Introduction

The amount of greenhouse gases (GHGs) in the atmosphere such as CO2 and N2O has increased worldwide because of anthropogenic emissions. Atmospheric carbon dioxide (CO2) increased substantially since the beginning of the industrial era (by more than 45% since about 1750). In particular, the rapid increase of anthropogenic CO2 emissions in South Asia contributes strongly to the acceleration of its growth rate, e.g., the anthropogenic CO2 emission rate from India was the fourth highest worldwide in 2017 (behind China, the USA and the European Union); For example India’s fossil CO2 emissions increased by +5.1% yr−1 for the last decade (2009–2018) compared to a global increase of +1.3% yr−11.

Global human-induced emissions of N2O which are dominated by contributions from the agricultural sector i.e., by the use of nitrogen-based fertilisers increased by 30% between 1980 and 2016 in particular in agriculture-oriented economies such as India and China2.

Due to the sparse availability of regional atmospheric observations over the Indian subcontinent there is currently a lack of sufficient coverage of continuous quality-controlled ground-based monitoring stations of GHGs such as CO2 and N2O. Continuous ground-based measurements of CO2 and N2O are essential to derive the spatial and temporal variations of CO2 and N2O emissions, for providing adequate boundary conditions for model simulations, and to infer long-term trends of their emissions3,4,5,6,7,8,9.

Our study shows the great impact that an expansion of this network with continuous measurements would have. This would allow for realistic 3D simulations of CO2 in the South Asian atmosphere and provides added value to atmospheric and climate modelling as well as to satellite-based CO2 monitoring.

Methods to constrain CO2 surface-atmosphere fluxes comprise a variety of bottom-up and top-down approaches10. The latter, also called inverse approaches employ a transport model with a priori fluxes that are adjusted so that simulated concentrations best fit observations11,12. Such approaches are frequently hampered by the limited temporal and spatial availability of ground based, airborne or space-borne observations.

In state-of-the-art chemistry transport models, the transport of air parcels differs because different methods (Eulerian, Lagrangian), different vertical velocities (kinematic, diabatic) and different meteorological reanalyses (e.g., ERA5, ERA-Interim, JRA-55) are used to drive the models13,14. Further, the implementation of convection and irreversible mixing differs from model to model.

Here we use a unique set of CO2 and N2O aircraft measurements at high temporal and vertical resolution up to ~ 20 km altitude (corresponding to ~ 55 hPa or ~ 475 K potential temperature) obtained in the Asian summer monsoon where CO2 and N2O airborne measurements were hitherto only available up to ~ 12 km ( ~ 180 hPa)15,16,17,18.

From about June to September, the Asian summer monsoon constitutes a seasonally persistent zonally restricted circulation pattern transporting climate-relevant emissions rapidly from the surface boundary layer to greater altitudes, i.e., to the lower stratosphere19,20,21,22,23. The Asian summer monsoon is associated with deep convection over the Indian subcontinent and an anticyclonic flow in the upper troposphere and lower stratosphere (UTLS) over the Asian monsoon region spanning from northeast Africa to the Pacific21. Air parcels are uplifted quickly by convection followed by slow diabatic uplift in the UTLS superimposed by the anticyclonic flow, while in other regions within the tropical transition layer (TTL) the heating rates are in general smaller during boreal summer23. Further, the thermal tropopause as well as isentropes (in log-pressure altitude coordinates) are enhanced in the region of the Asian monsoon anticyclone compared to the residual TTL24. The higher the air parcels are located above the level of maximum convective outflow ( ≈ 360 K ≈ 13 km), the larger the contribution of air masses is from outside the Asian monsoon anticyclone (i.e., from the stratospheric background) to the upward spiraling flow23.

We demonstrate that the combination of ground-based observations on the Indian Subcontinent (in particular at Nainital) and Lagrangian transport modelling provides realistic 3D CO2 distributions in the Asian Monsoon region up to 20 km altitude. This, in turn, is a prerequisite for a realistic representation of associated radiative effects in atmospheric and climate models and a solid foundation for satellite-based monitoring of CO2 fluxes based on inverse modelling approaches.

Results

Measurements

In general ground-based measurements of CO2 on the Indian subcontinent reflect the seasonality of carbon exchange in the northern terrestrial biosphere, which is mostly related to the seasonality of the vegetation activity by photosynthetic CO2 absorption by plants in this latitude range5,9. However measured CO2 values depend in detail strongly on local natural sources and sinks as well as—even more important—on anthropogenic emissions such as combustion of fossil fuels, land use change and biomass burning5,7,8,9,25.

The net CO2 uptake by plants is the net balance of photosynthesis and respiration; photosynthesis dominates over respiration in summer, but only during daytime and respiration dominates in winter and during nighttime. During the pre-monsoon period (March–May) a seasonal CO2 maximum and during the monsoon period (June–September) a seasonal minimum is found at different stations on the Indian subcontinent5,8,9,26 resulting in a negative phase (decreasing concentration of CO2) during April–August and a positive phase (increasing concentration of CO2) during September–March.

