The CONTRAIL project (http://www.cger.nies.go.jp/contrail/, last access: 7 October 2018) deploys two types of instruments on board aircraft: continuous CO2 measuring equipment (CME) and automatic air sampling equipment (ASE). We refer to Machida et al. (2008) for details, and only a brief description of the CME is given here. The CME measures CO2 mole fractions on board the aircraft using a non-dispersive infrared gas analyzer (NDIR; LI-840, LI-COR Biogeosciences). As of May 2018, installation of the CME is certified for eight Boeing 777-200ER and two Boeing 777-300ER aircraft of Japan Airlines (JAL). Once installed, the CME is operated automatically using the aircraft's flight navigation data until it is unloaded from the aircraft 2 months later. The CME samples air from the air conditioning system on the aircraft. The flow rate and the absolute pressure of the sample air in the NDIR cell are maintained at a constant level to minimize signal drift. The measured sample values are compared with two working standard gases (CO2 in air) in high-pressure cylinders (2 L) installed inside the CME and the measurements are traceable to the NIES 09 CO2 scale (National Institute for Environmental Studies). The mole fraction of CO2 in dry synthetic air in µmol mol-1 is reported in ppm in this paper. The latest results from the round robin intercomparison experiment show that the NIES-09 CO2 scale differs from the WMO-CO2-X2007 scale by less than 0.1 ppm (http://www.esrl.noaa.gov/gmd/ccgg/wmorr/wmorr_results.php, last access: 7 October 2018). The standard gases are currently introduced into the NDIR cell every 14 min during the ascent/descent portion of the flight and every 62 min during the constant altitude portion of the flight (cruise) typically at 8–12 km, i.e., during the ascent/descent and cruise measurement cycles, sample air is measured for 12 and 60 min, respectively, then standards 1 and 2 are measured for 1 min each. These standard gas intervals were initially 10 min during ascent/descent and 20 min during cruise until December 2005; the 20 min interval was then changed to 40 min until October–November 2011. The CME data are recorded as 10 s averaged measurements during ascent/descent (∼100 m intervals in altitude) and at 1-min intervals during cruise (∼15 km intervals horizontally). The data are rejected for 40 s after switching the gas stream and also when a standard deviation for the average period exceeds 3 ppm and when any failure in pressure or flow control is observed in the CME data record. To avoid heavy pollution around airports, the CME is not operated within 2000 ft (609.6 m) of the ground surface (this altitude was initially set to 1200 ft until March–June 2007). The overall analytical precision of the CME is estimated to be <0.2 ppm.

For the 10-year period from 2005 to 2015, we collected >7 million CO2 data points from >12 thousand flights all over the world. The CME measurements over the Asia-Pacific region are shown in Fig. 1. Flights from Japan to Southeast Asia (Bangkok, BKK; Singapore, SIN; and Jakarta, CGK) provide measurements over the East China Sea, the South China Sea, the Indochinese Peninsula, and the Maritime Continent. These measurement areas are substantially overlapped by flights to continental East Asia (Incheon, ICN; Shanghai, SHA; and Hong Kong, HKG) and to Taipei (TPE). Flights to Delhi (DEL) provide a unique opportunity for observations over continental Asia. In addition, extensive measurements from Japan to the north, to the east, and to the south are achieved from flights to Europe, to North America and Hawaii, and to Australia, respectively. The major airports where CONTRAIL CME measurements in Asia are made, along with the number of vertical profile measurements of CO2 over each airport, are listed in Table 1. Vertical profile data with less than 10 CO2 data points are not used in this study. As indicated in the table, the largest number of CO2 data has been obtained over the Tokyo Narita (NRT) airport with over 7000 vertical profiles, followed by Tokyo Haneda (HND) with over 3600 profiles. Figure 1b shows the number of monthly vertical profiles taken over the airports listed in Table 1. As seen in this figure, the CME measurements have acquired over 30 vertical profiles per month (colored red; i.e., at least one or multiple ascent/descent flights every day on average) over NRT and HND. Although measurements over other airports are less regular, data from sites where a substantial number of vertical profiles have been taken and cover much of the year are included in this study.

