Conclusions References Tables Figures

Abstract. We quantify the relative roles of natural and anthropogenic influences on the growth rate of atmospheric CO2 and the CO2 airborne fraction, considering both interdecadal trends and interannual variability. A combined ENSO-Volcanic Index (EVI) relates most (~75%) of the interannual variability in CO2 growth rate to the El-Nino-Southern-Oscillation (ENSO) climate mode and volcanic activity. Analysis of several CO2 data sets with removal of the EVI-correlated component confirms a previous finding of a detectable increasing trend in CO2 airborne fraction (defined using total anthropogenic emissions including fossil fuels and land use change) over the period 1959–2006, at a proportional growth rate 0.24% y−1 with probability ~0.9 of a positive trend. This implies that the atmospheric CO2 growth rate increased slightly faster than total anthropogenic CO2 emissions. To assess the combined roles of the biophysical and anthropogenic drivers of atmospheric CO2 growth, the increase in the CO2 growth rate (1.9% y−1 over 1959–2006) is expressed as the sum of the growth rates of four global driving factors: population (contributing +1.7% y−1); per capita income (+1.8% y−1); the total carbon intensity of the global economy (−1.7% y−1); and airborne fraction (averaging +0.2% y−1 with strong interannual variability). The first three of these factors, the anthropogenic drivers, have therefore dominated the last, biophysical driver as contributors to accelerating CO2 growth. Together, the recent (post-2000) increase in growth of per capita income and decline in the negative growth (improvement) in the carbon intensity of the economy will drive a significant further acceleration in the CO2 growth rate over coming decades, unless these recent trends reverse.


Introduction
Atmospheric CO 2 concentrations have risen over the last 200 years at an accelerating rate, in response to increasing anthropogenic CO 2 emissions.The resulting CO 2 dise-quilibrium has led to uptake of CO 2 from the atmosphere by land and ocean CO 2 sinks, which currently remove over half of all anthropogenic emissions and thereby provide a strong negative (stabilising) feedback on the carbon-climate system (Gruber et al., 2004;Sabine et al., 2004).The CO 2 airborne fraction (the fraction of total emissions from fossil fuels and land use change accumulating in the atmosphere) has averaged 0.43 since 1959, but has increased through that period at about 0.24% y −1 (Canadell et al., 2007).These interdecadal trends in CO 2 growth rate and the airborne fraction are the outcome of a race between two groups of forcing factors: the social, economic and technical drivers of anthropogenic emissions (including population, wealth and the carbon intensity of the economy), and the biophysical drivers of trends in land and ocean sinks.
The CO 2 growth rate also varies strongly at interannual (∼1 to ∼10 y) time scales, through mainly biophysical mechanisms.Fluctuations in CO 2 growth rate correlate with the El-Ni ño-Southern-Oscillation (ENSO) climate mode (Keeling and Revelle, 1985;Keeling et al., 1995;Jones and Cox, 2005), because the terrestrial carbon balance in tropical regions is tilted from uptake to release of CO 2 during dry, warm El-Ni ño events (Zeng et al., 2005;Knorr et al., 2005).Volcanic events are also significant: the CO 2 growth rate decreased for several years after the eruption of Mt.Pinatubo in June 1991 (Jones et al., 2001), probably because of increased net carbon uptake by terrestrial ecosystems due to higher diffuse solar radiation (Gu et al., 2003) and cooler temperatures (Jones and Cox, 2001) caused by volcanic aerosols.This interannual variability in the CO 2 growth rate is important for two reasons: it indicates mechanisms that govern the land and ocean CO 2 sinks, and it masks important longer-term trends in the CO 2 growth rate with strong variability at higher frequencies.
In this paper we investigate the combined anthropogenic and biophysical drivers of atmospheric CO 2 growth rates, with three aims.First, we obtain a simple quantification of the leverage of ENSO and volcanic signals on global CO 2 sinks at interannual time scales, using a combined ENSO-Volcanic Index (EVI).Second, we analyse observed interdecadal trends in the CO 2 airborne fraction by removing the interannual variabil-2869 ity associated with the EVI from several CO 2 records, confirming and extending the preliminary findings of Canadell et al. (2007).Third, we introduce an extended form of the Kaya identity which combines the biophysical and anthropogenic drivers of CO 2 growth, and use it both to diagnose the drivers of past trends and offer some indicative estimates of future CO 2 growth rates.

