In this study, we examine the role of the atmospheric circulation variability in reproducing the interannual variability of the observed atmospheric CO2 growth rate in the Northern Hemisphere. Results from a series of numerical experiments performed with a three-dimensional atmospheric transport model driven by European Centre for Medium-Range Weather Forecasts winds from 1979 to 1999 and forced at the bottom with an annually balanced, seasonally distributed (i.e., no interannual variations) terrestrial CO2 flux calculated by the Carnegie-Ames-Stanford Approach (CASA) ecosystem model show changes in atmospheric CO2 growth rates that are significantly correlated with those observed at many stations located in midlatitude to high-latitude regions in the Northern Hemisphere. Particularly, we find that the North Atlantic Oscillation (NAO) and the Pacific-North America (PNA) indices are correlated with the observed growth rates at Alert and Point Barrow. This suggests that where there is large variability in atmospheric circulation, it needs to be taken into account when changes in observed atmospheric CO2 at various stations are to be understood in terms of changes in CO2 sources. In numerical experiments in which biospheric CO2 fluxes from the CASA model are perturbed in a known arbitrary way, we find that the Northern Hemisphere observational sites do not appear to be sensitive to changes in the CO2 seasonal flux cycle except at close proximity and thus would not be useful for inferring a Northern Hemisphere biospheric sink. Results suggest sensitivity of the locations of some monitoring stations, with respect to atmospheric circulation patterns, in detecting major changes in the biospheric CO2 flux from certain regions in the Northern Hemisphere.
 The atmospheric carbon reservoir is in direct contact with three other major global reservoirs of oceans, land biosphere, and fossil fuel. Any change in atmospheric CO2 from year to year is a direct consequence of changes in carbon flux from each of these reservoirs. This interannual variation in the atmospheric CO2 concentration results in a year-to-year variation in its growth rate. Combustion of fossil fuel, with a resultant release of CO2 into the atmosphere, is an anthropogenic activity that has caused a steady exponential increase in the atmospheric CO2 content since the beginning of the Industrial Age in mid-1800s, from about 280 ppm to a present value of about 370 ppm (parts per million by volume) [Intergovernmental Panel on Climate Change (IPCC), 2001]. In recent years, fossil fuel contribution to the atmospheric CO2 increase trend has been around 1.4 ppm yr−1 [Ishizawa et al., 2002; IPCC, 2001]. However, it has contributed little to the interannual variation in the CO2 growth rate.
 Since the studies by Francey et al.  and Keeling et al. , the estimate of the magnitude of interannual variation in the net oceanic flux of CO2 has been decreasing [Francey et al., 1999; Morimoto et al., 2000], with the strongest statement of a relatively steady absorption by the oceans of around 2.0 GtC yr−1 coming from a study by Lee et al. . In a deconvolution calculation using a multibox INK model, Ishizawa et al.  showed that the net oceanic absorption of atmospheric CO2 has been increasing steadily from about 1.8 GtC yr−1 in 1978 to around 2.4 GtC yr−1 in 1998, with very little interannual variation. Therefore it is very likely that CO2 fluxes from the fossil fuel and ocean reservoirs have minimal impact on the interannual variability of atmospheric CO2.
 Previous studies have attempted to show that much of the interannual variability in atmospheric CO2 could be accounted for by the year-to-year change in the net biospheric flux [e.g., Francey et al., 1999; Lee et al., 1998; Morimoto et al., 2000; Ishizawa et al., 2002; Langenfelds et al., 2002]. Studies, such as Ito and Oikawa , have shown that existing global ecosystem models can produce interannual changes in the net ecosystem productivity (NEP) on the order of 2 GtC yr−1 and are often associated with occurrences of El Niño-Southern Oscillation (ENSO) events. Many of the El Niño events appear to be associated with a net transfer of CO2 from the biosphere to the atmosphere. ENSO events produce droughts, particularly in tropical regions, which can cause decrease in photosynthesis and increase in the frequency of wild forest fires [Keeling et al., 1989, 1995; Langenfelds et al., 2002].
 An event such as the eruption of Mount Pinatubo in June 1991 can cause a dramatic decrease in globally averaged surface temperature [Dutton and Christy, 1992]. This leads to a pronounced decrease in soil respiration [Ito and Oikawa, 2000], resulting in an abnormally reduced growth rate in the atmospheric CO2 concentration [Conway et al., 1994]. In a recent study, Gu et al.  found that a diffusion of solar radiation caused by an atmospheric aerosol loading due to the Mount Pinatubo eruption enhanced noontime photosynthesis, thereby contributing to the atmospheric CO2 growth rate decline after the eruption.
