Knowledge of the evolution of atmospheric carbon dioxide concentrations throughout the Earth’s history is important for a reconstruction of the links between climate and radiative forcing of the Earth’s surface temperatures. Although atmospheric carbon dioxide concentrations in the early Cenozoic era (about 60 Myr ago) are widely believed to have been higher than at present, there is disagreement regarding the exact carbon dioxide levels, the timing of the decline, and the mechanisms that are most important for the control of CO2 concentrations over geological timescales. Here we use the boron-isotope ratios of ancient planktonic foraminifer shells to estimate the pH of surface-layer sea water throughout the past 60 million years, which can be used to reconstruct atmospheric CO2 concentrations.
We estimate CO2 concentrations of more than 2,000 p.p.m. for the late Palaeocene and earliest Eocene periods (from about 60 to 52 Myr ago), and ®nd an erratic decline between 55 and 40 Myr ago that may have been caused by reduced CO2 outgassing from ocean ridges, volcanoes, and metamorphic belts and increased carbon burial. Since the early Miocene (about 24 Myr ago), atmospheric CO2 concentrations appear to have remained below 500 p.p.m. and were more stable than before, although transient intervals of CO2 reduction may have occurred during periods of rapid cooling approximately 15 and 3 Myr ago.
More than a century has passed since Arrhenius1 proposed that early Cenozoic warmth was caused by high levels of carbon dioxide in the atmosphere and Chamberlin2 suggested a variety of geological processes that could affect atmospheric carbon dioxide concentrations (pCO2 ). However, there is still disagreement as to the most
signi®cant controls on pCO2 over long timescales. Some authors have stressed the importance of changing inputs to the atmosphere such as volcanic and hydrothermal outgassing3,4 or metamorphic decarbonation reactions5 , while others have focused on outputs such as the weathering of silicate minerals and limestone formation6±8 or
organic carbon burial. The level of early Cenozoic pCO2 and when it might have declined is also still under debate.
The boron-isotope (d11B) approach to pCO2 estimation relies on the fact that a rise in the atmospheric concentration will mean that more CO2 is dissolved in the surface ocean, causing a reduction in its pH. We are able to estimate the pH of ancient sea water by measuring the boron-isotope composition of calcium carbonate (d11Bcc) precipitated from it. This is because boron in aqueous solution occurs as two species, B(OH)3 and B(OH)- 4, between which the equilibrium is strongly pH-dependent over the natural acidity range of sea water. Furthermore, there is a pronounced isotopic fractionation between the species of approximately -19.5½, so that the d11B of each species is highly dependent12 on pH. Because boron incorporation into marine carbonates is predominantly from B(OH)- 4, d11Bcc is a sensitive pH indicator13,14. The pH of sea water is governed by the carbonate equilibria, such that for a given pH value it is possible to calculate the aqueous CO2 concentration and thereby make quantitative estimates15±20 of atmospheric pCO2.
The pH and aqueous CO2 concentration of the surface ocean vary spatially because of factors such as deep-water upwelling, local productivity regimes, and freshwater in¯ows. To arrive at pH estimates that most closely re¯ect atmospheric pCO2, it is necessary to measure the d11B of carbonates that were precipitated far from coastal in¯uences and sources of upwelling. The ideal setting is in the low-latitude gyre systems, where a mixed layer of warm, lowdensity sea water in contact with the atmosphere generally overlies colder deep waters with little intermixing. Such environments support abundant planktonic foraminifera (a group of microscopic protists) that secrete calcite (CaCO3) shells. The shells fall to the sea ¯oor, from which a record of upper-ocean pH of many millions of years can be obtained.
We analysed the d11B of monospeci®c sample splits of surface mixed-layer dwelling foraminifera from 32 sediment samples from the open tropical Paci®c, spanning the past 60 Myr, augmenting data from six other previously studied samples19,20 (Table 1). It is necessary to use various species because none survived the entire time interval. Each monospeci®c sample consists of more than 100 individual shells that calci®ed at different times of the day and year
over a period of several thousand years, so the small deviations from equilibrium pH and aqueous CO2 that constantly occur in the upper ocean are likely to be averaged. Analytical methods were the same as previously published19 and boron-isotope ratios are quoted relative to standard NBS SRM 951.
