During the mid-1960s, large seasonal amplitudes were observed in surface measurements of Δ14C in the Northern Hemisphere. These seasonal oscillations were initially caused by stratosphere-troposphere exchange, with the injection of bomb 14C into the troposphere during winter and spring mixing. Here we show how fossil, ocean, and terrestrial biosphere fluxes modified the stratospheric signal during the 1960s, and the evolution of each of these components in the post bomb era. In our analysis, we used the Goddard Institute for Space Studies (GISS) atmospheric tracer model, gross ocean CO2 fluxes from the Lawrence Livermore National Laboratory (LLNL) ocean model, and terrestrial CO2 fluxes from a biosphere-atmosphere model driven by normalized difference vegetation index and surface air temperatures. We found that 14C-depeleted respiration from the terrestrial biosphere partially canceled the 14C-enriched stratosphere flux in the Northern Hemisphere in the mid and late 1960s. In more recent decades, our analysis suggested that the terrestrial biosphere contribution to the Δ14C seasonal cycle reversed phase, with the terrestrial biosphere currently releasing relatively 14C-enriched CO2 that mixes with relatively depleted troposphere CO2. The timing of this reversal depended on the residence times of carbon within the footprint of the observation station. Measurements of Δ14C in respiration from tundra and boreal ecosystems in Alaska provide evidence that some boreal forests have undergone this transition, while some tundra ecosystems have not. We predict that over the next century, several features of the latitudinal profile of Δ14C will substantially change because of continued fossil fuel emissions in the Northern Hemisphere, and the partial release of bomb 14C that has accumulated in Southern Hemisphere oceans.
 At the time of the Test Ban Treaty of 1963, 14CO2 was distributed unevenly through the atmosphere [Telegadas, 1971]. Nuclear fireballs had carried most bomb-produced 14C into the lower stratosphere, where it was quickly oxidized to 14CO and 14CO2 [Johnston et al., 1976; Glasstone and Dolan, 1977]. As many of the tests were conducted in the Northern Hemisphere (Table 1) [Enting, 1982], a significant amount of this 14CO2 subsequently reentered the Northern Hemisphere troposphere during winter and spring when mixing occurred between the stratosphere and troposphere [Rasch et al., 1994; Holton et al., 1995]. Southern Hemisphere troposphere 14CO2 concentrations lagged behind Northern Hemisphere values for several years in the mid-1960s, reflecting relatively slow inter-hemispheric mixing times and dilution from gross CO2 exchange with oceans and the terrestrial biosphere [Levin et al., 1980; Tans, 1981; Nydal and Lövseth, 1983].
Table 1. Hemisphere Distribution of Nuclear Weapon Testsa
 While the value of atmospheric Δ14C as a tracer of the carbon cycle has come largely from its mean annual secular trend, seasonal and regional variability in atmospheric Δ14C is clearly visible in many contemporary observations and has the potential to provide additional information about the source, age, and magnitude of regional fluxes. In general, four processes interact with atmospheric mixing and are responsible for temporal and spatial variability of tropospheric Δ14C: fossil fuel emissions, ocean-atmosphere exchange, stratosphere-troposphere mixing, and terrestrial ecosystem fluxes [Nydal and Lövseth, 1983; Enting and Mansbridge, 1987]. Over the last few decades, the relative contribution of each of these processes has undergone substantial modification, as much of the bomb 14C has moved from the stratosphere into the deep ocean and fossil fuel use has continued to increase.
 In the mid 1960s, the seasonal amplitude of Δ14C exceeded 150‰ across a wide range of northern stations, including Spitsbergen, Norway (78°N), Fruholmen, Norway (71°N), Point Barrow, Alaska (68°N), Trondheim, Norway (63°N), and Lindesnes, Norway (58°N) [Telegadas, 1971; Nydal and Lövseth, 1983]. At all these stations, the phase of the Δ14C seasonal cycle was similar, reaching a maximum in mid and late summer, and a minimum in mid winter. Further to the south, a seasonal cycle with similar amplitude and phase existed during this period at Rehovoth, Israel (32°N), Izana, Tenerife Island, Spain (28°N), and Mas Palomas, Grand Canary Island, Spain (27°N) [Nydal and Lövseth, 1983]. These extensive surface observations and those from aircraft [Telegadas, 1971] provide evidence that the seasonal cycle of Δ14C was coherent and widespread in the Northern Hemisphere troposphere in the mid and late 1960s, but became more subtle with time. As suggested by Broecker and Peng , the damping of this seasonal cycle may serve as a constraint on changes in the stratospheric 14C inventory.
 In later decades, stations in the Northern Hemisphere that were most exposed to fossil fuels tended to show larger seasonal amplitudes than those that were more remote [Levin et al., 1989, 1995; Meijer et al., 1995]. For example, during the 1980s the peak to trough amplitude of Δ14C measured at the surface near Groningen, the Netherlands was approximately 30‰, whereas at Izana in the Canary Islands, the amplitude was only 5‰ [Meijer et al., 1995]. Combining monthly Δ14C and radon measurements, Levin et al.  showed that fossil fuels emissions in Europe may have a seasonality that is substantially greater than that predicted by economic analyses of fuel use.
 In the tropics where fossil fuel effects are greatly reduced, large seasonal variations in Δ14C were observed in the early 1990s (∼40‰) that may have been caused by changes in ocean upwelling strength during an El Nino event [Rozanski et al., 1995]. At Wellington, New Zealand, Manning et al.  observed a decrease in the seasonal cycle of Δ14C from 1966 to 1980 and then a reversal of the phase and an increase in amplitude from 1980 to 1988. Model analyses of stratosphere-troposphere exchange could not explain this phase reversal, leaving open the possibility of ocean or land contributions [Manning et al., 1990].
 Measurements from Europe over the last 2 decades provide evidence for a strong continental Suess Effect, with fossil fuels driving down the mean annual Δ14C by over 20‰ in the European interior as compared with marine stations at similar latitudes [Levin and Hesshaimer, 2000; Meijer et al., 1995]. Measurements from Levin and Hesshaimer  provide evidence for a distinct latitudinal profile during the early 1990s with a minimum in mean annual Δ14C at mid latitudes in the Northern Hemisphere from fossil fuels, (∼−3‰), a maximum near the equator (∼+2.5‰), and a second minimum near 60°S (∼−3.0‰) caused by exchange with the Southern Ocean. This profile is markedly different from model projections of the preindustrial latitudinal profile that had atmospheric minimums in northern high latitudes and southern high latitudes, and a Southern Ocean to equator difference of over 7‰ [Braziunas et al., 1995].
 Using an atmospheric tracer model, here we explore the seasonal and latitudinal variability of Δ14C since the 1960s. A primary objective of our analysis was to develop a predictive understanding of fossil, ocean, stratosphere, and terrestrial biosphere contributions to Δ14C to more fully take advantage of past and future observations of this tracer. We also report experimental evidence from a transect of sites across Alaska that confine the range of possible carbon residence times in northern terrestrial ecosystems and consequently their contemporary impact on atmospheric Δ14C. Building on our analysis that isolated individual processes, we predict that several features of the latitudinal profile of atmospheric Δ14C will change over the duration of the 21st century.
