Interhemispheric (IH) exchange of CO2 occurs mainly by eddy transport in the boreal winter–spring and by mean convective and advective exchange in the boreal summer–autumn (Bowman and Cohen, 1997; Lintner et al., 2004; Miyazaki et al., 2008; and references therein).

On the basis of long-term (1949–2011) correlations of the upper tropospheric zonal wind with the Southern Oscillation Index (SOI), Francey and Frederiksen (2016; hereafter FF16) defined an index for the Pacific westerly duct, uduct, as a measure of IH eddy transport of CO2. This index is the average zonal wind in the region 5∘ N to 5∘ S, 140 to 170∘ W at 300 hPa, as summarized in Table 1. In this article the period of interest is 1992 to 2016 and the corresponding correlation is shown in Fig. S1 of the Supplement. There the role of the changing Walker circulation with the cycle of the El Niño–Southern Oscillation (ENSO) in determining the properties of the Pacific and Atlantic westerly ducts is also documented. The uduct index is an indicator of cross-equatorial Rossby wave dispersion and associated increases in near-equatorial upper tropospheric transient kinetic energy (Frederiksen and Webster, 1988), particularly between 300 and 100 hPa (∼9 to ∼16 km above sea level). The process normally occurs over the eastern Pacific Ocean during the boreal winter–spring, and the Rossby waves (Webster and Holton, 1982; Stan et al., 2017), generated downwind of thermal anomalies and continental influences, in particular the massive Himalayan orography, propagate in a south-east direction through the Pacific westerly duct generating and transporting turbulent kinetic energy. The generation of turbulent kinetic energy occurs through Rossby wave breaking in the Pacific and Atlantic ducts and enhances turbulent mixing (Ortega et al., 2018 and references therein). FF16 also considered the relationship of uduct and other trace gases including CH4. Indeed, recently Pandey et al. (2017) and Krol et al. (2018) also considered the implications of faster IH transfer of CH4 during the La Niña of 2011 when the Pacific westerly wind duct was open and uduct was large.

FF16 explained the exceptional step in CO2 IH difference between 2009 and 2010 as being due to a contribution from the large anomaly in uduct observed at the time. Recently results from the National Aeronautics and Space Administration (NASA) Orbiting Carbon Observatory-2 (OCO-2) during the 2015–2016 El Niño have been published (Chatterjee et al., 2017; and references therein). In particular release by NASA (2016) of data in a video “Following Carbon Dioxide through the Atmosphere” provides further direct evidence of the Pacific duct hypothesis. The NASA OCO-2 CO2 concentration in Fig. 1a is for 17 February 2015 and shows Rossby wave trains over the eastern Pacific and across South America associated with IH exchange as a typical example of OCO-2 images that coincide with the shaded 2015 period in Fig. 1b. The dynamical properties of these Rossby waves are further explored in the Supplement, including Figs. S3 and S4. Figure 1b uses the covariance between CO2 and uduct in early 2015, indicated by shading, to predict IH CO2 exchange through the Pacific duct. The atmospheric circulation data and indices used throughout this article are obtained from the National Centers for Environmental Prediction (NCEP) and National Center for Atmospheric Research (NCAR) reanalysis (NNR) data (Kalnay et al., 1996); in Sect. 6 we briefly consider the robustness of our results using another reanalysis data set. The results in Fig. 1a and b are also consistent with upper tropospheric (u,v) wind vectors. Between 5 and 23 February 2015 NNR wind vectors show that a strong Pacific North American height anomaly caused a split in the Pacific upper tropospheric winds, the Pacific westerly duct was open, and there were south-east cross-equatorial winds from the Northern to Southern Hemisphere. This is illustrated in Fig. 1c for 300 hPa wind vectors on 17 February 2015.

The focus here is on IH CO2 difference, anomalies in the mean convective and advective mode of IH CO2 exchange, and changes in the relative importance of the mean and eddy transport modes.

To represent the CO2 interhemispheric difference, we define Cmlo-cgo as the difference in Commonwealth Scientific and Industrial Research Organisation (CSIRO, 2018) analysed CO2 concentrations in baseline air sampled from Mauna Loa (mlo, 20∘ N, 156∘ W) and Cape Grim (cgo, 41∘ S, 145∘ E). FF16 discussed the measurement and sampling strategy, consistently applied over 25 years, which has been used to establish the data set with minimum uncertainty. They also examined the CO2 interhemispheric difference with the Scripps Institution of Oceanography (SIO) Mauna Loa and South Pole data (Keeling et al., 2009) and found broad agreement in the two data sets, in the period of overlap since the 1990s, in terms of CO2 changes and the relationships to the opening and closing of the Pacific westerly duct.

