ latitude by 5∘ longitude horizontal resolution, degraded from their native resolution of 1∘ latitude by 1.25∘ longitude. There are 30 vertical layers including 17 levels in the troposphere. Transport and chemical time steps are 10 and 20 min, respectively, as suggested by Philip et al. (2016). GEOS-4 uses the deep convection scheme of Zhang and McFarlane (1995) and the shallow convection scheme of Hack (1994). By comparing the GEOS-Chem simulations driven by GEOS-4 and GEOS-5 with satellite observations in the tropical troposphere, Liu et al. (2010) and Zhang et al. (2011) have found that GEOS-4 has stronger deep convection in the tropics than GEOS-5. In general, GEOS-Chem driven by GEOS-4 can simulate tropospheric ozone concentrations that are in good agreement with ozonesonde observations (Liu et al., 2010; Zhang et al., 2011; Choi et al., 2017; Y. Zhu et al., 2017).

GEOS-Chem has various modes that serve different simulation goals. In the full chemistry simulation, the emission inputs for a specific year are scaled from each inventory's base year. The global anthropogenic emission inventories are from EDGAR 3.2 for NOx, CO, and SOx (Olivier and Berdowski, 2001) and RETRO for VOC emissions (Pulles et al., 2007) in 2000, merged with the following regional inventories: the INTEX-B Asia emissions inventory in 2006 (Zhang et al., 2009), the US Environmental Protection Agency National Emission Inventory in 2005 (NEI05) for North America, the Cooperative Programme for Monitoring and Evaluation of the Long-range Transmission of Air Pollutants in Europe (EMEP) inventory for Europe (Vestreng and Klein, 2002) in 2005, the Big Bend Regional Aerosol and Visibility Observational (BRAVO) Study Emissions Inventory for 1999 in Mexico (Kuhns et al., 2003), and the Criteria Air Contaminants (CAC) inventory for Canada in 2005. Biofuel emissions are from Yevich and Logan (2003). Biomass burning and biogenic emissions are from the GFED3 inventory (van der Werf et al., 2010) and MEGAN 2.1 (Guenther et al., 2012), respectively. Lightning NOx emissions are calculated using the scheme of Allen et al. (2010) and the vertical distribution suggested by Ott et al. (2010). The annual global lightning NOx emissions are 5.97 TgNyr-1, comparable to 6 ± 2 TgNyr-1 in Martin et al. (2007) and 6.3 TgNyr-1 in Miyazaki et al. (2014). The annual total lightning NOx emission is 1.6 TgNyr-1 in Africa, 0.80 TgNyr-1 in northern hemispheric Africa (NHAF), and 0.79 TgNyr-1 in southern hemispheric Africa (SHAF) (see Figs. S1 and S2 in the Supplement). The anthropogenic CO and NOx emissions from GEOS-Chem for 2000 are compared with those in the HTAP2 emission inventories for 2008 (http://edgar.jrc.ec.europa.eu/htap_v2/) (Figs. S3 and S4). The annual anthropogenic CO emissions in Africa are 12.2 Tgyr-1 for 2000 in GEOS-Chem, lower than 62.5 Tgyr-1 for 2008 from the HTAP2 inventories. The anthropogenic NOx emissions in GEOS-Chem are 2.27 Tgyr-1 for 2000, also lower than 4.53 Tgyr-1 for 2008 from the HTAP2 inventories. Although the anthropogenic emissions contribute less significantly to the ozone generation in Africa than biogenic, biomass burning, and lightning emissions (Aghedo et al., 2007), the differences between these emission inventories imply that African ozone simulated by GEOS-Chem has some uncertainties.

