## 1. Introduction

Nearly 50 years of measurements identify several changes in the atmospheric concentration of carbon dioxide: an increase over time, an intra-annual cycle, and interannual changes in the intra-annual cycle. Between 1959 and 2004, the atmospheric concentration of carbon dioxide at Mauna Loa, Hawaii, increased from 316 to 377 ppm. Most of this increase is associated with the combustion of fossil fuels and anthropogenic changes in land use (Conway et al. 1994; Keeling et al. 1995; Houghton 2000; Marland et al. 2001; Prentice et al. 2000). Within each year, atmospheric concentrations of carbon dioxide rise and fall through an intra-annual cycle, with the amplitude increasing with latitude (Keeling et al. 1996). The intra-annual cycle is attributed to seasonal changes in the balance between net primary production and heterotrophic respiration in the terrestrial biota (Hall et al. 1975; Keeling et al. 1996; Francey et al. 1995; Morimoto et al. 2000). The intra-annual cycle changes from year to year, with a general increase in amplitude over time and changes in monthly values (Keeling et al. 1996; Dettinger and Ghil 1998).

Several analyses focus on interannual changes in the intra-annual cycle of atmospheric CO_{2}. Keeling et al. (Keeling et al. 1996) find that the amplitude of the intra-annual cycle increased most rapidly at high latitudes and that the phasing advanced by about seven days. Based on correlations between these changes and temperature, Keeling et al. (Keeling et al. 1996) argue that these observations are associated with temperature-driven changes in the terrestrial biota, specifically an earlier spring.

Dettinger and Ghil (Dettinger and Ghil 1998) use singular-spectrum analysis to analyze monthly patterns at both Mauna Loa and the South Pole. They identify several months in which the intra-annual cycle changes. Based on their timing and correlations with other time series, Dettinger and Ghil (Dettinger and Ghil 1998) write that these changes seem to be associated with tropical processes, especially sea surface temperature. Nonetheless, these correlations cannot be used to differentiate among possible causal variables, such as sea surface temperature, oceanic circulation, or changes in biotic activity.

Interannual changes in the intra-annual cycle are attributed to changes in 1) atmospheric circulation, 2) the flow of carbon between the atmosphere and ocean, and/or 3) the flow of carbon between the atmosphere and the terrestrial biota. The potential importance of atmospheric circulation is highlighted by Higuchi et al. (Higuchi et al. 2002), who argue that historical shifts in atmospheric circulation can cause annual variations in the amplitude of atmospheric CO_{2} that are similar to those observed at Mauna Loa and Point Barrow, Alaska. Murayama et al. (Murayama et al. 2004) conclude that interannual changes in atmospheric circulation disrupt the link between biotic changes and measurements taken at distant stations.

Bridging the link between atmospheric and oceanic circulation, a correlation between atmospheric concentrations of carbon dioxide and El Niño–Southern Oscillation (ENSO) events is first noted by Bacastow (Bacastow 1976) and is confirmed by Wignuth et al. (Wignuth et al. 1994), Keeling et al. (Keeling et al. 1995), and Kaufmann et al. (Kaufmann et al. 2006). The relative importance of oceanic and atmospheric circulation in the effect of ENSO events on atmospheric CO_{2} is uncertain. Wignuth et al. (Wignuth et al. 1994) suggest that changes in sea surface temperature may be responsible for correlations between ENSO events and atmospheric CO_{2}. Other analysts argue that changes in terrestrial vegetation generate the correlation between ENSO events and atmospheric CO_{2} (e.g., Ciais et al. 1995; Francey et al. 1995; Zeng et al. 2005). Siegenthaler (Siegenthaler 1990) suggests that ENSO events weaken the monsoon, which reduces carbon uptake by the terrestrial biota. Yang and Wang (Yang and Wang 2000) argue for the importance of clouds, which reduce solar radiation and may suppress net primary production. Rayner et al. (Rayner et al. 1999) argue for both ocean and terrestrial factors—oceanic changes generate negative anomalies while terrestrial responses generate positive anomalies.

Terrestrial biota affects the intra-annual cycle of atmospheric carbon dioxide via several mechanisms that include a fertilization effect due to the increased concentration of atmospheric CO_{2} (Kohlmaier et al. 1989), seasonal shifts in the phasing of photosynthesis and respiration (Chapin et al. 1996), and/or changes in the length or intensity of the growing season at mid- and high latitudes (Myneni et al. 1997).

