Earth Observing System (EOS) microwave limb sounder (MLS) is one of the four instruments on NASA's EOS Aura satellite flying in the polar sun-synchronous orbit. It measures the thermal emissions at millimetre and sub-millimetre wavelengths (Waters et al., 2006). It performs 240 limb scans per orbit with a footprint of ∼ 6 km across-track and ∼ 200 km along-track, providing ∼ 3500 profiles per day. MLS also measures vertical profiles of temperature, ozone, CO, H2O, and many other constituents in the mesosphere, stratosphere, and upper troposphere (Waters et al., 2006). In the UTLS, MLS has a vertical resolution of about 3 km. MLS vertical profiles of ozone show good agreements with the Stratospheric Aerosol and Gas Experiment II (SAGE-II), Halogen Occultation Experiment (HALOE), Atmospheric Chemistry Experiment (ACE), and ozonesonde measurements (Froidevaux et al., 2006). The MLS ozone profiles are considered to be useful in the range of 215–0.46 hPa (Livesey et al., 2005). In this study we analyse the MLS level 2 (version 4) ozone mixing ratios data for the period 2004–2013. The data have been gridded horizontally, within latitude bins of equal area (with the equatorial bin of 150 km width) and longitude bins of about 8.5∘. These data can be accessed from http://mls.jpl.nasa.gov/. For comparison, simulated ozone is convolved with the MLS averaging kernel (Livesey et al., 2011).

We employ the aerosol–chemistry–climate model ECHAM5-HAMMOZ, which comprises the general circulation model ECHAM5 (Roeckner et al., 2003), the tropospheric chemistry module MOZART2 (Horowitz et al., 2003), and the aerosol module Hamburg aerosol model (HAM) (Stier et al., 2005). It includes NOx, VOC, and aerosol chemistry. The gas-phase chemistry is based on the chemical scheme provided by the MOZART-2 model (Horowitz et al., 2003), which includes detailed chemistry of the Ox–NOx hydrocarbon system with 63 tracers and 168 reactions. The O(1D) quenching reaction rates used are taken from Sander et al. (2006) and isoprene nitrates chemistry taken from Fiore et al. (2005). The dry deposition in ECHAM5-HAMMOZ follows the scheme given by Ganzeveld and Lelieveld (1995). Soluble trace gases like HNO3 and SO2 are also subject to wet deposition. In-cloud and below-cloud scavenging follows the scheme given by Stier et al. (2005). Interactive calculation of cloud droplet number concentration is according to Lohmann et al. (1999) and ice crystal number concentrations are according to Kärcher and Lohmann (2002). The convection scheme is based on the mass flux scheme developed by Tiedke (1989). Lightning NOx emissions are parameterized following Grewe et al. (2001).

The model is run at a T42 spectral resolution corresponding to about 2.8∘ × 2.8∘ in the horizontal dimension and 31 vertical hybrid σ–p levels from the surface to 10 hPa. In our model simulations, emissions from anthropogenic sources and biomass burning are from the year 2000 RETRO project data set (available at http://eccad.sedoo.fr/) (Schultz et al., 2004, 2005, 2007, 2008). Emissions of SO2, black carbon, and organic carbon are based on the AEROCOM-II emission inventory, also for the year 2000 (Dentener et al., 2006). The distribution of NOx emission mass flux (kg m-2 s-1) averaged for the ASM season (June–September) is shown in Fig. S1 in the Supplement. It shows high values over the Indo-Gangetic Plain and East China. Other details of model parameterizations, emissions, and evaluation are described by Fadnavis et al. (2013, 2014, 2015) and Pozzoli et al. (2008a, b, 2011). Each of our model experiments consists of continuous simulations for 11 years from 2000 to 2010. The base year for emissions is taken as 2000 and emissions were repeated every year throughout the simulation period. Meteorology varied due to varying monthly mean sea surface temperature (SST) and sea ice concentration (SIC). The AMIP2 SSTs and SIC varying for the period 2000–2010 were specified as a lower boundary condition.