Figure 1 shows the seasonal variability of ground-based CO2 and N2O measurements at different sites during 2016 and 2017 (details see Methods; geographical positions are shown in Fig. 2). Stations in the northern hemisphere such as Nainital (India) and Mt. Waliguan (China) show one clear seasonal maximum in March–May and one seasonal minimum in June–September. CO2 in Comilla (Bangladesh) has two seasonal minima per year, in February–March and September, corresponding to crop cultivation activities that depend on regional climatic conditions in contrast to Nainital that only has one clear minimum in September9. Air masses over the Indian subcontinent were transported from the Indian Ocean region during summer (monsoon season) and from the inland during winter, therefore observations in Nainital are strongly affected by anthropogenic emissions from the Indo-Gangetic Plain during summer. Anthropogenic emissions, e.g., of CO2, in the Indo-Gangetic Plain are higher compared to other regions in India9 caused by the dense concentration of industries (e.g., thermal power plants, steel plants, refineries) as well as by the very high population density in this area27. Thus air masses transported long-range from the south to Nainital, can uptake these emission while passing over the Indo-Gangetic Plain9.

Fig. 1: Temporal variability of ground-based CO2 and N2O.
figure 1

The variability of ground-based CO2 a, b and N2O c is shown at different sites in Asia and the Pacific for 2016-2107 (details see Methods). Ground-based CO2 measurements from different continental stations in Asia a and from different stations in the Inter-Tropical Convergence Zone b are compared (geographical positions see Fig. 2). In addition, the seasonal variability of CO2 over the northern Indian subcontinent (mean value between 20–30°N and 75–95°E) of the lowest model level at 975 hPa of the GOSAT-L4B product (details see Methods) for comparison to ground-based CO2 measurements is shown. The pre-monsoon period (March–May) when a seasonal CO2 maximum is expected is high-lighted (light-grey) as well as the period of the StratoClim aircraft campaign during monsoon 2017 (dark-grey).

Fig. 2: Regional map of the measurement sites for greenhouse gases and of the aircraft measurements on the Indian subcontinent.
figure 2

The locations of the measurement sites for greenhouse gases in Nainital (NTL, India) Comilla (CLA, Bangladesh), Mt. Waliguan (WLG, China), Bukit Kototabang (BKT, Indonesia), Mauna Loa (MLO, Hawaii) and Samoa (SMO, Cape Matatula) a and the flight paths of the eight local scientific flights (F01-F08) by the high altitude research aircraft Geophysica b are shown. The scientific flights were carried out every second day from Kathmandu (Nepal) between 27 July and 10 August 2017.

The CO2 maximum at Mauna Loa is measured about 4 weeks later compared to measurements at sites in continental Asia (Fig. 1). The seasonal cycle of CO2 at Mauna Loa (Hawaii) is representative for marine northern-hemispheric background air. CO2 observations at Cape Matutala (Samoa) are representative for southern-hemispheric background air and are shifted about half a year compared to the northern hemisphere due to seasons determining vegetation growing periods.

The seasonal variability of N2O on the northern Indian subcontinent is consistent with the application of nitrogen fertiliser, biomass burning and change in monsoonal/trade winds9,28. Therefore, the N2O mixing ratios in Nainital and Comilla are in general higher compared to sites in the Pacific (Mauna Loa, Cape Matutala; see Fig. 1c). Higher N2O values are found in Comilla located in the eastern Indo-Gangetic Plain compared to Nainital located in the western Indo-Gangetic Plain (Fig. 1c).

Column-averaged CO2 from satellite measurements show also a distinct seasonal cycle of CO2 over India with a negative phase from northern hemisphere late spring to summer and a positive phase from autumn to spring26,29. The seasonal variability of CO2 over the Indian subcontinent (mean value between 10–35°N and 65–95°E) at the ground (the lowest model level at 975 hPa) as estimated by the GOSAT-L4B product (details see Methods) is compared to ground-based CO2 measurements in Fig. 1a, b. The GOSAT-L4B product is a model simulation using CO2 surface fluxes inferred from column-averaged satellite measurements (details see Methods); the lowest model level of GOSAT-L4B is closest to the inferred CO2 surface fluxes and is not strongly influenced by the tracer transport of the underlying transport model. The GOSAT-L4B mean value has a similar seasonal variability as other CO2 ground-based measurements on the northern hemisphere (Mt. Waliguan, Nainital, Mauna Loa), however its amplitude is lower than for the ground-based measurements in Nainital demonstrating the limitations of GOSAT-L4B data compared to in situ measurements.

In the frame of the StratoClim project funded by the European Commission, a measurement campaign using the Russian Geophysica high altitude research aircraft was conducted in Kathmandu (Nepal) in summer 2017 (see Fig. 2) to measure a variety of trace gases and aerosol characteristics for the first time in the Asian monsoon anticyclone up to 20 km altitude (corresponding to ~ 55 hPa or ~ 475 K potential temperature)30. These StratoClim measurements constitute a unique data set to characterise major processes which dominate particle and trace gas transport from one of the most polluted regions of the world into the lower stratosphere.