Many of these mechanisms may be linked to changes in climate. Changes in temperature and precipitation have the potential to affect the intra-annual cycle by extending the growing season and/or increasing summer greenness (Zhou et al. 2001). To date, there is no agreement about which factor, if either, predominates. Several analysts argue for the importance of longer growing seasons. Tanja et al. (Tanja et al. 2003) find that the start of photosynthesis in the spring is correlated with air temperature. Keyser et al. (Keyser et al. 2000) find that warmer spring temperatures elongate the growing season and increase net primary production by up to 20% for sites in Alaska. Hollinger et al. (Hollinger et al. 2004) find that the quantity of carbon stored by spruce forests in Maine is related positively to warmer spring and falls. This effect is simulated by Randerson et al. (Randerson et al. 1999), who find a correlation between spring temperatures and early-season net ecosystem uptake. Churkina et al. (Churkina et al. 2005) suggest a linear relationship between annual net ecosystem exchange and the carbon uptake period.

Other analyses point to the importance of summer conditions, especially soil moisture, which may decline due to higher summer temperatures and/or reduced precipitation. For example, Zhou et al. (Zhou et al. 2003) find that summer precipitation has a measurable effect on interannual variations in satellite measures of surface greenness. Consistent with this result, White and Nemani (White and Nemani 2003) find that precipitation during the growing season has a greater effect on net ecosystem exchange than the length of the growing season. Correlations between tree rings and the normalized difference vegetation index (NDVI) in June and July (as opposed to months in spring or fall) seem to suggest the importance of summer (Kaufmann et al. 2004). Angert et al. (Angert et al. 2005) argue that drier summers suppress carbon uptake.

Nor is the effect of climate limited to vegetation—soil carbon pools also respond to changes in climate (Trumbore et al. 1996). Studies of heterotrophic respiration find that the period of soil thaw (Goulden et al. 1998) and heterotrophic respiration (Shibistova et al. 2002) are very sensitive to temperature. The duration of these temperature effects on soil respiration are uncertain—some analysts argue that acclimatization reduces the effect of temperature over time (e.g., Luo et al. 2001). Yet others argue that carbon flows from mineral soils are relatively unaffected by temperature (e.g., Giardina and Ryan 2000).

To sort among these potential causes for interannual changes in the intra-annual cycle of atmospheric carbon dioxide at Mauna Loa, I use statistical techniques to identify months in which the intra-annual pattern changes and to evaluate whether these changes are associated with changes in atmospheric circulation. Results identify a previously unnoticed change between 1965 and 2004 during October in which October concentrations decline relative to the cycle’s mean value. Results confirm a previously noted change in spring (April), but unlike previous analyses, this result indicates that between 1965 and 2004, April concentrations rose relative to the cycle’s mean value. Additional results indicate that the timing of these changes is not caused by commonly recognized patterns of atmospheric circulation, such as ENSO events. Instead, these changes may be generated by asymmetric effects of warming. Warming in early spring may increase heterotrophic respiration relative to net primary production, thereby causing atmospheric concentrations of carbon dioxide in April to rise relative to the cycle’s mean value. Conversely, warming in late summer–early fall may enhance net primary production relative to heterotrophic respiration, thereby causing atmospheric concentrations of carbon dioxide in October to fall relative to the cycle’s mean value. This explanation is preliminary, and identifying causal mechanisms for changes in April and October will require additional analyses.

## 2. Methodology

To identify the month(s) in which the intra-annual cycle of atmospheric CO_{2} changes, I use a statistical methodology that is designed to detect seasonal changes in economic time series (Canova and Hansen 1995). The methodology estimates a statistical model for the intra-annual cycle of atmospheric concentrations of carbon dioxide and then evaluates whether this model is a satisfactory representation or whether the model changes in a statistically meaningful fashion during the sample period. The analysis proceeds in three steps. First, I create a stationary time series for the atmospheric concentration of carbon dioxide at Mauna Loa by removing year-to-year increases from monthly observations. In the second step, this time series is used as the dependent variable in a statistical model that specifies months as explanatory variables. In the third step, regression errors are analyzed to test whether the regression coefficients associated with individual months are stable over the sample used to estimate the statistical model.

*Y*is the monthly anomaly for the atmospheric concentration of carbon dioxide for month

*m*in year

*t*, PEAK is the largest monthly observation for the atmospheric concentration of CO

_{2}in year

*t*(usually in March or April), and TROUGH is the smallest monthly observation for the atmospheric concentration of CO

_{2}in year

*t*(usually September or October).