To track the transport of ozone generated in Africa, we use the tagged ozone mode in GEOS-Chem, in which odd oxygen is tagged (Ox=O3+NO2+2NO3+3N2O5+HNO3+HNO4 + PAN + PMN + PPN, Fiore et al., 2002; Zhang et al., 2008). Since ozone accounts for most of Ox, we refer to ozone instead of Ox for clarity. To prepare for the tagged ozone simulation, we first run GEOS-Chem in the full chemistry mode to generate the daily ozone production rate and loss frequency. Then we run GEOS-Chem in the tagged ozone mode to differentiate ozone produced in different source regions, tagged as different tracers, by using the archived daily ozone production rates and loss frequencies. As shown in Fig. 1, the source region, Africa, is defined as the region of 35∘ S–15∘ N, 20∘ W–55∘ E and 15–35∘ N, 20∘ W–30∘ E. Therefore, ozone produced in Africa below the tropopause from all natural and anthropogenic sources is tagged as a tracer. We further divide Africa into NHAF and SHAF, each being further separated by the tropospheric layers, the lower troposphere (LT, from the surface to 700 hPa), the middle troposphere (MT, 700–300 hPa), and the upper troposphere (UT, 300 hPa–tropopause), therefore adding six more tracers. The receptor region, Asia, is defined as the region of 5–60∘ N, 60–145∘ E.

In the simulation for a year, both seasonal variations of chemistry and meteorology are considered. To study the meteorological influence on the interannual variation in the transport of African ozone to Asia, the tagged ozone simulation is conducted for 21 years from January 1986 to December 2006 (using 1986 for model spin-up). As our purpose is focused on the meteorological influence on transport, we allow the meteorology to vary from year to year but fix the chemistry constantly, i.e., using the archived daily ozone production and loss data in one year, i.e., 2005, for the 20-year simulation. This treatment is similar to previous studies (Liu et al., 2005; Sekiya and Sudo, 2014). Note that meteorology affects both transport (a physical process) and chemistry, while this study only considers the former.

Numerical approaches to exploring the source–receptor relationships include (1) tagged tracer simulations, (2) trajectory simulations, (3) perturbation simulations, and (4) inverse simulations (Y. Zhu et al., 2017). Compared with the other approaches, the tagged tracer simulation can track ozone effectively from different source regions and quantify the contributions of ozone from different source regions to a receptor region. One issue with the tagged ozone simulation is the nonlinearity of chemistry. This nonlinearity can cause large differences between the full chemistry and the tagged ozone simulations. To test the scale of the nonlinearity in the simulations, ozone concentrations between the full chemistry and the tagged ozone simulations are compared at four ozonesonde sites in Africa and India (Fig. S5). The difference is found to generally be within ± 5 %, suggesting that this approach works in these regions without large bias.

GEOS-Chem has been used widely in studying pollution transport (Bey et al., 2001b; Koumoutsaris et al., 2008; Zhang et al., 2008; Liu et al., 2011; Wang et al., 2011; Ridder et al., 2012; Long et al., 2015; Jiang et al., 2016; Huang et al., 2017; Ikeda et al., 2017; J. Zhu et al., 2017; Y. Zhu et al., 2017). Tropospheric ozone simulated by GEOS-Chem has been extensively validated using ozonesonde and satellite data, such as in North America (Zhang et al., 2008; Y. Zhu et al., 2017), Europe (Kim et al., 2015), East Asia (Wang et al., 2011; J. Zhu et al., 2017; Y. Zhu et al., 2017), and other regions (Liu et al., 2009). These validation practices have suggested reasonable agreements between the model simulations and ozone measurements. In this study, for an enhanced confidence on the model performance, we compare the GEOS-Chem simulations with the ozonesonde data in Africa and India and with the satellite measurements from the Tropospheric Emission Spectrometer (TES) instrument. The ozonesonde data are acquired from the World Ozone and Ultraviolet Radiation Data Centre (WOUDC) (http://www.woudc.org/home.php) and the monthly TES product TL2O3LN is from the NASA Langley Atmospheric Science Data Center (https://eosweb.larc.nasa.gov/project/tes/tes_table).