High-resolution CO2 and N2O profiles measured in-situ (Fig. 3; details see Methods) reflect the seasonal variability of CO2 and N2O at ground level (see Fig. 1). Although CO2 has a strong diurnal cycle near the ground, CO2 concentrations are relatively independent from diurnal variations in the upper troposphere and lower stratosphere (UTLS). CO2 is chemically inert in the troposphere and stratosphere and can be used as an age tracer considering time periods of several months31,32,33.

Fig. 3: Airborne CO2 and N2O measurements from the StratoClim campaign in Kathmandu (Nepal) during July and August 2017.
figure 3

Each air parcel is coloured by the transport time from the model boundary layer (BL) to the time of measurements inferred by Lagrangian back-trajectory calculations. Air parcels located in the model BL as well as aged air (air located in the free atmosphere on 1 June 2016) are marked. The number of air parcels is determined by the different temporal resolution of the CO2 a and N2O b measurements (details see Methods). In addition, the mean WMO tropopause71 as well as the lowest and highest tropopause (grey dashed lines) over Kathmandu during the flight days are shown.

N2O is essentially inert in the troposphere and has no significant sinks at the surface of the Earth. The critical region for N2O loss is the tropical middle stratosphere (24–40 km)34 where destruction of N2O occurs via photolysis and reaction with excited atomic oxygen (O(1D)). The decrease of measured N2O profiles above 400 K potential temperature (Fig. 3) indicates mixing with older stratospheric air that has descended from higher altitudes35. The high-resolution CO2 and N2O vertical profiles up to 20 km altitude presented here yield a unique insight into their altitude dependency in the region of the Asian monsoon.

Both the seasonality of CO2 and its long lifetime make it an appropriate chemical tracer for a reconstruction along backward trajectories over several months because chemical processes can be ignored. Therefore, CO2 is very well suited to analyse in detail transport pathways, transport times and mixing in the Asian summer monsoon anticyclone and beyond using backward trajectory calculations over a simulation period of about one year.

Transport times and air mass origin

Trajectory calculations were performed based on the Chemical Lagrangian Model of the Stratosphere (CLaMS)36,37,38 (details see Methods). CLaMS diabatic backward trajectories driven by high resolution ERA5 reanalysis39 were started along the entire flight paths (every 1 sec) of all Geophysica flights to infer the transport time from the location of the measurement back to the time when the air parcel was released at the model boundary layer (BL; details see Methods). The trajectories are calculated back to 1 June 2016 and are analysed within different time periods to identify the source regions at the model BL depending on season (see Table 1). However, most air parcels encounter the model BL within a few months of backward transport (e.g., 64% of all trajectories reach the model BL during the monsoon season 2017).

Table 1 Time periods and age of air of considered seasons on Indian subcontinent.

As expected, simulated transport times increase with the altitude of sampled air parcels (Fig. 3). However, there is also a strong variability of transport times between individual air parcels at the same level of potential temperature indicating mixing of air masses of different transport times or of different ages.

For the CO2 reconstruction it is essential to determine the location where the back trajectories intersect the model BL so that they can be tagged with the closest ground-based measurement. Figure 4 shows the frequency distribution for different seasons of the locations where the air parcels were released at the model BL. Most air parcels were released at the model BL during monsoon 2017 (64%), pre-monsoon 2017 (14%), and winter 16/17 (6%). Minor fractions are from post-monsoon 2016 (3%) and monsoon 2016 (3%). In summary, 90% of the air parcels were released at the model BL after 1 June 2016, the other 10% is aged air.

Fig. 4: Frequency distribution (fd) of the air mass origins at the model boundary layer (BL).
figure 4

Frequency distribution (number of trajectories normalised by the total number of trajectories started along the flight path) of the locations where the air parcels were released at the model BL (see Methods). Trajectories driven by ERA5 reanalysis were started along the entire flight paths (every 1 sec) of all eight Geophysica flights. The frequency distributions are shown for different seasons (representing different ages of air; see Table 1): monsoon 2017 a (a zoom of Asia marked as grey box is shown right beside), pre-monsoon 2017 b, winter 16/17 c, post-monsoon 2016 d and monsoon 2016 e. The frequency distribution is calculated in longitude-latitude bins of 2.0° × 1.5°. The percentages indicate the fraction of air parcels released at the model BL within a certain season. In summary, 90% of the air parcels were released at the model BL after 1 June 2016, the other 10% originates from aged air. The patterns of the frequency distribution depend strongly on the considered season. Further, the locations of different ground-based measurement sites in Asia and the Pacific are shown (details see Table 2).