Equations (1)–(3) generate a monthly time series that fluctuates around zero (Figure 1). The absolute difference between the peak and trough in Figure 1 represents the amplitude of the waxing or waning portion of the intra-annual cycle. This time series is stationary, as indicated by test statistics [(*π*/2) = 38.7, (2*π*/3) = 24.1, (*π*/3) =21.9, (5*π*/3) = 34.9, (*π*/6) = 13.6] developed by Beaulieu and Miron (Beaulieu and Miron 1993).

_{2}in Figure 1 is given by Equation (4):in which Mon are categorical variables that represent each of the 12 months (the variable equals one when the dependent variable corresponds to that month; it is zero otherwise),

*β*and

*γ*are regression coefficients that are estimated using ordinary least squares, and ε is the regression error. The number of lags (

*s*) is chosen using a likelihood ratio statistic (Sims 1980). For the purpose of detecting changes in the intra-annual cycle, Equation (4) is equivalent to fitting a trigonometric specification (Canova and Hansen 1995).

Equation (4) specifies lagged values of the dependent variable to account for serial correlation—a change in the intra-annual cycle during one month may “carry over” and change the intra-annual cycle during the subsequent month(s). This type of persistence would make it appear as though the cycle is changing in many months—this methodology identifies the month(s) in which changes in the intra-annual cycle originate.

Regression coefficients associated with the categorical variables (*γ _{i}*) represent the pattern of the intra-annual cycle. That is, the value of (

*γ*) indicates the value for

_{i}*Y*, everything else being equal. For example,

*γ*

_{9}should be the most negative because September often is the trough in the intra-annual cycle.

To identify months in which the intra-annual cycle changes in a statistically significant fashion, the regression coefficients (*γ _{i}*) are tested to determine whether they are stable over the sample period. If the regression coefficients for individual months do not change over the sample period, the monthly pattern is said to be stable. That is, the value of (

*γ*) is the “best predictor” for that month’s value of

_{i}*Y*over the entire period. The intra-annual cycle is said to change in a given month if the regression coefficient for that month is unstable over the sample period. In this case, (

*γ*) is not the best predictor for a month’s value of

_{i}*Y*over the entire sample—a better prediction can be generated by increasing or decreasing the value of (

*γ*) over the sample period.

_{i}To evaluate whether the regression coefficient for an individual month (*γ _{i}*) is stable over the sample period, I use a test statistic that is designed to detect changes in the seasonality of economic times series (Canova and Hansen 1995). This method focuses on the regression errors, which are the difference between the observed value of

*Y*and the value predicted by Equation (4). The regression errors (ε) from Equation (4) are compiled by month to generate a time series of regression errors for January, February, etc., through December. For example, regression errors for January between 1965 and 2004 are given in Figure 2a, regression errors for February between 1965 and 2004 are given in Figure 2b, etc.

*L*) that is given byin which ε is a time series of the regression errors from Equation (4) for an individual month (e.g., the regression errors for January),

_{a}*T*is the number of observations for that month, and

*w*(

*k*/

*m*)1 is an optimal weighting function that corresponds to the Bartlett window (Andrews 1991). Calculating the test statistic requires a continuous time series; therefore,

*L*is calculated using residuals for the longest period for which data are available, January 1965 through 2004. The Mauna Loa record is missing observations for February through April 1964.

_{a}As described by Canova and Hansen (Canova and Hansen 1995), *L _{a}* is essentially the statistic developed by Kwiatkowski et al. (Kwiatkowski et al. 1992). The null hypothesis of the test statistic

*L*is that the regression coefficient is stable. This null hypothesis is evaluated against the generalized Von Mises distribution with one degree of freedom (monthly coefficients are tested individually). Values of

_{a}*L*that reject the null hypothesis indicate that the regression coefficient for that month is unstable. Such a result indicates that the intra-annual pattern “changes” during that month. The nature of this change is indicated by a general increase/decrease in the time series of regression errors. For example, if the regression error for April generally increases through the sample period (Figure 2d), this would indicate that a change in the intra-annual cycle in April causes

_{a}*γ*

_{4}in Equation (4) to overstate the observed value for April early in the sample period and understate the observed value for April late in the sample period.