Three ozonesonde stations in Africa are selected for their long record. The stations include Santa Cruz in northern Africa (28.42∘ N, 16.26∘ W; 36 m), Nairobi in eastern Africa (1.27∘ S, 36.8∘ E; 1745 m), and Irene in southern Africa (25.91∘ S, 28.21∘ E; 1524 m) (Fig. 1). These ozonesonde data have been used in studying tropospheric ozone in Africa widely (Clain et al., 2009; Thompson et al., 2012, 2014). In Asia, comparisons between GEOS-Chem simulations and ozonesonde observations have been made by Liu et al. (2002) and Y. Zhu et al. (2017) at stations over the Pacific Rim, showing that GEOS-Chem can generally capture the vertical and seasonal variations of ozone concentrations in the region. We further compare the GEOS-Chem simulations at three Indian sites, including New Delhi in northern India (28.3∘ N, 77.1∘ E; 273 m), Pune in western India (18.53∘ N, 73.85∘ E; 559 m), and Thiruvananthapuram in southern India (8.48∘ N, 76.95∘ E; 60 m) (Fig. 1).

Figure 2 shows the simulated and measured vertical ozone profiles by season, which are averaged from 1999 to 2003 at Santa Cruz, from 2003 to 2006 at Nairobi, and from 1999 to 2005 at Irene. Figure 3 compares the time series of monthly ozone concentrations between the model simulations and ozonesonde measurements at different tropospheric layers at the sites. The corresponding bias, root-mean-square error (RMSE), and correlation coefficient (r) between the two datasets are shown in Table 1 for the vertical profiles and in Table 2 for the time series, respectively.

For the ozone vertical profiles (Fig. 2 and Table 1), GEOS-Chem appears to reasonably capture the ozone vertical variation and its seasonality at the three sites. It appears that GEOS-Chem overestimates ozone in the upper troposphere at Santa Cruz in NH winter and spring and underestimates ozone in the upper troposphere at Nairobi in NH summer and autumn. The correlation coefficients between the two datasets are above 0.9 for most seasons at the three sites. The mean bias ranges from -9.3 % (at Nairobi in October) to 23.2 % (at Santa Cruz in January), while the RMSE ranges from 4.3 ppbv (at Nairobi in April) to 15.5 ppbv (at Santa Cruz in January). The bias and RMSE suggest that the model performs no better at a single station or in a single season than the other stations or seasons.

For the time series of ozone in the upper, middle, and lower troposphere, as well as the surface layer (Fig. 3 and Table 2), GEOS-Chem also performs reasonably well. The correlation coefficients between the two datasets in the tropospheric layers range within 0.57–0.79 at Santa Cruz, 0.76–0.90 at Nairobi, and 0.61–0.82 at Irene, all significant at a 95 % level. The correlation coefficient is 0.60, 0.82, and 0.39 for the surface layer at Santa Cruz, Nairobi, and Irene, respectively. However, the model underestimates upper tropospheric ozone in Nairobi (Table 2, bias = -14.2 %) and somewhat overestimates ozone in 500–300 hPa at Santa Cruz (Table 2, bias = 4.5 %) and near the surface at Irene (Table 2, bias = 60.3 %).

The comparison between the GEOS-Chem and ozonesonde data at the three India sites is shown in a seasonal altitude distribution in Fig. 4. The ozone concentrations are the means between 1994 and 2003. The time series of ozonesonde data at the sites are not shown because of inadequate records. The seasonal altitude patterns of ozone at the three sites are well simulated, although the model overestimates the ozone near the surface and underestimates the ozone in the middle and upper troposphere at the three sites. The annual mean bias at New Delhi is 5.6, -14.5, and -18.9 % in the LT, MT, and UT, respectively. At Pune, it is 2.1, -5.6, and -19.1 % in the LT, MT, and UT, respectively, while the annual mean bias at Thiruvananthapuram is -4.4 %, -13.3 %, and -27.4 % in the LT, MT, and UT, respectively.

