Variability in a four-network composite of atmospheric CO 2 differences between three primary baseline sites

. Spatial differences in the monthly baseline CO 2 since 1992 from Mauna Loa, (mlo, 19.5°N, 155.6°W, 3379m), Cape Grim (cgo, 40.7°S, 144.7°E, 94m) and South Pole (spo, 90°S, 2810m), are examined for consistency between four monitoring networks. For each site pair, a composite based on the average of NOAA, CSIRO and two independent SIO analysis methods is presented. Averages of the monthly standard deviations 10 are 0.25, 0.23 and 0.16 ppm for mlo-cgo, mlo-spo and cgo-spo respectively. This high degree of consistency and near-monthly temporal differentiation (compared to CO 2 growth rates) provides an opportunity to use the composite differences for verification of global carbon cycle model simulations. Interhemispheric CO 2 variation is predominantly imparted by the mlo data. The peaks and dips of the seasonal variation in interhemispheric difference act largely independently. The peaks mainly occur in May, near the 15 peak of Northern Hemisphere terrestrial photosynthesis/respiration cycle. Feb-Apr is when interhemispheric exchange via eddy processes dominates, with increasing contributions from mean transport via the Hadley circulation into boreal summer (May-Jul). The dips occur in September, when the CO 2 partial pressure difference is near zero. The cross-equatorial flux variation is large and sufficient to significantly influence short-term Northern Hemisphere growth rate variations. However, surface-air terrestrial flux anomalies would need 20 to be up to an order of magnitude larger than found to explain the peak and dip CO 2 difference variations. Features throughout the composite CO 2 difference records are inconsistent in timing and amplitude with air-surface fluxes but are largely consistent with interhemispheric transport variations. These include greater variability prior to 2010 compared to the remarkable stability in annual CO 2 inter-hemispheric


Introduction
Atmospheric CO2 measurements are normally introduced into global carbon budgets as a "global growth rate ... based on the average of multiple stations selected from the marine boundary layer sites with well mixed background air .., after fitting each station with a smoothed curve as a function of time, and averaging by latitude band …" (Le Quéré et al., 2018). This approach encourages sampling at multiple locations to seek atmospheric confirmation of national/continental emission changes. Particularly in the Northern Hemisphere (NH), with more complicated geography and atmospheric circulation, the influence of continental emissions on 35 marine boundary layer air can vary widely between sites.
A clearer indication of the global impact of regional emissions comes from sites demonstrating maximum spatial representation. In this case, global significance of biogeochemical CO2 exchanges between the surface will be informed by their impact on validated baseline data with the least continental influence. Such baseline data is more directly relevant to changes in global ocean acidification and climate change, but places heightened 40 demands on sampling criteria and calibration.
Sites selected to maximise spatial representation in their respective hemispheres, Mauna Loa, (mlo, 19.5°N, 155.6°W, 3379m) and South Pole (spo, 90°S, 2810m) also have the longest (multi-decadal) coherent trace gas monitoring data, based on flask sampling (Supplement S1). At these sites, and at Cape Grim (cgo, 40.7°S, 144.7°E, 94m) since 1991, co-sampled baseline air has been analysed at three different laboratories, using four 45 different methodologies summarized in Section 2.
To account for any persisting artefacts in the co-sampled data we examine, for each method, inter-site differences in the published monthly baseline data from the three sites. The standard deviation in the average of the co-sampled differences provides a practical uncertainty estimate. A key advantage compared to the growth rate approach, is that assumptions inherent in the growth rate smoothing (where for example 22-month 50 smoothing is used to separate interannual and seasonal variations) are avoided so that in this study near-monthly effective time resolution is achieved.
The inter-site difference approach was used by Francey and Frederiksen, 2016 (FF16)  The 2015/16 El Niño was stronger and has also been associated with unprecedented behaviour in the global carbon cycle (elsewhere attributed to the terrestrial biosphere anomalies, e.g. Yue et al., 2017). However, Frederiksen and Francey, 2018 (FF18), argued that the unprecedented strength in the Hadley circulation 60 increased IH exchange (reduced IH CO2 difference) late in 2016, overwhelming the earlier reduced eddy exchange linked to the strong 2015/16 El Niño. They also indicated dynamical contributions to IH CO2 during both El Niño and La Niña periods (e.g. FF16 Figure 5, and FF18 section 6.2, on multi-species IH differences).
While ENSO events are expected to impact on surface biology, it is also clear that they also influence atmospheric IH CO2 fluxes. The timing of the dynamical events suggests an alternate explanation for the CO2 composite CO2 record by comparing anomalies in the magnitude and phasing of composite IH CO2 variations with those in air-surface exchange model outputs, as well as in dynamics indices representing atmospheric IH exchange.

