Changes in carbon and sulfur cycling over geologic time may have caused considerable modification of atmospheric and oceanic composition and climate. Here we calculate pyrite sulfur (Spy) and organic carbon (Corg) burial rates from recently improved Cenozoic stable isotope records, and from these rates we infer global changes in Corg burial environments. Given predominantly normal shelf-delta organic carbon burial, the global Spy burial flux should be coupled to Corg burial. However, we find that the major early Cenozoic peak in Corg burial coincides with a minimum in Spy burial. Although the calculated magnitude of variations in global pyrite burial flux is sensitive to our assumptions about the concentration of sulfate in paleoseawater, a non-steady-state isotope mass balance model indicates very low Spy burial rates during the Paleocene and a dramatic increase starting near the Paleocene-Eocene boundary, dropping off to a fairly constant Cenozoic rate beginning in the middle Eocene. High Corg/Spy burial ratios (C/S mole ratio ≈15–30) coinciding with the Paleocene carbon isotope maximum most likely reflect enhanced accumulation of terrestrial organic carbon in Paleocene terrestrial swamps. We suggest that rapid burning of accumulated Paleocene terrestrial organic carbon could have significantly contributed to the short-lived negative carbon isotope excursion at the Paleocene-Eocene boundary in addition to or possibly even as an alternative to release of gas hydrates from the continental slopes. An early Eocene minimum in calculated Corg/Spy burial ratios (C/S mole ratio ≈2–4) suggests that the predominant locus of organic carbon burial shifted to euxinic environments in a warm early Eocene ocean.
 The most prominent feature of the Cenozoic carbon isotope record, the “Paleocene carbon isotope maximum,” began ∼2 m.y. after the Cretaceous-Tertiary boundary, peaked in the middle Paleocene (C25N–C26N), and waned through the late Paleocene and into the early Eocene [Shackleton and Hall, 1984; Miller et al., 1987]. This excursion has been interpreted as evidence for an increased rate of organic carbon (Corg) burial maintained for roughly 10 million years [e.g., Shackleton, 1987]. Superimposed on the waning stages of the carbon isotope maximum is a prominent short-term negative carbon isotope excursion at ∼55 Ma associated with the Paleocene-Eocene Thermal Maximum (PETM). One explanation for this feature is the rapid release of methane from gas hydrates in continental slope sediments [Dickens et al., 1995], resulting in a major short-term perturbation to the Earth's carbon cycle and climate.
 Because the marine and terrestrial carbon cycles are coupled through atmospheric CO2, the Paleocene carbon isotope maximum might reflect an increase in Corg burial in either marine or terrestrial environments. A number of studies [Shackleton et al., 1984; Oberhansli and Hsu, 1986; Corfield and Cartlidge, 1992; Corfield, 1994, 1998; Bralower et al., 1995] have shown that the Paleocene carbon isotope maximum is observed in both planktonic and benthic marine carbon isotope records. Furthermore, the magnitude of the oceanic vertical carbon isotope gradient may have increased as δ13C values become more positive. Corfield and Cartlidge  interpret this as an indication of increased marine productivity during the Paleocene. Thompson and Schmitz  argued, on the basis of marine sedimentary barium concentrations, for a large late Paleocene increase in marine organic carbon burial (6-fold) in oligotrophic regions of the oceans, and a much smaller increase (1.6-fold) in highly productive regions. An alternative explanation is that the Paleocene carbon isotope maximum was driven by an increase in terrestrial productivity and organic carbon burial [Oberhansli and Perch-Nielsen, 1990]. In this paper we argue that the Cenozoic sulfur isotope record [Paytan et al., 1998] can help us resolve the locus of organic carbon burial and the Paleocene-Eocene evolution of the global carbon cycle.
 Most previous studies of the relationship between the carbon and sulfur cycles [e.g., Veizer et al., 1980; Berner and Raiswell, 1983; Kump and Garrels, 1986; Carpenter and Lohmann, 1997; Petsch and Berner, 1998] have been concerned with long time scales (>10 m.y.), over which steady state assumptions are justified, and where carbon-sulfur-oxygen feedbacks are able to adjust to perturbations. Paytan and Arrigo  were the first to take advantage of the availability of relatively high-resolution (∼1 m.y.) Cenozoic carbon and sulfur isotope records to examine the evidence for coupling between the C, S, and O cycles on much shorter timescales. Here we focus on the relationship between early Cenozoic Corg and Spy burial fluxes and organic carbon burial environments as calculated from the respective seawater isotope curves for these elements.