 However, the role of atmospheric circulation changes (either internally or externally driven) cannot be neglected, particularly in midlatitude to high-latitude regions in the Northern Hemisphere where circulation variability reaches high values. Higuchi et al.  have demonstrated, for example, that a significant portion of the observed quasi-decadal variation in the amplitude of the seasonal CO2 cycle (an indicator of the metabolism of the terrestrial biological activity) at high-latitude stations (such as Point Barrow and Alert) in the Northern Hemisphere can be accounted for mostly by changes in the atmospheric circulation alone. In a similar vein, it is possible that a recognizable portion of the interannual variation in the growth rate of atmospheric CO2, particularly in midlatitude to high-latitude locations in the Northern Hemisphere, can be accounted for by changes in the atmospheric circulation. Another long-lived greenhouse gas CH4, with an atmospheric lifetime of 9–10 years, also shows notable interannual variation in its growth rate. Like CO2, atmospheric CH4 has both anthropogenic and biological sources that could potentially influence year-to-year change in its growth rate. However, in a recent study by Warwick et al. , it was found that interannual variations in meteorology have a significant contribution to the observed interannual changes in the atmospheric CH4 growth rate. In this study, we investigate the similar effect on the atmospheric CO2 growth rate. Our study focuses on the middle-high Northern Hemisphere region because it is here that we observe large interannual variability in the atmospheric circulation; it is here also that there appears to be some observational evidence to indicate increased land biological activities [e.g., Keeling et al., 1996; Myneni et al., 1997; Randerson et al., 1999] that can contribute to the observed changes in the CO2 growth rate. We will show, at least from 1979 to 1999, that a noticeable portion of the observed changes in the growth rate of atmospheric CO2 in midlatitude to high-latitude regions in the Northern Hemisphere could have been driven by changes in the atmospheric circulation. We also attempt to show that because of their locations with respect to the variability in the atmospheric circulation pattern, some of the monitoring stations in the Northern Hemisphere are likely not sensitive to large changes in the biospheric flux.
2. Methodology and Data
 As was the case in our previous study [Higuchi et al., 2002], we employ a validated three-dimensional atmospheric transport model. The period of model integration is from 1979 to 1999, the period for which we have the European Centre for Medium-Range Weather Forecasts (ECMWF) wind data. Interannual variations of atmospheric CO2 growth rate simulated by the model are compared to those obtained at various background monitoring stations (11 in total) located at various latitudes in the Northern Hemisphere. Stations used in this study are listed in Table 1 and their locations are shown in Figure 1. Except for Alert and Cape St. James, the CO2 data from the stations used in this study are obtained from the National Oceanic and Atmospheric Administration (NOAA) ftp sites (ftp.cmdl.noaa.gov/ccg/co2/in-situ for Point Barrow and Mauna Loa and ftp.cmdl.noaa.gov/ccg/co2/flask for other NOAA stations). The data for Alert (1988 to 1999) and Cape St. James (1979 to 1992) are obtained from the Meteorological Service of Canada (MSC). Daily mean CO2 data are used for analyses at Alert, Point Barrow and Mauna Loa where in-situ measurements are made. These daily data are based on only those hourly averages deemed to be “background” values (see the above NOAA ftp site for in-situ measurement). For other stations, the CO2 data obtained from flask sampling are used. We use only those flask data which are thought to be of “background” values (data regarded as “retained” values by NOAA (see the above NOAA ftp site for flask measurement)). On a sampling day, flask air samples are usually collected in pairs whose concentration values are averaged to obtain a single daily value for that day.
Table 1. List of Background CO2 Monitoring Stations Used in This Study
Location (Latitude, Longitude)
Mould Bay (MBC)
Point Barrow (BRW)
Station M (STM)
Cold Bay (CBA)
Shemya Island (SHM)
Cape St. James (CSJ)
Cape Meares (CMO)
Key Biscayne (KEY)
Mauna Loa (MLO)
 The three-dimensional atmospheric transport model NIRE-CTM-96 we use in this study is the same one that was used by Higuchi et al.  study, and has been used to investigate temporal trends and seasonal cycles of various trace gases in the atmosphere, including CO, CO2 and radon [Taguchi, 1996; Taguchi et al., 2002a, 2002b, 2003]. We perform three numerical experiments with the model; all with the same observed wind field that vary from year to year. In the first experiment, the model is driven by all the CO2 forcings (ocean, fossil fuel and biosphere) that do not contain interannual variability. In the second experiment, the model is driven only by the land biosphere. The third experiment is identical to the second one, except for specified perturbations in the seasonality of the terrestrial biospheric forcing.
 In our first experiment (Exp1), the model is driven from 1979 to 1999 by analyzed winds from ECMWF (1979–1993, reanalysis; 1994–1999, operational analysis). The atmospheric CO2 field in the model is driven by (1) the fossil fuel emission of 1995 (globally 6.17 GtC yr−1 emission), (2) terrestrial biospheric flux (annually balanced), and (3) oceanic flux (globally 2.19 GtC yr−1 net absorption). These CO2 forcing functions are identical to the ones that were prepared for the recent international atmospheric CO2 transport model intercomparison exercise (TransCom3) supported by IGBP as part of the GAIM activity, and are described by Gurney et al. . The three CO2 flux functions do not change from year to year, and therefore have no interannual variability. Therefore all the interannual variability in the atmospheric CO2 simulation in Exp1 is derived from the atmospheric transport alone. Therefore Exp1 will indicate the amount of contribution made by year-to-year changes in the atmospheric circulation pattern toward the observed interannual variation of the CO2 growth rate at various monitoring stations.