 We made separate atmospheric model runs for fossil fuel, ocean, terrestrial biosphere, and stratosphere fluxes of CO2 and 14CO2. We focused on the period from 1960 to 1990 when long-term records from Fruholmen, Norway [Nydal and Lövseth, 1996] and Wellington, New Zealand [Manning et al., 1990] allowed us to evaluate model scenarios. We also describe measurements of soil respiration Δ14C from tundra and boreal forest sites in Alaska.
 For our atmospheric model runs, we used pulse response (or Green's) functions from a three-dimensional (3-D) atmospheric general circulation tracer model. We made the pulse-response functions by releasing a gaseous “dye” tracer into the atmospheric model at a constant rate from a particular region (Table 2) for short interval of time (1 month). We then tracked the 3-D pattern of emitted tracer until it was well mixed (over a period of 3 years). In the model simulations described below, we combined and scaled these pulse functions according to the intensity of 14CO2 or 12CO2 fluxes in each region. In this way, we constructed atmospheric distributions of Δ14C that were consistent with our estimates of sources and sinks.
Table 2. Basis Regions Used to Generate Atmospheric Pulse Functions
 The advantage of the pulse-response approach is that it is computationally efficient: the atmospheric model has to be run only once to generate the pulse functions. This efficiency enables the analysis of multiple atmospheric model scenarios, each lasting over a period of several decades in the post bomb era. Using the pulse-response approach, it is possible to reproduce large-scale latitudinal and seasonal patterns of tropospheric CO2 and δ13CO2 [Gurney et al., 2002; Randerson et al., 2002], even though the approach largely neglects variability in fluxes within basis regions (i.e., there are only ∼20 discrete units interacting with the atmosphere with this approach, as compared with a fully coupled biosphere-atmosphere model that may have 20,000).
2.1. Atmosphere Model
 We generated pulse-response functions from the coarse resolution Goddard Institute for Space Studies (GISS) atmospheric transport model [Fung et al., 1991]. This version of the GISS tracer model has an 8° × 10° horizontal resolution, nine vertical levels, and uses the same winds and circulation statistics as a higher resolution (4° × 5°) atmospheric model [Tans et al., 1990]. The atmospheric tracer model uses 4-hour wind fields, and operates with a 1-hour time step. We directly compared the lowest level of the model with surface observations; this level had a thickness of ∼50 mb and extended from the surface to 949 mb, or ∼400 m.
 We constructed the atmospheric pulse functions by separately running the GISS tracer model for 36 months with an arbitrary initial pulse of 1 Pg C/month in each basis region listed in Table 2. Geographically, the fluxes used to generate the pulse were distributed evenly within each basis region for the case of the oceans and stratosphere, according to the spatial distribution of annual NPP for land [Randerson et al., 1997], and according to 1° × 1° maps of emissions [Andres et al., 1996] for fossil fuels.
 The impact of surface and stratosphere carbon fluxes on troposphere Δ14C depends strongly on background troposphere 14C and CO2 levels. For example, in the absence of other exchange, a 6 Pg C/yr fossil fuel flux in 1990 caused a decrease in Δ14C of 9.2‰ (in an atmosphere of 355 ppm and 145‰) whereas the same flux in a future atmospheric scenario of 500 ppm and −60‰ [Caldeira et al., 1998] would lead to a decrease of only 5.3‰. In the extreme, a flux has no impact on the isotopic composition of the atmosphere when its composition is equal to that of the background atmosphere. To capture these effects on Δ14C in our global model, we adjusted the annual mean background levels of CO2 and 14CO2 to match a global mean time series of observations (section 2.2). We then added the cumulative effect of atmospheric fluxes from the previous 36 months (including the current time step) using the Green's functions described above. We assumed that fluxes that occurred 36 months or more prior to the current model time step did not contribute to spatial or temporal gradients of troposphere Δ14C (i.e., these fluxes were well mixed and contributed only to the background).
 We started our atmospheric model runs in 1955, and then started our analysis of seasonal and spatial trends in the early 1960s. This allowed the troposphere a period of over 3 years to adjust to modeled surface fluxes. As described below, our flux models of the ocean, terrestrial biosphere, and stratosphere were initialized and forced with data from periods well before the tropospheric tracer model runs began in 1955.
 We compared our model results with seasonal observations from two stations: Fruholmen, Norway (71°N, 24°E) [Nydal and Lövseth, 1996] and Wellington, New Zealand (41°S, 175°E) [Manning et al., 1990]. To isolate seasonal variations in these two records, we removed the long-term secular trend using a smoothing spline [Enting, 1987]. The residual was used to construct seasonal cycles by grouping all the observations into 2-month intervals. Standard deviations were calculated using all the observations within each interval.
2.3. Fossil Fuel Fluxes
 We modeled monthly fossil fuel emissions using a sine function with a maximum in January and a minimum in July, and a peak to trough amplitude of 30%. This seasonality is consistent with the analysis by Rotty  for northern regions including Europe, Russia, Canada, the U.S. and Japan during the early 1980s [Randerson et al., 2002]. It is difficult to justify constructing a more complicated fossil fuel scenario due to the paucity of seasonal fossil fuel emissions data. Annual total emissions from 1955 to 2000 were taken from Marland et al. . We prescribed the spatial distribution of fossil fuels in our single fossil basis function from the 1° × 1° maps constructed by Andres et al.  for 1990.
2.4. Ocean Fluxes
 We used monthly ocean-atmosphere and atmosphere-ocean fluxes of 14CO2 and CO2 from the Lawrence Livermore National Laboratory (LLNL) three-dimensional ocean circulation model in our analysis [Duffy and Caldeira, 1995; Duffy et al., 1997]. Atmospheric Δ14C, winds, and natural and preindustrial radiocarbon spin up followed the protocol of the Ocean Carbon Model Intercomparison Project (OCMIP; http://www.ipsl.jussieu.fr/OCMIP/phase2/) [Orr et al., 2001]. Seasonal changes in the 14CO2 flux from the ocean to the atmosphere arise from month-to-month changes in the gross sea to air CO2 flux and from month-to-month changes in the Δ14C composition of the mixed layer (Figure 1). Monthly fluxes from the LLNL ocean model were averaged over eight regions for use with our atmospheric pulse functions (Table 2). The LLNL ocean model simulation used here in our troposphere analysis ended in 1990.
2.5. Stratosphere Fluxes
 Most of the large nuclear fireballs (from detonations at the surface or in the free troposphere) were lifted across the tropopause and into the lower stratosphere [Glasstone and Dolan, 1977]. The version of the GISS model we used here had nine vertical levels and did not have a well-defined stratosphere. We simulated stratosphere-troposphere exchange by creating separate box models for the Northern and Southern hemisphere stratosphere, each with monthly inputs from bomb radiocarbon [Enting, 1982], cosmogenic production, and entrainment of tropospheric air (representing upward flow in the tropics), and monthly losses (representing net stratosphere-troposphere mixing outside of the tropics) (Figure 2a). Stratospheric losses were evenly distributed in the GISS atmospheric tracer model in regions poleward of 30° using pulse functions at 200 mb (Table 2).
