Global Biogeochemical Cycles

Evidence for changes in carbon isotopic fractionation by phytoplankton between 1960 and 2010

Authors


  • These authors contributed equally to this work.

Abstract

[1] Rising CO2 is expected to drive a myriad of environmental changes in the surface ocean. Deciphering the phytoplankton response to this complex change is difficult. Here we determine whether a trend in the biological fractionation of stable carbon isotopes (εp) has occurred over the past 50 years. εp is primarily controlled by the acquisition and intracellular transport of inorganic carbon and the rate of carbon fixation. In turn, these processes are sensitive to phytoplankton physiology, community composition, and notably inorganic carbon availability. εp may therefore carry a signal of biological response to climate change. Temporal and spatial records of εp can be deciphered from the difference between the stable carbon isotopic composition of particulate organic matter (δ13CPOC) and that of the ambient inorganic carbon pool (δ13CCO2). Here we establish a global record of εp extending from the 1960s to today, extracted from a newly compiled data set of global measured δ13CPOC and part measured/part climatology δ13CCO2. We find that εp has changed significantly since the 1960s in the low- to mid-latitude surface ocean. The increase is most pronounced in the subtropics, where it is estimated at > 0.015‰ per year. Our findings of such rates of change are further supported by a high resolution temporal record from a single sediment trap near Bermuda. Our results are consistent with the idea that εp is affected by increased inorganic carbon availability driven by the rise in atmospheric CO2.

1 Introduction

[2] The isotopic composition of phytoplankton carbon is linked to their physiological function. Of the two stable carbon isotopes, 12C and 13C, phytoplankton preferentially utilize 12C due to the kinetics of enzymatic processes. The ratio of 13C to 12C, expressed in terms of δ13C as the relative difference (‰) with respect to the Pee Dee Belemnite (PDB) standard, is therefore lower in phytoplankton biomass (δ13CPOC) than in ambient inorganic carbon from which it is fixed (δ13CCO2). The fractionation between these δ13C values is termed the biological fractionation factor εp (εp ≈ δ13CCO2 - δ13CPOC) [Freeman and Hayes, 1992].

[3] A large component of εp is driven by CO2 fixation during photosynthesis by the enzyme, Ribulose 1,5 bisphosphate carboxylase oxygenase (Rubisco), which fractionates carbon by about −22 to −31‰ [Tcherkez et al., 2006]. However, εp rarely displays this maximum fractionation as there are many factors that limit the supply of CO2 to Rubisco, resulting in intracellular Rayleigh fractionation. Foremost, these factors include low external concentrations of CO2 or high growth rates [Baird et al., 2001; Laws et al., 1995; Popp et al., 1998; Rau et al., 1996; 1997], which in turn relate to temperature, nutrient availability [Bidigare et al., 1997], light intensity [Laws et al., 1995; Nimer and Merrett, 1993; 1996], and day length [Burkhardt et al., 1999a]. In addition, other factors can influence εp, such as utilization of isotopically heavier HCO3 [Burkhardt et al., 1999b; Sharkey and Berry, 1985], temperature effects on enzyme kinetics [Tcherkez et al., 2006], and interspecific variation [Burkhardt et al., 1999a; Falkowski, 1991]. Once the carbon has been fixed, further processing to lipids, carbohydrates and proteins also exert fractionation effects that alter δ13CPOC [Degens et al., 1968]. There may also be slight biomagnification through food chains [McConnaughey and McRoy, 1979].

[4] Many of the factors thought to influence εp (CO2 availability, mode of carbon acquisition, species composition, temperature, and nutrient availability) are expected to change as atmospheric CO2 increases [Behrenfeld et al., 2006]. This leads to the question: has there been a biological response by phytoplankton, either through a change in physiology and/or community structure to increasing levels of CO2, and is this reflected by a significant change in εp? The oceans are a major sink of CO2, absorbing an estimated 1.5–2.0 Pg C.yr−1 [Gruber et al., 2009] with the majority of anthropogenic carbon confined to water masses above the thermocline [Sabine et al., 2004]. It is uncertain how phytoplankton and therefore εp, may respond to the resulting changes in the surface ocean environment. Not only may they respond directly to increasing CO2 but also to indirect effects such as increasing temperature (a 0.3–0.5°C global increase has been observed between 1960 and 2005 [Solomon et al., 2007]), which results in stratification of surface waters that could restrict nutrient supply [Behrenfeld et al., 2006; Steinacher et al., 2010] and affect growth rates. The importance of the phytoplankton response extends to the global climate: phytoplankton are responsible for about half of net global photosynthesis, fixing an estimated 50 Pg C.yr−1 [Field et al., 1998]. About 30% of the fixed carbon ultimately sinks into deeper waters [Feely et al., 2001] removing the carbon from surface waters in exchange with the atmosphere and therefore the short term carbon cycle. Thus, the phytoplankton-derived carbon flux to depth constitutes a sink of atmospheric CO2. Any feedback between changes in atmospheric CO2 and phytoplankton carbon fixation will have important consequences for the climate.