In this study, we focus on CO2 variations in the troposphere. Observations in the UT are, however, quite often influenced by stratospheric air that has distinct characteristics in atmospheric composition (e.g., Hoor et al., 2002; Sawa et al., 2004, 2008, 2015). These data are excluded from the dataset based on potential vorticity (PV) values. PV at the location and time of each CO2 measurement taken by CONTRAIL is calculated from the JCDAS (Onogi et al., 2007) and the JRA-55 (Kobayashi et al., 2015) reanalysis datasets (the latter being used since 2014), and any data accompanied by PV values of >2 PVU (1 PVU = 10-6 m2 s-1 K kg-1) are excluded. It has been found that the 2-PVU criteria is relatively robust in separating out the CO2 measurements in the UT that are stratospherically influenced from those that are not (Sawa et al., 2008, 2015). In total, 33 % of the CONTRAIL CME CO2 data points collected at altitudes >8 km have been identified as stratospheric, although this fraction varies with altitude, latitude, and season (i.e., flight routes). In this study, the UT is defined as the region at altitudes of >8 km and with PV of <2 PVU. Note that most commercial airliners cruise at altitudes of 9–12 km, and that this cruising altitude region is deemed in large part stratospheric at higher latitudes (e.g., 86 % and 64 % of the data taken at >40∘ N was stratospheric in January and July, respectively), whereas it mostly resides in the UT at lower latitudes (<10 % of the data at <30∘ N was stratospheric throughout the year).

To calculate climatological distributions of CO2 in the troposphere, we apply a method similar to Sweeney et al. (2015). (1) The long-term trend of the flask-based CO2 mole fraction data at Mauna Loa (MLO; 19.54∘ N, 155.58∘ W; 3397 m a.s.l.), Hawaii, obtained from NOAA/ESRL/GMD (National Oceanic and Atmospheric Administration/Earth System Research Laboratory/Global Monitoring Division; available at ftp://aftp.cmdl.noaa.gov/data/, last access: 11 October 2018) is calculated using a digital filtering technique (Nakazawa et al., 1997). The dataset goes to the end of 2015. In general, the long-term CO2 trend at MLO is representative of the large-scale clean atmosphere and thus has been used as a reference site (Sweeney et al., 2015). (2) Deviations of individual CO2 data points from the long-term trend (ΔCO2) are calculated as ΔCO2(lat, long., alt,t)=CO2(lat, long., alt,t)-TrendCO2atMLO(t), where lat, long., alt, and t are latitude, longitude, altitude, and time of individual CONTRAIL CME data points, respectively, and “Trend CO2 at MLO” is the long-term trend curve derived as described above. The CONTRAIL CO2 data over 12 airports in Asia, color coded by altitude, are presented in Fig. 2, together with the MLO CO2 data and the calculated long-term trend. In this study, we present results from the statistical analysis of the ΔCO2 data (i.e., deviations of the individual data points from the black line in each panel of Fig. 2) for the years 2005–2015.

To better understand processes that generate the observed tropospheric distribution of CO2 over the Asia-Pacific region, we analyze CO2 simulated by the model NICAM-TM (Nonhydrostatic Icosahedral Atmospheric Model-based Transport Model; Satoh et al., 2014). Details of the NICAM-TM CO2 simulation and the evaluation of its performance have been presented by Niwa et al. (2011, 2012). The atmospheric CO2 transport is calculated using the 6-hourly meteorological data nudged to the JRA-55 reanalysis. The horizontal model grid interval is about 240 km and the number of vertical model layers is 40. For CO2 simulation, fossil fuel (FF) emissions are obtained from the CDIAC (Carbon Dioxide Information Analysis Center) database (version 2013; Andres et al., 2013), while fire emissions are from the GFED (Global Fire Emission Database version 3.1; van der Werf et al., 2010). A priori terrestrial biospheric (BIO) fluxes are derived from the CASA (Carnegie-Ames-Stanford Approach) model (Randerson et al., 1997). The air–sea exchange is based on the JMA (Japan Meteorological Agency) ocean flux data (Iida et al., 2015). The BIO fluxes are optimized in the NICAM-TM model inversion by using the GLOBALVIEW data (http://www.esrl.noaa.gov/gmd/ccgg/globalview/, last access: January 2012) and the CONTRAIL data in the FT (Niwa et al., 2012). Thus, the simulated atmospheric CO2 is obtained from the optimized fluxes. We also examine simulated CO2 fields driven by two different emission fluxes: one by FF and the other by BIO (hereafter referred to as FF CO2 and BIO CO2, respectively). For comparison, the simulated data are sampled at times and locations coincident with the individual CONTRAIL CME data points, and processed in the same manner; stratospheric data are excluded by the model PV values; all CO2 data points are detrended by the MLO long-term trend in the model.

