Is interannual fluctuation of atmospheric CO2 dominated by combined effects of ENSO and volcanic aerosols?



[1] The influence of ENSO, volcanic eruptions, and the North Atlantic Oscillation (NAO) on fluctuations of the atmospheric CO2 concentration were globally investigated on the ENSO timescale for the period 1958 to 1994. Two deseasonalized and detrended monthly time series of anomalous CO2 fluxes were generated: (1) modeled terrestrial biospheric CO2 flux anomalies calculated by the High Resolution Biosphere Model (HRBM) driven by realistic temperature and precipitation data but without considering radiation effects and (2) anomalous CO2 fluxes deduced from atmospheric measurements. While comparing the two time series, three types of periods could be distinguished: “a” periods with a phase shift close to zero between the two series, “b” periods with a phase shift of up to 11 months, and “c” periods with the two curves completely out of phase. During the “c” periods the modeled data show carbon release by the biosphere while the observed data show carbon uptake. The periods of type “c” are accompanied by major volcanic eruptions with considerable aerosol production. Enhanced atmospheric aerosol loading can increase the diffuse fraction of the solar radiation. Diffuse radiation in many cases penetrates better into plant canopies thus potentially enhancing photosynthesis. The resulting additional uptake of carbon may have overcompensated the carbon release caused by ENSO effects thus possibly bringing about a net uptake of CO2. The length variations of the time lags (“b” periods) may be attributed to impacts of vegetation fires. The influence of the NAO remained unclear.

1. Introduction

[2] The atmospheric CO2 concentrations as measured at the Mauna Loa Observatory [Keeling et al., 1995] or at other stations show variations in the increase rate on timescales of some months (seasonal) to several years [Kane and de Paula, 1996]. Interannual changes in the growth rate of the atmospheric CO2 concentration are superimposed on the seasonal variations. Knowing that the interannual variability of emissions from fossil fuel burning and from cement production are rather small [Marland et al., 1999], it must be concluded that variations (anomalies) in the uptake or release of carbon by the oceans and the terrestrial biosphere cause the fluctuations of the atmospheric CO2 content. In order to explain the interannual variations, it is important to know how the global weather conditions affect the carbon cycle.

[3] Studies applying inverse calculations based on atmospheric CO2, δ13C, and O2 (top-down approaches) have mostly been adopted to infer the partitioning of carbon sinks and sources between the oceans and the terrestrial biosphere. Some of these studies found the oceanic variability to be larger than the variability in the terrestrial biospheric CO2-balance [Francey et al., 1995]. Other results show similar variability in either flux [Rayner et al., 1999; Joos et al., 1999], but many of the studies show larger terrestrial variability [Keeling et al., 1995; Battle et al., 2000; Bousquet et al., 2000; Langenfelds et al., 2002]. The latter was supported by bottom-up approaches using data on the partial pressure of CO2 in the upper ocean waters [Lee et al., 1998; Feely et al., 1999] or applying ocean biogeochemical models [Le Quéré et al., 2000]. In addition, modeling studies provided evidence that fluctuations in the atmospheric growth rate of CO2 are primarily attributed to the terrestrial biosphere [Gérard et al., 1999; Rayner and Law, 1999; Houghton, 2000; Knorr, 2000]. On the other hand, in a recent study, Dargaville et al. [2002] pointed out that the variability in the ocean fluxes or the atmospheric transport may also be key factors in the atmospheric interannual variability.

[4] Several processes have been identified to influence the carbon balance of the terrestrial biosphere. The most important are variations in climate [e.g., McGuire et al., 2001; Schaefer et al., 2002], land use changes [Houghton, 1999], vegetation fires [Langenfelds et al., 2002], and anthropogenic nitrogen deposition [Townsend et al., 1996; Holland et al., 1997]. Climate is thought to have the most important influence on the interannual variability of the biospheric CO2 fluxes [Houghton, 2000; Dargaville et al., 2002].

[5] As a main part of the terrestrial carbon cycle the net exchange of CO2 between the biosphere and the atmosphere (NEP, net ecosystem production) is defined as the balance of three important processes: photosynthetic uptake of CO2 by plants, which on the canopy scale is referred to as gross primary productivity (GPP), and respiratory loss of carbon to the atmosphere, which is divided into the autotrophic respiration (RA) by plants and the heterotrophic respiration (RH) by decomposers of organic material. Often the balance of release and uptake of CO2 by plants (net primary productivity, NPP, NPP = GPP − RA) is addressed.

