Estimates of Holocene atmospheric CO₂ concentrations strongly rely on CO₂ measured in air extracted from Antarctic ice cores. Records for the past millennium indicate significantly reduced CO₂ levels from A.D. 1550 to 1800, which are temporally related to the historical Little Ice Age climate deterioration (1, 2). By contrast, Antarctic ice-core data do not clearly support a temperature–CO₂ correlation during the eight earlier short-term cooling pulses that punctuated Holocene climatic conditions in the North Atlantic region with a periodicity of ≈1,500 years (3). Only the so-called 8.2-ka-B.P. cooling event has been associated with a long-term, modestly declining CO₂ trend recognized in the Taylor Dome record (2). This most prominent century-scale instability of Holocene climate is reflected as a negative δ¹⁸O excursion in Greenland ice cores (4, 5) as well as in a variety of marine and terrestrial proxy records indicative of pronounced modifications of regional temperature and/or precipitation regimes in the North Atlantic region (3, 4, 6–11). The instability is related to a brief episode of massive release of meltwater associated with the final demise of the Laurentian ice sheet (12).

The conventional iced-based concept of relatively stabilized CO₂ concentrations during the greater part of the Holocene is challenged increasingly by stomatal frequency analysis of fossil leaves (13–15). Species of C₃ plants are often characterized by a plastic phenotype capable of consistent adjustment of numbers of leaf stomata in response to changes in ambient CO₂ concentration (16–18). Identification of a CO₂-sensitive gene involved in stomatal development in Arabidopsis thaliana demonstrates the genetic control of the response (19). As a corollary of this responsiveness, stomatal frequency analysis of fossil leaves enables the detection and quantification of atmospheric CO₂ changes at different time scales (14, 17–25). High-resolution analysis of tree leaves buried in Lateglacial and Holocene peat and lake deposits suggests temporal correlation between global CO₂ dynamics and Northern Hemisphere temperature. In Europe and North America, reconstructed CO₂ fluctuations (14, 15, 20) appear to parallel well documented δ¹⁸O and chrinomid-based temperature changes associated with the Younger Dryas stadial and the Preboreal Oscillation, the first of the Holocene climatic instabilities occurring at ≈11,000 calendar years B.P. (26). Both Younger Dryas and Preboreal Oscillation are not apparent in the CO₂ record of Antarctic ice (2, 27), which may be due to generally low temporal resolution of Lateglacial and early Holocene CO₂ data from Antarctica.

To corroborate the concept of a coupling between recurrent Holocene cooling pulses and CO₂ fluctuations, we document stomatal frequency data that constrain timing and magnitude of CO₂ shifts associated with the prominent 8.2-ka-B.P. cooling event. We analyzed leaves of European tree birches (Betula pubescens and Betula pendula) preserved in cores of gyttja deposits from Lake Lille Gribsø, North of Copenhagen, Denmark (55°58′43′′N; 12°18′52′′E). Well preserved leaf remains occur continually through an interval corresponding to the period between ≈8,700 and ≈6,800 calendar years B.P. Temporal control is provided by a series of six accelerator mass spectrometry ¹⁴C chronologies measured on single leaves (Table 1).

Stomatal frequency can be expressed in terms of stomatal density and stomatal index (SI): SI = (stomatal density/[stomatal density + epidermal cell density]) × 100. In contrast to stomatal density, SI reflects stomatal frequency independently of variation in epidermal cell size related to light intensity, temperature, or nutrient and water availability. In angiosperm leaves, therefore, mean SI is the more sensitive parameter for detecting stomatal frequency responses to changing CO₂ levels (28, 29). Effects of variation within and between leaves can be eliminated by using large data sets (17, 18, 28). Leaves of B. pendula and B. pubescens display essentially similar SI patterns and can be treated therefore as a single category in stomatal frequency analysis (18). There is sufficient evidence from field studies and experiments that CO₂-induced trends in mean SI for Betula leaves are not disturbed significantly by environmental factors other than CO₂ (30, 31).

Standardized, computer-aided determination of stomatal parameters on leaf cuticles was performed on a Leica Quantimet 500C/500+ image-analysis system. Measured parameters include stomatal density and epidermal cell density (including guard cells). Counting areas are restricted to stomata-bearing alveoles. Calculated SIs are mean values for up to seven leaves per data point. Seven digital images (field area, 0.035 mm²) per leaf were analyzed (standard deviations are constant after seven counts). We used the rate of historical CO₂ responsiveness of the European tree birches (Fig. 1) to quantify early Holocene SI-based CO₂ levels. Our approach of inferring an unknown value of a quantitative variable (CO₂/x₀) from the quantitative response of a single taxon (SI_birch/y) meets the basic requirements for a classical linear regression, which allows a better performance at the extremes and with slight extrapolation (32–34).

• Download figure
• Open in new tab
• Download powerpoint

Fig 1.

Historical response of SI for European tree birches (B. pendula and B. pubescens) to global atmospheric CO₂ increase from 287 to 356 ppmv. The training set includes mean SI values for herbarium material (•) collected in The Netherlands and Denmark in the period 1843–1995 and leaf remains from living peat (⋄) in The Netherlands, formed in the period 1952–1995 (17). Historical CO₂ concentrations of 315–356 ppmv are mean spring-season values from Mauna Loa monitoring (http://cdiac.esd.ornl.gov/ndps/ndp001.html), and concentrations of 287–315 ppmv are representative values derived from shallow Antarctic ice cores (http://cdiac.esd.ornl.gov/trends/co2/siple.htm). For analytical method, see text. Regression statistics: n = 63; slope = −0.0846. Goodness-of-fit linear model: R² = 0.78, R = 0.78. Analysis of variance results: F(1–62) = 225.11 (P = 0.000). Statistical analyses were performed with SPSS 8 for Windows (Chicago). B. pendula and B. pubescens display significantly similar SI patterns and can be treated therefore as a single category in stomatal frequency analysis. Species-specific regression statistics: B. pubescens: slope = −0.0871; goodness-of-fit linear model: R² = 0.63, R = 0.60; B. pendula: slope = −0.0724; goodness-of-fit linear model: R² = 0.80, R = 0.79. Trend lines are not given in the figure.