Atmospheric CO 2 budget and airborne fraction
The global atmospheric CO 2 budget is written as where C a = ν a [CO 2 ] is the mass of atmospheric CO 2 (with [CO 2 ] the atmospheric CO 2 mole fraction and ν a = 2.127 PgC ppm −1 ); C a = d C a /d t is the growth rate of atmospheric CO 2 (with primes denoting time derivatives); F E is the total anthropogenic CO 2 emission flux including emissions from fossil fuels (F Foss ) and net emissions from land use change (F LUC ); and F S is the total surface-air exchange flux including land-air and ocean-air fluxes (F LandAir and F OceanAir ).All fluxes are positive into the atmosphere, so F S < 0 in the current era and the total CO 2 sink is −F S .
The CO 2 airborne fraction, the fraction of emissions accumulating in the atmosphere, has two extant definitions based respectively on total anthropogenic emissions from both fossil fuels and land use change (F E = F Foss + F LUC ), and on fossil-fuel emissions only (F Foss ): where the subscript denotes the normalising flux.The former (a E ) is the "total" airborne fraction, while the latter (a Foss ) has been called the "apparent" airborne fraction (Oeschger et al., 1980;Enting, 2007).Similarly, a sink fraction (the fraction of emissions taken up by land and ocean sinks, −F S ) can be defined in two ways as s E = −F S /F E (total) and s Foss = −F S /F Foss (apparent).The relationships between the respective airborne and sink fractions are The total airborne fraction a E is preferable in principle to the apparent a Foss , for two reasons.First, a E is the ratio of total response of the atmospheric carbon cycle (C a ) to total forcing (F E ), whereas a Foss is the ratio of total response (C a ) to a partial forcing (F Foss ), omitting F LUC .Second (and in consequence), the total airborne and sink fractions add to 1, so trends in a E are always opposite to trends in s E and either fraction is a direct measure of the outcome of the combined influences of total emissions and total sinks on the CO 2 growth rate.The apparent airborne and sink fractions do not have this property because the additional forcing from land use change has to be included separately (Eq.3).Longstanding use of the apparent airborne fraction was originally motivated not from basic considerations but by the methodological problem of lack of knowledge of F LUC .
However, the situation has now changed with improved data, especially from satellites.Recent estimates of F LUC have converged on 1.5±0.5 PgC y −1 for 2000-2006, compared with F Foss ≈ 7.6±0.4and C a ≈ 4.1±0.1 PgC y −1 over the same period (Canadell et al., 2007).
In this paper a E is used as the primary measure of airborne fraction, but results are also given for a Foss .

Data
We used the following data for the period 1959 to 2006 (see Appendix A for sources and details): The analysis was done at a monthly time step, with slowly varying annual data (emissions, population, GDP-PPP) interpolated to monthly (details in Appendix A). 3 Interannual variability of CO 2 growth rate 3.1 Spectral structure of CO 2 growth and ENSO Figure 1 shows normalised cumulative spectra and cospectra of the CO 2 growth rate and each of the five ENSO indices (Ni ño3,Ni ño3.4,Ni ño4,SOI,and MEI).Normalised (co)spectra show the fractional contribution to the (co)variance from frequencies less than a given frequency (see Appendix B for analytical details).All of the covariance between C a and any of the five ENSO indices is spectrally band-limited to frequencies in a narrow window between ∼0.2 and ∼0.8 y −1 (periods from ∼5 to ∼1.25 y).Spectral components of C a and ENSO indices at higher frequencies add nothing to the covariance, their only effect being to degrade the correlation by adding uncorrelated high-frequency noise.It is therefore useful to filter out the high-frequency noise for diagnosis of the relationship between ENSO and carbon fluxes.Henceforth all time series are lowpass-filtered with a Fourier-transform filter which removes frequencies f >0.8 y −1 or periods <15 months (see Appendix B for details).