Figure 3 presents monthly averaged distributions of the UT ΔCO2 over the Asia-Pacific region (left panels) along with histograms in the respective 5∘ latitude bands (right panels). In the left panels, the black arrows indicate monthly averaged horizontal wind at 250 hPa pressure surface from the JCDAS reanalysis. We note that monthly CO2 distributions from the CONTRAIL data previously presented by Sawa et al. (2012) were calculated as averages in 20∘ (longitude) × 10∘ (latitude) bins. In this study we were able to increase the spatial resolution to 5∘ × 5∘ since we have more data. As seen in Fig. 3, the UT ΔCO2 undergoes a clear seasonal cycle that varies significantly with latitude and longitude.

In January–February, the UT ΔCO2 is relatively uniform in space (Fig. 3a and c), except in regions >35∘ N where the histograms show occurrences of higher ΔCO2 values (Fig. 3b and d). In March, high ΔCO2 values are apparent in regions >30∘ N over northern Japan and downwind (Fig. 3e), where a significantly increased frequency of high ΔCO2 up to 6 ppm is observed (Fig. 3f). This feature becomes more pronounced in April (Fig. 3h) with expanded areas of high ΔCO2 around Japan (Fig. 3g). By May, regions with high ΔCO2 extend to >20∘ N (Fig. 3i and j).

By June, the observed high ΔCO2 values over Japan and the northwestern Pacific nearly disappear (Fig. 3k). A significant fraction of the low ΔCO2 values down to -6 ppm and lower is observed at latitudes >35∘ N (Fig. 3l). Due to these low-CO2 values appearing at northern latitudes, the latitudinal gradient of UT ΔCO2 starts to reverse (i.e., northward positive to negative) after June, aided by moderately elevated CO2 observed at 15–30∘ N. In July, we begin to see very low ΔCO2 values below -6 ppm in high latitude regions (>40∘ N), particularly over boreal Eurasia (Fig. 3m and n). To the south, only very small spatial gradients are observed.

In August, we see the CO2 decrease broadly at all latitudes over the Asia-Pacific region, with distinctly low ΔCO2 values forming over South Asia to Southeast Asia (Fig. 3o). The UT wind field shows an anticyclonic wind circulation pattern over this region. This wind structure is coincident with the distinct low ΔCO2 observed over the continent, indicating that the low-CO2 air mass is confined within the UT anticyclone. This clear CO2 spatial structure associated with the anticyclone is for the first time depicted by the improved spatial resolution of the CONTRAIL data since Sawa et al. (2012). It is noted that such distinct low-CO2 structure does not appear until July, despite the fact that the anticyclonic wind pattern starts in June (Fig. 3k and m).

Moving into September, we see a further decrease in ΔCO2 across the wider Asia-Pacific region (Fig. 3q and r). The persistent UT anticyclonic structure is still observable in both ΔCO2 and wind fields, but the sharp boundary along the East Asian coast that was seen in August (i.e., longitudinal gradient or contrast between the continent and the ocean) is now to some degree blurred (see also Fig. 7a). In October, the anticyclonic low-ΔCO2 feature diminishes and ΔCO2 is now relatively uniform in the observation region. Thereafter ΔCO2 increases as a whole during the winter until the return of the spring.