The global distribution of ozone from the GEOS-Chem simulations is compared with the TES observation in 2005 in the four seasons at 464 hPa (Fig. S6), a layer around which the TES satellite data have the least bias. The GEOS-Chem simulations are smoothed with the TES averaging kernels and a priori profiles. The smoothed GEOS-Chem simulations resemble the GEOS-Chem simulations (not shown; see Jiang et al., 2016). Generally, GEOS-Chem can capture the global variation of ozone in space and by season, such as the elevated ozone concentrations over the NH middle latitudes in NH spring and summer and the plumes of elevated ozone from biomass burning over southern Africa in NH autumn. The smoothed ozone concentrations are generally lower than shown by the TES observations over Africa in all the seasons, with a maximum difference of 20 ppbv (or 20 %). Note that TES ozone retrievals generally are higher than ozonesonde data, having a mean positive bias of 3–11 ppbv in the troposphere (Nassar et al., 2008).

Overall, GEOS-Chem can reasonably capture the seasonality of the ozone vertical profiles in the ozonesonde data over Africa. In Asia, GEOS-Chem tends to overestimate ozone in the lower troposphere and underestimate ozone in the upper troposphere at three Indian sites. The GEOS-Chem simulations also compare well with satellite TES data in space and by season.

To supplement the analysis on the GEOS-Chem simulations, we use the Hybrid Single-Particle Lagrangian Integrated Trajectory model, version 4 (HYSPLIT-4) (Draxler and Hess, 1998; Stein et al., 2015), to simulate forward trajectories and examine the transport pathways of African ozone to Asia. The HYSPLIT model is one of the most extensively used atmospheric transport and dispersion models (Fleming et al., 2012). Meteorological inputs to HYSPLIT are from the NCEP reanalysis at a resolution of 2.5∘ × 2.5∘ (http://ready.arl.noaa.gov/archives.php). Six-day forward trajectories are calculated from 1987 to 2006 four times a day (00:00, 06:00, 12:00, and 18:00 UTC) at seven African sites, including Cairo (31.1∘ E, 30∘ N), Ghat (10.2∘ E, 24.6∘ N), Khartoum (32.3∘ E, 15.4∘ N), Abuja (7.3∘ E, 9∘ N), and Juba (31.4∘ E, 4.5∘ N) in NHAF, and Dar es Salaam (39.2∘ E, 6.5∘ S) and Luanda (13.1∘ E, 8.5∘ S) in SHAF (Fig. 1). The levels at 1.5, 5.5, and 11 km are selected to represent the trajector starting altitude at the lower, middle, and the upper troposphere, respectively. The seven sites are chosen to represent two longitudinal and 3–4 latitudinal zones in Africa.

The meteorological data include the NCEP/NCAR reanalysis I (Kalnay et al., 1996). The monthly wind fields are used to describe the climatology of atmospheric circulation during the study period from 1987 to 2006. The product is available on a 2.5∘ × 2.5∘ horizontal grid at 17 pressure levels from 1000 to 10 hPa (https://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.html).

Additionally, the monthly Outgoing Longwave Radiation (OLR) data from NCAR at 2.5∘ × 2.5∘ with temporal interpolation (Liebmann and Smith, 1996, https://www.esrl.noaa.gov/psd/data/gridded/data.interp_OLR.html) are used to indicate the intensity of the ITCZ. The dataset has been widely used for tropical studies on deep convection and rainfall (Mounier and Janicot, 2004).