 The total inventory of bomb 14C released into the tracer model was 625 radiocarbon units (1 RCU = 1 × 1026 atoms 14C) through 1975 (Table 1 and Figure 2). Bomb inputs consisted of detonation strength estimates (in Mt TNT) corrected for the height of detonation (and thus interception of neutrons by the surface instead of by atmospheric N) [Enting, 1982]. We used a conversion factor of 1.35 RCUs/Mt TNT to estimate bomb-produced 14C [Lassey et al., 1996]. Nuclear explosions in the Northern hemisphere entered the Northern hemisphere stratosphere, while explosions in the Southern hemisphere entered the Southern hemisphere stratosphere according to detonation sites provided by Enting . We did not allow mixing in the stratosphere across the equator; the Northern and Southern Hemisphere stratosphere box models were isolated from each other.
 Cosmogenic 14C production has been estimated at about 2.3 to 2.5 RCUs/yr [Lingenfelter, 1963]. We uniformly distributed this flux in the Northern and Southern Hemisphere stratosphere models. While cosmogenic 14C production represents a relatively small perturbation when compared with bomb inputs and fossil fuel dilution over the last few decades, it is becoming an increasingly important factor in determining the global decrease in atmospheric Δ14C [Caldeira et al., 1998]. We neglected 14C production from nuclear power plants, which was less than 0.5 RCUs/yr in 1990 [Hesshaimer et al., 1994].
 Each stratospheric box model had a lower and upper reservoir. In the Northern Hemisphere, these reservoirs were assigned residences times of 1.5 years and 5 years based on measurements of CO2 and N2O by Andrews et al. . Bomb inputs and upward troposphere fluxes entered the lower stratosphere reservoir, where they could either move into the upper stratosphere or vent directly into the extra-tropical troposphere (and thus into the GISS atmosphere model via the pulse response functions previously described). The lower and upper model reservoirs accounted for 78% and 22% of the stratosphere mass, respectively. With the two-reservoir stratosphere model used here, we were able to fit Northern Hemisphere stratosphere observations from Telegadas  (Figure 2b). With a 1-box stratosphere model (data not shown) it was difficult to reproduce this time series; modeled 14C decreased more rapidly than the observations in the late 1960s. The southern stratospheric model was identical to the northern one, but incoming and outgoing fluxes from the troposphere were reduced by 40% [Gettelman and Sobel, 2000]. The tropospheric air entrained into the stratosphere was assumed to have equal contributions from Northern and Southern hemispheres [Boering et al., 1996].
 We used seasonal estimates of the net stratospheric flux into the extratropical troposphere following the mass budget method developed by Appenzeller et al. . This seasonal distribution [Appenzeller et al., 1996, Figure 8] had a maximum in April and May (Figure 3) in the Northern Hemisphere and allowed us to approximately capture the phase of the seasonal cycle of Δ14C observed at Fruholmen, Norway, and Wellington, New Zealand. With other seasonal distributions of extra tropical (Northern Hemisphere) cross tropopause flux (CTF) that peaked in mid winter [i.e, the gross fluxes as estimated using the Wei method as reported in Gettelman and Sobel, 2000], we were unable to match the seasonal dynamics of the surface Δ14C observations.
2.6. Terrestrial Biosphere Fluxes
 We divided the land surface into eight regions (Table 2). In each region, terrestrial biosphere-atmosphere CO2 exchange was calculated as the approximate balance between net primary production (NPP) and heterotrophic respiration following a simple carbon model described by Randerson et al. . We estimated monthly NPP from the product of absorbed photosynthetically active radiation (APAR) and a globally uniform light use efficiency term. APAR was calculated for each region using satellite-derived vegetation indices (Advanced Very High Resolution Radiometer Simple Ratio) [Los et al., 1994] and solar insolation [Bishop and Rossow, 1991]. In the simulations presented here, we used a globally uniform light use efficiency of 0.52 g C/MJ PAR (which yielded a global annual NPP of 60 Pg C/yr). A simple description of carbon cycling and heterotrophic respiration within each region included a fixed plant allocation scheme and a temperature-dependent rate of microbial decomposition in litter and soil carbon pools. Allocation to leaves, fine roots, and wood had a fixed proportion of 0.4, 0.4, and 0.2. A Q10 of 1.6 described the sensitivity of heterotrophic respiration to mean monthly air temperature in each basis region. We neglected plant respiration in our model because we assumed it had a residence time much less than 1 year and thus imparted only a small isotopic anomaly to CO2 in the local atmosphere.
 The combination of the light use efficiency and Q10 values used here yielded realistic descriptions of the phase and amplitude of the seasonal cycle of CO2 and δ13CO2 at mid and high latitude NOAA/CMDL stations in the Northern Hemisphere (Figure 4). For all terrestrial biosphere scenarios presented below, we assumed that NPP remained in annual steady state at 60 Pg C/yr. The biosphere and atmospheric models were coupled so that isotopic fluxes in one region affected atmospheric composition and consequently isotope fluxes in downwind regions at later times [Randerson et al., 2002]. Soil and vegetation carbon pools in the terrestrial model were allowed to reach 14C steady state in 1750. From 1750 until the start of the atmospheric model runs in 1955, the Δ14C of NPP was forced to match the observed atmospheric record (section 2.2) in all basis regions. During each time step of the atmospheric model, carbon uptake by the terrestrial biosphere model had a prescribed Δ14C value from the atmospheric mean over the same basis region in the previous time step.
 We created three terrestrial biosphere scenarios by multiplying the turnover time of all the carbon pools in our model by a factor of 0.5, 1.0, or 2.0 (Table 3). We created these scenarios because terrestrial carbon residence times are relatively uncertain, and this range spanned previous estimates of 14C uptake by the terrestrial biosphere between 1965 and 1990 (Table 4). In the context of these scenarios, it is worth noting that while field measurements and satellite data provide a strong global constraint on NPP [Running et al., 1999], comparable global constraints on carbon residence times are relatively weak. It is doubtful the terrestrial biosphere isotopic disequilibrium forcing for δ13C or δ14C is known within a factor of 4 because this term critically depends on processes such as fine root allocation, mortality, and decomposition and only weakly on the mass of carbon in soils and vegetation [Trumbore, 2000]. Many of these processes are not well understood or well represented in models at regional and global scales that are used for disequilibria estimates. In our analysis of the combined processes, we used scenario 1 from the terrestrial biosphere because this scenario most closely matched a recent estimate (Table 4) that used satellite and climate data as constraints on the spatial distribution of NPP and decomposition across ecosystems with a 1° × 1° spatial resolution [Thompson and Randerson, 1999].
Table 3. Terrestrial Biosphere Model Carbon Cycling Scenarios
 Since our analysis focused on Δ14C, we did not consider the effects of mass dependent fractionation (against 13C or 14C). If we had considered δ14C, the seasonal amplitude arising from fractionation during photosynthesis would have added a seasonal cycle with an amplitude of ∼1.5‰ at mid and high northern latitudes, approximately double that observed for δ13C [Trolier et al., 1996].