Correlations between surface-air exchange flux, ENSO and volcanic activity
The mechanistic links between ENSO, volcanic activity and the CO 2 budget occur through the total (land plus ocean) surface-air exchange flux F S = C a − F E , rather than through C a .Therefore we examine lagged correlations between F S (rather than C a ) and ENSO and volcanic indices.The lagged correlation between time series X (t) and where τ is the time lag, • denotes an average over time t, and σ X and σ Y are the standard deviations of X and Y .Lagged correlations between the five ENSO indices and F S (Fig. 2, left) confirm the well-known relationship (Keeling and Revelle, 1985, Keeling et al., 1995, Jones and Cox, 2005) between ENSO and CO 2 growth rate.Peak correlations between ENSO and F S (using C a at MLO) depend on the choice of ENSO index, ranging between 0.62 for Ni ño3 and 0.45 for Ni ño4.The peak correlation is positive (so positive ENSO index anomalies, corresponding with dry, warm El-Ni ño events, are associated with positive anomalies in F S or negative anomalies in the total sink −F S ).The peak occurs when F S lags the ENSO index by 3±1 months.
To include the influences of both ENSO and volcanic activity on CO 2 fluxes and growth rate, we define an ENSO-Volcanic Index (EVI) as the linear combination where ENSOI is an ENSO index normalised to zero mean and unit variance; VAI is the global Volcanic Aerosol Index, a measure of volcanically-induced aerosol optical 2873 depth (Ammann et al., 2003); λ is the weight for VAI relative to ENSOI; and τ is the ENSO lag time, a measure of the time for ENSO to affect the CO 2 exchange flux F S .It is assumed that the VAI affects F S without time lag.Five alternative versions of the EVI are obtained, corresponding to the five ENSO indices.The EVI depends on two parameters, λ and τ, both of which are well constrained.From Fig. 2 (left) we used τ=3 months for all ENSO indices, so that the maximum correlation between EVI and F S occurs near t=0.The weight λ was chosen so that the EVI explains as much as possible of F S , which occurs when λ takes the value maximising the correlation between EVI and F S .For all five EVI this is close to λ = −16, the value used hereafter.Use of the EVI in place of an ENSO index increases the peak correlations with F S substantially (Fig. 2, right).With F S calculated from C a at MLO and an EVI defined from Ni ño3, the peak correlation is 0.75.Figure 3  The trend in a E was estimated using a stochastic method which accounts for temporal correlation in the time series (see Appendix C for details).The trend is expressed here as a proportional growth rate, defined for a time series X (t) as r(X ) = X /X , with units % y −1 .
The GLA series for 1959-2006 yielded a mean a E of 0.43 and a proportional growth rate r(a E )=0.24%y −1 (with 5% and 95% confidence limits −0.18 and 0.64% y −1 and probability P=0.81 of a positive trend).The result from the GLB series was nearly identical.This result does not provide an unambiguous, statistically robust determination of the trend in a E .

Noise reduction
Detection of trends in a E can be improved in statistical significance by removing the interannually varying component which is causally linked with ENSO and volcanic activity, using the EVI.
We write an arbitrary time series X (t) as the sum of trend (X T ), mean-annual-cycle (X C ) and anomaly (X , where X U is a noise component uncorrelated with the EVI and X E is linearly dependent on the EVI.This component is where µ is the sensitivity of X to the EVI, and use of the anomaly component EVI A ensures that X E (t) has zero mean, no trend and no annual cycle.The full decomposition is thus When X is a time series over N monthly time points t n (n=1,. . .,N), the components are given algorithmically by: where the trend is defined by fitting a polynomial P to X (t n ), • denotes an average over the record, and < • |condition> denotes a conditional average.
The noise-reduced version of X (t), denoted with a superscript (n), is given by subtracting out the externally-forced components X C and X E = µEVI A : The trends of the noise-reduced and original series are identical because the components removed have zero mean and no trend, but the variability of the new series is lower, improving the statistical significance of trends.This decomposition was applied to the CO 2 sink F S , yielding noise-reduced series and airborne fraction a and EVI, using an EVI defined from Ni ño3, the resulting sensitivity is µ = 0.9.With noise reduction, the GLA series for 1959-2006 yielded a proportional growth rate in total airborne fraction, r(a E (n) ), of 0.24% y −1 (5% and 95% confidence limits −0.04 and 0.50% y −1 ; probability P=0.92 of a positive trend), around a mean a (n) E of 0.43.The result with the GLB series is similar but with a slightly lower P of 0.88.Noise reduction therefore does not change the mean result from the above initial trend estimate but provides improved statistical reliability, raising P from 0.81 to about 0.9.This more complete analysis with multiple CO 2 series confirms our earlier result (Canadell et al., 2007) which was derived from the GLB series.

