The hourly output of WRF-Chem simulations has been spatially and temporally interpolated along the CARIBIC flight tracks. The observed and model simulated profiles have been averaged into vertical bins of 50 hPa for the comparison analysis. The comparison of O3 and CO profiles from CARIBIC measurements with standard WRF-Chem simulations (Std) is shown in Fig. . Here we only show the profiles collected during the descent of the aircraft as these have complete coverage until about 800 hPa, while the measurements start from about 600 hPa upwards in the ascending profiles. However for the analysis of model biases, all the ascending and descending profiles have been averaged to calculate the monthly profiles (Fig. ).

Higher levels of ozone and carbon monoxide occur in the lower troposphere (LT: 850–600 hPa) and Upper Troposphere (UT: above 300 hPa), while lower levels in the Middle Troposphere (MT: 600–400 hPa) cause a typical C-shape structure during July. This feature is suggested to be associated with the monsoonal convective uplifting of the lower tropospheric pollution and is captured by WRF-Chem.

Despite the qualitative agreement of the vertical distributions of O3 and CO, significant differences occur between model and measurements, particularly in lower tropospheric CO. For example, on 19 June the observational CO levels vary from 91.5±3.9 to 104.4±0.6 nmolmol-1 in the LT, whereas WRF-Chem simulated CO levels are significantly lower (75.4±1.0 to 85.8±0.7 nmolmol-1). The average underestimation (Mean Bias) of CO in the LT is found to be 12.6±4.4, 22.8±12.6 and 19.9±7.5 nmolmol-1 during June, July and August respectively, as calculated from all the ascent and descent profiles averaged for a month (Fig. ). WRF-Chem simulated average CO shows very good agreement with CARIBIC measurements during September in the LT (MB = -0.1±4.2 nmolmol-1).

The model underestimates a pollution event of strongly elevated ozone observed on 15 July 2008 (146.4±12.8 nmolmol-1 at 810 hPa). In contrast to CO which is typically underestimated in LT, the bias in model simulated O3 varies from an overestimation by 4.3±1.8 during June to an underestimation by 7.8±1.6 nmolmol-1 during August, except during the strong pollution event (-71.5± 25.9 nmolmol-1). The significantly higher levels of O3 (146.4±12.8 nmolmol-1) and CO (136.4±12.2 nmolmol-1) as observed during July are from two observational profiles on the same day (15 July), discussed separately as an event of strong pollution.

For the complete profiles from Standard WRF-Chem simulations (Std), the Root Mean Square Deviation (RMSD) values for O3 are found to vary from 6.5 to 12.6 nmolmol-1, except during a strong pollution event (RMSD = 48.1 nmolmol-1). RMSD values for CO are in the range of 5.5 to 18.2 nmolmol-1. Additional simulation Std_INTEX using a different emission inventory INTEX-B also shows similar results (Figs. , S2), as seen with simulation Std using HTAP emissions. The average vertical distribution of the water vapor mixing ratios from WRF-Chem is compared with the CARIBIC measurements in Fig. . Generally, WRF-Chem simulated H2O is in very good agreement with the observations, i.e., within the variability of 1 standard deviation. The observations are not available below 500 hPa in months other than during July, when the model tends to overestimate H2O in the lower troposphere.

WRF-chem simulated ozone profiles are compared with the monsoon-time climatology obtained from ozonesonde observations at Delhi, Pune and Thiruvananthapuram (Fig. ), as described in Sect. . WRF-Chem simulated ozone profiles in the lower and middle troposphere are generally observed to be within the 1 standard deviation variability of the observational climatology over the three stations. However, in the upper troposphere, WRF-Chem overestimates ozone mixing ratios over Delhi and Pune. The mean biases of the WRF-Chem are estimated against average ozonesonde climatology in summer monsoon in the LT (850–650 hPa) as calculated against CARIBIC observations in Sect. . MB in the LT are found to be lower at Delhi (-2.2±3.8 nmolmol-1) and Pune (-1.2±3.6 nmolmol-1), as compared to that over Thiruvananthapuram (-12.4±1.3 nmolmol-1). However, in the UT (e.g. at 150 hPa) ozone mixing ratios in WRF-Chem simulations at Delhi (94.1±31.1 nmolmol-1) and Pune (69.4±23.5 nmolmol-1) are found to be higher as compared to ozonesonde observations (61.1±34.0 and 31.3±17.5 nmolmol-1 respectively). The overestimation in upper troposphere by WRF-Chem has been reported earlier with a slightly different model setup (different convective parameterization ).

Figure  shows the monthly average CO profiles from simulation Std and the CO retrievals obtained from MOPITT over Chennai. For consistency with the comparison with CARIBIC observations (Sect. ), which are collected only during nighttime, we restrict the comparison of WRF-Chem and MOPITT to nighttime data, though we do not find large diel variability in free tropospheric CO in our simulations. The averaging kernel and the a priori profiles of MOPITT data have been applied on the monthly average CO profile from standard WRF-Chem simulation, denoted as Std(AK).