## 3. Results

The monthly frequency of the sample period from January 1965 to October 2004 contains 477 observations. Based on an econometric “rule of thumb” of *T*^{1/3}, a maximum lag length of 8 months is considered for estimating Equation (4). The Sims likelihood ratio test rejects a restriction that would reduce the lag length from four lags to three lags [*χ*^{2}(1) = 10.27, *p* < 0.01]. A restriction that would reduce the lag length from six lags to five lags is just shy of the 5% threshold [*χ*^{2}(1) = 3.74, *p* > 0.053], nonetheless, Equation (4) is estimated using four lags. The results described below do not change significantly if six lags are used to estimate Equation (4).

The large *r* ^{2} (0.98) indicates that Equation (4) is able to account for most of the variation in the intra-annual cycle during the sample period. The coefficients associated with 9 of the 12 monthly categorical variables (*γ _{i}*) are statistically different from zero (Table 1), which indicates that the statistical model is able to quantify the “average” monthly pattern. The exceptions are January, July, and December. These months often are the midpoint for the waning and waxing phase of the intra-annual cycle. As such, values of

*Y*for these months are close to zero, hence the value for

*γ*is not statistically distinguishable from zero. Regression coefficients (

*β*) associated with the lagged dependent variable are statistically different from zero (Table 1). This indicates that changes in the intra-annual cycle carry over from one month to the next. But the value associated with the previous month (

_{i}*β*

_{1}= 0.49) indicates that changes in the intra-annual cycle decay quickly.

Visual inspection of Figure 2 suggests several months in which the regression error increases or decrease over the sample period. This visual impression is confirmed by the test statistic *L _{a}*, which identifies months in the waxing and waning phase of the intra-annual cycle in which the monthly pattern changes (Table 1). For April, August, September, and October,

*L*rejects the null hypothesis that the regression coefficient (

_{a}*γ*) is stable at the 5% level (Table 1). For other months, the test statistic fails to reject the null hypothesis, even at the 10% level.

_{i}Information about the nature of the change in the monthly pattern can be gleaned from systematic movements in the regression errors (ε) over time. The regression errors for April (Figure 2d) tend to increase over the sample period, while the regression errors for August (Figure 2h) and October (Figure 2j) tend to decrease over the sample period. The increase in the April regression error indicates that the observed values of the April anomaly tend to be smaller than values predicted by (*γ*_{4}) in Equation (4) early in the sample period and that the observed values tend to be greater than the predicted values later in the sample period. This systematic change in the regression error for April indicates that between 1965 and 2004, the atmospheric concentration of carbon dioxide in April rose relative to the cycle’s mean value. This suggests an increase in carbon flows to the atmosphere and/or a reduction in carbon flows from the atmosphere.

Conversely, the general decrease in the August and October regression errors over the sample period indicates that the observed values of the August and October anomalies tend to be larger (less negative) than values predicted by Equation (4) early in the sample period and that the observed values tend to be smaller (more negative) than the predicted values later in the sample period. This systematic change in the regression errors indicates that between 1965 and 2004, the atmospheric concentration of carbon dioxide in August and October declined relative to the cycle’s mean value. This suggests a reduction in carbon flows to the atmosphere and/or an increase in carbon flows from the atmosphere.

These results are relatively insensitive to a variety of assumptions. The significance level of *L _{a}* changes little as the window used in Equation (5) is lengthened or shortened. Nor do changes in the sample period used to estimate Equation (4) affect the timing and direction of changes in the monthly pattern for April and October. For these months, values of

*L*that are calculated from regression errors, which are estimated with observations through 2000 or 1995, reject the null hypothesis at

_{a}*p*< 0.05 (Table 1). If Equation (4) is estimated with data from 1959 to 2004 and the regression errors from 1965 to 2004 are used to calculate

*L*, the test statistic rejects the null hypothesis at the 5% level for April and October, but the significance level of the test statistic for the August regression coefficient (

_{a}*γ*

_{8}) drops to the 10% level. Similarly, the significance level of the test statistic for the August regression coefficient (

*γ*

_{8}) drops to the 10% level if the sample period used to estimate Equation (4) is truncated in 2000 or 1995 (Table 1). Instead, there is some evidence for changes in the intra-annual cycle during September (1965–2000) or July (1965–95).