2. Cenozoic Organic Carbon and Pyrite Burial
2.1. Modeling the Carbon and Sulfur Cycles
 The principal control on δ13C of marine dissolved inorganic carbon (DIC) is the fraction of carbon delivered to the exogenic cycle that is buried as organic carbon (Corg) versus carbonate. Similarly, the major control over the δ34S of marine dissolved sulfate is the fractional burial flux of marine sulfur as biogenic pyrite (Spy) versus evaporite sulfate (gypsum/anhydrite) or pore water sulfate. We use simple models of the global carbon and sulfur cycles, each consisting of a single reservoir, one input, and two outputs (Figure 1) to interpret trends in the isotope records. In contrast to many previous treatments, the two cycles are not coupled through either a constant C/S sedimentary burial ratio [Kump and Garrels, 1986] or a productivity-anoxia function [Petsch and Berner, 1998]. Instead, we use the simple box models to independently invert Cenozoic C and S isotopic records for possible histories of the Cenozoic C and S cycles, and in particular Corg and Spy burial rates. The calculated histories are subject to the assumptions of the simple box models and are not unique. However, driving the models with their respective isotope records allows us to examine the relationship between the two cycles.
 The single box for each element represents the mass and isotopic composition of carbon or sulfur in the exogenic cycle (global ocean, atmosphere, and terrestrial reservoirs). Following Kump and Arthur , we group both volcanic and terrestrial weathering fluxes into a single input term for each that we refer to as “weathering” (FWC and FWS). Sedimentary outputs from the reservoir occur either as the oxidized form (carbonate (Fcarb) or sulfate (Fgyp)), or as the reduced form (organic carbon (Forg) or pyrite (Fpy)). Changes in the amount of carbon or sulfur in the exogenic system (M0C, M0S) result from long-term imbalances in weathering inputs and sedimentary outputs [Garrels and Lerman, 1984; Kump and Garrels, 1986]. This relationship is expressed in the equation below written in terms of carbon (equation (1a)) and sulfur (equation (1b)):
A change in the input/output balance or form of output (carbonate/sulfate versus organic carbon/pyrite) will change the isotopic composition of the ocean/atmosphere reservoir (δOC or δOS):
Here δW is the isotopic composition of the weathering (riverine) input of C or S. Delta (Δ) describes the average (biological) isotopic fractionation resulting from formation of organic carbon or pyrite from dissolved carbonate or sulfate, and is equivalent to the isotopic difference between sedimentary organic carbon versus carbonate or pyrite versus sulfate of the same geologic age. Substituting equation (1) into equation (2), we derive the time-dependent equation for the isotopic composition of the exogenic carbonate (equation (3a)) or sulfate (equation (3b)) reservoir:
Equation (3) expresses the assumption that the ocean is well mixed but does not require that either the mass or isotopic composition of the ocean be at steady state. It states that an imbalance in isotopic fluxes will cause an adjustment in the isotopic composition of the ocean. Note that the rate of change is inversely proportional to the amount of C or S in the reservoir. The implication is that the isotopic compositions of large reservoirs (e.g., oceanic sulfate) respond more slowly to a change in fluxes than relatively small reservoirs (e.g., oceanic inorganic C).