Because a substantial amount of ozone and ozone precursors are produced or emitted in Africa, tropospheric ozone in the other continents is largely influenced by ozone outflow from Africa (Aghedo et al., 2007; Zare et al., 2014). Figure 5 shows seasonal variations of imported African ozone in the Asian troposphere, varying with latitude and altitude. The values are the 20-year means (1987–2006) from the GEOS-Chem simulation. The largest African influences appear in the Asian middle and upper troposphere, i.e., 12–16 ppbv at 500 and 200 hPa. African ozone in the Asian lower troposphere is reduced to 2–10 ppbv at 700 hPa. Unlike imported European and North American ozone that influences Asia mostly over high altitudes, i.e., north of 30∘ N (Y. Zhu et al., 2017), imported African ozone prevails over lower latitudes, generally between 5 and 40∘ N, considerably larger than European and North American ozone south of 30∘ N (Wild et al., 2004; Sudo and Akimoto, 2007; Y. Zhu et al., 2017). North of 50∘ N, African ozone influence is small, i.e., less than 8 and 2 ppbv, respectively, above and below the Asian middle troposphere. Seasonally imported African ozone in the upper troposphere peaks in NH spring around 30∘ N (∼ 16 ppbv, Fig. 5a) and is at its minimum in NH summer south of 25∘ N. Owing to the high radiative forcing efficiency, the change in ozone concentrations in the upper troposphere can impact climate more significantly than that in the lower troposphere (Lacis et al., 1990). Therefore, the influence of African ozone on the climate in southern Asia is likely larger than that of European and North American ozone. In the middle troposphere (Fig. 5b), African ozone is at its maximum (∼ 16 ppbv) in NH winter between 20 and 25∘ N. In the lower troposphere (Fig. 5c), between 5 and 40∘ N, African ozone is high in NH winter (∼ 6–10 ppbv), while south of 10∘ N, African ozone has another peak in NH summer (∼ 4 ppbv). Near the surface (Fig. 5d), African ozone concentrations are low, i.e., below 4 ppbv.

The strong seasonality of imported African ozone can also be shown vertically in Fig. 6a, in which imported African ozone is averaged over Asia south of 40∘ N. The fractional contribution of imported African ozone to ozone in Asia is shown in Fig. 6b. In the Asian upper troposphere, imported African ozone is largest during March–May (over 10 ppbv) and least during July–October (below 6 ppbv). In the Asian middle troposphere, imported African ozone is at a minimum from July to September (∼ 4 ppbv, 6 % in the fractional contribution) and a maximum (over 10 ppbv, 14 % in the fractional contribution) from January to March, about 2 months earlier than in the upper troposphere. In the Asian lower troposphere, imported African ozone is largest in NH winter (∼ 4 ppbv, 8 % in the fractional contribution) and lowest in NH autumn (∼ 2 ppbv, 5 % in the fractional contribution).

Furthermore, the total imported African ozone over Asia is divided by tropospheric layer (UT, MT, and LT) and by hemisphere, and the fractional contribution for each region is shown in Fig. 7. Over the Asian upper troposphere, African ozone from the NHAF UT accounts for 60–70 % of the total imported African ozone (Fig. 7a), decreasing to ∼ 20 % in the Asian MT (Fig. 7b) and to below 20 % in the Asian LT (Fig. 7c). In the meantime, the influence of African ozone from the NHAF MT becomes larger in the Asian MT and LT (20–40 %), as does the influence of African ozone from the NHAF LT (20–40 %). African ozone from SHAF contributes to the total imported ozone throughout the year in the three tropospheric layers. The contribution is small, usually below 20 % of the total imported African ozone, except for in NH winter over the Asian UT (Fig. 7a) and in NH summer over the Asian LT (Fig. 7c). The two exceptions in interhemispheric transport will be discussed in detail in Sect. 4.2.

The seasonality of the transport of African ozone to Asia results from the collective impact of the emissions of ozone precursors in the source region and the meteorology and chemistry from the source region to the receptor region. The precursors of African ozone are mainly from biogenic, biomass burning, and lightning NOx sources (Piketh and Walton, 2004; Thompson, 2004; Aghedo et al., 2007; Giglio et al., 2013; Monks et al., 2015). The seasonalities of emissions from biogenic sources, biomass burning, lightning, and anthropogenic sources in Africa are characterized rather differently from the other continents (Williams et al., 2009; Guenther et al., 2012; Giglio et al., 2013; Albrecht et al., 2016). Since Africa covers areas in both hemispheres with a large portion in the tropics, the atmospheric circulation over Africa experiences obvious seasonal changes induced by the seasonality of the ITCZ and the Hadley cell (Nicholson, 2008, 2009; Žagar et al., 2011; Suzuki, 2011). To cast some light on the seasonal variations of imported African ozone over Asia presented in this section, the ITCZ and ozone precursor emissions over Africa are discussed in Sect. 3.2. Based on the discussion, the possible mechanisms that modulate the transport of African ozone to Asia are speculated in Sect. 3.3.