[6] Many global modeling studies on the terrestrial carbon balance of the past decades found climate variability to have caused a decrease in NEP in that period [Cao and Woodward, 1998b; Cramer et al., 2001] (for a contradictory view, see Dai and Fung [1993]). However, there is ongoing scientific discussion on which of the terrestrial processes dominates the interannual variability of NEP and which of the climate variables has the strongest effect.

[7] McGuire et al. [2001] applied four biogeochemical models and found that NPP dominated the interannual variability of NEP in three of them. This is confirmed by Kindermann et al. [1996] but differs from the findings of Knorr [2000] and to some extent from the results of Schaefer et al. [2002] and Ito and Oikawa [2000], who identified RH and RA to show more variability than GPP. The interrelations described above seem to differ between different parts of the Earth [Tian et al., 2000; Schaefer et al., 2002; Cao et al., 2002]. Kindermann et al. [1996] and Dai and Fung [1993] showed for extratropical ecosystems that NPP was the dominant process, whereas Wang and Polglase [1995], Cao and Woodward [1998b], and Oechel et al. [1993] found RH to have more impact on interannual variability, especially for cold ecosystems. In the tropics, Wang and Polglase [1995] and Tian et al. [1998] considered NPP to have the largest variability, while Townsend et al. [1992] and Dai and Fung [1993] attributed most of the variability to RH.

[8] In order to explain the variations in the increase rate of the atmospheric CO2 concentration, it is important to know how the global weather conditions (global circulation) affect the carbon cycle. In principle, physiological and physico-chemical processes of the biosphere on land transform the weather signals into carbon cycle disturbances.

[9] Weather conditions such as air temperature or precipitation affect the carbon metabolism in the biosphere via effects of soil and leaf temperatures and the availability of moisture on the processes mentioned above. There are different opinions about the response functions of plants to the weather conditions [see, e.g., Kindermann et al., 1996].

[10] Comparing the influences of precipitation and temperature, most studies found precipitation to have a larger influence on the interannual variability of biospheric fluxes [Dai and Fung, 1993; Kindermann et al., 1996; Cao and Woodward, 1998a].

[11] The variability of global climate conditions is largely caused by circulation anomalies. A well known circulation anomaly is the El Niño-Southern Oscillation (ENSO) phenomenon [e.g., Philander, 1990], but others exist as, for example, the North Atlantic Oscillation (NAO) [Hurrell, 1996; Hurrell and Van Loon, 1997] or the Antarctic Circumpolar Wave (ACW) [White and Peterson, 1996].

[12] Early evidence for the connection between ENSO and the interannual variability of the terrestrial carbon balance has been provided by Bacastow [1976] and Keeling et al. [1989]. There is extensive observational evidence that ENSO phenomena influence physiological processes in the terrestrial biosphere, documented by crop yield changes in Australia [Nicholls, 1985], Zimbabwe [Cane et al., 1994], North America [Hansen et al., 1999; Hsieh et al., 1999; Izaurralde et al., 1999], Europe [Bell et al., 1999; Gimeno et al., 1998], and many other parts of the world [Kiladis and Diaz, 1989; Bell et al., 1999]. Natural ecosystems have larger areas and higher fluxes of productivity and depletion than agricultural ecosystems [Esser, 1995]. It has to be concluded that ENSO-like phenomena may have an even stronger effect on the global carbon balance which is dominated by natural vegetation. Several model investigations corroborate this conclusion [e.g., Jones et al., 2001; Schaefer et al., 2002].

[13] Besides temperature and precipitation, the photosynthetic active radiation (PAR) absorbed by leaves is an important environmental control on photosynthesis. For whole canopies the relation of direct to diffuse radiation is of great importance. It has been long realized that to some extent diffuse radiation promotes plant productivity better than the direct beam [e.g., de Wit, 1965]. It is believed to penetrate better into a canopy and thus reduce the volume of shade [Roderick et al., 2001] whereby leaves which normally photosynthesize at undersaturated light conditions can enhance their productivity.

[14] Diffuse radiation arises mainly from the scattering of the incoming solar radiation at clouds and aerosols. The biggest enhancement of productivity has been found with moderate cloud cover and moderate atmospheric aerosol loading [e.g., Cohan et al., 2002]. Increased aerosol loading, for example, due to volcanic eruptions [see, e.g., McCormick et al., 1995] therefore potentially increases the terrestrial NPP [Roderick et al., 2001; Cohan et al., 2002; Gu et al., 2003].