2.7. Soil Respiration Δ14C Measurements
 Soil respiration and surface atmosphere Δ14C measurements were made at midday during July 2000 across a latitudinal transect in Alaska. Two sites were located in mature black spruce forest in interior Alaska, at the Caribou Poker Creek Research Watershed (65°07′N, 147°20′W, elevation ∼500 m), part of the Bonanza Creek Long Term Ecological Research (BNZ-LTER) site, and at Delta Junction, Alaska (63°55′N, 145°22′W, elevation ∼400 m). The third site was located in tussock tundra at the Toolik Lake LTER on the North Slope of the Brooks Range (68°38′N, 149°43′W, elevation 760 m).
 Soil respiration 14C/12C and 13C/12C isotopic ratios were measured using a modified dynamic flow chamber system [Gaudinski et al., 2000]. Air was drawn from chambers mounted on collars that had been permanently installed in the soil surface, through an infrared gas analyzer to determine the CO2 concentration, and then either through a soda lime CO2 scrubber or through a molecular sieve before returning to the chamber. To collect respired CO2 alone, CO2 was first scrubbed with soda lime from the chamber system to remove background atmospheric air. Scrubbing was maintained until the equivalent of 2–3 chamber volumes of air had passed through the scrubber. The airstream was then diverted through a molecular sieve that quantitatively trapped CO2 until 1.0–1.5 mg of CO2-C was adsorbed. Atmospheric samples were collected on molecular sieves using the same system, except that the air intake was placed above ground at a height of 2 m. Atmospheric samples were collected at the Toolik Lake and Delta sites. In the laboratory, the molecular sieve traps were heated to 675°C, which desorbs CO2 [Bauer et al., 1992]. Carbon dioxide was then cryogenically purified on a vacuum line, sub sampled for δ13C analysis, and reduced to graphite using titanium hydride, zinc, and a cobalt catalyst [Vogel, 1992]. Graphite targets were sent to the LLNL Center for Accelerator Mass Spectrometry (LLNL-CAMS) for Δ14C analysis. The 13C/12C isotopic ratios were used in the final analysis to correct the 14C/12C ratios for discrimination by photosynthesis [Stuiver and Polach, 1977], and to correct for remaining atmospheric CO2 in the chamber-sampler system.
 In this section, we first present model results for the individual processes. We then compare our combined model time series with observations from Fruholmen, Norway, and Wellington, New Zealand. Finally, we report soil respiration Δ14C measurements from tundra and boreal forest ecosystems in Alaska that help constrain the atmospheric Δ14C signal driven by northern terrestrial ecosystems.
3.1. Fossil Fuels
 The amplitude of the fossil fuel component did not substantially increase from 1965 to 1990 (Figures 5a and 6a). While total fossil fuel emissions almost doubled during this period, from 3.1 Pg C/yr in 1965 to 6.1 Pg C/yr in 1990 (a 49% increase) [Marland et al., 2000], atmospheric Δ14C decreased by almost the same amount, from 800‰ to 160‰ [Nydal and Lövseth, 1996]. This change in Δ14C corresponds to a 64% decrease in the 14C to 12C ratio of atmospheric CO2. Concurrently, atmospheric CO2 increased from 665 to 760 Pg C over this period (an 11% increase) [Keeling et al., 1989]. The increasing atmospheric CO2 mass and the decreasing 14C to 12C ratio of atmospheric CO2 caused the relative impact of fossil fuels on Δ14C to remain relatively constant. At Fruholmen, Norway, Δ14C seasonal cycle arising from fossil fuels had a minimum in February (corresponding to the time when fossil fuel derived CO2 concentrations were highest in the atmospheric model surface layer). The total peak-to-trough amplitude of the fossil fuel component was approximately 5‰ in 1990 (Figure 5a). At Wellington, New Zealand, the fossil fuel component was less than 1‰, with a maximum in March and a minimum in August, September, and October (Figure 6a).
 Fossil fuel CO2 plumes created increased seasonal amplitudes and depleted mean annual Δ14C values over Europe, the east coast of North America, and the east coast of Asia in 2000 (Figure 7). The Δ14C seasonal amplitude from fossil fuel emissions reached a maximum over Eastern Europe at 12‰ (Figure 7a), and was slightly offset to the east and north from the minimum in mean annual Δ14C. Globally, the mean annual Δ14C minimum occurred over Europe, with a value of −24‰ relative to the South Pole (Figure 7b).
 Despite relatively large fossil fuel fluxes from the west coast of North America, a large plume in Δ14C was not evident over western North America in our model simulations. We only present spatial maps of the seasonal amplitude and mean annual gradient for the year 2000 because, as described above, the impact of fossil fuel emissions on Δ14C did not rapidly change from the 1960s to the 1990s (Figure 5).
 The ocean contribution to the Δ14C seasonal cycle of atmospheric CO2 was greatest in the mid 1960s when the air-sea differences in Δ14C were large. At Fruholmen, peak-to-trough amplitudes in 1965 were approximately 9‰ in 1965, and decreased to approximately 1‰ in 1989 (Figure 5b). At Wellington, peak-to-trough amplitudes in the late 1960s were ∼4‰ and decreased to less than 1‰ by 1989.
 The mean annual regional pattern of troposphere Δ14C arising from ocean exchange changed substantially from 1965 to 1989 (Figure 8). In 1965, atmospheric Δ14C anomalies over much of the ocean surface were relatively uniform (with the exception of the Southern Ocean and parts of the Arctic Ocean) and the land/sea contrast was large (Figure 8d). By 1989, the rapid equilibration of tropical oceans and mid-ocean gyres to relatively high tropospheric Δ14C levels in previous decades led to a positive atmospheric Δ14C anomaly in equatorial regions (Figure 8b).
 In the eastern tropical Pacific, upwelling from the wind driven circulation now had a higher Δ14C content than surrounding surface waters, leading to a reversal of the seasonal cycle Δ14C in ocean surface water (Figure 1). This reversal in the seasonal phase of surface water Δ14C is consistent with coral measurements that show a rapid increase in the seasonal amplitude in the late 1960s and early 1970s, and then a decline in the late 1970s and early 1980s (as upwelling water became progressively more enriched) [Guilderson and Schrag, 1998; Rodgers et al., 2000]. In contrast, upwelling with minimal bomb 14C in the Southern Ocean continued to impart a negative Δ14C anomaly to the atmosphere in this region (Figure 8b).
 Northern Hemisphere stratosphere fluxes created seasonal amplitudes in excess of 250‰ in 1963 (Figure 5c). By 1970, the seasonal contribution had decreased to approximately 30‰. At Wellington, New Zealand, the seasonal amplitudes from the Northern hemisphere stratosphere were reduced by a factor of approximately 2 to 4 from those in the Northern Hemisphere during the late 1960s, and were delayed in phase by about 4–5 months (Figure 6c). During the 1960s, contributions from the southern stratosphere to the seasonal amplitudes at Fruholmen, Norway were small, in part because few nuclear detonations occurred in the Southern Hemisphere during this time (Figure 5d). At Wellington, fluxes from the Southern Hemisphere stratosphere were responsible for a seasonal amplitude of 2.5‰ in the late 1960s and ∼1‰ in 1989.