Based on Nicholson (2009, 2013), Suzuki (2011), and Žagar et al. (2011), the mean positions of the ITCZ in Africa in the four seasons are roughly illustrated in Fig. 8a–d. The latitudinal migration of the ITCZ with season varies with longitude and the migration is within a wider range of latitudes in eastern Africa (10–40∘ E) than in western Africa (west of 10∘ E). In eastern Africa, ITCZ shifts between ∼ 10∘ S in NH winter and ∼ 20∘ N in NH summer, while in western Africa, the center of the ITCZ swings from 5 to 20∘ N between NH winter and NH summer.

Seasonal variations of African ozone largely depend on biogenic, biomass burning, lightning, and anthropogenic emissions. The anthropogenic emissions are generally considered to be small and have weak seasonality (Aghedo et al., 2007; Williams et al., 2009). Based on the emission inventories in GEOS-Chem that are described in Sect. 2.1, the spatial distributions of ozone precursor emissions are shown by season in Fig. 8, including isoprene emissions from biogenic sources, CO emissions from biomass burning, and NOx emissions from lightning at 700 and 300 hPa. Seasonal variations of these emissions averaged over Africa, NHAF, and SHAF are shown in Fig. 9.

Isoprene (C5H8) is the dominant non-methane volatile organic compound (NMVOC) emitted by vegetation (Marais et al., 2012). Biogenic isoprene emissions in Africa are considered to be responsible for about 65 % of ozone enhancement in the African upper troposphere (Aghedo et al., 2007). Isoprene emissions shown in Fig. 8a–d are representative of biogenic emissions of ozone precursors. The magnitude and spatiotemporal pattern of the isoprene emissions in Africa in Figs. 8 and 9 are comparable with Marais et al. (2014). The maximum biogenic isoprene emissions are over central African rainforests throughout the year (Fig. 8a–d). The seasonal cycle of biogenic isoprene over Africa peaks in NH spring and autumn (Fig. 9a). Plenty of non-methane VOCs are emitted from biogenic sources that can be uplifted by strong convection. Aghedo et al. (2007) and Zare et al. (2014) suggested that biogenic emissions can lead to more ozone generation in the African upper troposphere than biomass burning and anthropogenic emissions.

In NH winter, fires in Africa are active in the NH between 0–10∘ N and 15∘ W to 40∘ E (Fig. 8e–h; see also Sauvage et al., 2005). From NH winter to autumn, regions with biomass burning shift southward (Fig. 8e–h; see also van der Werf et al., 2010; Giglio et al., 2013). In NH summer, fires are most active in the SH from the equator to 20∘ S. Therefore, the regional CO emissions from biomass burning peak during NH winter in NHAF and during NH summer in SHAF (Fig. 9b). Aghedo et al. (2007) stated that biomass burning has the largest impact on surface ozone in the vicinity of the African burning regions during the burning seasons.

In Africa, lightning NOx is produced mostly in the middle to upper troposphere (Fig. 8i–p; see also Pickering et al., 1998; Ott et al., 2010; Miyazaki et al., 2014). Miyazaki et al. (2014) estimated that the altitudes where the annual lightning NOx emissions maximize are around 11 km in northern Africa and 9.36 km in southern Africa. Aghedo et al. (2007) suggested that lightning emissions mainly enhance ozone production in the African middle and upper troposphere. Ascribed to the high efficiency of deep moist convection, frequent lightning activities occur in the ITCZ (Christian et al., 2003; Avila et al., 2010). Collier and Hughes (2011) suggested that the peak lightning activities are generally located on the southern border of the ITCZ in Africa. The seasonality of the geographic variation of lightning NOx emissions clearly shows the influence of the ITCZ over Africa (Collier and Hughes, 2011). When the ITCZ reaches its northernmost position in NH summer (Fig. 8c), lightning NOx emission over NHAF becomes the highest (Fig. 9c and d). Similarly, the lightning NOx emission over SHAF peaks in NH winter (Fig. 9c and d).