Equation (3) can be solved for a steady state condition where the weathering input is balanced by the two outputs for each system, so there is no instantaneous change in either the mass or the isotopic composition of the reservoir. However, a steady state assumption is not applicable to timescales of change similar to or shorter than an element's residence time [e.g., Kump and Garrels, 1986; Richter and Turekian, 1993]. This becomes important when considering short timescale (Cenozoic) variations in sulfur isotopes, an element with a relatively long (>10 m.y.) residence time. The steady state approximation [e.g., Paytan et al., 1998] results in gross underestimates of transient maxima and minima in pyrite burial indicated by rapid changes in the sulfur isotope record during the early Paleogene. Rearranging equation (3), we solve for Forg and Fpy burial fluxes with no assumption of steady state:
Because fractionation of δ13C and δ34S are small during precipitation of sedimentary carbonate and sulfate, we can use sedimentary records of δ13Ccarbonate and δ34Sbarite as proxies for the changing isotopic composition of the exogenic carbon (δOC) and sulfur (δOS) reservoirs respectively. Isotope records from carbonate and sulfate sediments can be differentiated to provide the time rate of change of seawater carbon and sulfur isotopic compositions, i.e., the time-dependent term in equation (4). M0 appears explicitly in equation (4) and in general must be approximated, resulting in one additional source of uncertainty.
 We establish the boundary conditions for the model by assuming an average Cenozoic Corg burial flux of 5 × 1018 mol C/m.y. [Kump and Arthur, 1997] (Figure 1). Given a normal marine Corg/Spy burial ratio, (e.g., 7.5 molar ratio [Berner and Raiswell, 1983; Raiswell and Berner, 1985, 1986]), we calculate an average Cenozoic Spy burial flux of 0.67 × 1018 mol S/m.y. This estimate is close to the value suggested by Holser et al. , determined independently from the pyrite content of marine sediments. FW and δW for carbon and sulfur are held constant at values consistent with this hypothetical Cenozoic steady state (Figure 1). Model values for ΔC, photosynthetic carbon isotopic fractionation, are based on the Hayes et al.  ϵTOC record. The curve is generally flat at −30‰ through the early Cenozoic and then increases approximately linearly to a modern value of −22.5‰ beginning around 30 Ma. A similar record is not available for sulfur, so we assume a constant value of −35‰ for ΔS [Kump and Garrels, 1986]. We consider two end-member cases of seawater [SO42−] evolution. In the first we assume that the sulfate concentration of seawater (MOS) has been constant (28 mM) throughout the Cenozoic. FWS is held constant and Fpy is calculated from isotopic mass balance. Fgyp is not explicitly considered because it does not affect the isotopic mass balance, and does not appear in equation (4). However, implicit in this calculation is that Fgyp varies inversely to Fpy to maintain constant MOS.
 The concentration of sulfate in seawater (e.g., MOS) could change on million year timescales as a result of an imbalance between the weathering input and the pyrite and sulfate outputs. Horita et al.  interpreted fluid inclusion data from early Eocene-Pliocene evaporites as evidence for a steady rise in seawater sulfate concentration from ∼18 mM to present values of ∼28 mM over the past 40 m.y. Similar data are not available for Paleocene evaporites, but Lowenstein et al.  argue that the major element chemistry of seawater has been evolving from a low-sulfate, low Mg/Ca composition toward its present composition since the middle Cretaceous. Accordingly, in the second end-member model, we constrain the Cenozoic evolution of MOS with an exponential fit to the data of Horita et al.  (Figure 2). As in the constant sulfate case, FWS is held constant, and Fpy is calculated from isotopic mass balance (equation (4)). Again it is not necessary to explicitly consider Fgyp, but the rate of increase in MOS implies low (but nonzero) sulfate burial fluxes during the Cenozoic.
 The Cenozoic sulfur isotope record used in our model is based on Paytan et al. , but with a revised age model (Figure 3). The Paytan et al.  record was constructed from δ34S measurements of sedimentary barite from eight DSDP and ODP cores plus Holocene core-top sediments. An updated age model (E. Thomas, personal communication, 2002) (Table 1) was constructed by consulting original DSDP/ODP biostratigraphic zone assignments for individual samples. Where necessary, zone assignments were updated for consistency with modern zonal concepts based on first and last appearance of key microfossil taxa. Numerical ages for datum levels are based on the work of Berggren et al. .
Table 1. Cenozoic Barite δ34S Record With Revised Age Modela
Location and Depth, m
Δ34S, ‰ CDT
Data from Paytan et al.  with revised biostratigraphic age model courtesy of E. Thomas (personal communication., 2002).
Depths for equatorial Pacific surface given in centimeters.