The study sites (Ocean Drilling Program Sites 865, 871 and 872) are all from the sedimentary caps of ¯at-topped seamounts in the tropical north Paci®c gyre, and are characterized by exceptionally good carbonate preservation. They have been the subject of considerable previous geochemical work, including isotopic, trace element and microstructural characterization of the foraminifera shells21±25. There is no evidence from this work that diagenetic alteration has affected shell chemistry. We do not know of any similarly well preserved carbonates from the open ocean in the age range of 25±33 Myr ago.
In order for a pH value to be calculated from a d11Bcc measurement, it is necessary to assume a value for the boron-isotope composition of sea water (d11Bsw); currently this value is +39.5½, but it may have changed through time. Fortunately, geologically rapid ¯uctuations in d11Bsw are unlikely to have occurred as dissolved boron has a long residence time26 in the ocean of approximately 20 Myr. We can estimate past d11Bsw values by utilizing the fact that different planktonic foraminifera species calcify from the surface mixed layer to the low pH conditions below the thermocline. The magnitude of the pH decline is controlled mainly by the local level of biological production, because
more productive locations have a greater ¯ux of sinking organic particles, which oxidize in the water column, reducing pH at depth. Although there is good evidence that the sites we analysed were always oligotrophic (with a maximum apparent oxygen utilization between 50 and 150 mM kg-1), in six previously studied time windows in the Cenozoic we have observed substantial differences in the d11Bcc differential between surface and deep planktonic environments19,20. As we have argued20, such differences can be explained as having been caused by variation in d11Bsw. Hence, we use the value of the d11Bcc differential to model d11Bsw for each time window and interpolate d11Bsw values for the other samples (Table 1).
Changes in sea surface temperature (SST) also affect pH and pCO2 estimates because the boric acid and carbonate equilibria are temperature dependent. However, oxygen isotopic studies have shown that the tropical SST has remained surprisingly constant through the Cenozoic despite pronounced cooling in the high latitudes27. We have assumed a SST of 27 8C for all our samples.
The d11B of surface-dwelling foraminifera may not provide a good estimate of pH in the mixed layer if the microenvironment of the foraminifer cell has a distinctive pH because of photosynthesis by algal symbionts. A symbiont “vital effect” on d11Bcc has been suggested for the analagous but much larger aragonitic skeletons of
corals28. Microsensor studies of the living symbiotic species Globigerinoides trilobus (also known as G. sacculifer) and related taxa show that pH increases over short distances as the foraminifera shell is approached29,30. However, there is evidence that G. trilobus precipitates its shell with no boron-isotope vital effect16±18. Furthermore, because our analyses of various extinct species produces an oceanographically consistent pH±depth pro®le20, we infer that
Palaeogene surface-dwelling species of the genera Acarinina and Morozovella also precipitated their shells at or close to boronisotopic equilibrium.
However, one of the species studied, upper Eocene Hantkenina alabamensis, yielded extremely negative d11Bcc values in all three samples analysed that cannot be interpreted using the pH model (see Table 1). As there is no textural or geochemical reason to suspect diagenetic alteration25, we conclude that a strong vital effect fractionation of the boron isotopes occurred in this species. One possibility is that H. alabamensis (which has a very involute shell
and large enveloping ®nal chambers) resorbed earlier-formed calcite late in its life cycle, thereby reworking isotopically light boron. These data are excluded from subsequent geochemical calculations.
Figure 1 illustrates the record of sea surface pH that we have reconstructed from the d11Bcc data. Our ®nding that the pH was signi®cantly lower in the early Palaeogene than in the Neogene period has implications for a wide range of biological and chemical processes in the ocean. For example, the speciation of a number of dissolved elements in sea water (such as P, N, Mn and Se) undergoes variations over the range of pH values (7.4±8.3) calculated in this study. The surface ocean would not have been so acidic as to inhibit biological calci®cation, but other more subtle effects on plankton ecology and sedimentation patterns may have occurred.
Reconstructing seawater carbon chemistry
The pH of the surface ocean is one of four key variables that de®ne its carbonate chemistry, the others being the concentration of aqueous CO2, the amount of total dissolved inorganic carbon (SCO2) and the total alkalinity31. In order to calculate a value for pCO2 it is necessary to assume a value for SCO2 or alkalinity.Unfortunately, no reliable record of variations in either parameter
exists over the timescale covered by this study. However, variation in the carbonate compensation depth (CCD) in the ocean has previously been inferred from historical patterns of carbonate sedimentation32, and this can be used to constrain alkalinity and SCO2.
