The 2009 – 2010 step in atmospheric CO 2 inter-hemispheric di ff erence

Introduction Conclusions References


Introduction
The 2009-2010 increase in annual mean CO 2 difference between hemispheres, ∆C N−S , was noted by Francey et al. (2013)  scaled to run parallel to the ∆C slope in order to emphasize the unusual nature of the 2009-2010 ∆C step. From this perspective the 0.8 ppm step, if the result of an anomalous flux, would equate to a 1.6 Pg C yr −1 (NH) source. Also obvious in Fig. 1a is the unusual post-2009 ∆C stability compared to the earlier record. The dC/dt in Fig. 1b show inter-annual variability on 3 to 5 year El Niño-Southern 5 Oscillation (ENSO) timeframes, forced primarily by climate variability on the equatorial land biosphere (Rayner et al., 2008). This variability is largely suppressed in ∆C when resulting CO 2 is mixed into both hemispheres. Methods to obtain ∆C and dC/dt from monthly data are descibed in Appendix A.
The hemispheric representativeness of extra-tropical baseline data from the selected monitoring sites is supported by a study of aircraft vertical profiles at 12 global sites, identifying mlo and cgo as being the least affected by surface CO 2 exchanges in their respective hemispheres (Stephens et al., 2007). While the spo data closely track cgo data and other mid-to-high southern latitude (SH) sites in the CSIRO network (Francey et al., 2013), the situation is less clear for mlo because of NH heteorogeneity and 15 downwind proximity to Asia. A possible contributing factor at mlo may result from geographical susceptibility to rapidly increasing SE Asian pollution, "rapidly transported to the deep tropics" (Ashford et al., 2015). However in the Supplement Fig. S1 we demonstrate similarity in year-to-year changes in ∆C using both Pacific and Atlantic extra-tropical NH sites from the NOAA network. The similarity is particularly signifi-20 cant in sign and magnitude for the two largest observed changes in 2009-2010 and 2002-2003, implying that especially for these periods mlo represents NH behaviour.

Isotopic evidence for the systematic nature of ∆C variation
If variations in ∆C involve CO 2 of terrestrial biosphere origin (which includes FF) then a strong relationship between the changes in CO 2 concentration and changes in its sta-Introduction for sample s and reference r: δ 13 C s = 13 C s / 12 C s − 13 C r / 12 C r / 13 C r / 12 C r Mass conservation in 13 C is approximated using the product of C and δ 13 C, i.e. δ 13 C is approximately inversely proportional to C (Enting et al., 1995). The CSIRO data in Fig. 2 reinforce the systematic nature of the ∆C variations with 5 a tight linear relationship between IH differences in the CO 2 stable carbon isotope ratio and in the inverse CO 2 concentration, (∆C) −1 , including both the step and pre-2010 year-to-year variations (r 2 = 0.95). However inconsistency between laboratories, shown in Table 1, is substantial. Concentration data, particularly from CSIRO and NOAA (Masarie et al., 2001;Francey et al., 2015), are in sufficiently close agreement 10 that the differences must lie with isotopic measurement (Appendix B). The random nature of the scatter in Fig. 2 is emphasised by a lack of correlation between NOAA and SIO ∆δ 13 C linear regression residuals (r 2 ∼ 0.1). We interpret the strong relationship using CSIRO isotope data as implying CO 2 dominated by a C3 photosynthetic signature (Farquhar et al., 1982), including FF, but ex-15 cluding significant contributions from oceans or equatorial C4 plants; also implied is that the CO 2 has had little opportunity for isotopic equilibration with natural reservoirs, i.e. < 1-2 years since release (Enting et al., 1995).

Reported source/sink anomalies
If ∆C variations in Fig. 1 are indeed systematic, then clues to the forcing should be 20 clarified by close examination of ∆C and dC/dt at the times of a number of anomalous surface flux events that have been reported over the period.
The largest such events since the 1960s that influence CO 2 are the 1991 Pinatubo volcanic eruption triggering increased removal of CO 2 from the amosphere in subsequent years (Frölicher et al., 2013(Frölicher et al., ) and 1997(Frölicher et al., -1998 Indonesian peat fires (Page BGD 12, 15087-15109, 2015 The 2009 , 2009). Taking into account that the differences between NOAA and CSIRO mlo dC/dt records are smaller after 2000 (reflecting general improvements in measurement, Francey et al., 2015) the hemispheric differences during these two events are generally small, as expected if their influences are distributed into both hemispheres. The major non-equatorial, terrestrial emission events reported over the last 2  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | by around 6 months, a phase difference not observed for other significant El Niño peaks in Fig. 1b. This implies either an undetected NH source or changes in IH transport.
Independent evidence for the NH origin of the 2009 to 2010 CO 2 ∆C step comes from a recent analysis of upper troposphere measurements for 11 latitude bands between 5 30 • N to 30 • S (Matsueda et al., 2015) where the step is evident north of 10 • N. These authors suggest a role for transport, as well as source/sinks, to explain their year-toyear variations in latitudinal differences.