Figure 4 presents a climatology of seasonal variations and vertical profiles of ΔCO2 over 12 airports in Asia, as uniquely obtained by CONTRAIL observation. We consider these figures to represent large-scale (regional) features in the FT as a result of detrending and binning of the data (500 m altitude and 14-day averages from multiple-year data). At lower altitudes, relatively local features can be visible due to boundary layer (BL) processes and flight route biases near the airports, but examining such smaller-scale phenomena in detail is beyond the scope of this study. We also calculate for each airport, altitude variation in the standard deviation (SD) of ΔCO2 calculated for each 2-week and 500 m bin (Fig. 5), as an extended update of Shirai et al. (2012) who addressed synoptic-scale CO2 variability over NRT. This type of analysis is made possible due to CONTRAIL's high-frequency measurements during ascent/descent over the airports.

Stephens et al. (2007) compiled CO2 data from flask-based aircraft observations at 12 sites around the world for comparison with model simulations. Flask CO2 data at 16 sites from the NOAA/ESRL aircraft program were reported by Sweeney et al. (2015), including some of the data analyzed by Stephens et al. (2007). The measurements by Sweeney et al. (2015) have revealed climatological CO2 variations over North America, whereas the present study focuses on Asia with more frequent in-flight observations. Since vertical profile measurements are relatively scarce over Asia (see supporting online material by Stephens et al., 2007), CONTRAIL observations provide greater spatiotemporal insight into regional carbon cycling processes. One of the remarkable features found in vertical CO2 profiles from other regions is the dramatic decrease in CO2 toward the ground in the summer period at mid-continental sites of the Northern Hemisphere (see Fig. S3 of Stephens et al., 2007 and Fig. 5 of Sweeney et al., 2015). Below we show that the vertical CO2 profiles and their seasonal changes observed by CONTRAIL in Asia are interestingly different from those reported by the previous measurements in other regions.

The seasonal CO2 cycles with spring maxima and summer minima, typical for the northern hemispheric troposphere (Stephens et al., 2007; Sweeney et al., 2015), are to some degree obvious across regions over the 12 airports (Fig. 4) in Asia. However, a clear difference from those outside Asia is the general absence of a dramatic decrease in CO2 near the ground in the summer. In other words, the contoured low ΔCO2 in the summer is apparently “floating” in the FT and not connected to the ground, implying that the observed vertical profiles in the summer are not dictated by overwhelming uptake underneath. This feature is observed at all airports except DEL. It is noteworthy that our measurement sites are all located in coastal areas except DEL, while previous measurements at inland sites generally showed more pronounced decreases of CO2 near the ground in the summer (Stephens et al., 2007; Sweeney et al., 2015). In contrast, the springtime maximum ΔCO2 extends from the ground to the UT, indicating that the surrounding or upwind regions of most airports are strong sources of CO2 during that season.

It is likely that some features shown in Fig. 4, especially in the BL, are due to the influence of nearby CO2 emissions. Indeed, at some airports, large elevations of CO2 values have been observed frequently in the BL. In order to reduce possible bias due to such pollution events, we did redraw Fig. 4 with median ΔCO2 values, instead of averaged values. We found no clear visual difference in the overall features discussed below. In fact, differences between average and median are mostly <1 ppm even below 2 km at all airports, except SHA and HKG where the value is ∼1.5 ppm on yearly average. Although pollution events are observed frequently over these two airports (as described below), we consider such “airport bias” in the climatological vertical profiles to be small within the scope of this study. Influence of nearby city emissions on the CONTRAIL observations will be addressed in our future publication.