[15] We have used the High Resolution Biosphere Model (HRBM) [Esser et al., 1994] to investigate its ability to predict the observed atmospheric CO2 flux anomalies on the ENSO timescale for the period 1958 to 1994. For that purpose the HRBM was driven by CO2 concentration data and monthly data of climate variables and their anomalies on a 0.5° grid. To account for effects of the ocean-atmosphere carbon exchange modeled ocean CO2 uptake data were used. Further on we have used ENSO index data to investigate the potential influence of El Niño and La Niña events on the fluctuations of the CO2 fluxes between the terrestrial biosphere and the atmosphere. An index for the state of the North Atlantic Oscillation (NAO, see below) and stratospheric aerosol data as an index for global volcanic activity were used to investigate the potential influences of these ENSO-scale phenomena.

2. Materials and Methods

[16] In order to investigate the ability of the High Resolution Biosphere Model (HRBM) [Esser et al., 1994] to predict atmospheric CO2 growth rate anomalies on the ENSO timescale, modeled net anomalous carbon fluxes between the biosphere and the atmosphere (Δabio) were compared with anomalous biospheric fluxes deduced from observations of the atmospheric CO2 growth rate (Δaano). The HRBM was designed to investigate the carbon balance of the terrestrial biosphere and the impacts of rising atmospheric CO2 concentrations and of climatic changes. It has been used to assess various diagnostic and prognostic issues [Esser and Lautenschlager, 1994; Esser, 1995; Kicklighter et al., 1999; Reichenau, 2000; McGuire et al., 2001]. The HRBM calculates biospheric balances (NEP) from NPP and the depletion fluxes (RH) of four litter fractions and a soil organic carbon fraction. It uses variable depletion coefficients for these five fractions which are calculated from monthly values of temperature and humidity. The influence of humidity follows a maximum function with the shape of the function shifting with temperature. The calculation of NPP is based on the MIAMI model [Lieth, 1975] extended by functions considering soil fertility, the CO2 fertilization effect, land use, and the impact of agricultural techniques on agricultural yield [Esser, 1991; Esser et al., 1994]. In the MIAMI model, annual precipitation and mean temperature are used to compute annual NPP. Influences of radiation are not taken into account. The annual NPP is then distributed over the 12 months using the ratio of the cube of the monthly relative actual evapotranspiration to the cube of the annual actual evapotranspiration [Esser et al., 1994]. The submodel for vegetation fires included in the HRBM [Esser et al., 1994] was not used for the present study. For an evaluation of HRBM simulations see, for example, Dargaville et al. [2002].

[17] For this study the HRBM was run on a global 0.5° grid from a steady state in 1860 until 1995. We have analyzed the period 1958 to 1994 since atmospheric CO2 data with high time resolution are available since the beginning of the Mauna Loa record in 1958. The end of the climate anomaly data determined the final year.

[18] The modeled anomalous carbon fluxes were calculated by

equation image

where Δbr is the monthly global terrestrial biospheric carbon balance calculated by a HRBM model run driven by climate data including monthly anomalies. The original climate anomaly data for temperature [Jones, 1994; Jones et al., 1997b] and precipitation [Hulme, 1992, 1994; Hulme et al., 1998] were interpolated to a complete terrestrial 0.5° grid. The base seasonal mean climate was taken from an update of the Leemans and Cramer [1991] data set. Terrestrial fluxes arising from long-term climatic trends (Δbt) were subtracted from Δbr. Δbt originates from a model run driven by 9-year symmetrical running means of the same climate data calculated separately for each of the 12 months, thus retaining long-term trends and the mean seasonal signal.

[19] Trends created by the CO2 fertilization effect [Esser et al., 1994] and by land-use changes (land-use matrix set according to McGuire et al. [2001], based upon work of Ramankutty and Foley [1998, 1999]) are included in both Δbr and Δbt and thus eliminated from Δabio. Finally, variations on timescales shorter than the ENSO timescale were removed from Δabio by calculating 13 months symmetrical running means.