Background information on flask networks
A historic overview of CO2 IH difference data is provided in Supplement S1.

75
By 1958 C.D. Keeling had identified mlo and spo as optimum sites to obtain background CO2 in the respective hemispheres and by the 1970s was obtaining a regular monthly supply of air admitted to 5L evacuated glass flasks from both sites (SIO1: Keeling C. D. et al., 2001). Since 1992, there are CO2 measurements as a byproduct of a global network focussed on O2/N2 ratios in baseline air (SIO2: Keeling R. F. and Schertz, 1992); this program uses 5L glass flasks flushed and filled to ambient pressure, with cryogenically dried air. While 80 there is commonality regarding calibration, in the context of spatial differences the SIO networks can be considered independent.
NOAA began sampling from all three sites, mlo, spo, cgo, (as part of a much larger network) from 1984, using a variety of flask and filling methods. From around 1992 the current system of Peltier-dried air in pressurized 2.5 L flasks (Tans et al., 1992, Conway et al., 1994, Dlugokencky et al., 2014 was phased in. NOAA has 85 maintained the World Meteorological Organization (WMO) Central CO2 Calibration Laboratory since 1996 (a role previously carried out by SIO). The NOAA atmospheric sampling is generally more frequent (typically 8-10 flasks per month) than is the case for the SIO or CSIRO programs (except for the CSIRO cgo program); however, the size and sampling frequency in the NOAA network amplifies calibration challenges due to shorter lifetimes of reference and calibration standards.

90
Both NOAA and SIO use non-dispersive infra-red analysers (NDIR) for CO2 measurement. (CSIRO flask sampling at cgo, spo and mlo in the early 1980s used NDIR for analysis of chemically dried air, pressurized into 5L glass flasks). However, analyses here are restricted to CSIRO's measurements from 1992 using chemicallydried, pressurized air in 0.5L glass flasks, but with retention of 5L flasks at spo (Francey et al., 1996). Gas chromatography with flame ionisation detection (GC/FID) was introduced to measure CO2 in flasks, a technique 95 providing a more linear response than NDIR (Supplement S2). Hourly radon measurements at Cape Grim (Chambers et al. 2016) were introduced around this time. Air mass history is further informed by a decade of vertical profiling (Langenfelds et al., 2003;Pak et al., 1996), back trajectory analysis, and other tracers (e.g. Dunse et al., 2001), demonstrating that selected cgo data can achieve a degree of spatial representation matching, or sometime exceeding, that at the more remote high-altitude sites at mlo and spo.
baseline sites, Figure 1 shows systematic behaviour in the SIO1, SIO2 and CSIRO monthly CO2 differences from NOAA. Five-month running means aid discussion. The CO2 mixing ratios used here are referred to in the commonly used units of parts per million (ppm) rather than the more strictly correct term of mole of CO2 per mole of dry air. Note that data independently flagged for sampling or measurement anomalies are rejected by individual laboratories prior to publication as monthly averages. Typically, a small number of gross outliers in 115 individual flask data (e.g. in flask-pair differences) are also rejected prior to publication.
In Figure 1, there is clear evidence of systematic differences: in mean offsets, seasonality, and between sites within one network. In the context of inter-hemispheric exchange, the typical 0.5 ppm range of variation remains relatively small compared to the 7-10 ppm maximum CO2 interhemispheric difference (IH CO2). Net IH exchange is proportional to IH partial pressure difference.

120
 Between 1991-1993, there is a marked inconsistency between NOAA mlo-spo and mlo-cgo, particularly in seasonal amplitude; CSIRO has comparable measurements that are more consistent (Supplement S4). This is a reason for caution when interpreting the data in this period.
 CSIRO records at cgo exhibit the smallest offset and scatter relative to NOAA (± 0.08 ppm) while SIO2 mlo data exhibit the largest scatter (± 0.37 ppm).
 Remnant seasonality is still evident in the CSIRO cgo differences from NOAA. While a small effect, the CSIRO GC/FID near-linear response for CO2 means results are not so sensitive to differences between 130 sample and reference CO2. This advantage is reinforced in the CSIRO SH data since reference gases use recent SH baseline air. This is generally not the case for non-linear NDIR measurement and particularly in the NH if relatively short-lived reference gases sourced in the NH have a less-than-optimum match with ambient CO2 from a site.