 The revised age model differs significantly from the originally published version [Paytan et al., 1998] only for the Eocene part of the record, but the difference is important. The updated age model indicates that the Eocene increase in δ34S from ∼17 to 22‰ occurs much more rapidly that originally indicated. This rapid rise is observed in both the Atlantic and Pacific oceans, recorded at three sites, east central North Atlantic site 366, and western North Pacific sites 305 and 577. Biostratigraphic ages at these three sites indicate a minimum in seawater δ34S of 17.3‰ at 55.5 Ma. All three record the early Eocene initiation of the rise in δ34S to 19.2‰ by 50 Ma. Site 366 sediments show that the steep rise in δ34S from 18.1 to 22‰ occurs entirely within early Eocene nannoplankton zone NP12 (52.85–50.6 Ma [Berggren et al., 1995]).
 The Cenozoic carbon isotope record (Figure 4) is based on the bulk carbonate δ13C data from DSDP Leg 74 sediments (sites 525, 527, 528, 529) [Shackleton and Hall, 1984]. This record roughly mimics long and short-term trends in both benthic and planktonic δ13C records and therefore provides a reasonable representation of how mean δ13C evolves with time. In order to compare the carbon and sulfur isotope records, we revised the age model for Leg 74 record using the updated age assignments for magnetostratigraphic datum levels of Cande and Kent .
 Solving equation (4a) requires both the isotopic records (δ0) and their first derivatives . Differentiation of noisy data is problematic because insignificant wiggles in the primary data are amplified in the derivative curve. We addressed this problem by smoothing the carbon (Figure 4) and sulfur isotope records (Figure 3) with cubic smoothing splines, which are easily differentiated [deBoor, 1999]. Because the sulfur curve is sparsely sampled, we weighted some data points to force the smoothed curve through the rapidly changing early Eocene part of the record. We experimented with a range of spline stiffness parameters to find a value that preserved the first order features of the isotope records while minimizing perceived noise. Lacking sufficient data for a statistically rigorous fit, the smoothed curves reflect a visual best fit to the data, and faithfully record the important features of the complete records.
2.2. Cenozoic Pyrite and Organic Carbon Burial Histories
 Organic carbon burial (calculated using equation (4a)) increases ∼25% through the Paleocene to a peak corresponding to the “Paleocene carbon isotope maximum” (Figure 5a). Integrating Forg through this peak, and subtracting a constant organic carbon weathering flux of 5 × 1018 mol/m.y., we calculate net burial of 1.25 × 1018 moles of organic carbon during the Paleocene. Following this peak, organic carbon burial drops rapidly to an early Eocene minimum, and then rises steadily throughout the Cenozoic to a middle Miocene maximum.
 The pyrite burial rate (equation (4b)) exhibits transient maxima and minima indicated by rapid changes in the sulfur isotope record during the early Paleogene (Figure 5b). Because of the long residence time of sulfur (∼25 m.y.), the Cenozoic sulfur isotope record represents a damped record of variations in global pyrite sulfur burial. Furthermore, the isotope record lags the pyrite-burial forcing by as much as 5 m.y. This is consistent with the arguments of Richter and Turekian , who showed that for long-residence-time reservoirs, the first derivative term dominates in equation (4). The pyrite burial calculation indicates a Paleocene minimum in pyrite burial, followed by an early Eocene peak (Figure 5b). Following the major early Paleogene perturbation, pyrite burial remains relatively constant through the rest of the Cenozoic. Our modeling shows that even assuming a low early Cenozoic sulfate concentration (variable sulfate case), the calculated pyrite burial curve retains its first order features: a significant decrease in pyrite burial in the Paleocene followed by a prominent maximum in the early Eocene and relative stability throughout the rest of the Cenozoic (Figure 5b).
2.3. Sensitivity Analyses
 Models used to calculate the evolution of organic carbon burial from isotopic mass balance are subject to many uncertainties including assumptions about paleoweathering fluxes and changing isotopic fractionation [e.g., Raymo, 1997]. Models of pyrite burial are subject to many of the same uncertainties, with added complications particularly when looking at relatively short-term changes, where steady state cannot be assumed. Given these uncertainties, the pyrite burial curves shown in Figure 5b are not unique solutions.