In order to understand the impact of enhanced anthropogenic NOx emissions on the distribution of ozone in the UTLS, sensitivity simulations were performed for the period 2000–2010. The experimental set-up is the same as described by Fadnavis et al. (2014). The four simulations analysed in this study are a reference experiment (CTRL) and three sensitivity experiments (referred to as experiments 2–4), where the anthropogenic NOx emissions over India and China are scaled in accordance with the observed trends. In experiment 2, anthropogenic NOx emissions are increased over India by 38 % (Ind38). In experiment 3, increases over China by 73 % (Chin73) are prescribed. In order to analyse the effects of similar NOx percentage increases over India and China, NOx emissions are increased over India by 73 % (Ind73) in experiment 4. The emission perturbations were obtained from observed NO2 trends of 3.8 % per year over India (Ghude et al., 2013) and 7.3 % per year over China (Schneider and van der A, 2012). Hiboll et al. (2013) also reported similar increasing NOx values over megacities in India and China. All four simulations use the same VOC and CO emissions and they all include NOx production due to lightning (lightning-on) and soil emissions. There may be indirect impact of lightning NOx emission. Since it is same in CTRL and sensitivity simulations its impact may be negligible.

In addition, a lightning-off simulation was performed for the same period and boundary conditions as experiments 1–4 (this simulation is the same as the one described in Fadnavis et al., 2014). The impact of lightning on NOx production is estimated by comparing the CTRL (lightning-on) with lightning-off simulations.

The accuracy of the simulation of the monsoon circulation probably depends on model resolution and an increased vertical resolution may improve the model performance (Druyan et al., 2008; Abhik et al., 2014). However, the model resolution of T42L31 is capable of reasonably simulating the general regional spatial pattern of precipitation and low-level circulation (Rajeevan et al., 2005) (see Fig. S2, showing simulated seasonal mean precipitation and circulation at 850 hPa in the CTRL simulation).

The heating rates and radiative forcings associated with the ozone changes in our three sensitivity simulations are calculated using the Edwards and Slingo (1996) radiative transfer model and the fixed dynamical heating approximation for stratospheric temperature adjustment. Similarly to previous studies (Riese et al., 2012; Bekki et al., 2013; Rap et al., 2015), we used the offline version of the model, with six shortwave and nine longwave bands, and a delta-Eddington two-stream scattering solver at all wavelengths.

The spatial distributions of ozone mixing ratios from MLS observations at 100 hPa and from the CTRL ECHAM5-HAMMOZ simulation at 90 hPa (the nearest model level) after smoothing with the averaging kernel of MLS are illustrated in Fig. 1a and b respectively. For comparison we have interpolated the model data to the MLS pressure grid, then applied the MLS averaging kernel and finally interpolated back to the model pressure grid. The climatological horizontal winds plotted in the figure clearly show the anticyclonic upper-level monsoon circulation. Recent attempts to characterize the extent of the anticyclone are based either on potential vorticity on isentropic surfaces or on geopotential height on pressure surfaces. Here we apply both characterizations of the anticyclone and show the PV contour related to the maximum PV gradient on 380 K (calculated from ERA-Interim reanalysis following Ploeger et al., 2015), and the 270 m geopotential height anomaly as proposed by Barret et al. (2016). The close agreement of both methods shows that from a climatological point of view the two criteria yield a very similar picture of the anticyclonic circulation and the related trace gas confinement. Locally and at particular dates, however, differences may be larger with potential vorticity correlating better with confined trace gas anomalies than geopotential height (e.g. Garny and Randel, 2013; Ploeger et al., 2015). The spatial pattern of low ozone concentrations in the monsoon anticyclone is well simulated in the model. It is in good agreement with MLS (90–140 ppbv), MIPAS (80–120 ppbv), and SAGE II (< 150 ppbv) measurements (Kunze et al., 2010; Randel et al., 2001; Randel and Park, 2006; Park et al., 2007).