## 4. Discussion

Instability, as indicated by a value of *L _{a}* that rejects the null hypothesis, identifies a month in which the intra-annual cycle changes over the sample period. Results presented above generally indicate that the intra-annual cycle of atmospheric carbon dioxide at Mauna Loa changes in April and October. April and October are the months in which the change in atmospheric concentrations is measured. Changes in carbon flows to and from the atmosphere may have occurred in months prior to April or October—a lag is possible because of the relatively long period that is required for air to move from North America or Eurasia to Mauna Loa.

*L*) using Equation (5).

_{a}Including an index for atmospheric circulation in Equation (6) can affect the results in three ways. If the pattern of atmospheric circulation represented by the index has no effect on the intra-annual cycle of atmospheric carbon dioxide at Mauna Loa, regression coefficients associated with Index (*δ _{i}*) will be statistically indistinguishable from zero. In this case, regression errors in Equation (6) will change little relative to those for Equation (4). Under these conditions, values of

*L*will not change significantly, nor will conclusions about the timing of changes in the intra-annual cycle. Together, these results would indicate that monthly changes in the intra-annual cycle of atmospheric carbon dioxide are not associated with the pattern of atmospheric circulation, represented by Index, used to estimate Equation (6).

_{a}Alternatively, the regression coefficients associated with Index (*δ _{i}*) could be statistically different from zero, which would indicate that the pattern of atmospheric circulation represented by Index has a statistically measurable effect on the atmospheric concentration of carbon dioxide at Mauna Loa. This effect could either strengthen or overturn results regarding the timing of monthly changes in the intra-annual cycle. Results generated using the regression errors from Equation (6) could strengthen results relative to those generated using the regression errors from Equation (4) if the pattern of atmospheric circulation represented by Index acts as “noise” to obfuscate changes associated with the flow of carbon between the atmosphere and ocean and/or terrestrial biota. Under these conditions, including Index in Equation (6) removes the noise associated with the pattern of atmospheric circulation, which makes it easier for the methodology to detect the signal associated with monthly changes in the intra-annual cycle generated by carbon flows to and from the terrestrial biota and/or the ocean.

Alternatively, including the index in Equation (6) could overturn the results generated using the regression errors from Equation (4) if the pattern of atmospheric circulation represented by the index is responsible for the monthly changes in the intra-annual cycle of atmospheric carbon dioxide identified by *L _{a}*. In this case, including Index in Equation (6) enhances the ability of Equation (6) to account for changes in the intra-annual cycle of atmospheric carbon dioxide. This improved explanatory power eliminates the systematic pattern in the monthly regression errors such that the value of

*L*no longer rejects the null hypothesis. Under these conditions, changes in atmospheric circulation as represented by Index are said to be responsible for monthly changes in the intra-annual cycle of atmospheric carbon dioxide.

_{a}Regression results for Equation (6) indicate that regression coefficients that are associated with one or more of the current or lagged values of the index for the SOI, NAO, EPI, WPI, and PDO are statistically different from zero (Table 2). This indicates that these patterns of atmospheric circulation have a statistically measurable effect on the intra-annual cycle for atmospheric carbon dioxide at Mauna Loa. Despite this effect, values for *L _{a}* that are calculated using the regression errors from Equation (6) generally are consistent with the values for

*L*that are calculated using regression errors from Equation (4). The monthly pattern of atmospheric carbon dioxide changes (

_{a}*p*< 0.05) in April and October. As before, the significance level changes little as the window used to calculate

*L*is lengthened or shortened. This indicates that changes in atmospheric circulation associated with the Southern Oscillation, the North Atlantic Oscillation, western Pacific oscillation, eastern Pacific oscillation, and Pacific decadal oscillation introduce noise that obfuscates monthly changes in the intra-annual cycle of atmospheric carbon dioxide at Mauna Loa, but do not cause changes in the intra-annual cycle between 1965 and 2004. This effect is consistent with arguments made by Higuchi et al. (Higuchi et al. 2002). They argue that historical shifts in atmospheric circulation can cause annual variations in amplitude, which are similar to those observed at Mauna Loa, but these shifts will not account for changes in the phasing of the cycle.

_{a}## 5. Conclusions

The inability to relate the timing of changes in the intra-annual cycle of atmospheric carbon dioxide at Mauna Loa to atmospheric circulation implies that April and October changes in the intra-annual cycle are generated by changes in the flow of carbon dioxide between the atmosphere and terrestrial biota and/or ocean. Mechanism(s) associated with these flows tend to be temporally and spatially heterogeneous. That is, rates of carbon uptake by terrestrial vegetation tend to be greatest in some biomes at certain times of the year. As such, their effect cannot be investigated using the methodology used here.