Anomalies in annual interhemispheric mixing
Meridional transport and eddy mixing due to large scale eddy motions are sources of 10 significant uncertainty in estimations of IH transport (Miyazaki et al., 2008). Here we examine the role of the opening and closing of the upper tropospheric equatorial westerly duct, and associated inter-hemispheric Rossby wave propagation, as a contributor to the 2009-2010 ∆C mlo-cgo shift, and other variations, shown in Fig. 1a. Extra-tropical NH Rossby waves, including a branch of the Himalayan wave-train, are 15 able to penetrate into the SH when near-equatorial zonal winds are westerly in the upper tropospheric duct centred on 140 to 170 • W and 5 • N to 5 • S (Webster and Holton, 1982;Frederiksen and Webster, 1988;Webster and Chang, 1988). This region is delineated and its tropospheric relevance revealed in Fig. 3a  other times, including El Ninos, the u duct are near zero or easterly, causing the Rossby wave eddies to be deflected northwards and dissipated in the equatorial regions, inhibiting inter-hemispheric exchange.
For the period July 2009 to June 2010 the average 300 hPa equatorial zonal winds in the duct region were easterly as shown in Fig. 3c, effectively closing the duct and 5 increasing the build-up of FF CO 2 in the NH. The July 2008 to June 2009 open duct pattern, with westerlies in the duct, is shown in Fig. 3d. (Appendix C addresses the altitude range involved in this process. Note also, the meridional wind may make a small contribution to IH transport in the duct region during this time). Figure 3b shows the 300 hPa zonal winds for July 2008 to June 2009 (Fig. 3d) minus 10 those for July 2009 to June 2010 (Fig. 3c) and the pattern bears strong similarities with the long-term zonal wind vs. SOI correlation in Fig. 3a.

Trace gas interhemispheric exchange through the duct
Inter-hemispheric exchange of a seasonally varying gas by this process depends on co-variance with u duct , and is represented in Fig. 4 Giglio et al., 2013). The next lowest seasonally integrated Σu duct ∼ 10 ms −1 in 2003, has the next largest ∆C increase (possibly complicating surface flux estimates from the inversion of CO 2 spatial differences by Rayner et al., 2008). For the next two occasions, when Σu duct ∼ 20 ms −1 , there is no clear ∆C response.
Compared to previous behaviour, the magnitude of exchange (∆C × u duct ) immedi-5 ately after the exended duct closure from July 2009 to June 2010 is the largest for each gas, in part reflecting the fact that 2010-2011 La Niña corresponds to the most intense Σu duct since 1990 (top panel Fig. 4). The species exchange at this time is most marked for CO 2 and H 2 , which we mainly attribute to the fact that these two gases exhibit the most significant ∆C trend (CO 2 positive, H 2 negative) over the two decades; also each 10 has seasonal concentration amplitudes that are the largest compared to mean annual gradients (Fig. S2). Through the four "duct-open" periods after 2010, Fig. 1a shows ∆CO 2 to be practically constant, a phenomena difficult to explain with known source/sink behaviour. During this period Σu duct monitonically decreases; the constant ∆C might be explained to missing data (particularly at spo) and measurement bias (Francey et al., 2015) and not considered further here. The 1986-1988 event most mirrors 2009-2010 being the next largest step, followed by four years of relatively stable ∆C.
We conclude from this that anomalies in the inter-hemispheric exchange through the duct have played a significant ongoing role in modifying spatial differences in CO 2 5 (and other trace species) at the surface. As NH FF CO 2 emissions increase further, the influence is expected to become more marked in ∆C × u duct . Peylin et al. (2013) describe conflict between groups of models in locating the major global terrestrial sink, whether mid-northern latitude or equatorial, and suggest atmo-10 spheric transport implementations may be involved. We have presented a variety of complementary evidence linking interhemispheric transport through the Pacific upper troposphere equatorial duct and the spatial and temporal difference in measured surface CO 2 concentrations. The observed patterns of CO 2 inter-hemispheric changes are not easily explained by observed source/sink behaviour.