During monsoon 2017 most air parcels were released in the northern part of the Indian subcontinent, the Tibetan Plateau, Bay of Bengal and eastern China (Fig. 4a). A cluster of air parcels at the model BL is also found in the western Pacific caused by typhoon activity (details see ref. 30). During pre-monsoon 2017 the origins are shifted towards the tropics to the northern Inter-Tropical Convergence Zone (ITCZ) e.g., over the Indian Ocean and the western Pacific. For winter 16/17, the origins move further to the south to the southern Inter-Tropical Convergence Zone (ITCZ) mostly over the Warm Pool region, northern Australia and western Pacific. The contributions from post-monsoon 2016 and monsoon 2016 are minor. In summary, we show that the patterns of the frequency distribution depend strongly on the considered season and hence from the age of air.

Based on the frequency distribution shown in Fig. 4 and on the limited availability of CO2 ground-based measurements in the region of the Asian monsoon and in the tropics from 2016 to 2017 a regional mask was developed where different BL regions (Fig. 5) are defined. This regional mask allows the CO2 at the model BL to be prescribed depending on the BL region.

Fig. 5: Regional mask to reconstruction CO2.
figure 5

Regional mask to reconstruction CO2 using CO2 ground-based measurements at different sites in Asia and the Pacific. In each model boundary layer (BL) region (marked by different colours) CO2 is prescribed from one specific measurement site: tropical southern hemisphere (tSH) by Samoa (SMO), Indian subcontinent (India) by Nainital (NTL), Bangladesh (BGD) by Comilla (CLA), Tibetan Plateau (TIB) by Nainital (NTL), maritime northern hemisphere (mNH) by Mauna Loa (MLO), continental northern hemisphere (cNH) by Mt. Waliguan (WLG) and Warm Pool region (Wpool) by Bukit Kototabang (BKT).

To estimate which back-trajectory length is best for CO2 reconstruction the fractions of air released at the model BL are inferred depending on different time intervals adjusted to the seasons on the Indian subcontinent (see Table 1). For our approach to reconstruct CO2 profiles from ground-based measurements it is important to use backward trajectories with a high fraction of air from the model BL. Figure 6 shows the fraction of air from the model BL, splitted into the BL regions shown in Fig. 5. Further, fractions of the free atmosphere are indicated (Fig. 6). The fractions of air are accumulated back to starting times of different seasons: monsoon 2017 (a), pre-monsoon 2017 (b), winter 16/17 (c), post-monsoon 2016 (d), monsoon 2016 (e). The longer the trajectories the higher are the contributions from the model BL and the lower are the fractions from the free atmosphere.

Fig. 6: The fraction of air from the model boundary layer (BL) and the free atmosphere.
figure 6

The fraction from the model BL and from the free atmosphere is calculated from all backward trajectories started along the Geophysica flight tracks averaged in 2 K intervals and accumulated back to the start times of different seasons: monsoon 2017 a, pre-monsoon 2017 b, winter 16/17 c, post-monsoon 2016 d, monsoon 2016 e (detailed start times are listed in Table 1). In e, the fraction of air referred to as the free atmosphere corresponds to the fraction of `aged air' defined in Table 1. The fraction of air from the model BL is divided in the different BL regions as shown in Fig. 5.

Considering all air parcels released at the model BL after the beginning of winter 16/17 yields a fraction greater than 90% up to levels of potential temperature of 400 K (Fig. 6c). Thus air masses from three seasons, monsoon 2017, pre-monsoon 2017 and winter 16/17, have to be taken into account. Above 400 K mixing with older air masses successively occurred and the fraction from the model BL rapidly decreases. Between 440 K and 480 K 50% of the air is younger than 1 June 2016; the other half is aged air.

At core altitudes of the Asian monsoon between ~ 360 K and 410 K the main contributions are from BL regions of the Indian Subcontinent, Bangladesh, the Tibetan Plateau and adjacent regions on the continental and maritime northern hemisphere. At the top of the Asian monsoon anticyclone (above 420 K) the contribution from the free atmosphere (stratospheric background) is dominating. The longer the trajectories the more contributions from model BL regions from the tropical southern hemisphere, the Warm Pool region, and the maritime northern hemisphere play a role. After a simulation period of ~ 14 months (until 1 June 2016) the contributions from the tropical southern hemisphere and the maritime northern hemisphere are roughly equal in the lower stratosphere.

Sensitivity of CO2 reconstruction on observation sites

Reconstructed vertical CO2 profiles using CLaMS Lagrangian trajectory calculations are determined by CO2 prescribed at the model BL and by the transport of air parcels along the trajectories driven by ERA5 reanalysis and diabatic vertical velocities. To infer the impact of different ground-based measurements, of mixing of air from inside the Asian monsoon anticyclone with air from the (stratospheric) background as well as of the trajectory lengths different sensitivity studies (cases) are performed for CO2 reconstruction.

To analyse how the seasonal variability of different CO2 ground-based measurements is reflected in CO2 reconstruction, CO2 is reconstructed prescribing CO2 for all air parcels released at the model BL using one specific ground-based site ignoring the origin of air parcels at the model BL. All air parcels that were released after 1 June 2016 at the model BL are used and all air parcels from the free atmosphere (mainly stratospheric background) are not considered in this first case (case S1; see Supplementary Discussion (Fig. S1) for the impact of the trajectory length).