 In 1965, almost the entire 14C anomaly in the Northern Hemisphere stratosphere was a direct result of nuclear detonations (Table 5 and Figure 2). Over the following decade, almost all original bomb 14C was flushed out of the stratosphere by atmospheric mixing and transport. By 1990, bomb 14C accounted for less than 2.3% of the total 14C anomaly in our modeled stratospheric flux (Table 5), and represented less than 0.05% of the original stratospheric 14C anomaly in 1965 (Figure 2b). Despite the rapid flushing of this bomb 14C, the stratosphere flux continued to play an important role in shaping the seasonal and latitudinal distribution of tropospheric Δ14C during the 1970 to 2000 period. During this second stage, the primary mechanism enriching the stratospheric flux was the entrainment of tropospheric air in the tropics, and the time delay associated with its return via stratosphere-troposphere exchange in the extra-tropics. This time delay in our model led to an enrichment in the stratosphere flux because fossil fuels emissions and ocean exchange were concurrently decreasing tropospheric Δ14C at the surface.
Table 5. Source of the 14C Anomaly in the Northern Hemisphere Stratosphere Flux
This refers to bomb 14C originating from the initial nuclear detonation and fireball. 14C that is entrained into the stratosphere by normal atmospheric circulation is included as tropospheric uplift.
These percent contributions were estimated for the component of 14C in the extratropical stratosphere to troposphere flux above a 0‰ baseline level (i.e., they represent the percent contribution to the 14C mass above that expected for a carbon flux with a isotope ratio equal to Rstandard, where Rstandard is equal to 1.176 × 10−12).
 When cosmogenic radiocarbon production was used as a sole input into our stratosphere model (no bomb inputs), we obtained a 14C mass anomaly of 3.75 RCUs in the northern stratosphere, an upper stratosphere Δ14C of 220‰, a lower stratosphere Δ14C of 60‰, and a seasonal amplitude of ∼1.3‰ at the surface at Fruholmen, Norway.
 With uniform cosmogenic production across both hemispheres (and a well mixed troposphere), our model generated no north-south gradient in tropospheric Δ14C. Reduced tropospheric exchange in the Southern Hemisphere stratosphere was compensated by the buildup of 14C in the stratosphere, and thus higher Δ14C levels in the extra-tropical flux. The reduced tropospheric exchange in the Southern Hemisphere stratosphere did, however, generate a north-south gradient when the troposphere was constrained to follow the observed post bomb trajectory, particularly during the 1970s and 1980s. More 14C-enriched air circulated through the Northern Hemisphere stratosphere, creating a positive anomaly in northern latitudes. Even though Δ14C levels were higher in the Southern Hemisphere stratosphere because of the longer residence times, this enrichment did not fully compensate for the reduction in the total stratosphere-troposphere fux in this hemisphere.
3.4. Terrestrial Biosphere
 During the 1960 to 2000 period, the terrestrial biosphere component of Δ14C seasonal cycle reversed phase, with the timing of the phase transition depending on the residence times of carbon in the regions surrounding the observation station. In scenarios with relatively rapid carbon cycling (Table 3), the amplitudes of the initial post bomb seasonal cycle were relatively small and decayed more rapidly (Figure 9b) as compared with scenarios with relatively slow carbon cycling (Figures 9c and 9d). During this initial phase of the terrestrial biosphere response, the Δ14C of respiration was much lower than the local atmosphere (Figure 9a). Consequently, respiration caused a decrease in Δ14C near the surface during summer and fall, when respiration rates were high. As litter and soil pools equilibrated with post bomb atmospheric Δ14C levels, the isotopic disequilibria between respiration and the atmosphere decreased, along with the terrestrial ecosystem contribution to the Δ14C seasonal cycle.
 The terrestrial biosphere component of seasonal amplitude dropped to zero when the Δ14C of heterotrophic respiration was equal to that of the local atmosphere. The timing of this phase transition depended on the residence time of carbon in nearby regions (Table 3). At observation stations closer to the equator than Fruholmen, Norway this phase transition occurred earlier, reflecting warmer temperatures and consequently more rapid carbon cycling within terrestrial ecosystems (data not shown).
 As heterotrophic respiration became more enriched than the local atmosphere, the terrestrial component of the seasonal cycle reversed phase, as compared with the initial post bomb terrestrial seasonal cycle. In this second stage, heterotrophic respiration was enriched in Δ14C (relative to the local atmosphere), causing a positive anomaly at the surface in summer and fall, which was erased by atmospheric mixing during winter and spring. The disequilibrium between respiration and the atmosphere in the second stage was much smaller than in the first stage because the annual rate of change in global atmospheric Δ14C had slowed. For the more rapid cycling scenarios, the seasonal amplitudes reached a maximum shortly after the phase transition time, whereas for the slow cycling time scenarios, the amplitudes continued to build out to the year 2000.
 The modeled effect of terrestrial ecosystem fluxes on the seasonal amplitude of Δ14C was greatest over eastern Siberia and western North America (Figures 10a and 10c). Over the oceans, including sites such as Fruholmen, Norway, this effect was considerably weaker, both in 1965 and in 2000.
 Just as the terrestrial biosphere reversed phase in terms of its contribution to the seasonal cycle from 1965 to 2000, it also reversed its contribution to the mean annual latitudinal profile of Δ14C. In the year 2000, terrestrial biosphere fluxes (Scenario 1) caused a north-south atmospheric gradient of ∼6‰, primarily because respiration from northern ecosystems was enriched relative to the contemporary atmosphere (Figure 10b). In contrast, in 1965, respiration from northern ecosystems had a Δ14C content that was depleted by several hundred per mil as compared to the troposphere (Figure 9a). This large isotopic disequilibrium created a north-south difference of approximately −40‰ (Figure 10d). Without terrestrial exchange, the observed differences between Northern and Southern Hemisphere troposphere Δ14C in the mid and late 1960s would have been considerably greater.
3.5. Combined Effects and Comparison With Observations
 At the time of the Test Ban treaty in 1963, stratospheric exchange was the dominant contributor to seasonal and latitudinal variation in troposphere Δ14C. With net extra-tropical stratosphere fluxes reaching a maximum during April and May in the Northern Hemisphere, we reproduced the phase and amplitude of the seasonal cycle of Δ14C fairly well at Fruholmen in 1963, 1964, and 1965 (Figure 11a). By the late 1960s and early 1970s, however, model simulations considerably underestimated the amplitude of the observed seasonal cycle (Figure 12a). During this initial period, fluxes from the terrestrial biosphere contributed substantially to the Δ14C seasonal cycle, with a phase that acted to cancel the stratospheric component (Figure 11a and Figure 12a).
 By the late 1980s, the stratospheric contribution had considerably weakened, but was still the dominant contributor to the seasonal cycle; the phase of the measured seasonal cycle at Fruholmen remained the same throughout the 30-year time series. Terrestrial biosphere (contributing 14C-enriched summer respiration at this point) and fossil fuel fluxes were in phase with observed Δ14C seasonal cycle, and amplified the stratospheric signal (Figure 11b). Again, the model simulations did fairly well at reproducing the phase of the seasonal cycle, but underestimated the amplitude in 1985, 1986, and 1987. During this latter period, fossil fuels were the second most important contributor, causing a decrease in Δ14C during winter months, and in particular during February.
 At Wellington, New Zealand, the Northern Hemisphere stratospheric fluxes generated a seasonal cycle in the late 1960s and early 1970s that had about the same amplitude as the observations, but with a phase that lagged behind the observations by approximately 1.5 months. By the late 1980s, the northern stratospheric contribution had substantially weakened (Figures 6c and 6d).