[5] Here we investigate whether any trend in εp has occurred over the past 50 years. We derived a community-integrated signal of εp in the global ocean, combining our newly collected data with historical records to create a global dataset of 525 surface water δ13CPOC measurements with corresponding δ13CCO2 values inferred from a compilation of measurements and/or climatology values for temperatures, salinity, carbonate system and δ13CDIC. To probe the underlying global trends in our spatially and temporally discontinuous data set, we develop a model for the spatiotemporal variation in εp. Using a variant of extended Multiple Linear Regression (eMTL) [Friis et al., 2005], we estimate the global distribution of εp and its rate of change between 1962 and 2010. To lend further support to our results, we compared our global results to a high resolution, sediment trap time series near Bermuda between 1978–2007.

2 Method

2.1 δ13CPOC Measurements

[6] We compiled and analyzed two data sets of δ13CPOC. One consists of surface ocean, globally distributed δ13CPOC spanning 1962 to 2010. The other is a high temporal resolution data set from one location, the Ocean Flux Program (OFP) sediment traps (31°50′N; 64°10′W; 4500 m water depth) extending from 1978–2007.

[7] The global distribution of δ13CPOC was measured on cruises AMT18 (2008), AMT19 (2009) and Icechaser II (2010). 5–10 L of surface water (5 m) from either underway or Niskin bottles was filtered onto precombusted GF/F filters, rinsed with deionized water, and dried at 60°C for 12 hrs. δ13C of POC was measured according to [Hadas et al., 2001]. The 106 measured values of δ13CPOC from the Atlantic Ocean were supplemented with 419 measurements from the literature to create a globally distributed, temporal data set of 525 δ13CPOC values. It is assumed that the different methods in POC collection (Table S1, supporting information) did not introduce significant bias [Brodie et al., 2011; Goericke and Fry, 1994; Lorrain et al., 2003; McConnaughey and McRoy, 1979]. We also assumed that the effect of post-fixation fractionation due to organic matter degradation or differential carbon assimilation in higher trophic levels is small.

[8] For the OFP dataset, δ13CPOC was measured as an annual mean from a sediment trap at 3200 m between 1978 and 2007. Samples consisted of the fine fraction of the trap material (<125 µm for 1990–2007, <37 µm for 1978–1988), which comprises >80 % of the flux material. 1–3 mg of sample was decalcified using HCl and gentle drying. The dried samples were measured in tin capsules by CF-IRMS using a Europa 20-20 CF-IRMS interfaced to a Europa ANCA-SL elemental analyzer. Standards analyzed along with the POC gave standard deviation of ±0.2‰ for carbon.

2.2 Derivation of εp

[9] εp quantifies the difference in isotopic composition between fixed organic carbon and ambient CO2:

display math(1)

[10] It is, therefore, completely specified by the combination of δ13C of POC and δ13C of dissolved CO2. The latter is not measured directly, but derived from δ13C of dissolved inorganic carbon (DIC). To account for fractionation between the pools of CO2, HCO3 and CO32−, δ13C of CO2 is calculated from the isotopic mass balance as in Freeman and Hayes [1992], using δ13C of DIC, the concentrations of the different inorganic carbon species and temperature-dependent equilibrium fractionation factors [Zhang et al., 1995]. In turn, the concentrations of individual carbon species are calculated from two parameters of the carbonate system ([CO2] or pCO2, alkalinity, total DIC), temperature and salinity.

[11] Values for temperature, salinity, and carbonate system parameters were taken from the original sources if available; climatological values [Antonov et al., 2010; Key et al., 2004; Locarnini et al., 2010; Quay et al., 2003; Takahashi et al., 2009] were substituted otherwise. δ13CDIC presents a greater challenge, as this variable is frequently unavailable, yet important for understanding the subtle temporal variation in δ13CPOC, as δ13CDIC itself changes in time [Bacastow et al., 1996]. During the recent AMT18 (2008), AMT19 (2009) and Icechaser II (2010) cruises, measurements of δ13CDIC were taken onboard: 30 mL seawater was filtered through 0.2 µm PES filter and poisoned with HgCl2 and δ13C of DIC was measured according to Hadas et al. [2001]. For samples reported in the literature without δ13CDIC, a basin-specific reconstruction of δ13C of DIC and its rate of change [Quay et al., 2003] was used to obtain a representative value. Quay et al. [2003] characterize the variability in surface ocean δ13CDIC with a basin- and latitude-specific value representative for 1995, and rate of change. To estimate δ13CDIC for a single sample, we inferred the appropriate offset and rate of change, which were combined to estimate δ13CDIC at the sample time.

[12] As Table 1 shows, many inferred values of εp are based on one or more approximate (climatological) values for environmental variables. The uncertainties associated with these approximate values translate into uncertainties in the value of εp. This was accounted for by estimating standard deviations for each εp value from the uncertainty in input variables and the sample-specific sensitivity to each of those inputs (variances of inputs are multiplied with the square of the associated sensitivity, and then summed to arrive at the variance of εp). If we review the average uncertainty in approximate inputs and the associated mean sensitivity in εp (Table 1), we find that the reconstructed value of εp is particularly sensitive to errors in temperature and δ13C of DIC; uncertainties in salinity, alkalinity and pCO2 do not appreciably reduce confidence in the reconstructed value of εp.