[20] The atmospheric flux anomaly data was deduced from atmospheric measurements, anthropogenic emission data, and model estimates. Hereinafter it will be referred to as the observed flux anomalies,

equation image

where Δa0 is the observed change of the atmospheric CO2 content since the previous month calculated as an average of measurements at Mauna Loa and the South Pole (update of Heimann [1997, Figure 8]), Δf is the sum of the emissions from fossil sources for the current month [Marland et al., 1999], and Δoc is the net flux between the oceans and the atmosphere [Orr et al., 2001] calculated by a run of the Large-Scale Geostrophic model of the Max Planck Institute for Meteorology in Hamburg, Germany [Maier-Reimer, 1993; Maier-Reimer et al., 1993]. The ocean model was driven by long-term means of the present seasonal climate. As for Δabio, the long-term biospheric trend Δbt was removed from the data, thus obtaining anomalies caused by the impacts of climate anomalies on oceanic and terrestrial CO2 fluxes. Prior to computing Δaano, a 13-month symmetrical running mean was applied to Δf, Δoc, and Δbt.

[21] The resulting monthly data series were smoothed using a b-spline technique (pda_curfit from Starlink,, July 2000). Positive values of Δabio and Δaano denote fluxes into the atmosphere.

[22] To compare Δabio and Δaano with the ENSO circulation anomaly, we used the multivariate ENSO Index (MEI [Wolter and Timlin, 1993]; for a comparison with the Southern Oscillation Index SOI see Ortiz-Tanchez et al. [2002]). The MEI integrates more information than other ENSO indices and therefore it probably reflects the state of the coupled ocean-atmosphere system better than any single component. For the comparison with the North Atlantic Oscillation the NAO index by Jones et al. [1997a] was used. We applied a 13-month symmetrical running mean and a b-spline smoothing to the NAO index as explained above.

[23] We used data on the optical thickness of the atmosphere at 550 nm (extended version of work by Sato et al. [1993], taken from, December 2001) as an index of the global atmospheric aerosol loading.

[24] A (moving box-car) correlation analysis was applied to the data to infer phase shifts between Δaano and Δabio. For that purpose, maximum positive coefficients of correlation (r) were obtained for time intervals of 47 months around each month of the investigation period. For each interval, coefficients of correlation were calculated for a set of time shifts between Δaano and Δabio. Since Δabio seemed to always lead Δaano (see Figure 2), only time lags of −24 to 0 months were investigated.

3. Results

[25] In Figure 1 we compare the modeled terrestrial monthly carbon balances for realistic climate Δbr and for long-term trend climate Δbt. For easier comparability, both curves were smoothed by applying a running mean and a b-spline as described in section 2.

Figure 1.

Monthly net carbon fluxes calculated by the High Resolution Biosphere Model (HRBM) driven by realistic monthly climate (Δbr, dashed line) or a 9-month symmetrical running mean of the same climate data for each month of the year separately (Δbt, solid line). A 13-month symmetrical running mean and a b-spline smoothing were applied to the data. Positive values denote fluxes into the atmosphere.

[26] Except for the last 3 years of the investigation period, Δbt shows a declining trend with only little interannual variation. A similar trend toward enhanced carbon uptake by the biosphere can be found in Δbr. In Δbr fluctuations on a timescale of 2–3 years are superimposed on the trend that, especially after 1973, cause periods of substantial carbon uptake by the biosphere.

[27] Time series of the carbon flux anomalies Δaano (observed anomalies) and Δabio (modeled anomalies, Δbt subtracted from Δbr) are presented in the middle panel of Figure 2. Δaano includes long-term influences of processes which occur on timescales much longer than that of ENSO, for example, feedbacks in the interaction of the carbon and nitrogen cycles or the 11-year solar cycle. These long-term influences are not explicitly considered in the models and, therefore, are not present in Δabio. As a consequence, Δaano in Figure 2 is more negative than Δabio before 1982, whereas thereafter it is sometimes more positive (1982–1983, 1987–1989). While the amplitude and the phase of either flux anomaly resemble each other, interesting differences occur at a more detailed view.

Figure 2.

(top) Multivariate ENSO index (MEI). Positive values denote El Niño events. (middle) Monthly modeled net anomalies of carbon fluxes from the biosphere to the atmosphere (Δabio, dashed line) and observed net anomalies of carbon fluxes into the atmosphere (Δaano, solid line). Positive values denote fluxes into the atmosphere. (bottom) Optical thickness of the atmosphere at 550 nm as an index of global atmospheric aerosol loading. Main volcanic sources (eruptions): Agung in 1963, El Chichón in 1982, and Mount Pinatubo in 1991.