 During the early 1970s, the latitudinal profile of Δ14C was driven largely by stratosphere inputs in the Northern Hemisphere (Figure 13). By the late 1980s, the latitudinal profile had two minima (one at mid northern latitudes caused by fossil fuel emissions and a second at high-southern latitudes caused by exchange with the Southern Ocean) and three local maxima (South Pole, tropics, and North Pole). This latitudinal profile is qualitatively consistent with a north-south transect of atmospheric Δ14C observations from the early 1990s [Levin and Hesshaimer, 2000]. In the tropics, both terrestrial ecosystems and tropical oceans created positive Δ14C anomalies. In tropical forests, the Δ14C of respiration in the late 1980s and 1990s was enriched because much of the carbon in vegetation and soils had residences times spanning several decades, and was fixed when the atmosphere Δ14C levels were much higher (Figure 10b). Likewise, in mid-ocean gyres, relatively shallow mixed layers and limited vertical entrainment of prebomb water masses has also led to mixed layer Δ14C values that more closely tracked atmospheric Δ14C levels as compared with high latitude oceans that were a stronger 14C sink during this period.
3.6. The Δ14C of Soil Respiration From Northern Ecosystems
 Measurements of Δ14C in soil respiration from boreal forest and tundra ecosystems in Alaska in July of 2000 are summarized in Table 6. Summer respiration measurements from arctic tundra ecosystems near Toolik Lake, Alaska were near or below local atmospheric levels (∼75‰), in contrast to boreal forest ecosystems in regions to the south (in interior Alaska) where soil respiration Δ14C values were significantly higher (∼112‰). One possible explanation for this difference in respired Δ14C is that the more northern tundra ecosystems had slower rates of carbon cycling (as a result of colder temperatures and a shorter growing season), and thus had not yet undergone the crossover in respiration Δ14C described in section 3.4 and shown in Figure 9a.
Table 6. Latitudinal Profile of Observed Δ14C in 2000 Summer Soil Respiration
The standard deviation reported here represents the variability across field samples. The precision of an individual AMS measurement is limited to ∼± 5‰.
Toolik Lake, AK
Caribou/Poker Creek, AK
Delta Junction, AK
64 and 68
4.1. Evolving Latitudinal Profile of Δ14C
 Over the last few decades, the processes that contribute to troposphere Δ14C variability have dramatically shifted. The north-south profile significantly weakened from 1970 to 1990 (Figure 13), and surface measurements from the early 1990s suggest the north-south Δ14C profile was nearly zero [Levin and Hesshaimer, 2000]. Based on our analysis of the component fluxes, we predict that the profile will continue to shift, with the Northern Hemisphere becoming progressively more 14C-depleted as compared with the Southern Hemisphere over the next few decades. In addition, we predict a persistent local maximum in tropical Δ14C.
 As fossil fuel emissions continue to grow, it is likely that Δ14C in the Northern Hemisphere will decrease from the direct infusion of fossil carbon with an isotopic signal of −1000‰. Gross CO2 exchange with the Southern Ocean will counteract the impact of fossil fuel dilution in the north, with the strength of this offset depending both on the magnitude of the gross exchange and the residence times of dissolved inorganic carbon (DIC) in surface waters. Given that fossil fuel emissions are relatively well known, a limited atmospheric sampling program monitoring the shift in the north-south profile of atmospheric Δ14C would have the potential to constrain the carbon dynamics of the Southern ocean [e.g., Manning et al., 1990; Levin and Hesshaimer, 2000]. These observations could then be used to determine the latitudinal distribution of land and ocean carbon sinks via constraints on isotopic disequilibria for the δ13C budget [e.g., Ciais et al., 1995] or via constraints on regional ocean carbon fluxes when used in concert with observations of wind speed and ocean Δ14C [Wanninkhof, 1992; Key, 1998].
 As compared with fossil fuel and ocean exchange, terrestrial ecosystems are likely to have a smaller, yet significant, impact on the north-south profile of Δ14C over the next few decades. Differences between terrestrial ecosystems and the atmosphere currently enrich the Northern Hemisphere troposphere by about 5‰ and offset dilution by fossil fuels by about 25% (Figures 10b and 7b). The terrestrial contribution to the north-south gradient is likely to persist over the next several decades based on observations of decadal-scale turnover of carbon in northern extratropical ecosystems [Trumbore, 2000], but may weaken with more gradual declines in tropospheric Δ14C [e.g, Caldeira et al., 1998].
 Cosmogenic 14C production, distributed approximately evenly within Northern and Southern hemisphere stratospheres, should have only a minor effect on the north-south profile as compared with fossil fuel emissions, and ocean and terrestrial biosphere isotopic disequilibria [Braziunas et al., 1995]. In addition, while asymmetries in the strength of stratosphere-troposphere exchange in the Northern and Southern Hemisphere may have induced a north-south gradient in Δ14C during the 1970s and 1980s (see section 3.3), our model analyses suggest that these effects were attenuated substantially by 2000, and will contribute far less than 1‰ variation to the north-south profile in the future.
 Over a period of several decades, basic features of the latitudinal profile visible in late 1980s in Figure 13 should remain intact (i.e., the 2 minima in northern mid latitudes and southern high latitudes, and the 3 maxima near the North Pole, equator, and South Pole). However, with continued fossil fuel burning over the next century, the oceans may ultimately become an atmospheric source of 14C [Caldeira et al., 1998], both regionally and globally. As this occurs, the contemporary minimum in Δ14C observed in Southern Hemisphere high latitudes (Figure 13) eventually may be replaced by a maximum, because Δ14C in these waters reflects exchange with intermediate waters that have a residence time of decades to centuries [Toggweiler and Samuels, 1993]. Specifically, the large mass of C involved with this circulation will equilibrate only very slowly with the atmosphere, and will impart a ocean to atmosphere flux with a Δ14C level somewhere between preindustrial levels [Braziunas et al., 1995], and the perturbation cause by the addition of bomb 14C (Figure 1). If atmospheric Δ14C drops below that of the ocean mixed layer in this region, then even the Southern Ocean will become a 14C source. Other areas of the ocean with less intense vertical mixing (and shorter carbon residence times) such as the mid-ocean gyres and the tropics would be expected to more closely track decreasing atmospheric Δ14C driven by fossil fuels.
4.2. Terrestrial Biosphere Process and Mechanism
 For terrestrial ecosystems, Δ14C signals are imparted to the troposphere in proportion to the isotopic disequilibrium between NPP and heterotrophic respiration. Fractionation against 13C and 14C by photosynthesis, does not affect troposphere Δ14C because Δ14C is specifically defined to be insensitive to mass dependent fractionation [Stuiver and Polach, 1977]. While photosynthesis does not immediately affect tropospheric Δ14C, it does, of course, regulate the mass and Δ14C content of carbon entering terrestrial ecosystems, and thus ultimately the magnitude of the respiration flux and the impact of the Δ14C disequilibrium on the atmosphere in the future.