Table 1. Use of Climatological Variables for Inference of εpa
VariableAvailabilityMean εp SensitivityClimatology
   Sources.d.Typical Res. Error in εp
  1. a

    Climatological values were used only if no measured values were available, as indicated by the “Availability” column. The sensitivity of εp to each variable was calculated numerically by perturbing the variable for every sample and observing the change inferred. The expected standard deviation (s.d.) of climatological variables was determined by comparing known values (see “Availability” column) with corresponding climatological estimates. The estimated error in εp resulting from the use of climatological data for each variable equals the product of the sensitivity and the climatological s.d. * For δ13C of DIC, a sample-specific standard deviation of inline image was used in order to account for the typical standard deviation of the offset (δ13CDIC in 1995; σa = 0.2‰) and slope (rate of change in δ13CDIC; σb = 0.005‰/yr) of the linear relationship used to reconstruct δ13CDIC.

temperature460/5250.12‰/°C[Locarnini et al., 2009; 2010]1.25°C0.14‰
salinity352/5250.003‰/PSU[Antonov et al., 2010]0.35 PSU0.00‰
δ13C of DIC83/5250.99‰/‰[Quay et al., 2003]±0.21‰*0.12‰
pCO2238/5250.0003‰/ppm[Takahashi et al., 2009]29 ppm0.01‰
alkalinity88/5250.0000‰/mEq[Key et al., 2004]29 mEq0.00‰

[13] To obtain values for environmental variables corresponding to the OFP sediment trap samples, we took temperature, salinity, alkalinity, total DIC, and δ13CDIC from Keeling and Guenther, [1994] collected from BATS between 1989–2002 and from Hydrostation “S” between 1982 and 1989. For the period following 2002, we obtained data from BATS (Nick Bates; personal communication).

2.3 Estimating the Change in εp

[14] Ideally, one would estimate the change in εp directly by comparing samples from different years, taken at the (approximate) same location and in the same season. However, as our data set is too sparse for this purpose, samples from different areas and seasons must be used to characterize εp in any given period. The consequence of this is that any long-term change in εp is likely to be obscured by natural variability. To solve this, we propose that part of the natural variability in εp may be explained by environmental variables, e.g., temperature. By formalizing the relationship between εp and environmental variables in a model, we can disentangle natural variability and long-term change.

[15] Given the plethora of factors that have been suggested to control εp, we do not invoke a mechanistic explanation for its natural variability. Instead, we describe its spatial pattern with a linear function of explanatory variables xi, which we will identify later and account for long-term change by allowing the regression coefficients βi to change at a constant rate:

display math(2)

[16] The long-term change in εp is then also given by a linear function of explanatory variables:

display math(3)

[17] This approach can be thought of as a continuous-time extension of extended Multiple Linear Regression (eMTL) [Friis et al., 2005], which models the spatial pattern in a target variable at two points in time with separate linear functions of environmental variables. In eMTL, the change in the target variable is given by the difference between the two linear equations. In our case, we have samples scattered over the world ocean over a period of five decades, rather than two comprehensive data sets at specific points in time. We therefore impose that the rate of change over the entire modeled period has a constant (linear) relationship with the explanatory variables and use a single linear equation to describe the global variability in εp.

[18] There are no a priori restrictions on the type of variables that could explain the variability in εp. An attractive choice may seem to be the concentration of CO2, which is often suggested to control εp [Rau et al., 1989]. However, the concentration of CO2 exhibits long-term change due to rising atmospheric CO2. While equation (2) does not prohibit a time-dependence of explanatory variables, using time-dependent variables has two drawbacks. First, a time-dependence causes spatial and temporal variability to become inextricably linked. This is most easily seen if we rederive the rate of change, p/dt, which then includes a dependence on ai. As a result, the estimated change in εp is sensitive to its spatial pattern and vice versa. Since εp exhibits pronounced spatial variability, but subtle temporal variability, a model based on time-dependent variables such as CO2 is likely to sacrifice temporal accuracy for a better representation of the spatial pattern. Experiments indicate that this problem indeed occurs when CO2 is used as the explanatory variable. The second problem with time-dependent explanatory variables is that they tend to increase the uncertainty in the estimated rate of change: If explanatory variables are time-dependent, the change in εp becomes a function of the change in explanatory variables. This change rarely is well-constrained. For these two reasons, we restrict ourselves to explanatory variables for which long-term temporal variability is thought to be small relative to their spatial variability.