Figure 2 summarizes annual covariations that motivated this study. In Fig. 2a the overall trend in the Cmlo-cgo reflects the increasing emissions, mainly in the Northern Hemisphere, of carbon from combustion of fossil fuels coupled with relatively slow transport into the Southern Hemisphere. The smooth dashed curve shows global annual anthropogenic emissions (Le Quéré et al., 2018) scaled by the coefficients of linear regression between Cmlo-cgo and emissions from 1992 to 2015 (0.36 ppm (PgC)-1 yr, n=24, r2=0.83). The year-to-year variations in Cmlo-cgo are more pronounced than the variations in emissions (for example, only 2009, corresponding to the global financial crisis, clearly interrupts the smooth emissions increase).

We examine here the concept of relatively rapid interhemispheric CO2 exchange through a spatially restricted Pacific duct and discuss issues of the uniqueness of the duct and the transport of both turbulent kinetic energy and CO2 to and through the Pacific duct region.

As noted in Sect. 2, the years 2010 and 2016 exhibit a similar anomalous eddy transport index, uduct, but have different Cmlo-cgo responses relative to previous years (Fig. 2a). Since the CO2 emitted in the Northern Hemisphere and tropics is also transported into the Southern Hemisphere by the mean divergent flow associated with the Hadley circulation, particularly during boreal summer (Miyazaki et al., 2008), this is now explored in more detail. The Pacific duct transfer in boreal winter–spring, with peaks in February–April, occurs when the CO2 IH partial pressure difference is near the maximum due to forest respiration. Likewise the mean IH transport related to the Hadley circulation occurs in boreal summer–autumn, with peaks in June–August, when the CO2 IH partial pressure difference has a proportionally larger contribution due to the accumulated fossil fuel CO2 from NH industrial emissions.

Figure 5 shows latitude–height cross sections, over the Pacific, averaged between 120 and 240∘ E, of June to August vertical wind in pressure coordinates, ω, while Fig. 6 shows the corresponding results for the meridional wind, v. Recall that negative ω corresponds to positive vertical velocity in height coordinates and negative v is north–south meridional wind. In the boreal summer–spring, average values for 1979 to 2016 show the uplift (negative ω) at low northern latitudes (Fig. 5a), while the advective Hadley cell meridional transfer (negative v) to the Southern Hemisphere at high altitude can be seen in Fig. 6a.

By subtracting the 1979–2016 average from the 2016 ω values and v values, the nature of the extreme 2016 anomaly becomes visible, with strong uplift including between 10 and 15∘ N shown in Fig. 5b and extensive meridional wind penetration into the Southern Hemisphere, particularly between 500 and 300 hPa, shown in Fig. 6b.

Figures 5c and 6c depict the difference between the anomaly years 2016 and 2010. Both the uplift between 10 and 15∘ N and penetration of the meridional wind into the Southern Hemisphere is stronger in the upper troposphere and mean transport through convection and advection into the Southern Hemisphere more extensive in 2016.

On the basis of these figures, and similar figures for the corresponding zonally averaged quantities, we have chosen four indices to characterize the mean circulation by the Hadley cell (Table 1). These are ωP, the vertical velocity in pressure coordinates over the Pacific Ocean at 300 hPa averaged between 120–240∘ E and 10–15∘ N, and vP, the meridional wind at 200 hPa averaged between 120–240∘ E and 5–10∘ N, as well as the corresponding zonally averaged indices ωH and vH introduced in Sect. 2.

The timing of a majority of short term variations in the 25-year baseline Cmlo-cgo corresponds to atmospheric transport changes that influence the interhemispheric exchange. To quantify relationships between Cmlo-cgo and the eddy and mean transport indices involved, we first suppress the Cmlo-cgo changes expected from reported anthropogenic emissions. The global annual average anthropogenic emissions (Le Quéré et al., 2018) are converted to ppm using the coefficient 0.36 ppm (PgC)-1 yr derived in Sect. 2 from Fig. 2 and subtracted from the observed Cmlo-cgo. In Fig. 6 we compare the FF-adjusted Cmlo-cgo, which we denote Cmlo-cgo∗, for two periods when Cmlo-cgo is positive; the first is February–April that best captures the eddy IH exchange, and the second is June–August when mean transfer related to the Hadley circulation is captured.