 The terrestrial biosphere contribution to the northern seasonal cycle of tropospheric Δ14C arises because the Δ14C composition of respiration and the atmosphere rapidly diverged in the 1960s (Figure 9a) and because in northern ecosystems heterotrophic respiration is primarily confined to summer and fall months, when microbial communities within the soil are warm and metabolically active [Goulden et al., 1998]. In the 1960s, summer and fall heterotrophic respiration initially released a pulse of Δ14C-depleted CO2 (relative to the local atmosphere), while more recently this pulse has become more Δ14C-enriched than the local atmosphere, because of lags associated with the decomposition of photosynthetic products fixed when atmospheric Δ14C levels were high. During winter and early spring, when respiration rates are low, the atmosphere relaxes back to its initial state; atmospheric mixing removes the signal imparted at the surface by respiration during the summer.
 Our measurements of Δ14C from black spruce forests in Alaska suggest that most terrestrial ecosystems now respire CO2 that has a higher Δ14C than the background atmosphere. We infer that most ecosystems are now a net 14C source because rates of decomposition and carbon turnover are much higher in temperate and tropical ecosystems to the south, and so these ecosystems would more rapidly acquire the Δ14C signal imparted by weapons testing (Figure 9a). Specifically, in high latitude ecosystems decomposition is limited by both temperature (a short growing season) and by chemically recalcitrant plant litter (e.g., needles and moss that have high lignin and low nitrogen content) [Trumbore, 1993; Hobbie et al., 2000]. Within mature black spruce forests, allocation to wood further decreases rates of carbon turnover. Thus, our field observations suggest that the positive Δ14C anomaly arising from northern terrestrial ecosystems in 2000 (Figure 10b) is basically correct (has the right sign); over 95% of heterotrophic respiration (and NPP) in the Northern Hemisphere extratropics (north of 20°N) originates from biomes other than Arctic tundra [Randerson et al., 1997].
 In contrast, our measurements from the Arctic tundra suggest that even by the year 2000, some ecosystems in the far north still had respiration that was depleted in Δ14C relative to the local atmosphere. This depletion was probably a result of slow carbon turnover rates and the decomposition of a significant amount of C fixed before the period of aboveground nuclear weapons testing. These conclusions are based on the assumption of more or less steady state conditions for ecosystem carbon cycling. Changes in ecosystem carbon cycling caused by local disturbance could also shift the Δ14C of respired CO2 to prebomb levels.
 The use of Δ14C to quantify rapid turnover pools (less than 5 years) within ecosystems requires accurate measurements of local atmospheric Δ14C [Gaudinski et al., 2000; Schuur et al., 2002]. Our analysis of spatial gradients in Δ14C induced by fossil fuels, ocean, and terrestrial biosphere fluxes suggests that the Δ14C of continental air may vary by 10‰ or more from season to season and from place to place. These results are consistent with observations from Europe that show large seasonal variability in the continental interior as compared with offshore ocean islands [Levin et al., 1989; Meijer et al., 1995]. For ecosystem studies exploring rapid turnover processes, our results confirm the idea that local atmospheric sampling made over the duration of the growing season, and weighted in proportion to rates of photosynthesis, are essential.
 Compared with any other place on Earth, it is possible that within the Amazon basin, the large one-way fluxes [Grace et al., 1995; Tian et al., 1998] and finite residence time of carbon within trees and soils [de Camargo et al., 1999] may have significantly diluted near-surface atmospheric Δ14C during the early 1960s. This temporary disequilibrium, lasting perhaps for a few months to a few years following the Test Ban Treaty, may have been recorded as a gradient in tree Δ14C (with relatively enriched values near the coast and depleted values in the continental interior). The mechanism for such a gradient may be as follows. Within the northern and equatorial regions of the Amazon basin, air travels along the surface from the Atlantic towards the western continental interior and the Andes until it is removed from the system by convection [Chou et al., in press]. At the coast, tree Δ14C would have been high in 1963–1965, with marine boundary layer air being relatively enriched in Δ14CO2. Interior tree Δ14C may have been exposed to significantly lower values of Δ14C because the CO2 in air had undergone exchange (and partial replacement) by ecosystems that were upwind. Enhanced recycling of CO2 within canopies in the Amazon interior would have enhanced the magnitude of this Δ14C gradient [Martinelli et al., 1991], even though it was not the primary mechanism.
 In our global analysis, we can see a trace of such a gradient (Figure 10d). However, several features of our coarse resolution modeling framework would cause us to underestimate its magnitude, including that we used only two tropical land basis regions and that the CO2 taken up by terrestrial ecosystems in the Amazon was drawn from an average of the troposphere over much larger area of the tropics and subtropics, including deserts. Nevertheless, the high levels of NPP within the Amazon led to visible Δ14C dilution in 1965 and enrichment in 2000 (Figures 10b and 10d). Trees from the interior Amazon have been measured and reflect Δ14C values in the mid 1960s that are close to remote Southern Hemisphere observation stations [Worbes and Junk, 1989]. However, tree rings are difficult to measure in this region, and so a systematic study of coast to interior Δ14C variability may be difficult; the mismatch of a single year (missing 1 growth increment) would significantly alter conclusions about the magnitude of the dilution.
4.3. Northern Hemisphere Seasonal Dynamics
 Even by the late 1980s, stratospheric fluxes were still the dominant contributor to the seasonal cycle at Fruholmen, Norway. Fluxes from the terrestrial biosphere initially attenuated the stratospheric signal, by ∼10% from 1963 to 1966, and by ∼25% from 1967 to 1971 (Figure 11a). By the late 1980s and early 1990s, the disequilibria of all the terrestrial scenarios had reversed sign, and were then in phase with the stratospheric tracer. Throughout the entire observational record at Fruholmen, we were able to reproduce the phase of the observations when northern stratospheric fluxes had a maximum in April and May, consistent with net flux estimates by Appenzeller et al.  but in disagreement with other studies that have suggested tracer mixing with the troposphere is greatest in the Northern Hemisphere winter. While this result is probably sensitive to our tropospheric model, the GISS model behaves similarly to other tropospheric models in terms of the phase delay in transmitting a CO2 tracer with a seasonal cycle from the surface to 200 mb [Rayner and Law, 1995]. Almost all of the models compared in the Rayner and Law  study had a time delay of approximately 1 month for transmitting peak CO2 levels in the spring from the surface to 200 mb, and between 1 and 2 months for the minimum in the fall.
 In contrast to our success in reproducing the phase of seasonal cycle, we substantially underestimated the amplitude at Fruholmen during several periods. Specifically, while we matched the amplitude from 1963 to 1966 and in 1988 and 1989, we underestimated the amplitude by at least a factor of 2 from 1967 to 1971 and from 1985 to 1988. There are several possible reasons for this, and probably more than one process is at work.
 First, the version of GISS atmospheric tracer model that had only nine vertical levels did not explicitly resolve planetary boundary layer (PBL) dynamics. Because the PBL traps surface fluxes at night and during winter [Denning et al., 1996], our model simulations may have underestimated the contribution of fossil fuel emissions and other surface sources during winter (and may have simultaneously overestimated contributions from the stratosphere). Along similar lines, Fruholmen is relatively close to the industrial center of Europe. If we underestimated the seasonality of the fossil fuel source, or its transport to the grid cell in our model containing the coordinates for Fruholmen, it would have had a large impact on the seasonal cycle of Δ14C because of its extreme isotopic composition (−1000‰), but only a minor impact on the seasonal cycle of CO2 (Figure 4) [Randerson et al., 1997].