[19] Several environmental variables exhibit little long-term change between 1962 and 2010 and could thus serve as explanatory variable. These variables include temperature, salinity, alkalinity, and nutrients. Although minor changes in these variables are known or suggested to have occurred (an estimated 0.3–0.5°C global increase in surface ocean temperature has been observed between 1960 and 2005 [Solomon et al., 2007]), these changes are small compared to their overall (spatial) variability. In view of the coverage of our data set, we select temperature, which is known for 88 % of samples (Table 1), as the primary explanatory variable. Temperature is known to correlate with numerous other environmental variables, including productivity and community composition. Notably, temperature has a well-defined relationship with the concentration of CO2 through the temperature-dependence of CO2 solubility [Weiss, 1974]. The commonly proposed dependence of εp on the concentration of CO2 should therefore be captured to some degree by a dependence on temperature as well. In addition to temperature, we have tested a range of secondary explanatory variables, including (climatological) salinity, nitrate, phosphate, silicate, alkalinity, apparent oxygen utilization, and temperature squared. The combination of temperature and nitrate was found to explain most of the observed variability in εp and is therefore used for the remainder of this work. Accordingly, the expected value of εp is given by a function of time t, temperature T and nitrate N:

display math(4)

[20] It should be noted that this model does not attempt to mechanistically explain the variability in εp in terms of temperature or nitrate; these variables were merely found to show the best correlation with εp, and accordingly produce the best constrained estimates of its rate of change.

[21] εp in the data set will display variability that is not captured by the model. We distinguish two sources of variability: errors in the reconstructed value of εp that arise from the use of approximate inputs (e.g., climatological variables, see Table 1) and residual errors (e.g., measurement error, variability not explained by temperature and nitrate). Accordingly, the variance of the difference between observed and modeled εp equals the sum of input-derived and residual variances. For the sake of simplicity, we assume that these differences are normally distributed and independent and that the residual variance inline image is equal for all samples. Input-derived variances are estimated for each sample as described in the previous section.

[22] Given the model for εp and its variability, the coefficients describing spatial and temporal variation (a0, a1, a2, b0, b1 b2) can be estimated. First, a Restricted Maximum Likelihood (REML) procedure is used to estimate residual variance inline image. This estimate is independent of the values of the model coefficients. With the variance for each sample known, model coefficients are given by their Weighted Least Squares (WLS) estimate. To minimize the influence of sampling bias (specifically, overrepresentation of the Southern hemisphere at high latitudes, and disproportional coverage of later years), a bootstrap-based resampling strategy is used [Efron and Tibshirani, 1998]. First, we separate low (0–30°), mid (30–60°) and high latitudes (60–90°). For each latitude range, we partition samples into four subsets by classifying according to hemisphere and time period (pre-1990, post-1990). From each of the four subsets, an equal number of samples is selected randomly with replacement. The selected number is equal to the size of the smallest subset. By combining selections from the three latitude ranges, a final set of 260 data points (out of the original 525) is obtained, which is then used to estimate model coefficients through REML and WLS. Estimated coefficients are combined with climatologies of temperature and nitrate [Garcia et al., 2010; Locarnini et al., 2009] to infer the annual mean distribution of εp in reference year 1990, as well as its time-independent rate of change b0 + b1T + b2N. Resampling, estimation and prediction is repeated 20,000 times to obtain robust estimates (the mean of results from all 20,000 trials) and confidence intervals (percentiles of trial results).

3 Results

3.1 The Observed Distribution of εp in the Surface Ocean

[23] The global distribution of δ13CPOC (cruise tracks shown in Figure 1a and δ13CPOC measurements shown in Figure 1b) shows a distinct latitudinal trend: the heaviest values occur in the tropics (−20‰), the lightest near the poles with a distinct asymmetry between the hemispheres (−24‰ in the Arctic Ocean,−28‰ in the Southern Ocean) (Figure 1b). After accounting for the mainly temperature-driven latitudinal variation in δ13CCO2 (Figure 1c), a weaker latitudinal trend in εp emerges (Figure 1d). Specifically, εp displays only 60% of the latitudinal variation of δ13CPOC. The reason for this lies in the covariation of δ13CPOC and δ13CCO2: The latitudinal pattern in δ13CCO2 mimics that in δ13CPOC, with the most negative values at high latitudes and less negative values at low latitudes. Moreover, δ13CCO2 exhibits little variability at any one latitude. As a result, much of the latitudinal trend in δ13CPOC disappears when δ13CCO2 is subtracted, while the residual (per-latitude) variability remains. Only a small fraction of this residual variability can be attributed to uncertainties in the δ13CCO2 that arise from the use of climatologies (error bars in Figure 1c and 1d). This suggests that much of the variability in εp cannot be explained by latitude; it must arise from latitude-independent variation in physical, chemical, or biological variables instead.

Figure 1.

Compiled dataset of δ13CPOC and εp. (a) Map showing location of samples, see supporting information, Table S1 for full references and dates. (b) Measured δ13CPOC, (c) measured/inferred values of δ13CCO2, and (d) inferred values of εp as a function of latitude (x-axis) and time (color). In Figures 1c and 1d, error bars indicate uncertainties due to the use of climatological variables (primarily temperature and δ13C of DIC, see Table 1).

3.2 The Estimated Change in εp

[24] From Figure 1d it is clear that deciphering any temporal change in εp from the observed data alone is difficult due to the latitudinal asymmetry, residual variability, and the temporal patchiness of the data set. At most, visual inspection could suggest a slight increase in εp at lower latitudes. This trend also emerges in areas where the tracks of cruises from different periods overlap (see S1, supporting information), such as in the North and Equatorial Atlantic. However, sample sizes are too small to conclusively prove that a change in εp has occurred.