As African ozone mainly peaks in the Asian middle and upper troposphere, the horizontal distributions of African ozone at 400 hPa in January, April, July, and October, overlaid with winds, are shown in Fig. 10 to illustrate the seasonality and the transport pathways of African ozone to Asia at this level. In NH winter, African ozone can be transported for a long distance along the subtropical westerlies in the two hemispheres, reaching the western Pacific in the NH and across Australia in the SH. In NH summer, the NH subtropical westerlies and tropical easterlies shift to their northernmost positions; less African ozone can be transported to Asia than in the other seasons. Furthermore, Fig. 11 shows the latitude–altitude cross sections of African ozone and winds averaged from 0 to 40∘ E. This is to show how African ozone in the source region varies vertically along different latitudes. Figure 12 is the same as Fig. 11 but for the longitude–altitude cross sections averaged from 20 to 35∘ N in order to show the transport pathway along longitudes from Africa to Asia. Figure 13 shows the 20-year mean paths for the trajectories that run from seven representative sites (Cairo, Ghat, Abuja, Khartoum, Juba, Dar es Salaam, and Luanda) to Asia. The mean paths are shown by season and by the original tropospheric layer, including the lapse day from the beginning of the trajectories. Additionally, we conduct three sensitivity experiments by “switching off” the biogenic, lightning, and biomass burning emissions, respectively, to assist our analysis. The separate contributions of the three sources to tropospheric ozone over Africa are shown in Fig. S7. In the following, we analyze the information from these figures, in combination with literature, to explore possible mechanisms responsible for the transport of African ozone to Asia in the four seasons.

As discussed, the ITCZ can impact meteorology in Africa (Sultan and Janicot, 2000; Xie, 2004; Hu et al., 2007; Collier and Hughes, 2011; Suzuki, 2011). The deep convection along the ITCZ can carry ozone precursors from biogenic, biomass burning, and anthropogenic emissions to the upper layers from the surface. The ITCZ is also a zone with large lightning activities and thus can impact the seasonality of NOx emission from lightning. The convective divergence in the upper troposphere over the ITCZ in Africa plays a significant role in output of African ozone and in the interhemispheric transport between SHAF and NHAF (Fig. 11 and see trajectories in Fig. 13).

To investigate the impact of ITCZ on the interannual variation of the transport of African ozone to Asia, we use the monthly OLR data from NCAR at 2.5∘ × 2.5∘ with temporal interpolation (Liebmann and Smith, 1996) as a proxy to indicate the intensity of the ITCZ. According to Waliser et al. (1993) and Fukutomi and Yasunari (2013), the number of the grids with OLR ≤ 260 Wm-2 in the region of 15∘ W–45∘ E, 20∘ S–20∘ N can indicate the intensity of the convection over the ITCZ in Africa.

We find that the intensity of the ITCZ in Africa is mostly related to imported African ozone over Asia in NH winter. Figure 14 shows the time series of anomalies of imported African ozone averaged over Asia, against the intensity of the ITCZ over Africa. The intensity of the ITCZ is normalized; i.e., the normalized value is equal to the original value minus the mean and then divided by the standard deviation in January from 1987 to 2006. The intensity of the ITCZ and anomalies of imported African ozone are correlated with the correlation coefficient (r) of 0.61, 0.46, and 0.64 in the Asian lower (975 hPa), middle (600 hPa), and upper (200 hPa) troposphere, respectively, all statistically significant at the 95 % level (p<0.05).

Differentiating between hemispheres, significant correlations are found between the ITCZ and imported NHAF ozone in the entire troposphere in Asia (Fig. 14). The interhemispheric transport of ozone from SHAF to Asia also correlates with the intensity of ITCZ in Africa well, with r being 0.56 in the Asian upper troposphere (Fig. 14a). The interhemispheric transport to the Asian middle and lower troposphere is minimal so these time series are not shown.