NRT and HND, Japan, are the two airports over which the largest number of CO2 measurements have been collected by the CONTRAIL CME, giving relatively smooth climatology of ΔCO2 (Fig. 4b and c). Seasonal and vertical characteristics of CO2 over HND and NRT are quite similar to each other. In the FT, ΔCO2 reaches its seasonal maximum in the spring (April–May) and minimum in the late summer to early autumn (September–October), with the seasonal amplitude in general decreasing with altitude. We also find substantially enhanced SD below ∼2 km over HND and NRT in the winter (November–April) and summer (June–August) (Fig. 5b and c). The high summer variability propagates up to higher altitudes (∼6 km), presumably associated with enhanced vertical mixing in the summer. The vertical gradient in CO2 is small (<2 ppm) during the summer period (June–September), but a clear gradient is detectable for the rest of the year. These features are commonly observed over the other two Japanese airports Nagoya (NGO; ∼260 km west of Tokyo) and Fukuoka (FUK; ∼880 km west-southwest of Tokyo and ∼850 km east-northeast of Shanghai). Also notable is that, in September, CO2 decreases with altitude, this feature being observed widely over these four Japanese cities. ΔCO2 undergoes a seasonal cycle with spring maximum and summer minimum also over ICN (∼570 km northwest of FUK; Fig. 4a), but the minimum occurs in late August to early September, about a month earlier than observed over the aforementioned Japanese airports. The low ΔCO2 in the BL is a characteristic that is not observed over Japan.

Along the east coast of continental East Asia, measurements are obtained over three cities: SHA, HKG, and TPE (Fig. 4d, h, and i, respectively). ΔCO2 increases from September until May when it reaches a seasonal maximum. The seasonal minimum in the UT appears in September–October, lagging the lower troposphere (LT) minimum by about a month. We see remarkably high ΔCO2 values in the BL over SHA and HKG, these phenomena being particularly pronounced over SHA where we frequently observe ΔCO2 enhancements of >20 ppm below 1 km. The elevated ΔCO2 in the winter season (November–April) is also characterized by high variability (Fig. 5d and h). Although the seasonal and vertical characteristics of CO2 over TPE appear to be essentially similar to those over SHA and HKG; our measurements are sparse during May–October.

ΔCO2 over DEL shows a unique seasonal variation. We note that DEL is the only inland site, whereas the all other sites presented in this study are located near the coast. Prominent is the strong CO2 drawdown throughout the troposphere in August–September, with very little vertical gradient in the FT due to vigorous vertical mixing (Fig. 4g). Another interesting feature is the relatively low ΔCO2 in the BL (<∼3 km) during January–March. This wintertime CO2 stagnation over DEL was recently attributed to uptake by crops (mainly wheat) grown in the winter season in the surrounding region (Umezawa et al., 2016).

Clear seasonal CO2 variations are also visible over BKK (Fig. 4j). The seasonal maximum happens in March–April in the LT and propagates upward. These 2 months correspond to a period of enhanced ΔCO2 variability near the ground (Fig. 5j). Over SIN in Southeast Asia, ΔCO2 exhibits measurable seasonal variation (Fig. 4k). The seasonal variation in the FT over SIN is similar in phase with that observed over BKK, but with comparatively reduced magnitude. It should be also noted that, over SIN, the vertical gradient of ΔCO2 is small throughout the year. A maximum vertical ΔCO2 difference is only ∼2 ppm observed in the boreal spring. Lastly, for CGK in tropical Asia (Fig. 4l), the observed seasonality in ΔCO2 in the LT is hard to characterize due to relatively large variability. But interestingly, the seasonal phases are apparently different below and above 2.5 km. Below that height, relatively high ΔCO2 values appear during August–October, while, over the same period, ΔCO2 in the FT decreases until the October minimum.

In Fig. 6, simulated (second column) monthly CO2 distributions in the UT are compared to the observations (first column). The model outputs are sampled at a location and time coincident with the observation and analyzed in the same manner as the measurements. The third and fourth column of panels show the simulated FF CO2 and BIO CO2, respectively. We do not present a contribution from biomass burning, since it is relatively minor (though not negligible) in evaluating the seasonal variation. Also shown are monthly CO2 distributions at 250 hPa pressure surface for 2011 (last column). We have chosen the model year 2011 as a representative year whose seasonal CO2 distribution patterns are not exceptional, although the simulated CO2 exhibits interannual variation due to year-by-year changes in meteorology and CO2 fluxes. Note that the model data in the last column are simple monthly averages at model resolutions; thus, both the UT and stratospheric model data are included and avoids sampling bias that might result from data availability as in the observations. However, the similarity in the features between the model and observed results, together with the model monthly averages, attests to the fact that the CME-based CO2 distribution is representative of the seasonal CO2 climatology in the UT.