15
The observed 2009-2010 changes in CO 2 IH difference in particular, because of the magnitude and absence of plausible reported source/sink changes (in a time of unprecedented monitoring of ecosystem and ocean exchanges), provide an unusual opportunity to test the implementation of atmospheric transport in inversion models and help remove current ambiguities between surface exchanges and transport. More 20 generally, this requires such models to demonstrate an ability to describe the spatial and temporal sytematic differences in selected high-quality baseline trace gas records that have well established large-scale representation, such as the mlo-cgo records used here. 12,2015 The 2009

Appendix A: Trace gas data processing
The analyses for both dC/dt and ∆C are based on monthly average mixing ratios (or δ 13 C isotopic ratios) obtained from a smooth curve through individual flask data (typically 4 month −1 ) with combined harmonic (seasonal) and 80 day smoothing spline (Thoning et al., 1989). At Cape Grim, selected data represent strong near-surface 5 winds (> 5 ms −1 , 164 m a.s.l.) with trajectories (typically > 10 days) over the Southern Ocean; at Mauna Loa samples are collected in moderate down-slope winds; South Pole samples are selected to avoid local (station) contamination. Conventional smoothing splines through de-seasonalised baseline-selected concentration data, with 50 % attenuation at 22 months, are differentiated to provide dC/dt since 1992; Francey et al. (2015) discuss dC/dt uncertainties. Annually averaged ∼ 80 day smoothed monthly baseline concentration data are used to provide ∆C with near-annual time resolution, i.e. potential ambiguity between seasonality and inter-annual variation is addressed differently by dC/dt and ∆C. CSIRO and NOAA data are processed identically. Scripps data used here are monthly data that are seasonally adjusted and filled 15 (http://scrippsco2.ucsd.edu).
(Note: Using the spatial differences from individual laboratories effectively removes most calibration issues that can complicate high precision comparisons of data between laboratories.) CO 2 differences between NOAA and CSIRO same-air comparisons since 1992 are 0.11 ± 0.13 ppm, with mean difference effectively cancelled in 20 mlo-cgo comparisons. This means the maximum δ 13 C measurement error due to CO 2 difference should be less than 0.005 ‰. −18 ‰ with respect to air, while C4 plant fractionation is ∼ −7 ‰ and fractionation for CO 2 entering the ocean is small ∼ −2 ‰). Over the whole period the slope reduces to 0.41 due to non-linearity exposed by the 0.8 ppm step. Using (∆C) −1 in Fig. 2 gives the slope 0.342(±0.057) ‰ ppm pre-2010 and 0.350(±0.018) ‰ ppm overall. Exact reasons for the varying quality of δ 13 C programs in Fig. 2 are not known.

5
However, reduced scatter in CSIRO program (Allison and Francey, 2007) is possibly related to feedback from regular quality assessment provided by unique method redundancy; the data in this report involve small subsamples of chemically dried whole flask air, from which CO 2 is extracted and analysed using a fully automated Finnigan-Matt 602 D Mass Spectrometer (MS) with MT Box-C extraction accessory, and bracketed 10 by extractions and analysis of cgo long-lived baseline air standards in high-pressure cylinders. Over most of the two decades a parallel cgo program involved unique largesample in situ extraction of CO 2 , which is returned and analysed on the same MS, but relative to independently propagated pure CO 2 standards. Unfortunately, substantial reduction in support means continuing quality of the CSIRO program is not assured 15 from 2014 (2014 data not included).

Appendix C: Atmospheric transport
In contrast to the situation in Fig. 3c Himalayan wave-train, are able to penetrate into the SH. Correlation analysis (Frederiksen and Webster, 1988) indicates increased upper tropospheric transient kinetic energy near the equator with facilitated IH transport of trace gases. Here we have focused on the 300 hPa level, but our results apply in broad terms to most of the upper troposphere. In particular, the correlation of the SOI with the zonal wind in the west- Allison, Paul Fraser and Marcel van der Schoot; many support staff in GASLAB, Cape Grim Baseline Air Pollution Station (The Australian Bureau of Meteorology with CSIRO), and measurement collaborators at NOAA, also contribute directly in this regard. The importance of the historic SIO records cannot be understated. Rachel Law provided global, and Ying Ping Wang with Chris Lu regional, CO 2 modelling advice.