135
While Figure 1 reveals some un-resolved systematic differences between data sets, Figure 2 emphasizes that they are generally small compared to the IH partial pressure differences that are a pre-requisite for IH net exchange. Data from each method are presented as 3-month seasonal averages in order to minimize potential influences related to network sample-frequency (by ensuring an adequate number of individual flask samples per period). As well, the particular 3-month seasonal selection distinguishes periods of distinct relatively stable 140 partial pressure differences between hemispheres and the selected seasons also distinguish eddy and mean IH transport mechanisms (FF18).


The largest IH CO2 variability is recorded in Feb-Apr and in May-Jul, both seasons having near-equally 155 large IH differences. The large seasonality in NH CO2 is widely linked to the photosynthesis/ respiration in NH forests. Feb-Apr is also when IH exchange by eddy processes is most influential (FF16), whereas mean transport via the Hadley circulation is the main dynamical influence in May-Jul (FF18).
Systematic differences due to sampling and measurement methodology can possibly arise from factors such as the linearity of instrument response, flask storage effects or undetected entrainment of laboratory air. Records 160 with the sparsest sample density (e.g.at spo and particularly in CSIRO spo data) may be more susceptible to undetected anomalies. Closer inspection of individual flask metadata, or of the less extensive in situ monitoring, may resolve some of these infrequent anomalies, but for the present, composite averaging of the flask data is relied on to moderate their influence.
4 Composite records of baseline station spatial differences 165 For each of mlo-cgo, mlo-spo and cgo-spo monthly CO2 differences, Table 1 shows the number of months between 1992 and 2017 contributing to a composite value, arranged in columns indicating the number of contributing networks; e.g. 266 of 312 months have 4 networks contributing to mlo-spo, while 279 months have three contributing networks at mlo-cgo.
The percentage of missing months for each network, and scatter in the composite differences for different 170 historic periods are tabulated in Supplement S1.
The monthly composite CO2 differences are shown in Figure 3 (and tabulated in Supplement S5). The small error bars represent the ensemble standard deviation. (The one exception is for cgo-spo in Feb 2009, with only the NOAA network contributing. It is arbitrarily assigned 100% uncertainty and appears as an outlier in Figure 3c). The seasonality at mlo, generally attributed to the NH forest photosynthesis/respiration cycle, is the 175 dominant variation in IH CO2. The composite uncertainties are small compared to seasonal amplitudes, especially for the IH differences. Average standard deviations of mlo-cgo, mlo-spo and cgo-spo are 0.25, 0.23 and 0.16 ppm respectively. Systematic inter-annual variability is well-defined and is reflected similarly in both IH records and is consistent with mlo driving most of the seasonal variation.


Unusually low boreal summer/autumn IH minima also occur in 1993-1994. Apart from being a period when measurement and calibration methods were consolidating (as discussed next section) the most significant volcanic influence (Pinatubo) is potentially an influence at this time.
A question arises as to how well mlo data represents the NH. Of more relevance to this study is how well do the mlo samples represent air that is transferred into the Southern Hemisphere. Flask samples are collected at mlo 195 above 3 km altitude in down-slope winds, close to the upper troposphere regions where the IH transfer processes defined in FF18 occur (see Figure 5 below), circumstances not shared by other NH surface monitoring sites.
Unlike in typical growth rate analyses, the peak and trough values are largely independent. This is visually explored in Figure 3 using plotting software spline polylines linking peaks (solid) and dips (dashed) months of IH CO2. Trace gas mixing within extra-tropical hemispheres is typically estimated at 1-2 months or less, and 200 inter-hemispheric exchange times estimated at 6-12 months or more (e.g. Bowman and Cowan, 1997;Jacob, 1999). Monthly changes in the peak and trough IH CO2 largely reflect flux changes in or out of the extratropical Northern Troposphere close to that month. The following sections seek similarities with possible causal forcing processes.

Processes influencing CO2 IH difference variations 205
Global carbon cycle models generally attribute short term variations in atmospheric CO2 to exchanges with the terrestrial biosphere (Le Quéré et al., 2018;Yue et al., 2017) and implicitly assume model atmospheric transport is correct on all time frames. While the models have demonstrated an impressive ability to predict mid-to-high latitude CO2 variations influenced by weather, it is less clear that short term variations in IH exchange (of a magnitude sufficient to influence hemispheric growth rates) have been 210 adequately captured. Exchange (CABLE) model (Kowalczyk et al., 2006;Haverd et al., 2018). In addition, extra-tropical (ET) NBP 220 from an ensemble of 16 Land Surface Models (shown in Figure 2 of Bastos et al., 2018) are considered. Because of the small SH contribution, the ensemble ET values are most comparable to CABLE NH NBP. Note: We do not discuss air-surface fluxes derived from CO2 data that are less spatially representative, and/or rely on atmospheric transport modelling. The latter introduce additional model degrees of freedom and potentially overestimate terrestrial variability if the variability in atmospheric IH transport is not adequately captured.