Overall, when the intensity of the African ITCZ is stronger, more ozone and ozone precursors are uplifted to the middle and upper troposphere and transported northward and then eastward to Asia by the NH subtropical westerlies. Additionally, driven by the enhanced convective divergence over the ITCZ, more ozone from SHAF is transported across the equator and to the NHAF upper troposphere. Consequently, African ozone increases in the Asian middle and upper troposphere. Meanwhile, carried by the downdrafts from the Asian winter monsoon (Y. Zhu et al., 2017), more ozone is transported to the surface in Asia.

As shown in Fig. 7 and discussed in earlier sections, ozone generated in SHAF can be transported across the equator and eventually to Asia. This is illustrated in more detail in Fig. 15, showing seasonal altitude variations of imported ozone from NHAF and SHAF over Asia and the fractional contributions of NHAF and SHAF ozone to the total imported African ozone. Ozone from NHAF accounts for over 80 % of the total imported African ozone in the Asian tropospheric column throughout the year, except for in the upper troposphere during NH winter and in the lower troposphere during NH summer (Fig. 15b and d). This represents two important interhemispheric transport pathways.

For the first transport pathway, Fig. 7 suggests that the SHAF ozone originates mainly from the SHAF UT. This ozone can be transported northward across the equator along the Hadley circulation and then eastward to Asia by the NH subtropical westerlies (Fig. 13a). The amount of ozone being transported is at a maximum in NH winter and at a minimum in NH summer (Figs. 7 and 15c) when the ITCZ is at its southernmost and northernmost position, respectively (Fig. 8a–d). This can be further illustrated in the horizontal distribution of SHAF ozone at 200 hPa (Fig. 16a). SHAF ozone is 2–4 ppbv over China south of 30∘ N and 4–6 ppbv over western India. As shown in Sect. 4.1 (Fig. 14a), the variation of the ITCZ intensity in Africa can explain 31 % of the interannual variation of the transport of SHAF ozone to Asia in NH winter along this pathway.

The second transport pathway is shown in Fig. 16b for SHAF ozone distribution at 850 hPa in July, as Fig. 7c also suggests that the interhemispheric transport mainly occurs from the SHAF lower and middle troposphere. SHAF ozone concentrations are 2–4 ppbv over the Arabian Sea and the west coast of the Indian subcontinent. This ozone is transported to India in NH summer by the Somali cross-equatorial flow (Fig. 16b), which is the strongest seasonal cross-equatorial flow in the lower troposphere, serving as an essential component of the Asian monsoon system (Zhu, 2012). This is the reason for the maximum SHAF ozone (∼ 4 ppbv) over the Asian lower troposphere south of 10∘ N in NH summer (Fig. 5c). One more piece of evidence for this transport is shown in the trajectories from Dar es Salaam, on the eastern coast of Africa (Fig. 13h and i). The interhemispheric ozone transport to western India takes more than 6 days. Furthermore, the signal of the transport is captured in the ozonesonde data in western India. The vertical distributions of the seasonal ozone variations at Pune and Thiruvananthapuram are shown for the ozonesonde and GEOS-Chem data (Fig. 4). A dip of lower tropospheric ozone concentrations in both datasets is apparent in NH summer, when the Somali jet carries clear air masses from the sea, which can be traced back to SHAF (Figs. 13h, i and 16b).

To search for a connection between the Somali jet and the imported SHAF ozone over western India, we use an index, proposed by Li and Li (2014), to indicate the intensity of the Somali jet. The index is calculated as the mean meridional wind at 850 hPa in the domain shown in Fig. 16b. Li and Li (2014) correlated the Somali jet and other cross-equatorial flows using the index. Figure 17 shows the anomaly of SHAF ozone averaged in the lower troposphere at Pune and Thiruvananthapuram during NH summer from 1987 to 2006. Positive correlations are found between the anomaly and the intensity of the Somali jet at both sites, with the correlation coefficients over 0.56, significant at the 95 % level.