[25] This situation improves when all measurements are integrated in the model that describes the spatial and temporal variability in εp with separate linear functions of temperature and nitrate. This model explains a significant fraction of the variability in εp; its correlation with observations equals 0.60 (95% confidence interval: 0.57–0.61). Best estimates and confidence intervals of the model coefficients are listed in Table 2. When combined with climatologies of temperature and nitrate, the model predicts the annual mean global distribution of εp and its rate of change, visualized in Figure 2. The model clearly captures the characteristic increase in εp from low to high latitudes as well as its North–South asymmetry (Figure 2a).

Table 2. Parameter Estimates and Associated 95% Confidence Intervals
ParameterUnitSymbolEstimate95% Confidence Interval
  1. a

    T denotes temperature in °C; N denotes nitrate concentration in µmol L−1; t denotes time, defined as the (decimal) number of years since 1990.

Trend offseta013.2512.27 − 14.23
Trend coefficient for Ta‰ °C−1a10.034−0.007 − 0.074
Trend coefficient for N‰ µmol−1 La20.150.10 − 0.20
Trend coefficient for t‰ yr−1b00.047−0.013 − 0.105
Trend coefficient for tT‰ yr−1 °C−1b1−0.0008−0.0032 − 0.0016
Trend coefficient for tN‰ yr−1 µmol−1 Lb2−0.0021−0.0053 − 0.0009
Residual s.d.σR1.711.61 − 1.82
Figure 2.

(a) Predicted εp for 1990, (b) the best estimate for its change in time, and (c) the minimum change that is significant. These values are calculated by applying the linear model of temperature and nitrate (Table 2) to the annual mean temperature and nitrate values from the World Ocean Atlas [Locarnini et al., 2010]: Figure 2a shows the mean estimate of a0 + a1T + a2N, Figure 2b shows the mean estimate of b0 + b1T + b2N, and Figure 2c shows the latter's fifth percentile. In Figure 2a, sampled values adjusted to the year 1990 by subtracting (b0 + b1T + b2N)t are overlaid. In Figures 2b and 2c, sample locations are denoted by black dots.

[26] The estimated rate of change in εp has a global mean value of 0.022% yr-1, but is subject to considerable spatial variability (Figure 2b). The increase in the equatorial region is modest (on average 0.019% yr-1) and becomes more pronounced near midlatitudes (0.030% yr-1). The model further hints at a strong asymmetry between Northern and Southern high latitudes, with highest rates of increase found in the Arctic Ocean (0.045% yr-1), while the Southern Ocean shows a decrease (−0.007% yr-1). However, estimates for these polar regions are highly uncertain: if we consider the minimum rate of change that is supported by 95% of the model trials (Figure 2c), we find that the change in εp in polar regions is not significant. This changes when moving to lower latitudes: near 30°North and South, the increase is estimated at a minimum of 0.016% yr-1. Finally, a large fraction of the equatorial region is also estimated to experience significant increases in εp. This is most pronounced in the Atlantic, where the minimum increase is estimated at 0.010% yr-1.

3.3 Supporting Global Results With the OFP Sediment Trap Dataset

[27] To test model predictions, we also analyzed a simple linear regression of δ13CPOC of organic matter, collected at OFP sediment traps located near Bermuda at 3200 m water depth. We measured δ13CPOC in spring samples, collected between Apr–Jun, every year between 1978 and 2007 (Figure 3). We compared this to the linear regression of δ13CCO2 from the surface ocean, assuming the bulk of the organic matter is formed in those surface waters. The OFP site is located in the North Atlantic Subtropical Gyre, which is one of the better-studied sites for changes driven by anthropogenic forces. We find that the δ13C of CO2 in the surface ocean is lightening significantly at a rate of 0.024% yr-1. This is comparable to the 0.025% yr-1 decrease in δ13C of DIC found by Gruber et al. [1999] for data prior to 1997. We find that δ13CPOC is lightening by 0.047% yr-1, resulting in εp lightening at a rate of 0.022% yr-1. Due to the small sample size for δ13CPOC, the rate of change for εp is not significant. However, our findings are compatible with the predicted increase of 0.027 (0.012–0.042)% yr-1 derived with the linear model of temperature and nitrate, assuming an annual mean temperature of 23°C and a nitrate concentration of 0.04 µmol L−1, representative for Bermuda.

Figure 3.

Temporal trends in δ13CPOC from spring bloom samples from the OFP sediment trap from 1978–2010 at 3200 m, rate of lightening is −0.047% yr-1 (confidence interval −0.095 − 0.001) (squares) and δ13CCO2 at the surface ocean (5 m), rate of lightening is −0.024% yr-1 (confidence interval −0.038− 0.010) (circles)

4 Discussion

4.1 Evidence for an Increase in εp

[28] We find that εp has increased significantly between 1960 and 2010 in the subtropical surface ocean and in some parts of the tropics, notably in the Atlantic (Figure 2c). The minimum rate of increase is estimated at 0.015% yr-1 in the subtropics and 0.010% yr-1 in the Equatorial Atlantic. No significant change in εp is found at high latitudes. Estimates are derived from a linear model that features temperature and nitrate as explanatory variables, but similar rates of increase are found with one- and two-variable models based on other explanatory variables (e.g., salinity, phosphate, apparent oxygen utilization, and temperature squared). This suggests that the inferred patterns are robust. Moreover, sediment trap samples from the Ocean Flux program near Bermuda suggest a comparable increase in εp of 0.022% yr-1.