Our approach was to model the ancient ocean by adjusting alkalinity so as to ensure that the water column is exactly saturated with respect to calcite at the lysocline (which we take to be 500 m shallower than the CCD). The depth of the lysocline depends on the calcium concentration of sea water, [Ca]sw, as well as the carbonate equilibria. The most important process that may have caused changes in [Ca]sw and alkalinity over the timescales considered in
this study is variation in the river water ¯ux. For example, continental weathering was probably more intense during
periods of warm climate and high pCO2, which would deliver more Ca and HCO3 to the ocean. Therefore we have assumed that [Ca]sw has remained proportional to alkalinity. We then calculate surfaceocean alkalinity and SCO2 by assuming that increases in these parameters with depth have remained the same as the modern western equatorial Paci®c Ocean, that is, DSCO2 from 0 to 4,000 m equals 340 mM kg-1 , D(alkalinity) from 0 to 4,000 m equals 75 meq kg-1
. Although this assumption may not be strictly correct, the relative homogeneity of sediment geochemistry in this
area over the past 60 Myr (refs 21±24) means it has undergone considerably less variation in the factors that control depth variation in SCO2 and alkalinity (principally productivity rates) than most other areas.
The calculated sea surface alkalinity record is shown in Fig. 2. The large amplitude variation in alkalinity over relatively short time periods in the Palaeogene is implied by the combined observations that the CCD was constant32 but surface ocean pH varied markedly. However, we note that the CCD record for the Palaeogene Paci®c Ocean is relatively poorly constrained, and it is possible that an updated study would imply less rapid alkalinity variation.
pdated study would imply less rapid alkalinity variation.Having established values for surface ocean pH and alkalinity, it is possible to calculate aqueous CO2 and atmospheric pCO2. The pCO2 record is shown in Fig. 3a, with an expansion for the Neogene in Fig. 3b. We emphasize that the error bars in Figs 1 to 3 re¯ect only the analytical uncertainties in determining d11Bcc. The absolute estimates must be treated with caution because there are other sources of error that are less easy to quantify. One important uncertainty is that the values of the various constants for carbonate equilibria in sea water have not been accurately measured at pCO2 levels above 500 p.p.m. (ref. 33), and so are subject to revision. Other problems,as discussed above, include taking correct values for SST, d11Bsw, [Ca]sw, local biological productivity and CCD, as well as accounting for possible species-speci®c vital-effect fractionation and diagenetic
alteration of the boron-isotope ratios. Despite these dif®culties, which are similar to those encountered in other lines of palaeoclimate research, we believe that the timing and direction of the changes in the properties illustrated in Figs 1 to 3 are robust and warrant discussion.
Causes of Cenozoic pCO2 variation
Modelling of the atmosphere suggests that pCO2 forces surface temperature in a logarithmic relationship. We use the equation of Kiehl and Dickinson34 to estimate the level of greenhouse heating predicted by our pCO2 estimates. We note that the relatively small changes in pCO2 that occurred in the Neogene would have had a disproportionately large effect on surface temperature because absorption of outgoing radiation by CO2 is further from saturation at low values. In Fig. 4, we compare this record with the oxygenisotope ratio (d18O) of deep sea benthic foraminifera35±37, which is an important climate proxy that re¯ects both cooling and ice growth. Also indicated on Fig. 4 are some of the most important palaeoclimate events of the Cenozoic.
It has long been recognized that the early to early-middle Eocene (55 to 45 Myr ago) was a pivotal time in global climate evolution when a long-term cooling trend was initiated that ultimately resulted in the present glaciated Earth38. There is abundant geological evidence for large reorganizations of the carbon cycle in this interval. Our data indicate that a substantial reduction in pCO2 occurred at this time, but the decline was not continuous, and the
relationship with deep-sea temperature and ice growth was not straightforward.
During the late Palaeocene and early Eocene there was large-scale volcanic activity associated with North Atlantic rifting39. Enhanced volcanic outgassing of CO2 may have been supplemented by magmatism and regional metamorphism in parts of the Himalayan belt11 and North America5 . Another source of CO2 may have been
the oxidation of methane released from storage either in wetlands40 or from large sea¯oor gas hydrate reservoirs, as may have occurred in the short-lived Late Palaeocene thermal maximum event at 55 Myr ago41 and similar subsequent events in the early Eocene42. In contrast, there is much less geological evidence for high levels of
CO2 emission in the later middle and upper Eocene.