 Cenozoic variations in ΔS, analogous to documented changes in ΔC [Popp et al., 1989; Hayes et al., 1999], would affect our interpretations of the sulfur isotope record. Habicht et al.  showed that ΔS can be sensitive to [SO42−] but only at extremely low concentrations (<1 mM), which are probably not relevant to the Cenozoic. Strauss  compiled available data for the Phanerozoic evolution of ΔS based on the difference between δ34S of gypsum sulfate and the average δ34S of pyrite sulfur at 10 to 100 m.y. resolution. Strauss inferred that ΔS has varied between −14 and −54‰ since the Precambrian. Holding all other parameters constant, we calculate the variability in ΔS required to attribute all Cenozoic changes in δ34S to ΔS alone [cf. Kump, 1989]. Figure 6 shows that the rapid Eocene rise in δ34S would require extreme values of ΔS (−70 to −170‰), values well outside of the observed range in Phanerozoic ΔS [Strauss, 1999]. Although improved knowledge of Cenozoic-scale variations in ΔS would improve interpretations of the sulfur isotopic record, it is unlikely that ΔS has exerted a dominant control on the Cenozoic sulfur curve.
 The isotopic value of the input sulfur flux (δWS) is determined by the relative weathering fluxes of sedimentary sulfides (pyrite) and sulfates (gypsum) on land, which are largely unknown [Bluth and Kump, 1994] and thus assumed to be invariant. Additionally, isotopic mass balance cannot distinguish between changes in burial terms (Fpy) and changes in weathering terms (FWS). Holding all other parameters constant, a decrease in the riverine sulfur flux (FWS) would drive the marine sulfur isotope mass balance (δoS) toward a more positive value without any change in the pyrite burial flux. Rearranging equation (2) to solve for FWS, we can test whether the sulfur isotope record can be explained by changes in the weathering sulfur flux under constant Fpy. Figure 7 shows that regardless of MOS evolution, the riverine flux would have to decrease dramatically near the Paleocene-Eocene boundary, and ultimately attain negative values of FWS, which is mathematically equivalent to net pyrite burial. This result shows that a change in the riverine flux cannot alone account for the sulfur isotope record.
 Furthermore, a change in global weathering fluxes would affect both the S and C cycles. For example, an increased weathering flux would simultaneously shift both S and C isotope records toward the values of the weathering flux (δWC = −4‰ and δWS = +7‰), with the sulfur response lagging the carbon response for the reasons given above. Such a forcing is inconsistent with the trends that we observe for the Paleogene. In any case, a major change in the riverine sulfur flux (presumably related to a decrease in the global weathering flux) should be evident from weathering proxy records. The marine Sr isotope record is an imperfect weathering proxy, but is nonetheless remarkably flat throughout this interval [Richter et al., 1992], suggesting no major long-term changes in weathering fluxes in the Paleocene-early Eocene. Peucker-Ehrenbrink et al.  suggest that the combination of unchanging 87Sr/86Sr with increasing 187Os/186Os from 65 to 40 Ma could reflect a gradual increase in the weathering flux of black shales without increased silicate weathering during the early Cenozoic. An increase in weathering fluxes at the PETM has been inferred from Os isotope records, but this was a short-lived event (∼104–105 years [Ravizza et al., 2001]). Regardless, a weathering-driven explanation of the early Cenozoic S curve would require a dramatic decrease in the terrestrial weathering flux near the Paleocene-Eocene boundary (Figure 7) which is inconsistent with existing radiogenic isotope proxy records. Our intention is not to argue for constant weathering fluxes over the entire Cenozoic, which would be unrealistic. However, our modeling indicates that the simplest explanation for the rapid changes in Paleogene δ34S is significant fluctuations in the global rate of pyrite sulfur burial. We believe changes in weathering fluxes had only a secondary effect on the C and S isotope curves. The constant weathering assumption is a simplification.