4.2 Why Does εp Increase? Indications for Control by CO2

[29] Natural variation in εp is often explained in terms of a varying degree of intracellular Rayleigh fractionation in phytoplankton, governed by the supply of inorganic carbon on the one hand and its removal due to biological fixation on the other [Baird et al., 2001; Farquhar et al., 1982; Laws et al., 1995; Popp et al., 1998; Rau et al., 1989; Rau et al., 1996; 1997]. Rising atmospheric CO2 primarily increases the supply of inorganic carbon; it arguably has less effect on fixation, which is also controlled by light and nutrient availability. An increase in carbon supply would cause more pronounced fractionation, i.e., an increase in εp. This trend is supported by our results: significant rates of change are positive without exception (Figure 2c).

[30] When εp is controlled by intracellular Rayleigh fractionation, the sensitivity of εp to perturbations in CO2 would be expected to decrease with increasing CO2 concentration: at high CO2, εp would already reflect the maximum biological fractionation factor and show little sensitivity to further CO2 increases. As a large fraction of the surface ocean is at near-equilibrium with the atmosphere, the concentration of CO2 is primarily governed by its solubility, which decreases with increasing temperature [Weiss, 1974]. Thus, warmer waters have lower concentrations of CO2 than colder waters, and accordingly should experience a greater increase in εp. To some degree, this trend is seen in our results: significant increases are only found in subtropical and tropical regions, not at higher latitudes. However, the predicted increase in the subtropics exceeds that in the tropics. One explanation for this is that in much of the tropics, upwelling of CO2-rich deep water causes a higher surface CO2 concentration than would be attained by equilibration with the atmosphere. As a result, the sensitivity of εp to rising CO2 would be reduced. Elevated concentrations of CO2 are most pronounced in the Equatorial Pacific (EP) and least in the Equatorial Atlantic (EA), which agrees with the model assigning the EP the lowest rate of increase of all tropical regions and the EA the highest.

4.3 Quantitative Support for a Link Between εp and CO2

[31] Predictions for the change in εp agree qualitatively with the hypothesis that it is mediated by rising atmospheric CO2. To further explore this, we compare with the quantitative predictions from a minimal mechanistic model of phytoplankton carbon acquisition [Laws et al., 1995]. This is a one-compartment model for phytoplankton that includes CO2 uptake and leakage over the cell membrane (proportional to external and internal [CO2], respectively) and internal carbon fixation; both CO2 uptake and fixation may fractionate carbon. The model links εp to its biological maximum εpm, the ratio of carbon fixation to CO2-specific carbon uptake ρ, and the concentration of CO2:

display math(5)

[32] To facilitate comparison with the original formulation by Laws et al., [1995] we note that εpm is the fractionation factor associated with carbon fixation (εR), diminished with the fractionation factor associated with carbon uptake (e.g., diffusion over the cell membrane). ρ represents the ratio of the rate carbon fixation to the rate CO2-specific carbon uptake, multiplied with εR/εpm.

[33] Rising atmospheric CO2 primarily increases [CO2]. We cannot exclude the possibility that this would also affect εpm and ρ, e.g., through changes in community structure, but such changes would be secondary biological responses to the change in [CO2]. Here we aim to establish whether the 1960–2010 change in [CO2] alone suffices to explain the observed change in εp. In this case, the change in εp is given by

display math(6)

[34] This is subject to spatial variability. Annual mean [CO2] varies spatially between 8 and 25 µmol L−1, primarily due to the temperature-dependence of CO2 solubility [Weiss, 1974]. The relative change in dissolved CO2, inline image, equals the relative change in pCO2 (provided local solubility does not change), which has an annual mean of 0.0031–0.0055 yr−1 if we adopt an increase in pCO2 of 1.5 ppm yr−1 [Takahashi et al., 2009]. To enable a more direct comparison between spatial variation of [CO2] and inline image we calculate their dimensionless coefficient of variation, i.e., their standard deviation divided by the mean, each taken over a 1° global climatology. Coefficients of variation are 0.33 for [CO2] and 0.06 for inline image, i.e., spatial variation in [CO2] exceeds that in inline image more than fivefold. If ρ and εpm do not vary systematically with [CO2], the change in εp would approximately be proportional to 1/[CO2]. This is the quantitative underpinning of the previous observation that the sensitivity of εp to changes in CO2 is expected to be greatest at low CO2.

[35] To test the relationship between εp and CO2, we fit the carbon acquisition model of equation (5) to the εp data set, using nonlinear least squares estimation in combination with the resampling procedure introduced previously. This renders (global mean) estimates of εpm = 17.6 (16.4 − 18.6)‰ and ρ = 2.02 (1.42 − 2.55) µmol L−1. The resulting relationship between [CO2] on the one hand, and εp (equation (5)) and its rate of change (equation (6)) on the other, is shown in Figure 4. Clearly, the carbon acquisition model captures only large-scale variability in εp, specifically its increase with increasing CO2 (Figure 4a). This is consistent with the conclusion reached by Goericke and Fry [1994]: εp and [CO2] show only weak covariation in the modern ocean.