Vertical profiles of ozonesonde measurements (averaged for the monsoon season during 2001–2009) at Indian stations, Delhi (28.61∘ N, 77.23∘ E), Pune (18.52∘ N, 73.85∘ E), and Thiruvananthapuram (8.48∘ N, 76.95∘ E) are compared with MLS measurements and ECHAM5-HAMMOZ simulated ozone mixing ratios in Fig. 1c–e. ECHAM5-HAMMOZ simulations show good agreement with MLS data between 200 and 50 hPa at all three stations. Comparison of ozonesonde observations with the ECHAM5-HAMMOZ simulation shows reasonably good agreement at Pune compared to Delhi and Thiruvananthapuram, where there are some discrepancies. The simulated ozone mixing ratios are lower than ozonesonde measurements by 10–40 ppb between 500 and 90 hPa at Pune and by ∼ 70–90 ppb in the upper troposphere (500–150 hPa) at Delhi. At Thiruvananthapuram, while at altitudes below 375 hPa simulated ozone mixing ratios show good agreement with ozonesonde data, at the altitudes above 375 hPa simulated values are lower than observations by ∼ 20–70 ppb. The differences between model and ozonesonde data may be due to different grid sizes: the ECHAM5-HAMMOZ model grid size is ∼ 280 km, while balloon observations are within ∼ 30–180 km spatial range (balloon typically drifts ∼ 30–180 km horizontally). In addition, previous work comparing these model simulations with various aircraft observations during the monsoon season found a reasonable agreement for PAN, NOx, HNO3, and O3 mixing ratios (Fadnavis et al., 2014).

Recent satellite observations and model simulations demonstrated the impact of convective transport of boundary layer pollution into the ASM anticyclone during the ASM season (Gettelman et al., 2004; Randel et al., 2010; Fadnavis et al., 2013, 2014, 2015). These pollutants are further transported across the tropopause as evident in satellite observations of e.g. water vapour (Bian et al., 2012), HCN (Randel et al., 2010), CO (Schoeberl et al., 2006), PAN (Fadnavis et al., 2014, 2015), and aerosols (Vernier et al., 2015; Fadnavis et al., 2013). To understand the influence of monsoon convection on the vertical distribution of NOx we show zonal and meridional cross sections over India and China. Vertical distributions of NOx averaged for the monsoon season over Indian latitudes (8–35∘ N) and Chinese latitudes (20–45∘ N) as obtained from CTRL simulations are shown in Fig. S3a and b respectively. These figures show elevated levels of NOx extending from the surface to the upper troposphere over India and China. The wind vectors along with the distribution of cloud droplet number concentration (CDNC) and ice crystal number concentration (ICNC) (Fig. S4a–c) indicate strong convective transport from the Bay of Bengal (BOB), South China Sea, and southern slopes of Himalayas, which might lift the boundary layer NOx to the upper troposphere.

During the monsoon season, the NOx distribution in the UTLS is also influenced by lightning, in addition to transport from anthropogenic sources. Lightning activity during this season was found to be more pronounced in Asia, compared to the other monsoon regions such as North America, South America, and Africa (Ranalkar and Chaudhari, 2009; Penki and Kamra, 2013). In our simulations, we find that lightning produces 40–70 % of NOx over northern India and BOB and 40–60 % over the Tibetan Plateau and western China region (Fig. S5).

Figure 2 shows the vertical distribution of anthropogenic NOx anomalies obtained from the Ind38, Ind73, and Chin73 simulations, compared with the CTRL simulation. Ind38 simulation shows that the convective winds over the BOB (80–90∘ E) (Fig. 2a) and at the southern flank of the Himalayas (Fig. 2d) lift up the enhanced Indian NOx emissions to the UT. Similarly the Chin73 simulation shows that the convective winds over the South China Sea (100–120∘ E) (Fig. 2c) and over the Himalayas (Fig. 2f) lift up the enhanced Chinese NOx emissions to the UT. While most transport is mainly into the UT, parts of it also occur into the lower stratosphere, with cross-tropopause transport being particularly evident in the Chin73 simulation (Fig. 2c and f). Randel and Park (2006) and Randel et al. (2010) also reported that pollution transported by Asian monsoon convection enters the stratosphere. Our results are also in good agreement with previous studies indicating significant vertical transport due to strong monsoon convection from the southern slopes of Himalayas (Fu et al., 2006; Fadnavis et al., 2013, 2015) and the South China Sea (Park et al., 2009; Chen et al., 2012). In the upper troposphere, NOx is transported over Iran and Saudi Arabia along the descending branch of the large scale monsoon circulation (Rodwell and Hoskins, 1995). However, the cross-tropopause transport is not present in the Ind73 simulation, where it is inhibited by the wind anomalies that show a descending branch over central India (∼ 20∘ N, 75∘ E) (Fig. 2b and e). These descending wind anomalies may also be related to the associated ozone radiative forcing and temperature changes, as discussed in Sect. 4.