3.1. Corg and Spy Burial Environments
 In the modern ocean, carbon and sulfur burial rates are coupled through burial of Corg and Spy in marine environments. Pyrite forms in sediments by the reduction of seawater sulfate at the expense of sedimentary organic carbon, and is a strictly anaerobic process. Sedimentary pyrite formation is limited to shelf, deltaic, estuarine, and hemipelagic muds [Berner, 1982]. Hedges and Keil  estimated that 45% of organic matter burial of the oceans takes place in continental shelf environments with an additional 45% in deltaic sediments. Berner  noted that sediments accumulating in shelf and deltaic environments tend to have a remarkably constant Corg/Spy ratio (∼7.5 molar, 2.8 weight). Raiswell and Berner  showed through analysis of shales that this ratio has been maintained throughout the Phanerozoic. Berner and Raiswell  attributed this ratio to fixed proportions of reactive (sulfide producing) versus refractory carbon in marine sediments. Alternatively, the constant C/S ratio in normal marine sediments may be related to fixed proportions of iron oxyhydroxides and organic carbon sorbed to fine grained minerals [Berner, 1984; Hedges and Keil, 1995]. Regardless of mechanism, the global rate of pyrite burial is largely determined by the rate of shelf-deltaic organic carbon burial.
 The relationship between organic carbon and pyrite burial can change when the locus of carbon burial shifts away from normal shelf-deltaic environments [Berner and Raiswell, 1983]. Several environments inhibit the burial of pyrite. Among these are deep ocean sediments, where sulfate reduction may be limited due to the presence of oxygen or lack of reactive organic matter, shallow water calcareous sediments, where pyrite formation may be limited by the availability of dissolved iron, and terrestrial environments (soils, swamps, and coal basins), where sulfate is in limited supply [Berner, 1982]. In contrast, pyrite burial rates are high in euxinic environments, where pyrite may form in the water column [Raiswell and Berner, 1985]. Raiswell and Berner  calculated a maximum in global Corg/Spy burial ratio (up to 50 molar ratio) in the Permian/Carboniferous based on modeling C and S isotopic records. The peak at the Permian/Carboniferous boundary coincides with a peak in recoverable coal resources of the same age, supporting the hypothesis that a change in the dominant locus of organic carbon burial to terrigenous environments should be reflected in marine carbon and sulfur isotope records.
 Changes in eustatic sea level should exert an important control over organic carbon burial environments. For example, Schlunz et al.  showed that organic carbon burial in the Amazon fan system is controlled by glacioeustatic sea level changes. During glacial sea level low stands, terrestrial organic carbon bypasses the continental shelf and is channeled through the Amazon Canyon and buried in the deep sea fan. During interglacial high stands, organic carbon burial is dominated by autochthonous marine organic carbon in the shelf environment. Weissert et al.  noted a correlation between marine transgressions and three positive carbon isotope excursions in the Cretaceous. In their model, these positive C excursions are related to both an increase in organic carbon burial (as black shales) and a decrease in carbonate burial, related to drowning of carbonate platforms by rising sea level.
 High-resolution eustatic sea level records for the Paleogene are now becoming available as a result of the New Jersey Coastal Plain Drilling Project [Miller et al., 1997, 1998]. These records show a long term lowering of 100–150 m over the whole Cenozoic, generally attributed to decreasing global mid-ocean ridge volume (but see Rowley ). The New Jersey Margin record, although preliminary [Miller et al., 1997], suggests that significant fluctuations in sea level may have occurred in the late Paleocene to early Eocene (Figure 8), coincident with the important variations in carbon and sulfur isotope records. Sea level dropped slightly from ∼60 m above present sea level to ∼40 m throughout the Paleocene, then abruptly rose to +110 m during the late Paleocene and early Eocene. Subsequently, sea level dropped back to +60 m during the early Eocene. Interestingly, the relationship between sea level and carbon isotopes is opposite that previously seen for Cretaceous excursions. The Paleocene carbon isotope maximum (and inferred peak in Corg burial) corresponds to a sea level low stand, and the waning of the isotope maximum (and inferred pulse of pyrite burial) corresponds to an abrupt rise in sea level of ∼70 m.
 We calculate the Cenozoic evolution of the global Corg/Spy burial ratio (Figure 9) based on the Spy and Corg burial histories in Figure 5. The calculated Corg/Spy burial ratio was at a maximum (∼15–30) in the Paleocene, resulting from both low pyrite sulfur burial and high organic carbon burial at this time. During the early Eocene, the situation reversed, and global Corg/Spy burial ratios attained a minimum (<4) driven by both low Corg and very high Spy burial fluxes. Simultaneous changes in Corg burial, Spy burial, Corg/Spy ratio, and sea level may point to changes in dominant organic carbon burial environments during the Paleocene and Eocene. High Corg/Spy burial ratios may represent terrestrial or open ocean burial environments, while low Corg/Spy ratios most likely represent burial in euxinic environments.