[28] With respect to the phasing of the two curves, three different types of periods can be visually distinguished: type a: the phase difference of the two curves is close to zero (periods 1965–1968 and 1978–1980); type b: Δaano lags 4–8 months behind Δabio (periods 1968–1977 and 1981–1990); and type c: the two curves are completely out of phase (periods 1963–1964 and 1991–1994).

[29] This classification of periods is affirmed by the correlation analysis. Figure 3 presents the time lags of Δaano behind Δabio at maximum positive coefficients of correlation (bottom panel) and the associated coefficients (r, top panel). No data are shown for months lacking positive coefficients. During periods “a,” maximum positive correlation was found at a time lag of 0 to −1 months. Coefficients of correlation range from 0.58 to 0.99 with the first period “a” showing generally lower values. For periods “b,” time lags range from −11 to −2 months with coefficients of correlation similar to those of periods “a.” In the periods “c,” no or very low positive correlation was found and the time lags associated with maximum positive correlation are the largest in the period of investigation.

Figure 3.

Results of a (moving box-car) correlation analysis of monthly modeled and observed flux anomalies (Δaano and Δabio, middle panel of Figure 2). For each month the coefficient of correlation for a symmetrical 4-year interval was computed for Δaano leading Δabio by 0 to 24 months. (bottom) Time lags at maximum positive coefficients of correlation and (top) the respective coefficients are presented. No data are shown for months lacking positive coefficients.

[30] Additional calculations were carried out which allowed for positive time lags (i.e., Δabio lagging behind Δaano, data not shown). High positive time lags occurred in the periods “c,” and time lags of up to 3 months were found in the beginning of the first period “a.” Over most of the period of investigation the results remained unaffected.

[31] Because of the driving climate data and the model structure, we known that for the modeled flux Δabio the variability on the ENSO timescale occurs mainly due to climate variability. We presume that the same may be the case for Δaano. Since the multivariate ENSO index (MEI) characterizes to some extent the actual state of the climate-ocean system and its variability, a correlation should exist between the anomaly fluxes (Δaano and Δabio) and the MEI.

[32] The MEI is shown in Figure 2 (top panel). According to common understanding, El Niño periods (denoted by positive MEI values) are generally correlated with more positive values of the global anomaly fluxes [Bacastow, 1976; Keeling et al., 1989, 1995; Sarmiento, 1993; Cao et al., 2002]. Thus the globally averaged effect of this circulation pattern on the terrestrial biosphere is a net release of carbon. Δabio follows this concept consequently with no or only minor phase differences to MEI. Δaano, however, deviates considerably from the MEI trend, especially during the above mentioned periods “c,” sometimes showing opposite trends.

[33] This indicates that the observed averaged carbon balance reaction of the Earth system to ENSO is not uniform. Mechanisms must exist which superimpose on the “normal” response of the earth system during periods of type “c.” What mechanisms or processes could have this effect?

[34] We have investigated the volcanic activity and the North Atlantic Oscillation (NAO), which represents another oscillation of the Earth system besides ENSO. The optical thickness of the atmosphere as volcanic eruption index is shown in the bottom panel of Figure 2. Three major and several minor volcanic eruptions occurred in the period under investigation. The Agung eruption in 1963 and the Mount Pinatubo eruption in 1991 coincide with the two periods “c” where Δaano does not show a release of carbon which would have been expected during El Niño conditions. The third major eruption, that of El Chichón in 1982, occurred during a period “b,” but was in co-occurrence with the strongest El Niño of the entire period of investigation (highest value of MEI). Therefore the ENSO impact could have overcompensated the volcanic impact bringing about a net global release of CO2.

[35] The North Atlantic Oscillation index (NAO index) characterizes to some extent the state of the North Atlantic ocean-atmosphere system. It therefore could potentially provide an indication of the influence of the North Atlantic's carbon balance on both the difference between the extrema of Δabio and Δaano and the phase shift between the curves (note that the ocean fluxes used to calculate Δaano do not include impacts of climate variability). The difference between Δaano and Δabio without considering phase shifts may be interpreted as an integrated measure for both discrepancies. Some correlation of the NAO index with the difference of the modeled minus observed anomaly flux (Δabio − Δaano, solid line in Figure 4) could be assumed in the periods 1963–1966 (r = 0.70) and 1987–1993 (r = 0.72) including the periods “c.” Within these periods, high values of both the NAO index and the difference between Δabio and Δaano occur simultaneously.

Figure 4.