In Fig. 7, CO2 mixing ratios reconstructed in this way are shown as median of 1 K intervals for several measurement sites. The averaging of reconstructed CO2 in 1 K intervals reflects mixing of air masses originating in different locations in the model BL or having different transport times.

Fig. 7: CO2 and N2O airborne measurements and reconstructed CO2 and N2O (case S1).
figure 7

Vertical profiles of CO2 a and N2O b airborne measurements (research flights F01-F08; HAGAR; light grey) are reconstructed using ground-based measurements from different sites in Asia and the Pacific indicated by different colours (locations of the different sites are shown in Fig. 2). CO2 mixing ratios from ground-based measurements are prescribed at the time when each trajectory reach the model BL and subsequently passively transported along each trajectory to the location of the measurement (details see Methods). The reconstructed CO2 values are shown as median in 1 K intervals (based on trajectories started every one second along the flight path). The seasonal variability of CO2 ground-based measurements is visible in the vertical profile of reconstructed CO2. Trajectories that do not reach the model BL until the beginning of the monsoon season 2016 on 1 June 2016 (i.e., air from the free atmosphere; mainly from the stratospheric background) are not considered here.

The CO2 maximum in the UTLS is best reconstructed using Nainital measurements as expected because the majority of the air parcels originate at the BL region of the Indian subcontinent. In Fig. 6c, it is shown that up to 410 K mainly air parcels released at the model BL after 1 December 2016 contribute to measured CO2 profiles. Above 410 K, air masses from the (stratospheric) background and from the northern and southern Intertropical Convergence Zone (ITCZ) contribute strongly to the composition of air probed during StratoClim within the Asian monsoon anticyclone in July and August 2017 at its top and beyond (see Fig. 4). Contributions from India are minor above ~ 410 K (Fig. 6), therefore reconstructed CO2 at this altitude has to be prescribed by measurements from other regions (e.g., Mouna Loa, Samoa and Bukit Kototabang). A mixture of air from different origins and with different ages needs to be considered for a full CO2 reconstruction which will be discussed in the next Section.

N2O can be reconstructed in a similar way as CO2 using the ground-based measurements from Nainital, Comilla, Mauna Loa and Samoa. Because of the in 2016/17 low seasonal variability of N2O at the ground (Fig. 1c) compared to variability of vertical N2O profiles below 400 K (Fig. 3b), the reconstruction below 400 K results in a constant vertical profile (Fig. 7b). Because chemical loss of N2O in the stratosphere cannot be represented in the approach of back-trajectory reconstruction, the reconstructed and measured N2O profiles start to diverge between 400 K and 410 K indicating mixing with aged stratospheric air above this altitude.

CO2 reconstruction

For a reliable reconstruction of measured vertical CO2 profiles over the entire altitude range, both accurate back-trajectory calculations are required as well as precise CO2 concentrations at the ground. For the latter purpose, a regional mask was developed (case S2) where CO2 is prescribed in the model BL depending on different BL regions (Fig. 5).

Figure 8a shows reconstructed CO2 using a regional mask for back-trajectory calculations until 1 December 2016 neglecting the contributions from the free atmosphere (case S2a). The comparison with measured in situ CO2 profiles shows a very good agreement from the model BL up to ~ 410 K. Above ~ 410 K, the fraction of trajectories from the free atmosphere (mainly stratospheric background) has to be taken into account (Fig. 6c). Figure 8b shows reconstructed CO2 as in case S2a but using in addition GOSAT-L4B CO2 data for the fraction of air parcels from the free atmosphere (case S2b). Here, for each 1 K interval the median of all air parcels considering both the fraction from the model BL as well as from the free atmosphere is calculated. This approach allows the mixing of air at the top of the Asian monsoon anticyclone between air mass from the boundary layer and air from (stratospheric) background to be considered. Air from the boundary layer above 400 K originates mainly in the southern and northern ITCZ. Extreme low CO2 values from ground-based measurements in the Warm Pool region have to be taken into account to reconstruct CO2 in this altitude range.

Fig. 8: Reconstructed CO2 using back-trajectory calculations until 1 December 2016 compared to HAGAR CO2 airborne measurements.
figure 8

Reconstructed CO2 is shown as median calculated from all trajectories until 1 December 2016 in 1 K intervals for research Flights F01-F08. a Reconstructed using regional mask shown in Fig. 5 for the fraction of trajectories ending in the model BL (case S2a). b Reconstructed CO2 as in case S2a but using in addition GOSAT-L4B CO2 data for the fraction of trajectories ending in the free atmosphere, mainly from stratospheric background (case S2b). Bars indicate the range between the 25 and 75 percentile.