Figure 4.

(a) εp and (b) its rate of change as a function of the concentration of CO2, as predicted by the carbon acquisition model of equations (5) and (6), shown in red. The rate of change is based on an increase of 1.5 ppm CO2 yr−1 [Takahashi et al. 2009]; it shows minor variation independent of the concentration of CO2 due to variation in pCO2 between samples. In Figure 4a, measured combinations of εp and CO2 are shown by open circles (Northern hemisphere) and triangles (Southern hemisphere). In Figure 4b, open symbols indicate the rates of change predicted by the linear model of temperature and nitrate, which are shown along with error bars representing 95% confidence intervals.

[36] If we set aside the issue of residual variability in εp for the moment, the proposed relationship between εp and CO2 (Figure 4a) can be used to infer the change in εp induced by rising CO2. Figure 4b shows the resulting increase in εp (red symbols), on top of predictions from the linear model based on temperature and nitrate (open symbols and error bars). The rates of increase derived with the carbon acquisition model fall within the confidence interval of the linear model. In other words, the rates of change predicted by the two models are statistically indistinguishable. This may seem trivial, as both models have been calibrated against the same data set and confidence intervals are large, but it is in fact an important result. The carbon acquisition model relates all variability in εp to dissolved CO2. As spatial variability in εp and [CO2] is far greater than the long-term change between 1960 and 2010 (Figure 1), the carbon acquisition model is effectively tuned to reproduce to the spatial pattern of εp—its temporal change follows simply from the imposed change in CO2. In contrast, the linear model does not impose any relationship between spatial and temporal variability in εp: it separately describes spatial variation of εp (a0 + a1T + a2N) and its rate of change (b0 + b1T + b2N). The fact that its estimated rate of change is in all cases indistinguishable from that predicted by the carbon acquisition model provides further support for the hypothesis that the temporal change in εp is driven by CO2. While confidence intervals of the linear model are large, we would argue that the agreement between models remains encouraging, particularly at low CO2 (<12 µmol L−1), where estimates of the linear model are best constrained.

4.4 Distinguishing Between Physiology and CO2 as Drivers of εp

[37] Clearly, the carbon acquisition model cannot explain all variability in εp in terms of variation in CO2 (Figure 4a). This is understandable, as experiments that have shown that temperature, nutrient availability [Bidigare et al., 1997], day length [Burkhardt et al., 1999a], species-specific variation [Burkhardt et al., 1999a; Falkowski, 1991], and light intensity [Laws et al., 1995; Nimer and Merrett, 1993; 1996] all influence εp. Most of these factors are expected to act on εp by changing ρ: the ratio of carbon fixation to CO2-specific carbon uptake. Thus, much of the residual variability in Figure 4a may be due to variation in the rates of fixation or uptake. Fortunately, we can recover a simple linear relationship between εp and its rate of change by combining equations (5) and (6):

display math(7)

[38] This relationship is independent of ρ and should therefore hold under varying rates of carbon acquisition and fixation. Accordingly, it is more likely than equation (6) to hold across the world ocean. The slope of the linear relationship between inline image and εp equals (minus) the relative change in CO2, inline image, which we determined previously to be relatively constant across the globe (coefficient of variation of 0.06); it has a global mean of 0.0043 yr−1. Linear regression of the rate of change in εp predicted by the linear temperature + nitrate model on measured εp strikingly produces an identical slope of −0.0043 yr−1, albeit that this slope is not highly constrained (95% confidence interval −0.010 to 0.001). To summarize, if we assume long-term change in εp is exclusively the direct result of increased availability of CO2 (and not of changes in carbon acquisition, fixation, or the maximum fractionation factor εpm), the carbon acquisition model states that the change in εp must be proportional to εp itself. The model also specifies the value of the proportionality constant. The fact that a CO2-independent statistical analysis of the data set recovers this same constant of proportionality strongly suggests that rising CO2 directly affects εp.

4.5 Phytoplankton Carbon Acquisition in More Detail

[39] The carbon acquisition model of equation (5) arguably gives the simplest possible mechanistic description of the link between εp and the carbonate system. Several authors consider further detail, notably the common use of HCO3 as a carbon source by phytoplankton [Burkhardt et al., 1999b; Sharkey and Berry, 1985]. This would affect εp as HCO3 is isotopically heavier than CO2. To account for HCO3 uptake in the model, εpm must be re-interpreted as εR − b, where εR is the fractionation factor associated with carbon fixation, f the fraction of inorganic carbon taken up as HCO3, and εb the fractionation of HCO3 relative to dissolved CO2. Does this change our interpretation? εR is 22–31‰ [Tcherkez et al., 2006], while εb varies globally between 8.5 and 12.3‰ according to its temperature dependency [Zhang et al., 1995]. Under these constraints, variation in εpm remains modest: at most 3.8‰ around a mean of 11.6‰, obtained at minimum εR and f = 1. This suggests that equation (7) holds approximately, even if phytoplankton use HCO3 and even if HCO3 use is spatially variable. Accordingly, equation (7) could be a robust indicator of CO2 sensitivity in εp.