Difference of monthly modeled and observed CO2 flux anomalies (Δabio − Δaano, solid line, left axis) compared to the NAO index (dashed line, right axis). Solid line: difference without considering transport lag (coefficient of correlation with NAO: r = 0.26); dotted line: difference computed with Δabio leading Δaano by 6 months (r = 0.31).

[36] From the time lag during periods “b” a mean time lag of 6 months was estimated. That time lag was considered in an additional calculation of the difference signal (Figure 4, dotted line). Discrepancies between the difference signals calculated with and without considering the potential time lag occur mainly from 1971 till 1981. The large differences in 1971/1972, 1974/1975 and 1976/1977 disappear, while the difference in 1973/1974 turns from a positive to a negative sign with a similar absolute value (−0.787 · 1014 and 0.709 · 1014 without and with considering a lag of 6 months, respectively). From 1978 until 1981 the difference signal considering the time lag shows larger differences. Neither of the two difference signals shows good correlation with the NAO (r = 0.31 and r = 0.26 for the signal with and without considering the time lag, respectively).

4. Discussion and Conclusions

[37] The comparison of the observed flux anomaly (Δaano) with the MEI (Figure 2) provides rather similar patterns for the data series: Rising atmospheric CO2 concentrations are found during periods of positive MEI (El Niño periods) which is in agreement with earlier results [Bacastow, 1976; Bacastow et al., 1980; Keeling et al., 1989]. Apart from the observed phase shifts, the similarity of the trends and amplitudes of the modeled anomaly Δabio (except for periods “c”) and the MEI confirms earlier findings that the terrestrial biosphere is mainly responsible for the atmospheric CO2 variations on the ENSO timescale [Keeling et al., 1995; Lee et al., 1998; Battle et al., 2000; Bousquet et al., 2000; Feely et al., 1999; Gérard et al., 1999; Rayner and Law, 1999; Houghton, 2000; Knorr, 2000; Le Quéré et al., 2000; Langenfelds et al., 2002].

[38] Besides these considerations, the behavior of Δabio and Δaano raises two important questions: (1) Why do we see varying time lags of up to 11 months between these two values during longer periods of time (periods “a” and “b”); (2) why are the two fluxes completely out of phase in some shorter periods (periods “c”)?

[39] With respect to the first question, first of all, the influence of atmospheric transport has to be considered. Depending on the region the original biospheric CO2 signal comes from, it takes some time until its impact on the atmospheric concentration can be seen in the records of the two sampling stations used in this study. As inferred from the correlation analysis (Figure 3), the atmospheric record lagged the biospheric record by 2 to 11 months. This is in accordance with Law et al. [1992], who found a time of 300 days for extratropical and 150 days for tropical signals to become globally uniformly distributed. Law et al. [1992] also showed that the time needed by a tracer originating from the tropics or from 30°N to reach the South Pole is approximately 2 to 3 months and 8 months, respectively, giving an average of 5–6 months, which is similar to the mean lag of about 6 months found in this study.

[40] Interannual variability of the atmospheric transport could provide an opportunity to explain the variations of the phase shifts between the observed (Δaano) and the modeled (Δabio) anomalies. However, Dargaville et al. [2000] showed that transport variability may be of some importance only within a hemisphere whereas, on a global scale, it is of minor importance [see also Law and Simmonds, 1996; Bousquet et al., 1996]. Furthermore, according to Dargaville et al. [2000], the inclusion of interannual atmospheric transport variability has more effect on the mass of carbon transported than on the phasing of the fluxes.

[41] Nevertheless, the time a signal takes to reach the sampling site is different for either station and origin. Therefore a strong peak in the CO2 record of Mauna Loa, for example, may reach the South Pole station a few months later, thus potentially causing a phase shift in the observed anomaly Δaano which was derived from averaged records of both stations.

[42] Apart from transport effects, other processes may exist in nature that were not considered in the model which bring about fluctuations of the time lag between Δaano and Δabio on the ENSO timescale. Vegetation fires are known to have an impact on the carbon cycle but were not considered in the terrestrial carbon cycle model (HRBM) runs used to calculate Δabio.

[43] The mean annual carbon emissions from burning were computed earlier by the HRBM's fire routines. These model runs yielded an annual mean of 4.14 PgC·yr−1 carbon emitted to the atmosphere [Mack et al., 1996]. This value is in line with Andreae [1993] and Olson [1981], who estimated this carbon flux to be 4.08 PgC·yr−1 and 4.98 PgC·yr−1, respectively. These fluxes seem too large in comparison to the average amplitude of Δabio (1.8 PgC·tyr−1). However, biomass burning does not necessarily enhance the CO2 flux from the biosphere to the atmosphere. The main effect is a time displacement of carbon fluxes due to a CO2 release from material that otherwise would have been converted to CO2 anyway by biological decomposition processes (RH).