225
NBP signs are reversed and are described as terrestrial-to-air carbon fluxes. Global wildfire emissions from the Global Fire Emissions Database (Randerson et al., 2018, GFED4.1) from 1997-2015 are classified as NH, EQ and EQ/SH. Seasonal anthropogenic emission anomalies are calculated as differences from the detrended 2000 to 2016 monthly data of Oda et al., (2017). For each data set, anomalies (in PgC month -1 ) in seasonal behaviour for each latitude band were determined by subtracting the mean seasonality from the monthly values.

230
The major seasonal anomalies in NBP and Wildfire emissions that potentially influence IH CO2 are shown in Figures 4(a) and 4(b). The largest anomalous surface-to-air flux is the extreme equatorial emission anomaly from equatorial wildfire in late 1997 (~0.9 PgC over 3 months); it is not associated with unusual behaviour in the IH CO2 records.
Despite mixing of CO2 within the extra tropical Northern Hemisphere being as rapid as 1-2 weeks (Jacob, 1999) 235 compared to IH exchange times of greater than 6 months (Bowman and Cowan, 1997), we see strong correlations with transport for unlagged 3-month averages. And since IH CO2 peaks re-occur within 1 month of the same time each year, close correspondence in timing of terrestrial anomalies and the IH CO2 peaks would be expected if NH terrestrial exchange was the main determinant. This is not evident in Figure 4.
More importantly, the amplitude range of terrestrial anomalies appears to be far too small to account for the 240 magnitude of the changes in the peaks and dips of IH CO2.
Over the last 25 years the annual relationship between global (mainly NH) fossil fuel combustion emissions and IH CO2 has been 2.8 PgC ppm -1 (equivalent to the 0.36 ppm (PgC) -1 used by FF18). This is applicable when northern fossil fuel emissions effectively mix globally. The volume of the troposphere north of Mauna Loa is around 33% of the global troposphere, so that on the shorter time frame of within-hemisphere mixing, only 245 ~0.92 PgC is required to change the NH background CO2 by 1 ppm. In Figure 4(c) we round this to 1 PgC = 1 ppm for simplicity.
The variability in the air-surface fluxes, relative to that in IH CO2, is displayed in Figure 4(d), which plots the standard deviations of residuals from the mean seasonality, for each month over the available record. As for the peaks and dips, we assume a 1:1 relationship between ppm and PgC month -1 in IH CO2. The main variation in 250 IH CO2 occurs in Mar-Apr, when variability in surface-air fluxes is small but variability in eddy IH exchange is large (see below). A second peak in IH CO2 standard deviation occurs Aug-Sep, around the time of the dips (but also when equatorial wild fires are more active suggesting a possible contribution from the equatorial emissions at this time?).
Accepting the precision and near-hemispheric spatial representation of the composite IH CO2 records, these 255 inconsistences with surface emissions in both timing and magnitude suggest that there are other short-term influences on IH CO2 of greater magnitude than air-surface exchange.