By comparing with the observations, we see that NICAM-TM (second column) is able to reproduce the overall general seasonal features of the observed CO2 distribution pattern in the UT over the Asia-Pacific region. The model simulation (second column) shows seasonal CO2 elevations centered at 20–40∘ N in April–June, depletion of CO2 over boreal Eurasia starting from June, and a distinct decrease in CO2 over South Asia to Southeast Asia in August–September, all of which are in agreement with the observation (first column). In Sect. 4, we discuss how these features constitute the large-scale seasonal CO2 distributions, as depicted in the last column. One notable feature that is not well reproduced by the model is the high ΔCO2 values observed over northern Japan in April, the cause of which is yet to be determined.

In Sect. 3.1, we presented two major features in the CO2 distribution in the UT over the Asia-Pacific region in the boreal summer: (1) the distinct low-CO2 values associated with the monsoon anticyclone over South Asia to Southeast Asia during August–September and (2) the highly variable low-CO2 values at northern latitudes (>40∘ N) during June–August (Fig. 3). These summertime low-CO2 phenomena are hereafter referred to as the “monsoon low CO2” and “boreal low CO2”, respectively.

We have shown in Sect. 3 that seasonally elevated CO2 is observed throughout the whole troposphere over the East Asia region in April–May (Figs. 3 and 4). This elevated CO2 is accompanied by an increased spread of ΔCO2 values in the UT north of 30∘ N in March–April (Fig. 3f and h). As shown by the NICAM-TM simulation, seasonally high CO2 can be explained mostly by the BIO emission fluxes, but a significant portion of the associated variability is due to enhanced synoptic-scale meteorological variability.

One of the most likely factors in meteorology is the active passage of eastward-tracking synoptic systems. In East Asia, cyclonic activity is most frequent in the spring (Chen et al., 1991; Adachi and Kimura, 2007). In association with the eastward moving springtime cyclonic activity, two major transport pathways have been suggested for pollutant outflow from the continental East Asia to different tropospheric layers over the northwestern Pacific. (1) The first mechanism involves the advection of polluted BL air behind the cyclonic cold front as it moves eastward over the East Asian continent out to the Pacific (Liu et al., 2003; Sawa et al., 2007). Consequently, periodic passages of cyclones produce episodic variations in anthropogenic trace gases in the BL across the northwestern Pacific (Liu et al., 1997; Liang et al., 2004; Sawa et al., 2007; Tohjima et al., 2010, 2014). (2) The second mechanism involves frontal uplift of air in front of a moving cold front, in what is called the warm conveyor belt. The uplift frequently takes place over south and central China and the plume travels northeastward along the warm conveyor belt to the northwestern Pacific (Bey et al., 2001; Liu et al., 2003; Miyazaki et al., 2003; Liang et al., 2004). Convective uplift along a frontal zone over central China and Southeast Asia also transports BL air to the FT (Miyazaki et al., 2003; Oshima et al., 2004). In addition to the above transport processes associated with cyclones, orographic forcing over south and central China has also been observed to uplift the BL air to the FT (Liu et al., 2003). Once in the FT, the plume can be easily exported to the Pacific by the midlatitude westerly winds (Bey et al., 2001; Liu et al., 2003).

In summary, periodic and episodic cyclonic uplifting of BL air over continental East Asia, with strong surface CO2 emissions, could explain the seasonal maximum level of CO2 and increased variability in the FT in the spring (Fig. 3). Using the CONTRAIL data, Shirai et al. (2012) also showed that the observed synoptic-scale variability in the FT CO2 over NRT increases in the spring, as air influenced by the continental East Asian CO2 emissions is advected towards Japan.