Uncertainties and implications
The most uncertain quantity affecting the growth rate r(a E ) is F LUC , for which the current best estimate is F LUC ≈1.5±0.5 PgC y −1 for 2000-2006 (Canadell et al., 2007).Our observed trend in a E changes from positive to negative if F LUC is reduced to 40% or less of the best estimate, that is, to 0.6 PgC y −1 or less for 2000-2006 with equivalent proportional reductions in earlier years.However, such a large reduction is well outside the present uncertainty range for F LUC .
We also determined the trend in the apparent airborne fraction (a Foss ), even though a E is the more fundamental carbon-cycle attribute for reasons given in Sect. 2. The proportional growth rate of a Foss for 1959-2006 is small and negative, with r(a Foss ) ≈ −0.2±0.2%y −1 around a mean of 0.57.The different trends in a E and a Foss are easily understandable by noting that r(a Foss ) is the sum of r(a E ) and the growth rate r(a Foss /a E ) in the ratio of the two airborne fractions.This ratio, a Foss /a E = 1 + F LUC /F Foss , decreased fairly steadily through 1959-2006 at a rate r(a Foss /a E ) ≈ −0.4 % y −1 (around an average a Foss /a E of 1.32) because F Foss grew more quickly than F LUC (Canadell et al., 2007).The decreasing trend in F LUC /F Foss therefore accounts fully for the observed different signs in the growth rates of a Foss and a E .
Two further methodological checks were applied to all estimates of airborne-fraction growth rates.First, estimates of growth rates like r(a E ) were found to have some sensitivity to the exact starting and ending times of the CO 2 series used to determine C a .The extent of this sensitivity was investigated with an enhanced stochastic trend estimation method, in which bootstrap subsampling of the time series under test was used to reduce sensitivity to starting and ending times (see Appendix C for details).Results from this method were statistically consistent with those given above, confirming the robustness of the estimated trends.
Second, the entire analysis was also carried out using individual-station CO 2 series from MLO and SPO instead of the globally-averaged series GLA and GLB.Results

2877
were similar to those with the globally-averaged series, despite the fact that [CO 2 ] at MLO was higher than at SPO by an offset which increased from ∼1 ppm in the 1960s to ∼3 ppm in 2000-2005.By using an exponential-growth model for C a it can be shown that this offset accounts for a statistically insignificant difference in r(a E ) of about 0.06% y −1 .
The result that total airborne fraction increased over 1959-2006 implies that total sinks increased slower than total emissions.Using Eq. ( 1) to write the relationship between the growth rate −r(F S ) of total sinks (−F S ) and the growth rate r(F E ) of total emissions, we obtain: The numbers beneath each term give average values in % y −1 for 1959-2006.Both total sinks and total emissions grew significantly, but the observation of an increasing airborne fraction shows that sinks grew slightly slower than emissions.Use of the observed atmospheric CO 2 budget to partition the sinks into land and ocean components shows that the ocean fraction of the total sinks decreased substantially whereas the land fraction did not (Canadell et al., 2007).The observed increase in the airborne fraction is in not in agreement with available predictions of coupled carbon-climate models.The average prediction of 11 models (Friedlingstein et al., 2006(Friedlingstein et al., ) for 1959(Friedlingstein et al., -2006 is a decrease in a E at −0.27±0.36%y −1 , suggesting that these models tend to overestimate the rate of increase in total sinks as CO 2 concentrations rise.Equation ( 9) shows that this is a sensitive test for carbonclimate model predictions of trends in total sinks, because the sign of r(a E ) is determined by the small difference between the two larger quantities r(F E ) and −r(F S ).