In contrast to the comparison with the in situ vertical profiles from CARIBIC, the WRF-Chem simulated CO shows very good agreement with the satellite data in the lower troposphere during June. The mean bias value between WRF-Chem and MOPITT is found to be 1.5±0.8 nmolmol-1 in the LT during June as compared to the WRF-Chem and CARIBIC data comparison (-12.6±4.4 nmolmol-1). Interestingly, in comparison to the satellite data, WRF-Chem is found to overestimate CO in the LT by 21.4±2.8, 37.8±5.0, and 26.9±4.0 nmolmol-1 during July, August and September respectively. Middle tropospheric CO is also significantly overestimated by WRF-Chem as compared to MOPITT during July–September. This could be partially associated with the unscreened-out cloud contamination in the satellite retrievals during the summer monsoon season. The a priori CO data from the global chemistry transport model could be another potential source of the discrepancy .

WRF-Chem profiles, after applying the satellite operator become very similar to the satellite a priori, especially in the lower and middle troposphere. During this period, averaging kernels in the lower troposphere are found be smaller (less than 0.1) as compared to the values reported for example during spring . This indicates relatively lower sensitivity of MOPITT for lower tropospheric CO over this region during the summer monsoon. The different results regarding the WRF-Chem evaluation against the in situ measurements and satellite data clearly highlight the need of more in situ measurements of vertical profiles for validation of chemistry-transport models as well as the satellite retrievals over this region, particularly during the monsoon, when the sky is obscured by clouds. Such studies would be invaluable for addressing the discrepancies due to limited overpassing time for MOPITT, retrieval errors due to sensor degradation, not updated CO a priori, cloud-contamination, systematic errors as well as errors in model simulations.

In order to understand if the observed discrepancies between WRF-Chem and CARIBIC observations are associated with emissions and processes at the surface in India, we analyze the variations in surface ozone and CO over this region. WRF-Chem (Std) simulated average distributions of surface O3 and CO over the Indian region are shown in Fig.  for the 4 months of the summer monsoon in 2008. The distribution of O3 as well as CO shows large spatial heterogeneity across the region in all 4 months.

Surface ozone levels are typically lower (<30 nmolmol-1) than aloft over most of the domain. The ozone levels are found to be highest over the polluted Indo-Gangetic Plain (IGP), in northeastern India, and also over the eastern coastal region (40–50 nmolmol-1). Average surface ozone levels over most of the Indian region are relatively low, mostly below 40 nmolmol-1, which is mainly due to the inflow of marine air masses and suppressed photochemistry in cloudy and rainyconditions. The highest levels of surface ozone are simulated over the northern part of the domain, where the influences of marine air/monsoon are relatively smallest. While vertical trace gas distributions are affected by monsoon convection, both CO and O3 are not soluble and not directly affected by precipitation scavenging. Wet scavenging of O3 precursors and prevailing cloudy-rainy meteorological conditions, however, could suppress the ozone production, particularly near the surface. CO mixing ratios vary from about 50 to 300 nmolmol-1, except over the IGP where a high CO belt (400 nmolmol-1 and more) accumulates throughout the monsoon season. Towards the end of the monsoon period in September, ozone and CO levels show most pronounced enhancements over the IGP and also a tendency of pollution buildup in the surrounding regions.

The WRF-Chem simulated spatial distributions of surface ozone and CO are found to be consistent with previous studies over the Indian region mostly based on satellite observations , simulations with a global chemistry transport model and a previous study evaluating WRF-Chem simulations over the Indian region . WRF-Chem simulations were found to significantly overestimate surface O3 and underestimate CO at an urban site in the Indo-Gangetic Plain towards the onset of monsoon, while the model was in better agreement during May .

Figure  shows a comparison of surface ozone and CO variations from WRF-Chem with ground-based observations. Unfortunately simultaneous measurements of ozone and CO are sparse over this region and therefore observations of ozone are utilized from Gadanki (79.2∘ E, 13.5∘ N) , a rural site in southern India and observations of CO are used from the coastal site Cape Rama (73.8∘ E 15.1∘ N; ). O3 and CO model results from the Std simulation are found to be within the 1σ standard deviation of the measurements at Gadanki and Cape Rama.

The significant underestimation of CO by WRF-Chem in the free troposphere (Sect. ) as compared to CARIBIC measurements is not evident at the surface. It is suggested that the discrepancies between WRF-Chem and CARIBIC observations are likely not caused directly by surface emissions and chemistry and may be associated with the influence of large-scale air mass transports. We further investigate this by conducting a sensitivity simulation with 50 % higher CO emissions (Sect. ) over the Indian region. The possible role of transport is investigated by backward air trajectory analysis and conducting a sensitivity run with 25 % higher influx of CO from the domain boundary based on trajectories (Sect. ).