Nonetheless, changes in Figures 2d and 2h suggest an asymmetric effect of warming. The general increase in April regression errors suggests that warming in the early spring may increase heterotrophic respiration relative to net primary production. The net effect of this warming is consistent with the sensitivity of heterotrophic respiration to temperature and the lack of leaves in much of the mid- and high-latitude temperate forests in the Northern Hemisphere. Nor does this interpretation contradict previous results, which indicate that the timing of the spring downturn has advanced by 7 days (Keeling et al. 1996). Spring warming may enhance heterotrophic respiration prior to the downturn such that the “height” of the peak increases even though the date at which net primary production exceeds heterotrophic respiration comes earlier, thereby advancing the date of the downturn.

Conversely, the general decrease in the October regression errors suggests that warming may elongate the growing season at a time when deciduous trees have their full complement of leaves. Under these conditions, fall warming may extend the period during which net primary production is greater than heterotrophic respiration. Extending this period increases the amount of carbon removed from the atmosphere during the growing season, which lowers the trough of the intra-annual cycle.

To investigate whether the asymmetric effects of warming are responsible for changes in the intra-annual carbon cycle identified here, further research will use the notion of Granger causality to examine the statistical ordering of spatial and temporal changes in proxies for terrestrial biota, aquatic biotic activity, or oceanic circulation, relative to station measures of atmospheric carbon dioxide. Preliminary evidence indicates that it is possible to identify locations where and months when changes in the terrestrial biota, as measured by the NDVI, generate changes in atmospheric concentrations of carbon dioxide at Mauna Loa and Point Barrow (Kaufmann et al. 2007, manuscript submitted to *Earth Interactions*). Expanding this analysis to include land and sea surface temperatures and precipitation may identify the physical mechanisms that drive changes in biotic activity or oceanic circulation, which generate changes in station measures of atmospheric carbon dioxide.

## Acknowledgments

This research was funded by the National Science Foundation’s Ecological Rates of Change Program (NSF DEB 02-11216). I thank Bruce Anderson, Stephen Hall, Bruce Hansen, and Michael Dettinger for their comments on preliminary versions of the manuscript. I am solely responsible for any errors that remain.

## REFERENCES

Allan, R. J., , N. Nicholls, , P. D. Jones, , and I. J. Butterworth. 1991. A further extension of the Tahiti–Darwin SOI, early ENSO events and Darwin pressure.

*J. Climate*4:743–749.Andrews, D. A. 1991. Heteroscedasticity and autocorrelation consistent covariance matrix estimators.

*Econometrica*59:817–858.Angert, A., , S. Biraud, , C. Bonfils, , C. C. Henning, , W. Buermann, , J. Pinzon, , C. J. Tucker, , and I. Fung. 2005. Drier summers cancel out the CO2 uptake enhancement induced by earlier springs.

*Proc. Natl. Acad. Sci. USA*102:10823–10827.Bacastow, R. B. 1976. Modulation of atmospheric carbon dioxide by the southern oscillation.

*Nature*261:116–118.Beaulieu, J. J., and J. A. Miron. 1993. Seasonal unit roots in aggregate US data.

*J. Econ.*55:305–328.Canova, F., and B. E. Hansen. 1995. Are seasonal patterns constant over time? A test for seasonal stability.

*J. Bus. Econ. Stat.*12:292–349.Chapin, F. S., , S. A. Zimov, , G. R. Shaver, , and S. E. Hobbie. 1996. CO2 fluctuations at high latitudes.

*Nature*383:585–586.Churkina, G., , D. Schimel, , B. Braswell, , and X. Ziao. 2005. Spatial analysis of growing season length control over net ecosystem exchange.

*Global Change Biol.*11:1777–1787.Ciais, P., , P. P. Tans, , M. Trolier, , J. W. C. White, , and R. J. Francey. 1995. A large northern-hemisphere terrestrial CO2 sink indicated by the C-13/C-12 ratio of atmospheric CO2.

*Science*269:1098–1102.Climate Prediction Center cited. 2005. Northern Hemisphere teleconnection patterns. [Available online at http://www.cpc.ncep.noaa.gov/data/teledoc/telecontents.shtml.].