Wind indices reflecting CO2 IH transport:
In contrast to the case for air-surface exchanges, there are a number of prominent features in the composite IH CO2 records that are shared with behaviour in the dynamical indices of FF18. Inter-hemispheric exchange of 260 CO2 occurs mainly by eddy processes in the boreal winter-spring and by mean convection and advection associated with the Hadley circulation in the boreal summer-autumn (FF18 and references therein). FF18 developed wind indices that characterize both types of IH transport based on reanalysis data sets focusing on the National Center for Environmental Prediction (NCEP) and National Center for Atmospheric Research (NCAR) reanalysis (NRR) data (Kalnay et al., 1996). Eddy transport is described by uduct, the average 300 hPa zonal 265 velocity in the Pacific Westerly duct region (Frederiksen and Webster, 1988) of 5⁰N to 5⁰S, 140 to 170⁰W (FF16, FF18). Here we use that index and two of the four indices for mean transport introduced in FF18. These are P, the average 300 hPa vertical velocity in pressure coordinates in the region 10-15⁰N, 120 to 240⁰E, and vP the average 200 hPa meridional velocity in the region 5-10⁰N, 120 to 240⁰E. Figure 5  The top panel in Figure 6a shows a 3-decade time series of the uduct index which characterizes cross-equatorial Rossby wave dispersion, Rossby wave breaking and corresponding increases in transient kinetic energy and eddy transport in the near-equatorial upper troposphere (Webster and Holton, 1982, Frederiksen and Webster, 1988, Ortega et al., 2018. The large scale Rossby waves are generated by thermal anomalies and topographic 275 features including the Himalayan mountains from which they propagate south-eastward and are able to penetrate into the SH when uduct is positive, corresponding to an open Pacific Westerly duct.
The P and vP indices in Figures 6b and 6c describe the strength of the mean transport by the Hadley cell in the Pacific region with negative P corresponding to uplift and negative vP to north to south transport.
Net interhemispheric trace gas exchange requires a partial pressure difference between hemispheres. For CO2 280 the average seasonal cycle of 25-year mean partial pressure difference, represented here by monthly baseline mlo-cgo, is shown in Figure 7(a) (mlo-spo is not shown here since, reflecting on data quality, it is effectively identical).
The positive mean IH CO2 is largely due to fossil fuel emissions. Months of positive (north-south) IH difference are shaded green and only in Sep-Oct is there a small reverse gradient. Transport of CO2 from the 285 Northern to the Southern Hemisphere occurs when green shaded areas in Fig. 7(a) coincide (on average) with blue shaded areas ( Fig. 7(b), via eddy transfer with index uduct), or with red shaded areas (Figs. 7(c) and 7(d), via mean transport with indices P and vP). In 2010, the IH CO2 exceeds the average between Feb-Jul ( Fig. 7(a)) with reduced eddy transfer between Feb-Apr, associated with lower that average uduct ( Fig. 7(b)). Further, between Jun-Sep, there is weaker ascent (Fig. 7(c)) and north to south upper tropospheric wind ( Fig. 7(d)) in the key regions defining P and vP. As noted in FF18, the IH CO2 eddy and mean transports reinforce to contribute to the unprecedented 2009 to 2010 step in IH CO2.

295
 In 2016, the IH CO2 is larger than average between Feb-Jun and smaller than average between Jul-Oct ( Fig. 7(a)). These results are again consistent with the behaviour of the dynamical indices. There is reduced IH CO2 eddy transfer in the first half of the year (Fig. 7(b)) but very strong mean transport in the second half of the year (Figs. 7(c) and(d)) that accounts for the annual IH CO2, as noted in FF18.


In 1998, the IH CO2 exceeds the average from May-Dec and is close to the mean annual cycle for the 300 rest of the year. We note from Figure 7 that the annual increase in IH CO2, also shown in Figure 2 of FF18, is largely induced by the Jun-Aug mean Hadley circulation.
 It is suggestive that the relative variation in IH CO2 Feb-May for the three big El Niño years matches that in uduct, however it is puzzling that the largest uduct anomaly, 1998, is when IH CO2 is closest to the mean behaviour. The fact that the mean transport indices at this time of year are also consistently well 305 below their long-term average is also of note, since with uduct close to zero and -P, -vP indicating descent and south to north meridional winds, there is no obvious mechanism for IH exchange in this season. Yet, over the 25 years correlation of the Apr-May IH CO2 peaks with -P-vP is significant, r ≈ 0.4. One possible explanation for these behaviours in the early part of the Boreal winter/Austral summer may be found in changes in the volume of the well-mixed portion of the Northern Hemisphere (see Discussion, Section 7).
Different responses of IH CO2 to wind indices at different ENSO events, and from non-ENSO periods, are discussed in Section 7.
As an aside, we also include a similar plot for the average SH cgo-spo differences in Figure 8. Despite some concerns about artefacts in spo data (e.g. due to long flask-air storage times), all networks indicate that on average spo baseline CO2 exceeds that at cgo in the austral summer months. The minimum cgo-spo appears to precede inversion estimates of Southern Ocean CO2 uptake south of 30⁰S (Lenton et al., 2013). High precision continuous CO2 monitoring across the Southern Ocean (Stavert et al., 2019, and personal communication) confirm small and relatively smooth seasonal variation. The earlier Nov-Dec minimum in CO2 difference coincides with a seasonal dip in fossil fuel emissions (Oda et al., 2017) perhaps indicating an alternative 320 explanation.
6 Year-to-year variation in the composite records The annual net impacts of the various potential influences on site IH CO2 (when typical terrestrial biosphere seasonal variations are balanced) appear in Figure 9.