 The second experiment (Exp2) is identical to Exp1, except that the fossil fuel and the oceanic fluxes are not used. The only CO2 flux forcing is obtained from the annually balanced, seasonally distributed (the growing season net CO2 absorption in each continental region ranges from 0.56 to 1.65 GtC yr−1) terrestrial biospheric CO2 source function estimated by the Carnegie-Ames-Stanford Approach (CASA) ecosystem model [Potter et al., 1993] for the 11 continental regions that were identified by Gurney et al. . Seasonally varying (but not interannually varying) biospheric flux for each region is identical to the one used in the Gurney et al.  study. These biospheric CO2 flux regions are listed in Table 2. Exp2 identifies the impact of changing atmospheric circulation on changes in the modeled CO2 growth rate due only to biospheric CO2 flux that does not have interannual variation.
Table 2. Eleven Biospheric CO2 Flux Regions That Were Employed by Gurney et al.  and are Used in This Study to Drive the Atmospheric Transport Modela
Correspondence of the names of each biospheric region used by Gurney et al.  to those in this study is also shown.
Boreal North America
Temperate North America
 In the third experiment (Exp3), the design is similar to Exp2 except that the seasonal distribution of the CASA biospheric flux in each of the 11 different vegetation regions is perturbed in a known arbitrary way to examine the sensitivity of biospheric flux variability on the interannual variation of the CO2 growth rate simulated for selected monitoring stations. We perform four simulation runs, each with different amount of biospheric flux perturbation; the wind field is the same as the one used in Exp2. In the first simulation run, the biospheric emission/absorption (initially balanced annually) in each region is forced to change its net CO2 flux by +0.1 GtC in the first year, and then by −0.1 GtC the second year. (We will designate this run as the 0.1 GtC run.) That is, each flux region is forced to go through a 2-year sink-source cycle. This is accomplished by shifting the “zero line” in the modeled seasonal cycle by an appropriate amount. Physically, for each year, this introduces a change in the length of the growing season, as well as in the source-sink strength. By shifting the zero line upward, growing season and the magnitude of the summer CO2 absorption are increased, while a reverse situation is achieved by shifting the zero line downward. In the remaining three simulation runs, the annual net flux is changed by ±0.2 GtC (0.2 GtC run), ±0.5 GtC (0.5 GtC run), and ±1.0 GtC (1.0 GtC run). The effect due to changes in the atmospheric circulation is eliminated by subtracting the results of Exp2 from those obtained in Exp3. The reason for imposing a 2-year oscillation in the biospheric carbon flux is motivated by the presence of quasi-biennial oscillations in the growth rate detectable in Figure 2 discussed in section 3.1. If we assume the latter to be “noise” caused by changes in the atmospheric circulation, we want to obtain a qualitative estimate of how much perturbation in the biospheric forcing of similar periodicity is necessary to detect a signal of this perturbation in the modeled growth rate.
 The daily simulated values are obtained by averaging the 6-hourly outputs in the same day. In Exp1 and Exp2, to compare with the observed CO2 values, we use only the daily simulated values for the days when the observed data are available. On the other hand, in Exp3, all the daily simulated values for 1979–1999 are used since comparison with observed data is not made.
 For both the model simulations and the observations, the CO2 concentration values are passed through a curve fitting method [Nakazawa et al., 1997] to obtain atmospheric CO2 growth rate. In this iterative procedure, the fundamental and its 1st and 2nd harmonics are used, in what is called a three-harmonic fit. Growth rate is obtained by differentiating the trend.
3. Results and Discussion
Figure 2 shows the results of Exp1. For each station used in this study, the observed growth rate is compared to the model growth rate forced by the fossil fuel, the oceans and the biosphere (M-ALL). Also shown are the relative contributions by the fossil fuel, the oceans and the biosphere individually. (Hereinafter, M-BIO will be used to designate model growth rate forced by the CASA biospheric flux from all the biospheric regions.) In this figure, interannual variations with a period of about 2 to 4 years appear in observed atmospheric CO2 growth rate at each station. Also, the model growth rate (M-ALL) at each station (especially those located at northern middle to high latitudes) shows interannual variation with a similar period, although magnitude and phasing of respective peaks and troughs do not always agree between the observation and the simulation. Agreement between observation and simulation worsens at low-latitude stations such as Key Biscayne and Mauna Loa, and very high growth rates observed in 1998 at all stations associated with the anomalous El Niño event are not fully simulated. Since there are no interannual variations in the CO2 forcing functions (fossil fuel, oceans, biosphere), any agreement between the M-ALL results and the observation demonstrates the importance of the role variation in the atmospheric circulation plays in the observed changes in the atmospheric CO2 growth rate. The differences in the growth rate variation between the simulation and the observation can be partially explained by contributions from interannual variations in CO2 sources/sinks, as have been suggested in many previous studies. For example, discrepancies in the magnitude of the peak in 1998 and the trough in 1992 suggest contributions of anomalous net CO2 emission and absorption, respectively. Many researchers have related anomalies in the growth rate associated with ENSO events and volcanic eruptions to changes in the CO2 exchange between the atmosphere and the terrestrial biosphere due to climate change induced by these events. However, the fact that some of the growth rate anomalies associated with these events can be simulated by M-ALL to some degree (to a large degree at some stations) shows that changes in the atmospheric circulation due to climate change induced by these events do make nonnegligible contributions to the CO2 growth rate anomalies.
 In Figure 2, it can also be seen that both the fossil fuel and the oceans contribute relatively little to the overall interannual variation in the growth rate and that at each station, transport of CO2 flux signal from various biospheric regions (M-BIO) contributes a significant portion. Therefore the above discussions can also be applied to M-BIO.