We calculate the change in ozone production over India and China due to enhanced NOx emissions in the Ind38, Ind73, and Chin73 simulations with respect to the CTRL simulation. Figure 3, showing longitude–pressure cross sections of net ozone production (ppt day-1) changes, indicates that the majority of this additional ozone production occurs in the lower troposphere. At altitudes below 300 hPa, the ozone production and loss vary between -15 and 15 ppt day-1. In the upper troposphere (300–150 hPa), the estimated amount of additional net ozone production in Ind38 and Ind73 simulation is 3–7 ppt day-1, while in the Chin73 simulation it is ∼ 3–13 ppt day-1. We also simulate ozone loss near the tropopause in the Ind73 simulation (Fig. 3b). We note that these ozone anomalies are not driven by lightning NOx, as this is included in all simulations. It is interesting to understand ozone production over the highly populated Indo-Gangetic Plain and Tibetan Plateau region (these regions are marked in Fig. S4). A longitude–pressure cross section over this region show that ozone production over the Indo-Gangetic Plain and Tibetan Plateau in Ind73 is (20–25 ppt day-1) is much larger than Ind38 (6–20 ppt day-1) in the lower troposphere (Fig. S6).

Figure 4 shows the vertical distribution of ozone anomalies induced by enhanced anthropogenic NOx emissions in the three perturbation experiments compared to the CTRL simulation, averaged over India and China. Although the air mass in the monsoon anticyclone is relatively poor in ozone (Fig. 1b), the elevated amounts of ozone anomalies in response to enhanced anthropogenic NOx emissions are clearly seen in Fig. 4. This may be partially due to convective transport of enhanced-NOx-emission-induced ozone anomalies produced in the lower troposphere and partially due to chemical ozone production from convectively transported boundary layer ozone precursors. Ozone anomalies are enhanced near 300–200 hPa over western Asia (40–60∘ E) (Fig. 4a–c), possibly due to the vertical convective transport of ozone anomalies and precursors and also from subsequent horizontal transport in the monsoon anticyclone (Barret et al., 2016).

Latitude–pressure cross sections of enhanced-NOx-emission-induced ozone anomalies plotted in Fig. 4d and f illustrate how convection over the BOB, the southern slopes of the Himalayas and the South China Sea lifts the enhanced ozone anomalies from India and China into the upper troposphere. These ozone anomalies are also transported further across the tropopause and into the lower stratosphere, where ozone production is also driven by photolysis and NOx anomalies.

In the Ind73 simulation, similarly to the NOx anomaly distribution (Fig. 2b and e), the descending branch of circulation over central India also suppresses the vertical transport of ozone anomalies across the tropopause (Fig. 4b and e). This subsidence may be related to ozone heating rate changes, as there is significant increase in ozone production over the Indo-Gangetic Plain and Tibetan Plateau in the lower troposphere due to enhanced anthropogenic NOx emissions (Sect. 4).

The distributions of NOx and ozone anomalies in the monsoon anticyclone region in the Ind38, Ind73, and Chin73 simulations with respect to the CTRL simulation are shown in Fig. 5a–f. A maximum in the NOx anomalies in the ASM anticyclone (60 to 120∘ E) is seen in all the simulations. NOx anomalies are high at the eastern part of the monsoon anticyclone since convective injection into the anticyclone occurs mainly in that region (Fadnavis et al., 2013). Increase in NOx anomalies in the Ind38 simulation is higher (Fig. 5a) than that in the Ind73 simulation (Fig. 5b), mainly due to descending motion over central India in the Ind73 simulation, as seen in the previous sections. In contrast to NOx anomalies, ozone anomalies in Ind38 are lower than Ind73, especially in the north-eastern part of anticyclone. Satellite observations also show high ozone precursors and low ozone amounts in the anticyclone (Park et al., 2007; Barret et al., 2016). Similarly, the Chin73 simulation shows higher values of NOx anomalies (> 18 %) and strong negative ozone anomalies (∼ -8 %) in the north-eastern region of the monsoon anticyclone (Fig. 5c and f). Figure 5 also shows that the tropical easterly jet transports NOx and ozone (from India and China) to Saudi Arabia, Iran, and Iraq.