[44] The variability of the time lag between Δaano and Δabio may therefore originate from residues of the disturbed seasonal cycle in Δaanoa0, equation (2)), due to influences of fires. The overall seasonal signal is dominated by the Northern Hemisphere, while fires occur mainly in the tropics and the Southern Hemisphere [Wittenberg et al., 1998]. Hence, due to the time shift between the Northern and Southern Hemisphere's vegetation periods, vegetation fires could make the Southern Hemisphere dominate the averaged signal, causing the phase shift. This assumption is supported by the results of Wittenberg et al. [1998] and van der Werf et al. [2003], who report an enhanced amplitude of the seasonal cycle of NEP due to biomass burning for several parts of the world.

[45] Recent literature corroborates these considerations. Schultz [2002] showed that emissions from vegetation fires show significant interannual variability and according to Langenfelds et al. [2002], a substantial body of evidence indicates that fires influence the interannual variability of the atmospheric CO2. The fire effects on the flux anomalies require further investigation to clarify if they are indeed partly responsible for the observed phase shifts.

[46] With respect to the second question concerning the periods “c,” where Δabio and Δaano were completely out of phase, we examined potential influences of volcanic eruptions and of the North Atlantic Oscillation (NAO). Both periods “c” were accompanied by major volcanic eruptions indicated by a strong increase of the atmospheric optical thickness (Figure 2, bottom panel). After the eruption of Mount Pinatubo in 1991, Sarmiento [1993] reported an enhanced uptake of CO2 by the oceans. Le Quéré et al. [2000] in a modeling analysis also found an enhanced oceanic uptake between 1991 and 1994 but with a smaller magnitude and mainly attributed to effects of the subsequent El Niño events during that period. Besides the enhanced uptake by the oceans, atmospheric δ13C measurements show that most of the decline in the growth rate of atmospheric CO2 after 1991 was caused by terrestrial biospheric uptake [Ciais et al., 1995]. Keeling et al. [1995] or Jones and Cox [2001] attributed this uptake to a cooling of the atmosphere after the eruption of Mount Pinatubo (Pinatubo cooling; see Jones and Kelly [1996]).

[47] According to studies by Keeling et al. [1995], Braswell et al. [1997], or Vukićević et al. [2001], there is a time lag between the occurrence of a temperature anomaly and its impact on the biospheric carbon balance. Thus it is unlikely that temperature effects alone caused the sink immediately after the Mount Pinatubo eruption. Furthermore, the cooling in the early 1990s is of similar magnitude as other cooling events in the second half of the twentieth century which can be seen in the temperature data presented by Ito and Oikawa [2000] or Cao et al. [2002]. The temperature anomaly data used to drive the HRBM also show these other cooling events. During these periods, modeled and observed CO2 anomalies resemble each other, while in the early 1990s a similar terrestrial mean cooling caused a period “c” situation. Thus temperature effects of volcanic eruptions have probably not mainly caused the decline of the atmospheric CO2 growth rate. In our opinion, effects not included in the HRBM most likely have been the main cause of the anomalous CO2 sink.

[48] Knorr [2000] stated that the influence of radiation anomalies on the tropical biosphere may have been the main cause of the anomalous uptake of carbon in 1992. Especially, the diffuse fraction of the incoming radiation can be of greater importance to the photosynthetic uptake of CO2 by plants [Gu et al., 2002]. Important sources of diffuse radiation are volcanic aerosols originating from large eruptions and spreading around the globe [McCormick et al., 1995]. Larger amounts of aerosols reduce the absolute level of radiation that reaches the biosphere, while its diffuse fraction which in many cases penetrates better into canopies [Roderick et al., 2001] is enhanced, thus reducing the volume and intensity of shade. Leaves inside a plant canopy which photosynthesize below the point of light saturation may therefore enhance their photosynthesis. On the other hand, leaves on top of the canopy which normally are light saturated often still receive enough light on lower radiation levels to saturate their light harvesting systems [Cohan et al., 2002; Gu et al., 2002].