The linear regression through the 25-year mlo-cgo annual data gives a slope of 0.067 ± 0.006 ppm yr -1 compared to that through monthly values of 0.56 ± 0.021 ppm yr -1 in Figure 3, or through the peaks of 0.087 ± 0.011 ppm yr -1 or the dips of 0.049 ± 0.011 ppm yr -1 . We interpret this as indicating the combined long-term influence of both eddy and mean transport on the annual mean IH CO2.


In 2017, the IH difference is close to the 3-decade trend, with the duct open and Hadley strength returning to be close to its long-term mean. measurement and sampling error (as indicated by the composite standard deviations) thus requiring biogeochemical explanation. This discussion focusses on the potential of IH transport measured by wind indices to explain major features in IH CO2 variation, with emphasis on periods and events when they are likely to be the dominant influence on IH CO2. It complements the more general statistical analyses in FF18. In Figure 6, decreasing uduct acts to lessen eddy IH exchange and increase IH CO2, while the increasing Hadley circulation (decreasing vP and P) decreases IH CO2.  Frederiksen and Webster 1988 and discussed in FF16 and FF18. Note that in Figure 6, there is no precedent for similar sustained opposing behaviour in the two modes of IH transfer. The trend and lack of scatter in 2010-2014 IH CO2 can be understood by the IH CO2 fluxes being significantly larger than air-surface exchanges at the time.

8 Conclusions
Over the last 25 years there is a high degree of agreement in the measurement of monthly spatial differences in background CO2 levels by three measurement laboratories using four different sampling methodologies and sampling frequencies. Geographic isolation of sample collection sites and consistent sophisticated background selection over the 25 years, as well as coincident monitoring of a wide range of atmospheric species, excludes local and regional influence on CO2 at mlo, spo and cgo to an extent not generally available at other surface monitoring sites.
The temporal variation in the composite IH CO2 exhibit several systematic features on monthly to multi-year timeframes that are not reflected in independent evidence of air-surface exchange but do correspond to features in dynamical indices selected to represent both eddy and mean IH exchange. The comparisons in this paper 400 imply a major role for IH exchange of CO2 in NH growth rate variations.
The evidence for a significant influence of atmospheric dynamics on the CO2 IH gradient has relevance for global carbon cycle studies. It implies that both eddy and mean transport processes, and volume effects, need to be specifically included in transport model simulations, since the balance between the two is constantly changing, particularly in El Niño periods when eddy transport is reduced. It also means that El Niño events may 405 be a poor predictor of the carbon cycle behaviour in non-ENSO years.
Global carbon cycle model simulations should be able to reproduce the major features identified here in the composite IH records if the re-analyses transport is correctly implemented. In attempting to simulate the composite differences, one complication is model selection of baseline that matches the flask sampling criteria.
While monthly baseline averages appear to succeed in this respect, a more comprehensive treatment (outside the 410 scope of this study) based on individual flask measurements rather than monthly averages, and other trace gas observations (FF16, FF18), and in particular radon (Chambers et al., 2016), could possibly improve this process.  (Kalnay et al., 1996) The Supplement related to this article is provided
RJF generated the composite records and their analyses while JSF provided information on atmospheric dynamics and the roles of transport mechanisms. LPS and RLL contributed CO2 measurement quality assessments. All four authors contributed to the written document.
work. From CSIRO: Vanessa Haverd was generous with her time in providing regional groupings and updates of the CABLE DVM data; Ying-Ping Wang also provided other DVM data for scrutiny; Paul Krummel and Nada Derek advised on data processing and graphics, and Cathy Trudinger and Rachel Law on the global carbon budget. The dynamics contributions were prepared using data and software from the NOAA/ESRL Physical Sciences Division website at http://www.esrl.noaa.gov/psd/. Helpful comments from two anonymous referees are acknowledged.       605 Figure 9: Annual changes in the baseline CO2 difference between sites. Interhemispheric differences are plotted on the left axis. The peak magnitudes of strong El Niños (brown, ONI index > 1) and strong La Niñas (purple, ONI index <-1) are indicated. The cgo-spo annual differences are plotted on a doubled right-hand scale. Annual Fossil Fuel emissions from FF18, are shown on the top right axis.     : Annual changes in the baseline CO2 difference between sites. Interhemispheric differences are plotted on the left axis. The peak magnitudes of strong El Niños (brown, ONI index > 1) and strong La Ninas (purple,