Conway, T. J., , P. P. Tans, , L. S. Waterman, , K. W. Thoning, , D. R. Ktzis, , K. A. Masarie, , and N. Zhang. 1994. Evidence for interannual variability of the carbon cycle from the National Oceanic and Atmospheric Administration/Climate Monitoring and Diagnostics Laboratory Global Air Sampling Network.

*J. Geophys. Res.*99:D11. 22831–22856.Dettinger, M. D., and M. Ghil. 1998. Seasonal and interannual variations of atmospheric CO2 and climate.

*Tellus*50B:1–24.Francey, R. J., , P. P. Trans, , C. E. Allison, , I. G. Enting, , J. W. C. White, , and M. Trolier. 1995. Changes in oceanic and terrestrial carbon uptake since 1982.

*Nature*373:326–330.Giardina, C. P., and M. G. Ryan. 2000. Evidence that decomposition rates of organic carbon in mineral soil do not vary with temperature.

*Nature*404:858–861.Goulden, M. L. Coauthors 1998. Sensitivity of boreal forest carbon balance to soil thaw.

*Science*27:214–217.Hall, C. A. S., , C. A. Ekdahl, , and D. E. Wartenberg. 1975. A fifteen year record of biotic metabolism in the Northern Hemisphere.

*Nature*225:136–138.Higuchi, K., , S. Murayama, , and S. Taguchi. 2002. Quasi-decadal variation of the atmospheric CO2 seasonal cycle due to atmospheric circulation changes: 1979–1998.

*Geophys. Res. Lett.*29.1173, doi:10.1029/2001GL013751.Hollinger, D. Y. Coauthors 2004. Spatial and temporal variability in forest-atmosphere CO2 exchange.

*Global Change Biol.*10:1689–1706.Houghton, R. A. 2000. Interannual variability in the global carbon cycle.

*J. Geophys. Res.*105:D15. 20121–20130.Hurrell, J. W. 1995. Decadal trends in the North Atlantic Oscillation: Regional temperatures and precipitation.

*Science*269:676–679.Kaufmann, R. K., , R. D. D’Arrigo, , C. Laskowski, , R. B. Myneni, , L. Zhou, , and N. Davi. 2004. The effect of growing season and summer greenness on northern forests.

*Geophys. Res. Lett.*31.L09205, doi:10.1029/2004GL019608.Kaufmann, R. K., , H. Kauppi, , and J. H. Stock. 2006. Emissions, concentrations, and temperature: A time series analysis.

*Climatic Change*77:249–278.Keeling, C. D., , T. P. Whorf, , M. Whalen, , and J. van der Plicht. 1995. Interannual extremes in the rate of rise of atmospheric carbon dioxide since 1980.

*Nature*375:666–670.Keeling, C. D., , J. F. S. Chin, , and T. P. Whorf. 1996. Increased activity of northern vegetation inferred from atmospheric CO2 measurements.

*Nature*382:146–149.Keeling, C. D., and T. P. Whorf. the Carbon Dioxide Research Group cited. 2003. Atmospheric CO2 concentrations (ppmv) derived from in situ air samples collected at Mauna Loa Observatory, Hawaii. [Available online at http://cdiac.ornl.gov/ftp/trends/co2/maunaloa.co2.].

Keyser, A. R., , J. S. Kimball, , R. A. Nemani, , and S. W. Running. 2000. Simulating the effects of climate change on the carbon balance of North American high latitude forests.

*Global Change Biol.*6:S1. 185–195.Kohlmaier, G. F., , E. O. Sire, , and A. Janecek. 1989. Modeling the seasonal contribution of a CO2 fertilization effect of the terrestrial vegetation to the amplitude increase in atmospheric CO2 at Mauna Loa observatory.

*Tellus*41B:487–510.Kwiatkowski, D., , P. Phillips, , P. Schmidt, , and Y. Shin. 1992. Testing the null hypothesis of stationarity against the alternative of a unit root: How sure are we that economic time series have a unit root?

*J. Econ.*44:215–238.Lintner, B. R. 2002. Characterizing global CO2 interannual variability with empirical orthogonal function/principal component (EOR/PC) analysis.

*Geophys. Res. Lett.*29.1921, doi:10.1029/2001GL014419.Luo, Y., , S. Wan, , D. Hui, , and L. L. Wallace. 2001. Acclimatization of soil respiration to warming in tall grass prairie.