We focus first on the FF-adjusted Cmlo-cgo∗ plots in Fig. 7a and b. The mean and year-to-year variation is very much larger in (a) compared to (b) and is also larger in (a) compared to the annual averaged values in Fig. 2a. The contrasting behaviour between the two periods after 2012 is also more marked.

To emphasize the similarities between Cmlo-cgo∗ and Pacific duct winds, we plot -uduct in Fig. 7a, so that easterlies are shown as positive and the more frequent westerlies as negative; the timings of peaks in both panels then correspond to each other. When winds in the Pacific duct are easterly or near zero, FF-adjusted Cmlo-cgo∗ peak or are above average; this is now more obvious in 2016 compared with the corresponding results for Cmlo-cgo in December–May shown in Fig. 2b and c. In fact the FF-adjusted Cmlo-cgo∗ has very similar behaviour to the detrended Cmlo-cgo, with pattern correlations of anomalies of r=0.931, r=0.954, and r=0.981, for January–December, June–August, and February–April respectively. The similarity can also be seen by comparing the top panel of Fig. 7b with that of Fig. 8. Despite persistent agreement in timing, the magnitude of the Cmlo-cgo response to the uduct anomaly is more variable. This is reflected in the correlation between the detrended Cmlo-cgo anomalies and the detrended uduct anomalies, which is r=-0.500 for February–April and r=-0.228 for December–May. These results confirm the preferential Pacific duct transfer in late boreal winter and early spring (February–April). They also indicate that although there is an important relationship between Cmlo-cgo and the zonal wind in the Pacific duct, other processes detailed in Sect. 2, such as changes in direct advective transport by the mean winds and emissions, also play roles in year-to-year IH variations. This is also confirmed by a regression analysis of Cmlo-cgo anomalies onto uduct anomalies (not shown), where there is significant scatter about the regression line.

In particular, during 2009–2010 there were a number of complicating factors that most likely contributed to this. The unusually low Cmlo-cgo in 2008 and 2009 coincide with the global financial crisis when global emissions dipped (and recent estimates of emissions by British Petroleum, 2018, suggest an even larger 2008–2009 anomaly than in data used here). Terrestrial net biosphere production south of 30∘ S was also anomalously low in 2009 and anomalously high in 2010 (FF16; Trudinger et al., 2016; though by amounts not sufficient to impact on the Cape Grim baseline CO2 records). The low Cmlo-cgo also align with near-record strong westerlies in the Pacific duct, and associated larger eddy transport, in 2008; both potentially contribute to an increase in the magnitude of the subsequent CO2 step.

In Fig. 7b the post-2010 decrease in June–August FF-adjusted Cmlo-cgo∗ is clearly mirrored in the decreasing ω and v indices (indicating strengthened Hadley circulation), particularly in the 120–240∘ E Pacific sector (ωP and vP) compared to the zonal average (ωH and vH). The considerably weaker strength of the Hadley circulation in 2010 compared with 2016 is shown quite distinctly. The correlations between the detrended Cmlo-cgo anomalies and indices of mean transport are shown in Table 2.

Generally June–August correlations are stronger than the June–November correlations, and correlations over the Pacific sector 120–240∘ E are generally larger than for the zonally averaged quantities. This is clearly the case for ω, while vH is an exception, being larger for the longer time period. The Cmlo-cgo correlations for June–August involving ωP and vP have roughly similar magnitudes to those for February–April, involving uduct, and ωP and vP provide similar predictability of the role of the Hadley circulation in mean IH CO2 transport as uduct does for eddy transport. Interestingly, during 2009–2010 the effects of uduct and ωP and vP reinforce one another to make the step in Cmlo-cgo large, while for 2015–2016 ωP and vP counteract uduct, and the exceptionally strong Hadley circulation becomes the dominant feature in determining the annual Cmlo-cgo (Fig. 1a). These results show that there is an important connection between the Cmlo-cgo and the indices that characterizes the strength of the Hadley circulation and mean transport. Again, as also suggested by regression analysis (not shown), other processes, detailed in Sect. 2 and above, also play important roles.