 Second, NaOH solution traps that were used extensively to measure tropospheric Δ14C at the surface sampled the atmosphere in a way that is different from what might be expected from flask sampling in the middle of the day or from a monthly mean extracted from an atmospheric model [Hesshaimer and Levin, 2000]. Averaging day and night (typically for a period of a week or two) NaOH traps would have integrated over both turbulent and stable atmospheric periods. However, since the total CO2 taken up by the NaOH trap is partly controlled by diffusion, periods with higher levels of atmospheric CO2 in the boundary layer (i.e., night) may have contributed to more of the total dissolved inorganic carbon in solution than well-mixed periods (i.e., day). This effect may have amplified the contribution of fossil fuel emissions during winter months. We did not try to estimate or account for this potential model-data mismatch by re-sampling our atmospheric model: monthly mean concentrations from the atmospheric model were directly compared with observations derived from NaOH traps.
 Third, cosmogenic 14C production is affected by solar cycles on decadal and centennial timescales [Bard et al., 1997]. In addition, the magnitude of stratospheric mixing with the troposphere is quite variable from year to year [Gettelman and Sobel, 2000]. Our stratosphere and troposphere models did not attempt to account for these interannually varying processes.
 Finally, our terrestrial biosphere model was simplified so that all the carbon pools in the model (plants, litter, and soils) were driven with a single monthly value of air temperature. Thus, the relative contribution of different pools to the respiration flux did not change much from month-to-month (there was a small change from the increase in total biomass during the growing season in rapidly exchanging pools). The impact of terrestrial ecosystems on the seasonal dynamics of atmospheric Δ14C in our model was caused largely by the magnitude of respiration being much greater in summer than in winter. In contrast to our model structure, observations from the boreal forest show that as plants and surface soils freeze during fall and winter, the Δ14C of soil respiration drops below atmospheric levels [Winston et al., 1997]. This drop reflects the very slow turnover of soils at depth that are not yet frozen and still metabolically active. However, this mechanism should contribute only minimally to atmospheric forcing due to the relatively low CO2 flux rates during this time.
4.4. Southern Hemisphere Seasonal Dynamics
 As mentioned in section 1, the seasonal amplitude of Δ14C from Wellington, New Zealand decreased from 1967 to 1980 from about 20‰ to 3‰ [Manning et al., 1990]. Model simulations by Manning et al.  in which bomb 14C was distributed evenly across the stratosphere in both the Northern and Southern Hemispheres generated a seasonal cycle at Wellington that was almost completely out of phase with the observations. Here, we were able to reproduce the amplitude and most of the phase of the Wellington Δ14C seasonal cycle in the late 1960s and early 1970s by distributing 14C in the Northern and Southern Hemisphere stratosphere according to the latitude of the detonations, and by assuming that no north-south mixing occurred across the equator in the stratosphere. The importance of Northern hemisphere extratropical stratosphere flux (in term of explaining the seasonal signal at Wellington) is consistent with hemispheric asymmetries in weapons detonation (Table 1), the large seasonal amplitudes observed at mid and high northern latitudes [Telegadas, 1971], and the time delays and attenuation associated with the transmission of seasonal signals across the equator observed for other tracers (i.e., CO2).
 From 1980 to 1987, a second seasonal cycle with a smaller amplitude emerged at Wellington, with a peak in July and August and a minimum in January and an amplitude of ∼6‰ [Manning et al., 1990]. The emergence of a second seasonal cycle is consistent with the decline of the Northern Hemisphere stratosphere component in the mid 1970s and the emergence of underlying seasonal contribution from the Southern Hemisphere stratosphere or Southern Hemisphere oceans (Figure 6). We were unable to resolve this transition with our model. This may have been a limitation of aggregating our ocean fluxes (we averaged over the entire South Pacific from 0°S to 48°S for the pulse model) or misrepresentation of the Southern Hemisphere stratospheric flux.
 While atmospheric CO2 and δ13C distributions are sensitive to net CO2 fluxes, atmospheric Δ14C is a unique tracer because its atmospheric distribution depends mostly on gross exchange with ocean and terrestrial biosphere reservoirs, fossil fuel use, and stratospheric inputs. Net carbon sinks have almost no impact on tropospheric Δ14C, making this tracer ideal for constraining the residence times of carbon reservoirs interacting with the atmosphere. The distribution of tropospheric Δ14C near the surface depends largely on three factors: 1) the magnitude of gross CO2 fluxes from the stratosphere and surface, 2) the Δ14C content of these gross fluxes (e.g., the age of the respired carbon from terrestrial ecosystems), and 3) the time a packet of air remains in contact with the surface (i.e., the strength of planetary boundary layer mixing). While preliminary, we draw the following conclusions from our analysis.
 First, the full 30-year time series of Δ14C observations from the surface at Fruholmen, Norway provides an important and underutilized constraint on the seasonality of extra-tropical stratospheric fluxes in the Northern Hemisphere. Net stratospheric fluxes that peak in late spring (April and May) in the Northern hemisphere extra tropics are most consistent with the observed distribution of the Δ14C tracer in the troposphere. Using the spatially resolved 14C flux estimates from the terrestrial biosphere developed here, along with recently revised Δ14C measurements from the free troposphere and stratosphere [Hesshaimer and Levin, 2000], it now feasible to formally invert for the monthly distribution of the stratospheric flux.
 Second, heterotrophic respiration from terrestrial ecosystems (that was initially 14C depleted) partially cancelled seasonal oscillations of Δ14C caused by stratospheric fluxes during the mid and late 1960s. Similarly, terrestrial biosphere fluxes significantly counteracted the large north-south difference in Δ14C caused by the injection of 14C from Northern Hemisphere stratosphere-troposphere exchange during this period.
 Third, the Δ14C content of fluxes in many terrestrial ecosystems and in some ocean regions has (or will) reverse seasonal phase as bomb 14C cycles through internal reservoirs with decadal timescales.
 Fourth, it is possible to account for much of the amplitude and phase of the Δ14C seasonal cycle at Wellington, New Zealand during the 1960s and 1970s by distributing most of the bomb radiocarbon in the Northern hemisphere stratosphere.
 Fifth, the latitudinal profile of Δ14C will continue to shift over the next few decades, with the Southern Hemisphere troposphere becoming progressively more 14C-enriched as compared with the Northern Hemisphere. Measuring the profile would provide strong constraints on the latitudinal distribution of combined ocean and land carbon residence times, and thus ultimately the potential of these reservoirs to serve as carbon sinks.
 We wish to thank J. Adkins, C. Masiello, and K. Treseder for discussion and for comments on previous versions of this manuscript, two anonymous reviewers for constructive suggestions, and J. John for technical assistance with the GISS tracer model. This work was supported by a NASA IDS grant NAG5-9462. We are also grateful to I. Levin, K. Lövseth, M. Manning, R Nydal, and others for the very long time series of atmospheric Δ14C that have been made publicly available at the Carbon Dioxide Information Analysis Center.