Unified assessment of the drivers of CO 2 growth
To assess the relative effects on CO 2 growth of changes in airborne fraction and anthropogenic drivers of CO 2 emissions, we use an extended form of the Kaya identity.In its usual form (Nakicenovic et al., 2000, Nakicenovic, 2004, Raupach et al., 2007), the Kaya identity expresses global fossil-fuel CO 2 emissions as F Foss = P gef , where P is global population, g = G/P is per capita income or per capita GDP, e = E/G is the energy intensity of GDP, f = F Foss /E is the fossil-carbon intensity of energy, G is global GDP-PPP, and E is global primary energy consumption.An equivalent expression is F Foss = P gh Foss , where h Foss = F Foss /G = ef is the fossil-fuel carbon intensity of the global economy.
We modify this identity in two ways, first to describe total emissions (F E = F Foss + F LUC ) rather than F Foss .Land use change emissions can be written in Kaya form as F LUC = P gh LUC , where h LUC = F LUC /G is the land-use-change carbon intensity of the global economy, corresponding to h Foss above.The Kaya identity for total CO 2 emissions is then where h E = F E /G = h Foss + h LUC is the total carbon intensity of the global economy, accounting for both fossil fuels and land use change.Second, we describe the atmospheric CO 2 growth rate (C a ) by introducing the airborne fraction a E = C a /F E into Eq.( 10).The end result after both these changes is an extended Kaya identity in which a E appears as an extra factor: The proportional growth rates of factors in Eqs. ( 10) and ( 11) are related by 2879 because r(X ) = X /X yields r(X Y Z) = r(X )+r(Y )+r(Z) for any X , Y and Z.All terms in Eq. ( 12) have units time −1 .Note that r(C a ) = C a /C a is the proportional growth rate of the CO 2 growth rate, a measure of the second derivative of C a .Figures 4a and b respectively show time series of the factors in the Kaya identity for F E , (Eq.10) and the extended Kaya identity for C a (Eq.11) for the period 1959-2006, with series are normalised to 1 in 1980 so that trends can be compared.Figures 5a and  b show the corresponding proportional growth rates (Eq.12), with 7-year smoothing for clarity.Average growth rates of all factors, with 5% to 95% confidence intervals, are given in Table 1.
For F E (Figs. 4a and 5a) the average growth rate r(F E ) over 1959-2006 was 1.8% y −1 , with interannual variability from less than 0.5 to over 3% y −1 .This growth was driven by additive contributions of +1.7% y −1 from r(P ) (growth in population), +1.8% y −1 from r(g) (growth in income), and −1.7% y −1 from r(h E ) (reduction or improvement in the total carbon intensity of the global economy).Uncertainties in all these growth rates are low (0.1% y −1 or less; Table 1).
The CO 2 growth rate C a (Fig. 4b) is noisy, because of the interannual variability discussed above.Over the last five decades C a increased inexorably, reaching an average of C a ≈ 4.1 ±0.1 PgC y −1 or [CO 2 ] = 1.9 ppm y −1 through 2000-2006 (Canadell et al., 2007).The drivers of this increase can be expressed (Fig. 5b and Eq. ( 12) as contributions from the growth rates r(P ), r(g), r(h E ) and r(a E ) to r(C a ) = C a /C a , the growth rate of the CO 2 growth rate.Even with the 7-year smoothing used here, r(C a ) fluctuated strongly around a mean of +1.9% y −1 , with contributions from r(P ), r(g), r(h E ) and r(a E ) given in Table 1.

Discussion and conclusions
There were significant interdecadal trends in the emissions drivers P , g and h E through 1959-2006 (Fig. 5a and Table 1).Growth in population (P ) slowed from 2 to 1.2% y −1 .
Per capita income (g) grew more rapidly since 2000 than over the previous four decades, with r(g)=3.0%As noted elsewhere (Raupach et al., 2007, Canadell et al., 2007), these trends have together driven a substantial recent increase in the growth rate of total emissions, with r(F E )=3.0% y −1 over 2000-2006 compared with 1.9% y −1 over 1959-1999.The growth rate in F E (= F Foss +F LUC ) is slightly lower than the recent growth rate in fossil-fuel emissions (r(F Foss )=3.3% y −1 over [2000][2001][2002][2003][2004][2005][2006] because there has been no recent growth in the land-use-change emission (F LUC ).Similar trends appear in the growth rate of the CO 2 growth rate, r(C a ) = C a /C a (Fig. 5b and Table 1).Averaged over the whole period 1959-2006, most of the interdecadal trend (r(C a ) ≈ 1.9% y −1 ) was attributable to increasing emissions (r(F E ) ≈1.8% y −1 ), caused in turn by the growth rates of P , g and h E .A small component of r(C a ), about 0.2% y −1 out of 1.9% y −1 , was caused by the interdecadal growth in airborne fraction, r(a E ) (these figures do not satisfy Eq. ( 12) exactly because of statistical uncertainties and roundoff errors).
Most of the strong interannual variability in r(C a )) originates from variability in the CO 2 exchange flux F S and thence the airborne-fraction term in Eq. ( 11).Much of this variability in turn is associated with the EVI.Subtracting the EVI-correlated fluctuating component out of C a and a E as in Sect.4.2, we obtain a noise-reduced form of the extended Kaya identity, C (n) 5c shows the growth rates of extended Kaya factors with this noise reduction.The variability in each of r(C a (n) ) and r(a ) is about half of the equivalent variability without noise reduction (Fig. 5b).
The extended Kaya identity allows estimation of the relative impacts on future [CO 2 ] 2881 of likely trends in airborne fraction and the drivers of total emissions (P , g and h E ).To do this we consider the time interval ∆t x to reach a specified future concentration [CO 2 ] x at a given r(C a ) (the growth rate of the CO 2 growth rate).The interval ∆t x can be determined analytically (Appendix D).We take To reduce emissions and thence atmospheric CO 2 , it is necessary to reduce the growth rates of the emissions drivers P , g and h E in some combination.Growth in population (P ) is presently just over 1% y −1 and is forecast to decline to zero in the second half of the 21st century (Lutz et al., 2001).Growth in global per capita income (g) is needed to improve quality of life in the developing world.This leaves the primary option as increasing the negative growth rate in carbon intensity (h E ).To achieve a reduction rate in total emissions of −2% y −1 (which would halve emissions in 35 years) in the presence of global growth rates of 2% y −1 in g and 1% y −1 in P , it is necessary to achieve a decline in h E at a rate of around −5% y At a given r C , the time to reach [CO 2 ] x is x when the rate of increase in [CO 2 ] is held steady at its initial value [CO 2 ] 0 .If r C > 0, then ∆t x is less than T .In the limit r C → 0, ∆t x approaches T .
Table 1.Proportional growth rates (r(X ) = X /X , in % y −1 ) of factors in the Kaya identity (F E = P gh E ) and the extended Kaya identity (C a = P gh E a E ), for periods 1959-2006, 1959-1999 and 2000-2006 (inclusive of end years).Errors denote approximate 5% to 95% confidence intervals.Where not shown, errors are less than 0.1% y −1 .Roundoff errors are responsible for slight departures from Eq. ( 12).
Period 1959-2006 1959-1999 2000-2006 r(F E ) 1.8 1.9 3.0 r(P ) 1.7 1.7 1.2 r(g) The sensitivity µ was chosen to minimise the variance of F A , thus placing as much as possible of the anomaly F A S into the EVI-correlated component.With lowpass-filtered series F S