The somewhat different behaviours of Cmlo-cgo and the dynamical indices, particularly during the El Niños of 2009–2010 and 2015–2016 and of 1997–1998, may partly reflect the diversity of El Niños and whether the heating is focussed in the eastern Pacific or in the central Pacific (Capotondi et al., 2015; L'Heureux et al., 2017 and references therein). The strong 1997–1998 event, like the 1982–1983 event, was a classic eastern Pacific El Niño, with maximum temperature anomalies of nearly +4 ∘C (L'Heureux et al., 2017). The 2009–2010 event, in contrast, was a central Pacific El Niño, with record-breaking warming in the central Pacific (Kim et al., 2011). The 2015–2016 El Niño fell between these two canonical cases, with less warming in the eastern Pacific Ocean than the 1997–1998 event, but similar warming to the 2009–2010 event in the central Pacific (L'Heureux et al., 2017).

The broadly increasing magnitude of the negative ω and v indices since 2012 is associated with both increasing global temperatures, breaking the record in 2016, and the large El Niño of 2015 and 2016. This has resulted in the increasing importance of the mean convective and advective CO2 transport by the Hadley circulation relative to the eddy transport including through the Pacific duct. It will be interesting to see whether this favouring of the mean over the eddy IH CO2 transport will become increasingly important with further global warming and the extent to which it depends on extreme El Niños (Cai et al., 2014; Freitas et al., 2017; Yeh et al., 2018).

The dynamical indices that we have used for this study are based on the NCEP-NCAR reanalysis (NNR) data (Kalnay et al., 1996). There is generally close correspondence between the major global atmospheric circulation data sets that, like the NNR data, use full data assimilation throughout the atmosphere (Frederiksen and Frederiksen, 2007; Frederiksen et al., 2017a; Rikus, 2018). We have confirmed this by recalculating our dynamical indices and main correlations with Cmlo-cgo based on the NASA Modern Era Retrospective-analysis for Research and Applications (MERRA) data (Rienecker et al., 2011). For example, the 1992 to 2016 correlation between MERRA and NNR data for uduct in February–April is r=0.974, for ωP in June–August is r=0.899, and for vP in June–August is r=0.931. The corresponding correlations between detrended anomalies of Cmlo-cgo and the MERRA-based dynamical indices are also very similar. The correlations are r=-0.512 with uduct for February–April (compared with r=-0.500 for the NNR index), r=0.504 with ωP for June–August (compared with r=0.522 based on NNR), and r=0.538 with vP for June–August (compared with r=0.539 based on NNR).

Next, we consider the eddy and mean IH exchange of other trace gas species and their correlations with CO2 and dynamical indices of transport. We focus on February–April for eddy transport and June–August for mean transport since these periods were the peaks for correlations of CO2 IH difference with eddy and mean transport indices respectively. However, there are differences in the seasonal variability of the interhemispheric gradient in the different trace gas species that are reflected in their transport, and for that reason we also briefly mention the results for other time periods. We begin by further examining Mauna Loa minus Cape Grim (mlo–cgo) differences, between 1992–2016, in the routinely monitored CSIRO species (CSIRO, 2018) CH4, CO, and H2 in addition to CO2 that were briefly considered by Francey and Frederiksen (FF16), as well as N2O (for 1993–2016). Thereafter we discuss mlo–cgo differences in SF6 data sourced from the NOAA Halocarbons and other Atmospheric Trace Species Group (HATS) program from 1998 (NOAA, 2018).

The major El Niño of 2015 and 2016 coincided with record global warming, with 2016 having the highest global average surface temperatures and 2015 the third highest (2017 had the second highest). The strength of the Hadley circulation also increased to unprecedented levels during 2015–2016 and had a major impact on the mean interhemispheric (IH) transport of CO2 and on the difference in CO2 concentration between Mauna Loa and Cape Grim (Cmlo-cgo). This study has focussed on the roles of IH transient eddy and mean transport of CO2 on interannual variations in Cmlo-cgo and has established dynamical indices that characterize the broad features of this transfer (Table 1). Interestingly, some of these indices are based on regions that lie close to or overlap the region of the Niño 3.4 sea surface temperature (SST) index (the average SST in the region 5∘ N–5∘ S, 120–170∘ W), where ENSO is strongly coupled to the overlying atmosphere (L'Heureux et al., 2017).