 Latitudinal distribution of the standard deviation of the observed growth rate, along with those obtained from M-ALL and M-BIO, is shown in Figure 3. Overall, the observed standard deviation, which is indicative of the magnitude of interannual variability in the growth rate, shows a distribution with most values ≥1.0 ppm yr−1 and appears to be independent of latitude, except perhaps near subtropical regions where there is an indication of tendency toward lower values. M-ALL shows a similar pattern of latitudinal distribution (i.e., independent of latitude), but with a more pronounced indication of lower standard deviations in the subtropics (Mauna Loa, Key Biscayne, Midway). M-ALL also shows that it can produce variability in the growth rate that is more than half, on average, of the observed growth rate variability in the polar region (Alert, Mould Bay, Point Barrow) but less so toward the equator. Sandwiched between these two regions, M-ALL displays less consistency in its agreement with the observation, showing near agreement at Station M and Cape Meares but noticeably less so at Cape St. James, Shemya Island and Cold Bay. A large portion of the standard deviation simulated by M-ALL can be accounted for by M-BIO, indicating that at least in the model, the year-to-year change in the atmospheric transport of CO2 from biospheric sources without any interannual variations makes a significant contribution to the interannual variability of the CO2 growth rate.
Figure 4 shows the correlation of the observed growth rate with those simulated by M-ALL and M-BIO at each monitoring station as a function of latitude. It is interesting to note that the correlation between the observed and M-ALL decreases from about 0.5 in the polar region to near −0.2 in the subtropical region; using F test, we find that only 4 stations (Alert, Mould Bay, Point Barrow, and Shemya Island) show statistical significance at 95% confidence level. With the CASA-biospheric forcing only, M-BIO shows a similar latitudinal distribution pattern in general but with an improvement in correlation values, and the number of stations showing statistical significance increase to seven, mostly in the high-latitude region near 60°N (additional stations, besides the four given above, are as follows: Station M, Cold Bay, and Midway). A significant correlation indicates a certain degree of phase matching between any two time series being compared. Therefore stations Cape Meares (45°N) and Cape St. James (52°N), which showed similar standard deviations between the observed and M-ALL (and M-BIO) but have failed to show any correlation, indicate that there are other processes involved (which are not simulated by the present model) that contribute to the observed changes in the growth rate. (One possibility is that these stations see certain biospheric source regions that are not simulated very well (in terms of phasing of the seasonal flux) by the CASA model.)
 In addition to the same-order-of-magnitude standard deviations between the observed and the M-BIO, all the stations to the north of these two stations (Cape St. James and Cape Meares) show significant correlations. This means that on the basis of the values shown in Figure 3, at least about half of the interannual variability observed in regions north of ∼50°N can be accounted for by the phasing of the year-to-year change in the atmospheric circulation with the seasonal phasing of the biospheric CO2 flux.
3.2.1. Identification of Biospheric Source Regions
 We have shown the importance of atmospheric transport of biospheric CO2, compared to the fossil fuel and the oceanic fluxes, in the amount of contribution it makes to the observed changes in the growth rate at various monitoring stations in northern midlatitude to high-latitude regions. Therefore we now examine in more detail how the growth rate simulated by M-BIO at a particular monitoring station is influenced by 11 biospheric regions, using the standard deviations of the growth rate as a reference. A map of the 11 biospheric regions (listed in Table 2) is graphically displayed in Figure 5. We choose those stations that have shown statistically significant correlations with the observed growth rate (see Figure 4). Table 3 shows relative contributions (in terms of standard deviation) from individual biospheric sources to the variability of the growth rate simulated by Exp2 at each of the stations. Values (Ai,j) in Table 3 are calculated as
where σi,j is the standard deviation of the CO2 growth rate simulated in Exp2 for station i contributed by biospheric region j. Note that a standard deviation of M-BIO for station i is not equal to σi,j.
Table 3. Relative Contributions (Percentage of Standard Deviation) of Individual Biospheric CO2 Flux Regions to the Variability of the Growth Rate Simulated in M-BIO at Each of the Stations That Show Statistically Significant Correlation With the Observation
 All seven stations are influenced, to a varying degree, by biospheric CO2 from Canada, boreal Asia, Europe and the United States. In relative terms, biospheric CO2 from boreal Asia (Siberia) makes the largest contribution to the growth rate simulated at all the stations except Station M. Not surprisingly, given its geographical location, Station M experiences the greatest contribution from the European flux. A relatively large portion of the growth rate simulated for Shemya Island and Mould Bay is accounted for by boreal Asia. While Alert and Midway are influenced also by boreal Asia, they also come under the influence of the United States and Temperate Asia, respectively. Canada makes the largest contributions at Mould Bay, Cold Bay and Point Barrow. It is interesting to note that the effect at Cold Bay from boreal Asia is relatively small compared to the effect at Shemya Island, even though these two stations are geographically close. This points to the sensitivity of the location of a monitoring station as to which biospheric source its growth rate change is influenced by. In the case of Cold Bay and Shemya Island, the counter-clockwise circulation of the Aleutian Low located over the Gulf of Alaska advects air from Canada more to Cold Bay than to Shemya Island, allowing the latter to come under the influence of westerly flow from boreal Asia.