Fig. 4 .
Fig. 4. (a) factors in the Kaya identity, F E = P gh E , with F E (total emissions) in black, P (population) in red, g (per capita GDP-PPP) in green and h E (carbon intensity of GDP-PPP) in dark blue.(b) factors in the extended Kaya identity, C a = P gh E a E , with C a in black, a E in sky blue and other colours as in left plot.All factors are normalised to 1 in 1980. 2895

Fig. 5 .
Fig. 5. (a) proportional growth rates (% y −1 ) of factors in the Kaya identity, F E = P gh E ; (b) growth rates of factors in the extended Kaya identity, C a = P gh E a E ; (c) growth rates of factors in the noise-reduced version of the extended Kaya identity, C (n) a = P gh E a (n) E , where ((n)) denotes removal of the EVI-correlated fluctuating component.All growth rates are smoothed with a 7-year running mean.Colours match Fig. 4.
annual global CO 2 emissions F Foss and F LUC ; compares peak correlations between the ENSO indices and F at both MLO and SPO.Correlations are slightly lower at SPO than MLO, but are still increased by using the EVI rather than corresponding ENSO index.Since λ is negative, a positive anomaly in the VAI component of the EVI is associated with a positive anomaly in the sink −F S (while a positive anomaly in the ENSO component is associated with negative anomaly in −F S as noted above).Two estimates were used (see Sect. 2 and Appendix A): from the average of the MLO and SPO CO 2 series with annual cycle removed (denoted GLA), and from a globally-averaged CO 2 series available from 1980 onward, augmented with MLO data before 1980 (denoted GLB).
S , and between the corresponding EVI and F S , usingC a E = C a /F E = 1 + F S /F E )provides a measure of the relationship between total CO 2 emissions and sinks.We estimated trends in monthly series of a E inferred from C a records from 1959 to 2006.Since a E is inherently globally aggregated, it is necessary to use estimates of a globally-averaged C a .
y −1 over 2000-2006 compared with 1.8% y −1 over 1959-1999.Also, the negative growth rate (improvement) in the carbon intensity of the economy (h E ) declined since 2000: r(h E ) was −1.2% y −1 over 2000-2006, compared with a mean of −1.7% y −1 over 1959-1999.(Figures for r(h E ) differ from Canadell et al., 2007 for two reasons: the use of GDP-PPP here and GDP-MER (Market Exchange Rate) there, and the inclusion of F LUC in h E here).