We note that the termination of North Atlantic volcanism at about 54±53 Myr ago40 corresponds approximately to the initial drop that we record in pCO2 (Fig. 2). However, despite growing evidence for a short phase of cooling in the earliest Eocene43, diverse geochemical and palaeontological indices suggest that the peak of global warmth was not reached until about 54±50 Myr ago35±37, a time for which we estimate relatively lower pCO2 values (700±900 p.p.m.). This was a period of maximum warmth and continental humidity combined with intermittently low sea level, causing extensive mangrove swamp and coal formation10. Deep weathering on the exposed continents may have accelerated the drawdown of CO2. In addition, it has been proposed that continuing collision between India and Asia would have caused a progressive reduction in arc volcanism through this interval, combined with a switch from organic carbon (Corg) exhumation on the Indian continental shelf to Corg burial in large foreland basins, bringing about a reduction11 in pCO2.
High-latitude cooling began at about the time of a sharp sea-level regression at 50 Myr ago that corresponds to a widespread hiatus in marine sediments35±37. This event is the ®rst of four major steps36 (as highlighted on Fig. 4) that punctuate the long-term history of Cenozoic cooling. McGowran10 has proposed that a reverse greenhouse effect operated at that time, caused by high levels of siliceous plankton productivity in the oceans and correspondingly enhanced rates of marine Corg burial. Our data do not support a precise covariation of pCO2 and temperature; indeed we record a pCO2 peak during the cooling phase at approximately 45.5 Myr ago. Nevertheless a longer-term link between Corg burial and pCO2 cannot be ruled out. We note that a major transition in Cenozoic d34Ssw occurred 55±45 Myr ago as would result from extended deposition of high levels of pyrite in anoxic marine sediments44, and this interval corresponds to a general decline in pCO2.
Overall, the pCO2 values for the early Cenozoic show considerably more variability than do the values for the late Cenozoic. This may be partly due to the magni®cation of errors that occurs at low pH values, but it may also re¯ect greater instability of the global carbon system during warm periods. The relative constancy of pCO2 values in the later Cenozoic is circumstantial evidence that pCO2 is stabilized by a homeostatic feedback mechanism involving the
greenhouse effect45. This is because radiative forcing is ampli®ed at low pCO2, hence small variations would be damped more ef®ciently by homeostasis.Our oldest samples from this set indicate that there was a
reduction in pCO2 in the lowermost Miocene. This was a time of high-latitude cooling that may have been associated with enhanced glaciation on Antarctica36. However, the most dramatic cooling phase in the Miocene was at 16±14 Myr ago, which has been interpreted as a major expansion of the East Antarctic Ice Sheet and increased formation of Antarctic Bottom Water46. This intensi- ®cation of bottom-water formation would have been accompanied by accelerated upwelling of nutrient-rich water, which we would expect to have boosted productivity and caused a temporary reduction in atmospheric pCO2 . A major increase in Corg burial in diatomites around the Paci®c rim has been recorded at this time9.Our d11Bcc data suggest a transient pCO2 decline from about 300 p.p.m. to 140 p.p.m.
However, there is little support in our data (or the similar pCO2 record produced recently47 on the basis of biomarker carbon isotopes) for a large and permanent drop in pCO2 during the middle Miocene, as has previously been suggested8,9,15. Another prominent step in the global cooling trend occurred in the late Pliocene (between about 4 and 2 Myr ago)36 which culminated in the onset of major glaciation in the Northern Hemisphere. Once again, this step was probably associated with enhanced deepwater formation (this time in the North Atlantic), which would have
led to a transient increase in surface-ocean productivity and a lowering of pCO2. We note a downturn in our pCO2 record at this time, from about 280 to 210 p.p.m. Denser sampling will obviously be needed to test all these possible linkages.
Change in the carbon dioxide concentration of the atmosphere is commonly regarded as a likely forcing mechanism on global climate over geological time because of its large and predictable effect on temperature48. Our d11Bcc proxy for pCO2 broadly con®rms the prediction of Arrhenius1 that early Cenozoic pCO2 levels were often several times modern values, and that a strong greenhouse effect probably contributed to global warmth at that time. These supergreenhouse’ conditions (pCO2 . 1,000 p.p.m.) also imply considerably lower surface-ocean pH, higher alkalinity and higher levels of SCO2. We ®nd that there was considerable ¯uctuation in these variables in the Palaeogene, but since the earliest Miocene the system seems to have been much more constant and more closely comparable to the present, despite continuing climate cooling. This suggests that other factors, such as complex feedbacks initiated by
tectonic alteration of the ocean basins, were also important in determining global climate change.
.
.
.