3.2.2. Growth Rate Changes at Alert and Point Barrow
 It was shown in Table 3 that high-latitude stations such as Alert and Point Barrow are influenced to various degrees by the way biospheric CO2 is transported from boreal Asia, Canada, United States and Europe. These are the four main biospheric regions that surround these stations. It is of interest, therefore, to explore further the relationship between the interannual variability in the atmospheric circulation and the CO2 growth rate at Alert and Point Barrow derived from continuous measurements. Two major low-frequency variability modes that influence the interannual variability of the atmospheric circulation in middle to high latitudes in the Northern Hemisphere are the Pacific-North America (PNA) and the North Atlantic Oscillation (NAO) teleconnection patterns [Wallace and Gutzler, 1981; Barnston and Livezey, 1987].
Figure 6 shows the normalized PNA and the NAO indices, along with the observed growth rates at Alert and Point Barrow. The plotted PNA and the NAO indices are obtained by running a 2-year weighted running mean on monthly data downloaded from the University of Washington website (http://tao.atmos.washington.edu/data_sets/). Monthly values of observed growth rate at Alert are correlated with the NAO and the PNA at −0.35 and +0.40, respectively, both statistically significant at 95% confidence level. Monthly values of observed growth rate at Point Barrow are similarly correlated at the statistically significant values of −0.25 and +0.42 with the NAO and the PNA, respectively. These relationships provide a strong indication of an important control the atmospheric transport has on the rate of interannual variation of atmospheric CO2 observed at these stations. The line of argument provided so far leads to the interpretation that the interannual variations of the growth rates at Alert and Point Barrow depend significantly on which of the four major biospheric regions (boreal Asia, Canada, United States and Europe) the two stations are exposed to from year to year. This depends on the atmospheric circulation patterns. (How actual change in the biospheric CO2 emission/absorption itself in each region impacts on the stations will be addressed in the next section.)
 Although there seems to be some statistical evidence to suggest that the intensity of Aleutian Low is negatively correlated with that of the Icelandic Low in winter [Honda et al., 2001] and that there is a limited relationship between the NAO and the PNA modes of oscillation [Huang et al., 1998], positive and negative phases of the PNA and the NAO can occur in different combinations with various intensities. This produces a more complicated relationship between the circulation tendencies induced by the low-frequency variability modes and their impact on the CO2 growth rates at Alert and Point Barrow. In addition, the seasonal cycle of the biospheric flux is present throughout the year, but both the PNA and the NAO are dominant during the winter season only. This can produce a problem related to the phasing of the seasonal CO2 flux (emission in winter and absorption in summer) with the atmospheric transport associated with the low-frequency variability modes.
 Although the relationship between the low-frequency variability modes and the interannual variation in the growth rate observed at Alert and Point Barrow is not always clear because of the problems identified above, it is interesting to note that the PNA and the NAO that cause significant anomalies in the wintertime circulation are correlated with the observed growth rate at these stations. The dominance of the circulation anomalies caused by the PNA and the NAO in advecting CO2 during wintertime that contributes to changes in the growth rates at Alert and Point Barrow suggests the importance of biospheric respiration during the winter season in contributing to the interannual variability of the growth rate in high-latitude regions. The importance of wintertime biogenic sources of CO2 in high-latitude regions has been identified by Yuen et al.  and Brandefelt and Holmén , and our results are consistent with their interpretation.
 Ideally, during the positive phase of the PNA, with anomalous low over the Gulf of Alaska and anomalous high over western Canada, there is an increased tendency for Alert and Point Barrow to be influenced by the biospheric CO2 from Canada and the United States. During the negative phase, the anomalous centers change sign, with high over the Gulf of Alaska and low over western Canada. With this anomalous flow configuration, Alert and Point Barrow increasingly come under the influence of air transported from boreal Asia (Siberia).
 The positive phase of the NAO over the Atlantic Ocean is associated with a strengthened Icelandic Low. This provides a very strong cyclonic circulation over the eastern Arctic, with a resulting tendency to transport CO2 from Europe and boreal Asia to Alert and Point Barrow. During the negative phase, the Icelandic Low is weaker and the biospheric influence from Europe and boreal Asia (Siberia) is decreased, while there is an increased influence from the North American biospheric sources. It is also interesting to note that although the NAO is most prominent during the winter season, as noted above, its influence on the subsequent summer circulation over the high-latitude Northern Hemisphere region has been detected [Ogi et al., 2003]. The authors have found that this phenomenon is caused by the “memory retention” of the wintertime NAO by the sea-surface temperature, sea-ice and snow cover. For the wintertime positive NAO, the summertime geopotential heights over British Isles, central and eastern Siberia and subarctic North America are higher than normal. With anomalous positive heights over Siberia, the boreal Asia biospheric source continues to influence Point Barrow and Alert into the summer season. Opposite effect is seen for the wintertime negative phase of the NAO.