We begin by examining a pollution event observed on 15 July 2008 over Chennai to investigate its origin. Ozone and carbon monoxide levels were observed to be very high during the month of July but are substantially underestimated by WRF-Chem (Figs.  and ). Since there were only two observational profiles during July and both on the same day (15 July 2008), this observation is suggested to be more representative of a pollution event rather than the monthly average conditions over this region. During this event, O3 (146.4±12.8 nmolmol-1) and CO (136.4±12.2 nmolmol-1) mixing ratios are found to be very high in the lower troposphere (∼805 hPa), indicating that these concentrations are associated with the transport of polluted air with ample time for photochemical ozone build up, while significant influence of transport of ozone-rich air from the stratosphere is unlikely. It is found that the residence time of this air mass is more than 3 days over southern India during this event, much longer than during CARIBIC flight times in other months. Moreover, the air masses are found to be influenced by boundary layer pollution as indicated by significantly higher mean pressure along the trajectory (915±43 hPa). To investigate the underprediction of the event in the model, we analyzed the wind fields over Chennai from radiosonde measurements. The model is found to generally reproduce the variations in wind speed over Chennai at different altitude levels (e.g., in the range of 4–10 m s-1 at 980 hPa; Figs. S3, S4). However, the model does not capture the occurrences of low-wind speed (1–3 m s-1) and overestimates systematically the wind speed during the July period. Therefore the air parcels could not possibly collect enough pollutants from the boundary layer leading to the underprediction of the strong pollution event in model. Additionally, no indication of underestimation of emissions is found as the model performance did not improve in reproducing the event when emissions were increased by 50 %.

Long-range transport of pollution in regional models is controlled by the chemical boundary conditions, generally provided from a global model. Previous studies investigated the impact and uncertainties in long-range transport in regional model simulations. In WRF-Chem simulations in this study the long-range transport is controlled by the time varying chemical boundary conditions from a global model MOZART/GEOS5 simulations.

We assess the contribution of long-range transport of CO in the lower troposphere over Chennai by conducting a sensitivity simulation with increased CO at the domain boundary. Backward air trajectories suggest that CO is significantly underestimated in the lower troposphere in westerly air masses (Figs.  and ). Therefore, we increase the CO mixing ratios by 25% in the MOZART/GEOS5 data, over a region (7.5∘ N < lat <16.5∘ N) on the western boundary as shown in Fig. , chosen suitably based on the backward trajectories (Fig. ).

Figure  shows a comparison of average CO profiles from CARIBIC measurements, WRF Chem standard simulations (Std) and the sensitivity simulation with increased CO at the western boundary (1.25×_BDY). In contrast to the sensitivity run with increased emissions (1.5×_EM), here we find significant improvement in the WRF-Chem simulated CO in the free troposphere. For example during June, WRF-Chem simulated CO mixing ratios from 1.25×_BDY simulation (95.8±4.3 nmolmol-1) are comparable to the observations (96.6±9.1 nmolmol-1) at ∼800 hPa. The improvements are also significant in other months in the lower free troposphere. In contrast to the June–August period, the air masses over Chennai show influences of higher emissions on the free troposphere. This could be associated with the transport from the continental Indian region as shown by backward trajectories (Fig. ). The enhancements due to higher CO in the boundary conditions are significantly less during September as compared to June–August. We suggest that since air masses over Chennai during September are more influenced by the regional emissions, the influence of uncertainty in boundary conditions is not evident here. Further it is noted that such dominance of regional impacts on CO vertical distributions during September is captured better by WRF-Chem, as compared to the global model simulation (Fig. S5).

This study suggests that anticyclonic advection plays a very important role which could transport polluted air masses from outside the region (domain) during the summer monsoon. This complements conventional thinking that convected regional emissions dominate the tropospheric composition during the monsoon season and points to a potentially significant external source of pollution to the monsoon anticyclone. We show that this transport is generally very fast, i.e., the residence time of air masses is 1–2 days over southern India, except during the strong pollution event (Sect. ). This rapid transport could advect CO-rich air masses from more strongly polluted upwind regions. As indicated by the backward air trajectories and a sensitivity run, CO-rich air masses could originate in central Africa and the Persian Gulf region. During the summer monsoon, CO mixing ratios have been found to be highest over central Africa associated with biomass burning emissions and references therein. A recent study utilizing the trajectory-mapping technique and aircraft observations also indicated elevated CO mixing ratios over the western boundary of our model domain. We suggest that improvements in the global fire emissions input to the global models and data assimilation would be helpful in better constraining the effects of long-range transport during the monsoon. Regional emissions from continental India are shown to significantly influence the free troposphere over southern India towards the end of the monsoon (September).