Thus at this altitude range, there is mixing of air from the model BL with air from the stratospheric background. Reconstructed CO2 from the model BL is higher than the measured CO2 profile and air from the stratospheric background has lower CO2 values (not shown here) than the measured CO2 profile. Thus only the mixing of these two different air masses allows measured CO2 profiles to be reconstructed accurately. Using this approach our findings yield a good overall agreement between measured and reconstructed CO2 profiles, however differences are found at potential temperature levels between 430 K and 470 K, but still within the range of the 25 and 75 percentile.

The sensitivity of the quality of the reconstruction of CO2 (case S2b) on the employed trajectory length was tested. They can be too short (and thus miss contributions from the model BL) or too long (resulting in higher uncertainties). The longer the back-trajectory calculations the higher the altitudes of the end points of the trajectories from the free atmosphere. Based on the latter trajectories CO2 is reconstructed from GOSAT-L4B data that are providing CO2 values up to 10 hPa. The longer the trajectories the more the altitudes of the end points exceeds the altitude of the pressure level of 10 hPa and the CO2 values are here extrapolated to higher pressure levels which increases the uncertainties of reconstructed CO2 (see Supplementary Discussion (Fig. S1) for a detailed discussion on this issue). We decided to show back-trajectories to 1 December 2016, because for this date up to 410 K reconstructed CO2 is determined solely by CO2 prescribed at the model BL and by the transport of air parcels along the back-trajectories. Here, the uncertainties regarding the CO2 extrapolation to higher pressure levels are negligible.

However, above 410 K the quality of the GOSAT-L4B needs also to be taken into account for an assessment of the quality of CO2 reconstruction. GOSAT-L4B data depend on CO2 fluxes at the Earth’s surface (GOSAT-L4A data), on model resolution as well as on vertical transport in the used atmospheric transport model (NIES-TM; details see Methods). In Fig. 1a, b, it is shown that GOSAT-L4B data of the lowest model level at 975 hPa over the Indian subcontinent during the seasonal maximum in March-May 2017 are somewhat lower than the ground-based measurements in Nainital. Therefore, reconstructing CO2 from GOSAT-L4B (lowest level) CO2 by CLaMS trajectories yield lower CO2 values compared to case S2 in the UTLS caused by the lower CO2 seasonal maximum at the ground; see Supplementary Discussion (Fig. S2) for a detailed discussion on this issue as well as a direct comparison of GOSAT-L4B CO2 data with HAGAR CO2.

The contribution of CH4 oxidation in the stratosphere is estimated to be much lower (0.09% at 470 K) than the variability of reconstructed CO2 in this altitude region, therefore, CO2 from CH4 oxidation is not considered in our approach.

Discussion/conclusions

Unique airborne measurements of the GHGs CO2 and N2O at altitudes of the Asian monsoon anticyclone are presented. High-resolution in situ CO2 profiles are observed up to 20 km altitude and are reconstructed by back-trajectory calculations. Below 410 K, reconstructed CO2 using CLaMS Lagrangian trajectory calculations is determined solely by CO2 prescribed at the model BL based on measurements and by the transport of air parcels along the trajectories using high-resolution ERA5 reanalysis and diabatic vertical velocities. Above 410 K, the agreement with in situ CO2 profiles is improved by taking into account the stratospheric background. Our findings show that the Lagrangian transport in CLaMS using diabatic vertical velocities and driven by the European Centre for Medium-Range Weather Forecasts’ new high-resolution reanalysis ERA5 is very well suited for CO2 reconstruction (see Supplementary Discussion Fig. S2 for further details on this issue) and could be applied to other CO2 aircraft observations.

The good agreement of reconstructed CO2 with in situ CO2 profiles speaks for the benefits of uninterrupted surface measurements in Nainital and Comilla. It implies that a greater number of continuous ground-based measurements of CO2 and also other GHGs in South Asia, in particular on the Indian subcontinent, would be a great asset for atmospheric and climate modelling and for improving estimates of regional-scale surface-atmosphere CO2 fluxes, which are needed to develop policies mitigating the continued growth of fossil fuel emissions40.

State-of-the art global inversion systems assimilating mostly ground-based in situ observations currently do not well constrain the annual net CO2 flux of South Asia, their estimates ranging from − 0.5 to + 0.4 Pg C/year41, which reflects the lack of CO2 observations in that region. For example, the current release of the Carbon-Tracker (CT2019B)42, though based on 460 time series datasets from around the world, does not include ground-based measurements from the Indian subcontinent after 2013 (when CO2 monitoring in Cape Rama ended). Thus not surprisingly, the CO2 distribution at the ground over South Asia during summer 2017 is not well represented in CarbonTracker (see Supplementary Discussion Fig. S3 for further discussion on this issue). As a consequence, when using CarbonTracker CO2 as the lower boundary condition, the vertical distribution of CO2 over South Asia during summer 2017 could not be well represented in 3-dimensional model simulations43 in contrast to our approach (see Supplementary Discussion Figs. S4 and S5 for reconstruction of each StratoClim research flight F01-F08).