[40] Further complications arise if f, the fraction of inorganic carbon taken up as HCO3, is affected by a change in [CO2]. This would require modification of earlier expressions for the change in εp (equations (6) and (7). A change in f in response to rising [CO2] might be expected both because phytoplankton carbon acquisition might respond passively to increased availability of CO2 relative to HCO3 and because HCO3 uptake could be actively downregulated. Specifically, the elevation of CO2 could alleviate the requirement for carbon concentrating mechanisms (CCMs) found in most phytoplankton. A reduction in CCM machinery would shift the ratio of HCO3 to CO2 utilization and free up more energy for growth [Raven et al., 2011]. Evidence for CO2-sensitivity of f is conflicting: it is not consistently found across taxa [Rost et al., 2003] and is seen in some field studies [Neven et al., 2011] but not all [Cassar et al., 2004; Tortell et al., 2010; Tortell et al., 2008]. We therefore believe present-day knowledge to be insufficient to introduce a quantitative relationship between f and [CO2], as required by equations (6) and (7). Nevertheless, we may note that if increases in [CO2] alleviate nutrient demands for operating CCMs [Beardall and Giordano, 2002; Hopkinson et al., 2011], it is likely that the biggest response will be in regions of limiting nutrients, i.e., the subtropical gyres. Here an alleviation of nutrient requirements due to a downregulation of CCMs could have a large benefit. In nutrient-rich upwelling areas, such as equator and polar regions, the advantage is less obvious. Downregulation of CCMs increases the uptake of CO2 relative of HCO3, and thus increases εp. If most pronounced in oligotrophic areas, CCM downregulation might produce a similar spatial pattern to that shown in Figures 2b and 2c. Unlike the CO2-only model of equation (5), however, this spatial pattern would then not necessarily indicate changes in the intracellular concentrations of carbon species.

[41] Our results suggest that the increase in εp is likely a result of enhanced CO2 availability for phytoplankton. This could benefit phytoplankton growth by directly increasing intracellular CO2 [Laws et al., 1995] or by freeing energy and nutrients through downregulation of CCMs. In turn, increased growth would affect εp by raising the relative rate of carbon fixation (ρ in equation (5)). However, it seems unlikely that this mechanism played a major role between 1960 and today: while some phytoplankton, such as Emiliania huxleyi, may be limited under modern day CO2 conditions [Iglesias-Rodriguez et al., 2008; Rost and Riebesell, 2004], other taxa (e.g., diatoms) are not [Riebesell et al., 1993; Rost et al., 2003]. Moreover, any direct benefit of rising CO2 may be obscured by its indirect consequences, e.g., an increase in temperature leading to enhanced stratification and nutrient limitation [Behrenfeld et al., 2006; Steinacher et al., 2010].

5 Conclusion

[42] We find a significant change in the fractionation of stable carbon isotopes by phytoplankton in the surface ocean at low and mid latitudes between the 1960s and today. The change is strongest in the subtropics, with the rate of increase estimated at a minimum of 0.015% yr-1. Results qualitatively and quantitatively support the hypothesis that anthropogenic CO2 emissions increase the availability of inorganic carbon for phytoplankton, which directly causes more pronounced fractionation. While we cannot rule out the possibility that physiological parameters such as the rate of carbon fixation, CO2-specific uptake,HCO3 use, or fractionation during fixation have changed as well, our conclusion is that such secondary changes are not required to explain the dominant variability in εp. Phytoplankton appear to be already responding to a changing environment.

Acknowledgments

[43] This research was supported by ERC grant SP2-GA-2008-200915 and a Phillip Leverhulme Award to R.E.M.R.. J.N.Y. acknowledges financial support through the Clarendon Scholarship, Oxford. The work of J.B. was supported by the Netherlands Organisation for Scientific Research (NWO) through the Rubicon program, grant 825.09.018, and by a Junior Research Fellowship at St. John's College, Oxford. Comments of Paul Quay and one anonymous reviewer helped to further refine statistical methods and improve the manuscript text. We thank the BODC, Tony Bale (PML) and Yoshihisha Mino for providing AMT3 underway δ13CPOC data; Heather Bouman, Thomas Jackson (Oxford University), and Martine Couapel (NHM) for POC and DIC sample collection onboard Icechaser and AMT18 cruises, respectively; Nick Bates for unpublished BATS time-series (http://bats.bios.edu/) between 2006–2009. We thank the captain and crew of the RRS James Cook and RRS James Clark Ross for their assistance in AMT and Ice chaser cruises, respectively. We thank the captain and crew of the R/V Atlantic Explorer for their assistance on OFP cruises, and JC Weber and Marshall Otter for assistance with OFP sample analyses. We thank the NSF Chemical Oceanography Program for continuous support of the Oceanic Flux Program time-series over more than 30 years, mostly recently by grants OCE 092285 and OCE 0623505.

Ancillary