The reconstructed CO₂ record shows a fluctuating pattern (Fig. 2). Inferred CO₂ minima with averages of ≈275 ppm by volume (ppmv) occur at ≈8,680 years B.P. and between ≈8,430 and ≈8,040 years B.P.; prominent maxima with values of 300–325 ppmv occur at ≈8,640 years B.P. and between ≈7,920 and ≈7,270 years B.P. The series of low CO₂ concentrations around 8,300 years B.P. follow a declining trend of ≈25 ppmv within a time interval of <100 years. This interval ends with a return to levels >300 ppmv after ≈300 years. Timing and duration of the century-scale CO₂ excursion are in harmony with the proxy records for the 8.2-ka-B.P. cooling event embracing a period of 200–300 years between ≈8,400 and ≈8,100 years B.P. (3–12). The SI signals thus indicate a temporal association between the 8.2-ka-B.P. cooling and atmospheric CO₂ concentration. The exact phase relationship between changes in temperature and CO₂ cannot be determined yet.

• Download figure
• Open in new tab
• Download powerpoint

Fig 2.

Reconstructed CO₂ concentrations for the time interval between ≈8,700 and ≈6,800 calendar years B.P. based on CO₂ extracted from air in Antarctic ice of Taylor Dome (left curve; ref. 2; raw data available via www.ngdc.noaa.gov/paleo/taylor/taylor.html) and SI data for fossil B. pendula and B. pubescens from Lake Lille Gribsø, Denmark (right curve; see Table 1). The arrows indicate accelerator mass spectrometry ¹⁴C chronologies used for temporal control (Table 1). The shaded time interval corresponds to the 8.2-ka-B.P. cooling event (3–12). Quantification of mean CO₂ concentrations is based on the rate of historical CO₂ responsiveness of the European tree birches (Fig. 1); ±1σ CO₂ estimates are derived from the standard deviation of the SI mean values.

Our CO₂ reconstructions reflect rapid changes with a significantly greater magnitude than the smooth and modest atmospheric CO₂ decline to values ≈260 ppmv inferred from the low-resolution Taylor Dome ice-core record (Fig. 2). The data also confirm the regular occurrence of early Holocene atmospheric CO₂ concentrations well above 300 ppmv, unknown from Antarctic ice cores but common in leaf-based time series (9, 14). These apparent controversies between leaf-based and ice-based CO₂ data have not been resolved yet (ref. 35; see text at www.sciencemag.org/cgi/content/full/286/5446/1815a). It should be noted that early Holocene records from Greenland ice cores have repeatedly indicated rapidly fluctuating CO₂ levels including values >300 ppmv (36, 37). At present, the Antarctic record is usually considered to be reliable, so that discrepancies are ascribed to CO₂ enrichment within the Greenland ice (38, 39). However, there is evidence that in polar ice also postdepositional CO₂ depletion could occur, but underlying chemical processes of this potential source of error have not yet been investigated in detail (38, 39).

The documented coupling between CO₂ fluctuations and the 8.2-ka-B.P. cooling implies a distinctive involvement of the oceans, where short-term perturbations of sea-surface temperature and/or salinity allow rapid CO₂ transfer between the atmosphere and surface waters. Holocene cooling events are generally related to a reduction in the thermohaline circulation, resulting in sea-surface temperature lowering in large parts of the North Atlantic (3, 6). This forcing mechanism is evident particularly for the 8.2-ka-B.P. cooling event, where weakening of the thermohaline circulation was triggered by catastrophic release of Laurentide meltwater (12). By applying a freshwater perturbation over 20 years, a global atmosphere-sea-ice-ocean model simulation of the 8.2-ka-B.P. event produces a 320-year lasting weakening of the North Atlantic thermohaline circulation and 1–5°C surface cooling over the adjacent continents (40).

The reconstructed atmospheric CO₂ reduction of ≈25 ppmv indicates a temporarily enhanced North Atlantic sink for CO₂ at the time of the 8.2-ka-B.P. cooling event. While regional palynological data support temperature changes (11), vegetation reconstructions do not provide evidence of an extended terrestrial sink. The occurrence of global CO₂ fluctuations substantiates the interpretation of δ¹⁸O records in ice cores from Antarctica and Greenland in terms of globally parallel climate changes on the Northern and Southern Hemispheres (41). The CO₂ fluctuations falsify conclusively the concept of an approximate antiphase relation between short-term temperature changes on the two hemispheres that would cause buffering of North Atlantic CO₂ drawdown by the effects of synchronous sea-surface temperature increase in the Southern Ocean (42). It thus may be concluded that leaf-based CO₂ data support a much more dynamic evolution of the Holocene CO₂ regime than previously thought. In effect, there seems to be every indication that the occurrence of Holocene CO₂ fluctuations is more consistent with current observations and models of past global temperature changes than the common notion of a relatively stable CO₂ regime until the onset of the Industrial Revolution.

• Received April 20, 2002.
• Accepted July 15, 2002.

• Copyright © 2002, The National Academy of Sciences