3.2. Role of Marine Organic Carbon Burial
 Our model suggests that global organic carbon burial increased ∼20–30% during the Paleocene Carbon Isotope Maximum while pyrite burial decreased at the same time. Below we will argue that a terrestrial mechanism best explains this scenario. Because previous workers have explained the Paleocene carbon isotope maximum via changes in the marine carbon cycle [Corfield and Cartlidge, 1992; Thompson and Schmitz, 1997], we discuss our reasoning for discounting several marine-based scenarios.
 Positive carbon isotope excursions in the Cretaceous record tend to correlate with widespread deposition of black shales [Arthur et al., 1985]. In contrast, lack of evidence for abundant organic-carbon rich marine sediments was an early indication that the Paleocene carbon isotope maximum might reflect a pulse of terrestrial, rather than marine organic carbon burial [Oberhansli and Hsu, 1986; Oberhansli and Perch-Nielsen, 1990]. Meyers and Dickens  noted that the Paleocene was a time of low organic carbon accumulation throughout the Indian Ocean basin. Interestingly, the only notable exceptions are the occurrence of sediments with ∼0.5% organic carbon on Broken Ridge, which was inferred to be of terrigenous origin, and deposits of Paleocene brown coal on a now submerged portion of Ninetyeast Ridge [Meyers and Dickens, 1992]. Some examples of Paleocene organic carbon rich marine sediments do exist, among these sediments of the western North Pacific recovered by DSDP Leg 43. Black clays, which are abundant in mid-Cretaceous sediments of the western North Atlantic, reappear at some sites in mid-Paleocene sediments [Tucholke and Vogt, 1979]. These abyssal sediments contain up to 1.3% organic carbon and are interpreted to reflect poorly ventilated deep waters [Tucholke and Vogt, 1979]. The late Paleocene Waipawa Formation in New Zealand [Killops et al., 2000] is another example. These organic carbon rich rocks (locally up to 9 wt% organic carbon) are inferred to have formed in a dysaerobic shelf environment that developed in response to regional upwelling [Killops et al., 2000]. The Waipawa Formation contains abundant sulfur [Killops et al., 2000] and therefore does not represent the high organic carbon, low sulfide burial environment inferred from our model to dominate the Paleocene. Perhaps the most notable example of Paleocene black shales occurs in the southern Tethyan margin. These rocks contain up to 2.7% organic carbon but are entirely restricted to the PETM event [Speijer and Wagner, 2002]. These black shales are therefore not pertinent to the discussion of the broader Paleocene carbon cycle, but are nonetheless interesting as they may reflect a response of the ocean-atmosphere system to massive input of carbon at the PETM [Speijer and Wagner, 2002].
Corfield and Cartlidge  cited an increase in planktonic-benthic foraminiferal δ13C gradients as evidence for increased marine productivity during the Paleocene. Their arguments were made on the basis of changes in the δ13C gradient between Morozovella (planktic) and Subbotina (deeper water planktic) or Nuttallides (benthic). More recent work by D'Hondt et al.  has shown that Paleocene planktic foram species Morozovella and Acarinina were likely photosymbiotic, and argued that tests of these foraminifers may have been consistently 13C enriched relative to contemporaneous seawater. They cautioned strongly against paleoproductivity interpretations based on these species. Interestingly, the late Paleocene radiation of photosymbiont-bearing planktonic foraminifera (i.e., Morozovella and Acarinina [Norris, 1991]) might be viewed as supporting a transition to globally lower marine productivity rather than higher. Photosymbiosis as a survival strategy is generally viewed as advantageous in oligotrophic environments where nutrient availability is limited [Norris, 1996].