Recent advances in space-based remote sensing have greatly increased spatial and temporal coverage of column-averaged CO2 and driven the development of inverse systems employing these satellite data, alone or in conjunction with in situ data44,45,46,47. In fact, GOSAT-L4B data is a simulation product based on CO2 fluxes derived from a joint inversion of column-averaged GOSAT and ground-based CO2 data. However, the GOSAT and OCO-2 satellites do not provide measurements in persistently cloudy regions, including South Asia during the monsoon season44,46. The need for improvements in this region is obvious from the comparison of GOSAT-L4B with CO2 ground-based measurements in Nainital and Comilla as well as with CO2 vertical profiles measured by the HAGAR instrument during the StratoClim campaign (Fig. S2). The misrepresentation of the observed 3D structure in the GOSAT-L4B and Carbon-Tracker simulation reflects the large void of data in the South Asia region and/or limitations in model transport, and thus casts doubt on the reliability of the derived regional CO2 fluxes.

Our study shows that during the Asian Monsoon spatio-temporal patterns of CO2 on the Indian Subcontinent driven by regional flux variations rapidly propagate to approximately 13 km with slower ascent above. Enhanced CO2 compared to the stratospheric background can be detected up to 20 km. However in the stratosphere, the fraction of air originating on the Indian subcontinent is low compared to contributions from the tropics and of aged air from the stratosphere. We suggest that the propagation of these signals from the surface to the stratosphere constitutes a stringent test for atmospheric transport simulations and thus the data presented here provide an unprecedented opportunity for CO2 inversion systems to critically evaluate model transport and assess the derived CO2 fluxes in South Asia. High-resolution CO2 profiles can further be used to study stratosphere-troposphere-exchange processes as well as the intra-seasonal variability during the Asian monsoon season.

South Asia is the most densely populated part of the world and here further increasing anthropogenic emissions are expected in the future due to the strong growth of Asian economies. We conclude, the quantification of CO2 and other GHG surface fluxes and their temporal changes would highly benefit from an expansion of the ground-based GHG measuring network in South Asia complemented by regular vertical CO2 soundings, which could be achieved at comparatively low cost by AirCore sampling and subsequent laboratory analysis (a method requiring only moderate instrumentation collecting air in a very long lightweight stainless-steel tube, usable on a variety of platforms including small balloons)48. This would also provide a solid base for policy action.

Methods

Ground-based measurements

Ground-based measurements of atmospheric mole fractions of CO2 and N2O at different observation sites in South Asia, in particular, on the Indian subcontinent, and from the western Pacific (locations of all sites are shown in Fig. 2) are used to reconstruct observed CO2 vertical profiles during the StratoClim campaign. Measurements of CO2 and N2O at Nainital (India) and Comilla (Bangladesh)9,49,50,51,52 were provided through the Global Environmental Database, Center for Global Environmental Research, National Institute for Environmental Research (NIES). Measurements at Mt. Waliguan (China)53, Bukit Kototabang (Indonesia)54, Mauna Loa (Hawaii)55, https://gml.noaa.gov/aftp/data/trace_gases/ and Samoa (Cape Matatula)55, https://gml.noaa.gov/aftp/data/trace_gases/ are provided by the World Data Centre for Greenhouse Gases (WDCGG) (https://gaw.kishou.go.jp). Further details about each site such as location, elevation and measurement method are summarised in Table 2.

Table 2 Observational sites of CO2 and N2O in South Asia and in the western Pacific.

CO2 and N2O data at different sites are provided on different time scales (daily, weekly or monthly). For CO2 and N2O reconstruction it is required to interpolate all ground-based measurements on a daily time grid over the time period from 2016 to 2017. To get a comparable variability of the interpolated data, they are smoothed with a boxcar average of a width of 30 days as shown in Fig. 1.

Airborne measurements

HAGAR56,57 is a multi-tracer in situ instrument operated by the University of Wuppertal. Apart from CO2 and N2O, it also provides simultaneous in situ measurements of CH4, CFC-12, CFC-11, H-1211, SF6, and H2. Except for CO2, which is measured at high time resolution (3 to 5 s) by non-dispersive infrared absorption (NDIR), all the other species were measured by gas chromatography with electron capture detection (GC/ECD) every 90 s. The instrument is calibrated every 7.5 min during flight with either of two standard gases, which are inter-calibrated in the laboratory with standards provided by NOAA GML. For StratoClim the accuracy of the measurements was estimated to be about 2 ppb for N2O and about 0.2 ppm for CO2.

HAGAR data were referenced to standards provided by NOAA and are based on the CO2 WMO X2007 scale and the N2O NOAA-2006 scale. The data can be converted to the current CO2 WMO X2019 and N2O NOAA-2006a scales using the following equations, which are based on reassigned standard values on the current scales:

$$X2019=1.00033* X2007+0.0467$$
(1)
$$2006a=0.99841* X2006+0.587$$
(2)

These conversions amount to small positive shifts of about 0.18 ppm for CO2 and 0.05 to 0.17 ppm for N2O.