[49] In a recent paper, Gu et al. [2003] emphasize that enhanced diffuse radiation has a strong global impact on NEP. Roderick et al. [2001] explicitly studied the effect of the changed ratio of direct to diffuse radiation, due to the higher aerosol content in the atmosphere after the eruption of Mount Pinatubo. They calculated an additional carbon uptake by the terrestrial vegetation of 2.5 PgC·yr−1 caused by the increased diffuse fraction. This agrees with our findings that the difference between Δabio and Δaano is 2.54 PgC·yr−1 for 1992. The same mechanism can be proposed for the eruption of Agung (1963) in the first period “c” where the observed atmospheric growth rate of CO2 was reduced in a similar manner [Keeling et al., 1995].

[50] The El Chichón eruption in 1982 did not result in a period “c” situation. Δaano shows a release of CO2 in a similar magnitude as during other El Niño events not associated with volcanic eruptions. Probably, the severe concomitant El Niño event with the highest MEI value during the period of investigation enhanced the biospheric carbon release to an extent that overcompensated the aerosol effect [Gu et al., 2003]. Furthermore, the strong vegetation fires in 1983 [Malingreau et al., 1985] may have promoted the enhanced growth rate of the atmospheric CO2 concentration.

[51] A crucial point in the discussion of the large terrestrial sink in the early 1990s is that the reduction of the atmospheric CO2 growth rate began before the eruption of Mount Pinatubo [Keeling et al., 1995]. According to our results, the observed decline of the atmospheric CO2 growth rate previous to the eruption was due to the normal reaction to ENSO. These conditions last until the atmospheric aerosol content rises. Only after the eruption, MEI and Δaano show opposite trends indicating a volcanic effect.

[52] The effect of the NAO remains unclear. Nevertheless, it is worth mentioning that during part of the investigation period, the NAO pattern appears to correlate with the difference between Δaano and Δabio (Figure 4). Effects of the NAO on the terrestrial biosphere have already been considered in the HRBM through the climate data sets while impacts on the oceans' CO2 balance have not been accounted for. Gruber et al. [2001] assumed that low values of the NAO index correspond with higher CO2 uptake by the oceans. This would result, in our investigations, in a diminished Δoc and thus increased Δaano (equation (2)) and therefore reduce the difference between Δaano and Δabio. On the other hand, Le Quéré et al. [2000] state that NAO induces only little variability in the global sea-air flux of CO2.

[53] Calculating coefficients of correlation between the difference signal and the MEI yields weak negative values (r = −0.22 and r = −0.11 for the difference computed with and without taking into account a mean of 6 months for the transport lag, respectively). One could suppose that the temporary inverse behavior of the difference signal and the MEI is due to impacts of ENSO on the ocean-atmosphere carbon flux. That flux is often believed to react to ENSO in an opposite manner as the terrestrial biosphere thus taking up CO2 from the atmosphere during El Niño events [Francey et al., 1995; Keeling et al., 1995; Feely et al., 1999]. Additional calculations with our data using an atmospheric transport model are necessary to prove this hypotheses.

[54] We conclude that the atmospheric CO2 variations on the ENSO timescale are most probably dominated by the influences of global circulation anomalies because of their climatic effects. Further, we found indications that volcanic eruptions with considerable aerosol production (high atmospheric optical thickness) may create disturbances of the (biospheric) carbon cycle by increasing the photosynthetic carbon uptake due to the enhanced diffuse fraction of the incoming radiation. This effect may sometimes cancel out or even overcompensate the fluctuations of the atmospheric CO2 concentration caused by influences of ENSO. Finally, the variability of the time lag or phase shifts observed between the modeled and the observed CO2 anomalies may be due to influences of vegetation fires, but further investigations are required on this assumption.


[55] The authors wish to thank all who contributed to their work, especially E. Maier-Reimer, Max Planck Institute for Meteorology, Hamburg, Germany, for the ocean CO2 flux data, P. D. Jones and D. E. Parker, Climate Research Unit, University of East Anglia, for the temperature anomaly data, Klaus E. Wolter, NOAA-CIRES Climate Diagnostics Center, for the Multivariate ENSO Index data, and Corinne Le Quéré, Max Planck Institute for Biogeochemistry, Jena, Germany, who provided information and literature on the NAO. The transient precipitation data set G55WLD0098.DAT (Version 1.0) was constructed and supplied by Mike Hulme at the Climatic Research Unit, University of East Anglia, Norwich, U. K. His work has been supported by the U. K. Department of the Environment (contract EPG 1/1/14). We also gratefully acknowledge two anonymous reviewers whose constructive comments helped to improve the original manuscript.