Figure 7 shows the relative contributions the four biospheric regions make to the M-BIO simulated growth rates at Alert and Point Barrow. Table 4 shows the correlation of the growth rate due to each biosphere (Exp2) with that of M-BIO, for both Alert and Point Barrow. Not only does boreal Asia make the largest contribution to the interannual variation of the M-BIO growth rate at both locations (see standard deviation values in Table 4), it also has the highest correlation with the M-BIO growth rate. The phasing of the seasonal biospheric CO2 flux and the atmospheric transport makes an unambiguous and clear interpretation of Figure 7 difficult, but it is possible to delineate certain predominant features. It appears, for example, that large growth rate values (both positive and negative) at Alert occur when the contributions from all four regions are in relatively good phase. Such is the case from 1988 to 1990, and from 1995 to 1997. It is also interesting to note in Table 4 that the correlations between Canada, United States, Europe and M-BIO are high (0.61, 0.60 and 0.64, respectively) at Alert, but contributions from these three biospheric regions to the interannual variation of the growth rate are relatively low (see Figure 7 also). Table 5 shows that the growth rate contributions from Canada and the United States are in phase (correlation at 0.78) and those from Europe and boreal Asia are in phase (correlation at 0.80) at Alert. This is consistent with the above discussion related to anomalous circulations induced by the low-frequency variability modes. However, for the 1988–1999 period, there seems to be no clear relationship between the Canada/United States and the boreal Asia/Europe regions.
Table 4. Correlation of Growth Rate Produced by Each of the Four Biospheric Sources With That Produced by M-BIO for Alert and Point Barrowa
All values are statistically significant at 95% confidence level. Numbers given in parentheses are standard deviations of the growth rates produced by the four biospheric sources. Standard deviations of the M-BIO growth rates are 0.57 and 0.59 for Alert and Point Barrow, respectively.
Table 5. Growth Rate Produced by a Biospheric Region Correlated With Those Produced by the Other Regionsa
Bold numbers indicate statistical significance at 95% confidence level.
 Given the geographical proximity, Point Barrow provides similar results to those obtained for Alert. As was the case at Alert, large positive and negative growth rate values at Point Barrow occur when the contributions from all four regions are in relatively good phase (Figure 7). For example, from 1980 to around 1985, growth rate contributions from all the biospheric regions are in relatively good phase, particularly in the early part, producing a growth rate cycle with a period of about 2 years. As indicated in Table 4, boreal Asia plays the most significant part in influencing the magnitude of the growth rate at Point Barrow. From around 1986/1987 to 1994, Canada/United States (correlated at 0.67 with each other) and boreal Asia/Europe (correlated at 0.52 with each other) become out of phase. Table 5 shows that Canada/United States and boreal Asia/Europe are negatively correlated at around −0.31 to −0.40 (statistically significant at 95% confidence level) for that period. Because of this, the growth rate during the 1986/1987–1994 period is characteristically different from the time before and after this period. The large positive and negative growth rates in 1995 and 1997, respectively, are due mostly to the boreal Asia influence.
 The importance of the phasing of the seasonality of the biospheric CO2 flux and the interannual variation in the atmospheric transport in determining the year-to-year change in the growth rate can be appreciated by examining the contribution boreal Asia makes to the seasonal atmospheric CO2 cycle at Alert. Recall that among the 11 biospheric regions, boreal Asia contributes the most to the simulated interannual variation of the growth rate for Alert. In Figure 8, we show the modeled daily CO2 values at Alert due to contribution by air transported from the boreal Asia region. The growth rate and the fitted seasonal curves are obtained by applying the method developed by Nakazawa et al.  to the daily values. It is important to note in Figure 8 that for ease in comparison between the seasonal cycle and the growth rate, the CO2 seasonal cycle has been shifted by −6 months, to compensate for the lag introduced by the calculation of the growth rate from the differentiation of the trend in the atmospheric CO2 concentration. That is, there is a lag between the rate of change of atmospheric CO2 and the actual CO2 concentration at a particular time. Figure 8 suggests a relationship between the growth rate and the degree to which Alert is influenced by the CO2 from boreal Asia during different seasons. If atmospheric transport is such that influence of air from boreal Asia on Alert is enhanced during the summer season in one year, and then suppressed during the summer in the following year, the resulting growth rate cycle will show a rise over the period. For example, increase in the summer minimum from one year to the next has resulted in an increase in the growth rate in early 1990. This event is marked as event 1 in Figure 8. Same interpretation can be applied to a maximum growth rate in early 1996, identified as event 2 in the figure. Event 3 shows an opposite situation in which the summer minimum decreases from one year to the next, resulting in a minimum growth rate during 1991 and 1992. It is interesting to note that the negative growth rate in 1994 (marked as event 4) is associated with a decrease in winter CO2 arriving from boreal Asia. In this way, growth rate changes at a monitoring station can be effected by changes in atmospheric transport alone, without invoking any change in the biospheric CO2 flux.
 The role of atmospheric transport is significant in explaining a large portion of the interannual variation observed in the atmospheric CO2 growth rate at midlatitude to high-latitude regions in the Northern Hemisphere. Changes in the atmospheric transport dictate when and which biospheric CO2 source influences atmospheric CO2 observed at a monitoring station. Given the relative impact of changes in the atmospheric circulation plays in the interannual variation of the CO2 growth rate, we now pose the following question: Which, and by how much, biospheric flux needs to be perturbed for the monitoring stations to detect a change in the growth rate above that produced by changes in the circulation? In order to detect interannual changes in biospheric CO2 flux in atmospheric CO2, it is necessary to eliminate the effect of interannual atmospheric circulation variability. This leads us to our third experiment.