GOSAT-L4B data product

CO2 mixing ratios are used from the Greenhouse gases observing satellite (GOSAT) (http://www.gosat.nies.go.jp/en/about_5_products.html) launched 2009 by the Japan Aerospace Exploration Agency (JAXA)58 L4B data product to reconstruct measured CO2 profiles.

GOSAT observes infrared light using the Thermal And Near-infrared Sensor for carbon Observation (TANSO) instrument which consists of a Fourier Transform Spectrometer (FTS) and a Cloud and Aerosol Imager (CAI). The FTS measures the solar radiation reflected from the ground by sensors at three short-wave infrared bands (0.76, 1.6 and 2.0 μm) as well as the ground and atmospheric radiation at a wide thermal infrared band (5.5–14.3 μm). FTS has a instantaneous field of view determined by the nadir footprint size of 10.5 km in diameter. Absorption spectra are obtained from TANSO where interferences of clouds and aerosols are small and column abundances of CO2 over observation points can be calculated. Thus vertically integrated concentrations of CO2 were used to estimate sources and sinks of CO2, i.e., surface fluxes (GOSAT-L4A data product), by performing inverse simulations. Monthly fluxes of CO2 in GOSAT-L4A data are estimated for 42 terrestrial and 22 oceanic regions (64 regions total)44.

We use the GOSAT-L4B data product (V02.07) of atmospheric CO2 concentrations which has 17 vertical levels up to 10 hPa (in atmosphere hybrid sigma pressure coordinates), a horizontal resolution of 2.5° × 2. 5° and a time step of six hours driven by GOSAT-L4A surface fluxes58.

The GOSAT-L4B data product is a result of a global atmospheric tracer transport simulation using NIES atmospheric tracer transport model (NIES-TM v08.1i) using a 32-level hybrid isentropic grid59. Vertical velocities are calculated in isentropic coordinates above 350 K using a climatology of adiabatic heating/cooling rates derived from monthly mean values on pressure levels of the JRA-25 reanalysis60 provided by the Japan Meteorological Agency (JMA).

CLaMS trajectory calculations

Trajectory calculations were performed using the the Chemical Lagrangian Model of the Stratosphere (CLaMS)36,37,38 which was developed with the aim to study transport and chemical processes throughout the troposphere and stratosphere in the presence of strong tracer gradients. Here, CLaMS diabatic backward trajectories were started along the entire flight paths (every 1 sec) of all Geophysica flights conducted over the northeastern part of the Indian subcontinent. Overall ~ 110,000 back-trajectories are calculated between 9000 and 16000 per flight depending on the flight lengths.

In the CLaMS model, potential temperature is used as the vertical coordinate when the pressure is less than about 300 hPa, (i.e., in the upper troposphere and in the stratosphere); when the pressure is greater than about 300 hPa (more accurately, for pressure p exceeding a reference level of p/psurface = 0.3), a pressure-based orography-following hybrid coordinate (in units of K) is used38. Above about 300 hPa, the vertical velocity is determined solely by the total heating rate38,61. Total diabatic heating rates include clear-sky radiative heating, cloud radiation, latent heat release, as well as turbulent and diffusive heat transport for the upper troposphere and stratosphere are used from ERA5 reanalysis61. ERA5 reanalysis39 is a high-resolution atmospheric data set with 137 vertical levels up to 0.01 hPa, a horizontal resolution of ~  31 km (TL 639) and a hourly time resolution. We retrieved the data at 0.3° × 0. 3° horizontal grid. Caused by the much higher spatio-temporal resolution of ERA5 reanalysis39 compared to earlier reanalyses, a much better representation of convective updraft and tropical cyclones is realised62,63,64. Trajectories are considered ending in the model boundary layer, when they are located for the first time below about 2–3 km above surface considering orography (i.e., the vertical hybrid pressure–potential–temperature coordinate ζ≤ 120 K) in the paper referred to as ‘model boundary layer (BL)’details see e.g., 22,23. CLaMS trajectory calculations are very well suited to analyse the transport in the region of the Asian monsoon and were utilised using both ECMWF’ prior reanalysis ERA-Interim23,65,66,67,68,69,70 as well as the new ERA5 data set30,63.

CO2 reconstruction approach

CO2 mixing ratios from ground-based observations measured during the time when the CLaMS back-trajectories reach the model BL are used. For that the ground-based observations are interpolated on a daily grid. This calculated CO2 defines the reconstructed CO2 at the start point of the trajectory along the Geophysica flight path.

A regional mask was developed (setup S2) where CO2 is prescribed in the model BL depending on different geographical regions (see Fig. 5). In each of these geographical regions (in the paper referred to as ‘BL region’) CO2 is prescribed using one specific measurement site, e.g., trajectories ending in the BL region marked in green and dark-red (roughly Indian Subcontinent and Tibetan Plateau) are prescribed using ground-based measurements from Nainital and the BL region marked in yellow (roughly Bangladesh) is prescribed using ground-based measurements from Comilla. Unfortunately the coverage of ground-based measurements of CO2 over the Indian subcontinent in 2016 to 2017 is sparse, therefore only data from Nainital and Comilla are available.