*Nature*413:622–625.Marland, G., , T. A. Boden, , and R. J. Andres. cited. 2001. Global CO2 Emissions from Fossil-Fuel Burning, Cement Manufacture, and Gas Flaring: 1751–2004. [Available online at http://cdiac.ornl.gov/ftp/ndp030/global.1751_2004.ems.].

Morimoto, S., , T. Nakazawa, , K. Higuchi, , and S. Aoki. 2000. Latitudinal distribution of atmospheric CO2 sources and sinks inferred by

*δ*13 C measurements from 1985 to 1991.*J. Geophys. Res.*105:24315–24326.Murayama, S., , S. Taguchi, , and K. Higuchi. 2004. Interannual variation in the atmospheric CO2 growth rate: Role of atmospheric transport in the Northern Hemisphere.

*J. Geophys. Res.*109.D02305, doi:10.1029/2003JD003729.Myneni, R., , C. Keeling, , C. Tucker, , G. Asrar, , and R. Nemani. 1997. Increased plant growth in the northern latitudes from 1981 to 1991.

*Nature*386:698–702.Newey, W. K., and K. D. West. 1987. A simple positive semi-definite heteroskedasticity and autocorrelation consistent covariance matrix.

*Econometrica*55:703–708.Prentice, I. C., , M. Heimann, , and S. Sich. 2000. The carbon balance of the terrestrial biosphere: Ecosystem models and atmospheric observations.

*Ecol. Appl.*10:1553–1573.Randerson, J. T., , C. B. Field, , I. Y. Fung, , and P. P. Tans. 1999. Increases in early season ecosystem uptake explain recent changes in the seasonal cycle of atmospheric CO2 at high northern latitudes.

*Geophys. Res. Lett.*26:2765–2768.Rayner, P. J., , R. M. Law, , and R. Dargaville. 1999. The relationship between tropical CO2 fluxes and the El-Niño-Southern Oscillation.

*Geophys. Res. Lett.*26:493–496.Shibistova, O. Coauthors 2002. Annual ecosystem respiration budget for a

*Pinus sylvestris*stand in central Siberia.*Tellus*54B:568–589.Siegenthaler, U. 1990. Biogeochemical cycles—El-Niño and atmospheric CO2.

*Nature*345:295–296.Sims, C. 1980. Macroeconomics and reality.

*Econometrica*48:1–49.Tanja, S. Coauthors 2003. Air temperature triggers the recovery of evergreen boreal forest photosynthesis in spring.

*Global Change Biol.*9:1410–1429.Trumbore, S. E., , O. A. Chadwick, , and R. Amundson. 1996. Rapid exchange between soil carbon and atmospheric carbon dioxide driven by temperature change.

*Science*272:393–396.White, M. A., and R. R. Nemani. 2003. Canopy duration has little influence on annual carbon storage in the deciduous broad leaf forest.

*Global Change Biol.*9:967–972.Wignuth, A. M. E., , H. Heiman, , K. D. Kurz, , E. Maier-Riemer, , U. Mikolajewicz, , and J. Segschneider. 1994. El Niño-Southern Oscillation related fluctuations of the marine carbon cycle.

*Global Biogeochem. Cycles*8:39–63.Yang, X., and M. Wang. 2000. Monsoon ecosystems control on atmospheric CO2 interannual variability: Inferred from a significant positive correlation between year-to-year changes in land precipitation and atmospheric CO2 growth rate.

*Geophys. Res. Lett.*27:1671–1674.Zeng, N., , A. Mariotti, , and P. Wetzel. 2005. Terrestrial mechanisms of interannual CO2 variability.

*Global Biogeochem. Cycles*19.GB1016, doi:10.1029/2004GB002273.Zhang, Y., , J. M. Wallace, , and D. S. Battisti. 1997. ENSO-like interdecadal variability: 1900–1993.

*J. Climate*10:1004–1020.Zhou, L., , C. J. Tucker, , R. K. Kaufmann, , D. Slayback, , N. V. Shabanov, , and R. B. Myneni. 2001. Variations in northern vegetation activity inferred from satellite data of vegetation index during 1981 to 1999.

*J. Geophys. Res.*106:20069–20083.Zhou, L., , R. K. Kaufmann, , Y. Tian, , R. Myneni, , and C. Tucker. 2003. Relation between interannual variations in satellite measures of northern forest greenness and climate between 1982 and 1999.

*J. Geophys. Res.*108.4004, doi:10.1029/2002JD002510.