 The purpose of the third experiment (Exp3) is to identify the effect of CO2 flux changes in each of the 11 biospheric regions on the growth rate simulated by M-BIO at each of the 11 monitoring stations. We perform four simulations in Exp3 and, as described in section 2, these are identified as the 0.1, 0.2, 0.5, and 1.0 GtC runs. In order to eliminate the effect of interannual variation in the atmospheric transport, we subtract the results of M-BIO from those of Exp3. Results from the 0.2 to 1.0 GtC runs are presented in Table 6 as relative changes with respect to the M-BIO result at 11 monitoring stations. The results from the 0.1 GtC run are not shown since they indicate smaller changes than those resulting from the ±0.2 GtC perturbation. The values (Ai,k,l) shown in Table 6 are calculated as
where i (=1, 2, 3, …, 11) denotes the monitoring station influenced by the kth biospheric region. The parameter l indicates the experiment number (0.1, 0.2, 0.5, and 1.0); l = 0 denotes the normal case (M-BIO) in which the biospheric CO2 source function is not perturbed. The variables σ(MBi,k,l) and σ(MBIOi) are the standard deviation of the time series MBi,k,l and MBIOi, respectively, for the period 1979–1999 and are calculated as
where Ci,j,l is the CO2 transported to station i from biospheric region j (= 1, 2,…, 11) in experiment l, and G is a function to obtain the CO2 growth rate from the CO2 values using the curve fitting method described above [Nakazawa et al., 1997]. MBIOi represents contributions from all the 11 biospheric regions at station i for the normal case (l = 0).
Table 6a. Relative Changes (Expressed in Terms of Percentage) in the Growth Rate Variation With Respect to the M-BIO Results at 11 Monitoring Stations, as a Result of Imposed 2-Year Sink-Source Cycle With Amplitude 0.2 GtC at Each of the 11 Biospheric Regions
Table 6b. As in Table 6a, but With Sink-Source Cycle With Amplitude 0.5 GtCa
Bold numbers indicate statistical significant at 95% confidence level.
Table 6c. As in Table 6a, but With Sink-Source Cycle With Amplitude 1.0 GtCa
Bold numbers indicate statistical significant at 95% confidence level.
 By imposing a source-sink CO2 flux of ±0.2 GtC yr−1 with a 2-year cycle in each of the 11 biospheric regions, we see that changes relative to the M-BIO results are small (Table 6a). The 0.2 GtC run shows a pattern of effect similar to that produced by the 0.1 GtC run, except that the magnitude of the effect is slightly greater. The results obtained from the 0.1 GtC and 0.2 GtC runs are not significantly different from the M-BIO results.
 With the 0.5 GtC run (Table 6b), Canada produces the greatest impact at Alert and Mould Bay (with about 40% increase in the magnitude of the growth rate), with smaller impact (on the order of 20–30% increase) at Point Barrow. Effects from other biospheric regions do not change very much from those detected in the 0.1 and 0.2 GtC runs.
 With a relatively large perturbation in the 1.0 GtC run, we see a noticeable contribution from a few Northern Hemisphere biospheric regions other than Canada (Table 6c). Very high-latitude stations (Alert, Mould Bay, Point Barrow) continue to be affected by the perturbation in the Canada-biospheric emission/absorption, with the growth rate at Mould Bay experiencing a near 100% increase in magnitude. The growth rate at Cape St. James (CSJ) increases by about 30% because of the contribution from the Canada region. However, we begin to see noticeable contributions from the boreal Asia-biospheric region at these stations, as well as at Shemya Island. Consistent with the results of M-BIO shown in Table 3, the effect at Cold Bay from boreal Asia is small compared to the effect at Shemya Island. The effect from the United States-biospheric region is noticeably visible at Cape Meares (CMO), with greater than 20% increase in the growth rate amplitude, but nowhere else. Again, it is interesting to note that even though Cape St. James is geographically close to Cape Meares, it is not affected by the United States-biospheric perturbation to the extent Cape Meares is. In some cases, the location of a monitoring station appears to be very sensitive as to which biospheric source perturbation its growth rate change is influenced by. Station M shows no noticeable change, even from the perturbation in the European flux. (In the M-BIO results shown in Table 3, Station M shows the greatest contribution from the European flux.) Perturbations in CO2 fluxes in other biospheric regions have no significant effect on the growth rate amplitudes at the 11 stations.
 Although the 0.5 GtC and 1.0 GtC runs show appreciable increase in the amplitude of the growth rate change at some stations, some of these may not be significantly different from the M-BIO results. Using the F test to detect any significant changes in the standard deviation of the growth rate time series from those obtained in M-BIO, we find that for the 0.5 GtC run, Canada produces statistically significant (at 95% confidence level) impact at Alert, Mould Bay and Point Barrow. For the 1.0 GtC run, statistically detectable effects are seen at Alert, Mould Bay, and Point Barrow (due to the Canada-biospheric region), Cape Meares (due to the United States-biospheric region), and Alert, Mould Bay, Point Barrow and Shemya Island (due to the boreal Asia-biosphere region). Tropical/subtropical biospheric regions do not have any statistically significant impact on the subtropical stations in the Northern Hemisphere stations. This interesting result might indicate sensitivity of the locations of some low-latitude monitoring stations, with respect to low-latitude atmospheric circulation patterns, in detecting major changes in the biospheric flux from certain tropical/subtropical regions. This issue of location sensitivity has already been mentioned above in reference to Cold Bay and Shemya Island in detecting perturbations in the boreal Asia CO2 flux emission/absorption.