One of these indices, uduct, which is a measure of eddy IH transport of CO2, was introduced in FF16. This index is the 300 hPa Pacific zonal wind averaged between 5∘ N–5∘ S and 140–170∘ W and is strongly correlated with the Southern Oscillation (SOI) index (r∼0.8 in Fig. 4a of FF16). A particular focus of that study was to propose an explanation for the record step in CO2 IH difference between 2009 and 2010 and it was concluded that the closing of the Pacific duct (negative uduct) during the El Niño of 2010 was a significant contributing factor. It was also noted that there were half a dozen other occasions going back to the 1960s when the closing of the Pacific duct was related to an increase in CO2 IH difference (Keeling et al., 2009).

Here, we have extended the analysis of the relationship between uduct and Cmlo-cgo to 2016. We again find that during boreal winter–spring, and particularly during February–April when eddy transport of CO2 from the Northern to Southern Hemisphere is most active, there is an increase in Cmlo-cgo during the El Niño of 2015–2016. However, while the timing of the increases in these years, and for other occasions going back to 1992, agree with the closing of the Pacific duct, the magnitude is more variable, indicating the contribution of other processes discussed in Sect. 2. We have analysed the intermittent nature of the opening and closing of the Pacific westerly duct. In particular, episodes in February 2015 have been related to results from NASA (2016) data in the video “Following Carbon Dioxide through the Atmosphere”. The video provides further evidence of the propagation of Rossby waves through the open Pacific westerly duct and the transfer of CO2 into the Southern Hemisphere. We have also noted that large-scale uplift slightly downstream of Asia occurs when the Pacific duct is open, allowing these substantial emissions to be transported directly, via Rossby wave dispersion, through the duct.

A major focus of this article has also been the role of changes in the mean IH CO2 transport from the Northern to Southern Hemisphere due to variability in the Hadley circulation. We have introduced indices (Table 1) that measure this transfer based on the 300 hPa ω, the vertical velocity in pressure coordinates, between 10 and 15∘ N and 200 hPa v, the meridional wind between 5 and 10∘ N, both zonally averaged (ωH and vH) and with averaging restricted to the Pacific sector 120–240∘ E (ωP and vP). The correlations for June–August between Cmlo-cgo and ωP or vP(r∼0.5) have roughly similar magnitudes to those in February–April involving uduct. The indices ωP and vP provide similar predictability of the role of the Hadley circulation in mean IH CO2 transport as uduct does for eddy transport. We have also found that during 2009–2010, the effects of uduct and ωP and vP reinforce one another to make the step in Cmlo-cgo large. In contrast, for 2015–2016 ωP and vP counteract uduct, and the record Hadley circulation primarily determines the annual Mauna Loa and Cape Grim CO2 difference. The effects of interannual changes in mean and eddy transport on IH gradients in CO2 (and CH4, CO, H2, N2O, and SF6) have been examined for the period 1992 to 2016.

The sign and strength of zonal winds in the Pacific westerly duct (uduct) are related to Rossby wave dispersion and breaking and are correlated with corresponding changes in near-equatorial transient kinetic energy (Fig. 6; Frederiksen and Webster, 1988), resulting in intermittent changes in the mixing of trace gases. This effect may not be adequately represented in the parameterizations (Frederiksen et al., 2017b) used in atmospheric circulation and transport models. Model determinations of short-term variations in the Hadley circulation exchange are also susceptible to uncertainties in representations of the equatorial convective dynamics (Lintner et al., 2004). Over at least 25 years, much of the variability in CO2 between the two surface monitoring sites of Mauna Loa and Cape Grim can be associated with dynamical near-equatorial atmospheric indices of global significance in a changing climate. The changing nature of the seasonal and inter-annual changes in CO2 IH Pacific duct eddy and mean Hadley circulation transfer between 1992 and 2016 provides an interesting case study and potential test of inversion models of atmospheric transport.

We plan to further explore trace gas IH transfer focussing on Southern Hemisphere CO2 stable isotope data in a study that distinguishes between mean IH transfer and eddy transfers of both current season emissions and accumulated Northern Hemisphere fossil fuel emissions.