 Previous studies, mainly relying on analyses of stable isotopes of CO2, have invoked changes in the biospheric CO2 flux to account for interannual variations in the atmospheric CO2 growth rate observed at background monitoring stations. In the present study, we have provided evidence to suggest that changes in the atmospheric circulation, and thus in CO2 transport, can contribute significantly to the observed year-to-year change in the growth rate, particularly at certain sites in midlatitude to high-latitude regions in the Northern Hemisphere.
 Using a three-dimensional atmospheric transport model, we have conducted numerical experiments to separate the relative contributions changes in the atmospheric transport and the biospheric emission/absorption make to the interannual variation in the growth rate. In one experiment, the model was driven from 1979 to 1999 by observed winds; the atmospheric CO2 field was forced by annually balanced, seasonally changing biospheric CO2 flux calculated by the CASA ecosystem model for the 11 biospheric regions that were identified in the TransCom 3 project [Gurney et al., 2002]. The experiment, therefore, identified the amount of contribution made toward the observed interannual variation in the atmospheric CO2 growth rate by the year-to-year change in the atmospheric circulation in transporting biospheric CO2 (with the same seasonal cycle year after year) to various monitoring stations. Eleven stations were selected to cover latitudinal distribution from high Arctic (Alert) to subtropical region (Mauna Loa). It was found that at most stations, simulated standard deviation of the growth rate amounted to at least 50% of the observed standard deviation. Furthermore, changes in the growth rate observed at high-latitude stations are more susceptible to changes in the atmospheric circulation; in particular, we found that the NAO and the PNA indices are correlated with the observed growth rates at Alert and Point Barrow. These and other northerly stations were found to be influenced, to a varying degree, by biospheric CO2 transport from Canada, boreal Asia, Europe and the United States. The significance of the role of the atmospheric transport results from a phase relationship between changes in the circulation regime and the seasonal biospheric CO2 flux. It is therefore a mistake to invoke interannual changes in the biospheric CO2 flux as the only cause for the observed changes in the atmospheric CO2 growth rate.
 In another experiment, we imposed a 2-year sink-source cycle with various amplitudes to each of the 11 CASA biospheric emission/absorption vegetation regions, to examine the effect of interannual biospheric flux variability on the atmospheric CO2 growth rate change. For relatively small perturbations (cycle amplitudes of 0.1 to 0.2 GtC), all the biospheric regions did not produce, at the stations used in this study, effects that were significantly different from the M-BIO results statistically. However, for a relatively large perturbation of 1.0 GtC amplitude, only the biospheric emission/absorption from Canada and boreal Asia showed any relatively significant impact, and only at those stations located mostly in high-latitude regions. The results of Exp3 suggest that interannual variation in the net biospheric flux needs to be very large (on the order of 0.5 to 1.0 GtC yr−1) for it to be detectable in the growth rate change as being significantly different from that produced solely by changes in the atmospheric transport. Even then, only the high-latitude stations can detect changes in the CO2 flux from just the Canada and boreal Asia biospheric regions. Further investigation with different experimental designs is needed to verify the robustness of the Exp3 results and their interpretation.
 One of the poignant messages of our study is that many of the monitoring stations located in the midlatitude to high-latitude regions in the Northern Hemisphere may not be optimally located to detect changes in CO2 fluxes from biospheric ecosystems in these regions. The “noise” caused by interannual variability in the atmospheric circulation could interfere with the detection of changes in the biospheric CO2 flux that might be taking place right now. A reduction of the uncertainty in estimations of biospheric CO2 sources and sinks from atmospheric CO2 concentration measurements requires, among other things, a relatively accurate reproduction of low-frequency variability modes, such as the NAO and the PNA, with some defined level of agreement with the observation. The results of our study show that these variability modes in the atmosphere make significant contributions to the interannual variations in the atmospheric CO2 growth rate in the midlatitude to high-latitude monitoring stations like Alert and Point Barrow in the Northern Hemisphere. If we want the “background” stations to “see” changes in the biospheric CO2 sources, then our results indicate, at least qualitatively, that a placement of “background” stations must be considered within the context of the pattern of atmospheric circulation variability, and in particular with respect to low-frequency variability modes. In the discussion related to the results obtained for Cold Bay and Shemya Island in Exp2, for example, we showed that these stations, which are closely located, are influenced by different biospheric source regions. We also showed that in Exp3, Cape Meares is affected more by the CO2 flux perturbation in the United States-biospheric region than Cape St. James, even though these two stations are located geographically close. A strategy for station placement to detect biospheric CO2 emission changes should involve a careful diagnosis of atmospheric circulation variability.
 We would like to acknowledge Doug Worthy (Meteorological Service of Canada) for making available the CO2 data from Alert and Cape St. James. We would also like to thank Hidekazu Matsueda (Meteorological Research Institute, Japan) for his useful comments and discussions. Thoughtful and constructive suggestions and comments by the anonymous reviewers on the original submission have helped to improve the paper.