 Taking pyrite burial as a proxy for burial of marine organic carbon in shelf-delta environments [Kump, 1993] we interpret the sulfur isotope record as evidence for a ∼50% decrease in shelf carbon burial during the Paleocene. However, a net shift of organic carbon sedimentation to pyrite-lean pelagic settings might explain the combination of increased organic carbon burial and very high Corg/Spy burial ratios. Such a scenario would require a roughly 4-fold increase in global open-ocean carbon burial to offset a 50% decrease in shelf carbon burial while maintaining a 20–30% net increase in Corg burial. This scenario is in a sense consistent with the dramatic early Paleocene to late Paleocene increase (6-fold) in organic carbon burial in oligotrophic regions of the ocean proposed by Thompson and Schmitz . However, Thompson and Schmitz  argued for an overall global doubling in marine organic carbon burial, which is much greater than our estimate. Their conclusions are based on an observed increase in the paleoproductivity proxy “excess Ba” at several DSDP sites, which provides indirect, qualitative evidence for changes in organic carbon burial. A more troubling complication with this interpretation is that they are essentially arguing for a doubling in the global Ba sink that was maintained for at least 4 m.y. Given that the marine residence time for Ba is only 8 k.y., it is not clear what a sustained increase Baexcess might mean in terms of productivity [e.g., Dickens et al., 2003]. Increased Baexcess would require an increase in the Ba input (weathering) flux, unless the observed increase in Baexcess was balanced by Baexcess decreases elsewhere in the Paleocene ocean that have yet to be measured. Finally, this explanation (a quadrupling of the open-ocean organic carbon burial rate) is viable only if it doesn't provide sufficient organic matter to support significant pyrite production, in which case the C/S ratio would be diminished.
 Finally, tropical river-dominated ocean margins (e.g., Amazon [Aller et al., 1986, 1996; Aller and Blair, 1996] and Gulf of Papua [Aller et al., 2003]) are an additional carbon burial environment that tend to have Corg/Spy burial ratios higher than the canonical marine shelf sediment value [Berner, 1982]. Sedimentary C/S ratios of ∼20 (molar) are typical of the Amazon shelf [Aller and Blair, 1996]. Ratios of 10–20 are also seen in the Gulf of Papua, another tropical shelf dominated by riverine processes and physical reworking [Aller et al., 2003]. However, in this case, C/S ratios decrease offshore, and at depths of >45 m, are more typical of normal shelf sediments (e.g., 8 mole ratio). Aller et al.  suggest that these high Corg/Spy burial environments are most likely to be associated with tropical weathering during periods of high sea level stand. This mechanism is inconsistent with our observation that organic carbon burial waned during the late Paleocene as sea level rose. It is also not clear that increased organic carbon burial in these environments could produce global C/S ratios as high as 15–30 (Figure 9).
3.3. Role of Terrestrial Organic Carbon Burial
 Dominance of terrestrial organic carbon burial could explain why the Paleocene Carbon Isotope Maximum does not apparently correlate to widespread black shale deposition, and does coincide to a calculated maximum in global Corg/Spy burial ratio. We can test this hypothesis for consistency with the geologic record of worldwide coal distribution. Although compiling global statistics on coal resources is notoriously difficult, Duff  estimates that half of the world's known coal resources are of Mesozoic-Tertiary age. Important Tertiary coal deposits are found in Europe, east-central and southern Australia, southern and southeastern Asia, the Urals, Siberia, and Pakistan [Ross and Ross, 1984; Shah et al., 1993; Baqri 1997]. Ross and Ross  estimate that 60% of economic Tertiary coals are North American, dominantly located in the western plains and Cordillera. Late Paleocene coals are among the thickest in the entire geologic record [Shearer et al., 1995; Retallack et al., 1996]. The late Paleocene Fort Union Formation, located in the Montana-Wyoming Powder River Basin contains some of the most significant post-Paleozoic coals in North America [Ellis et al., 1999]. The Wyodak-Anderson member alone contains an estimated 550 Gt economically recoverable coal in beds as much as 60 m thick [Ellis et al., 1999]. Documented Paleocene coal fields in Russia [Crosdale et al., 2002] and Pakistan [Jaleel et al., 1999] tend to have somewhat thinner seams (up to 20 m) and smaller reserves (175 Gt, Thar coalfield, Pakistan) than the Wyodak example but are still significant examples of Paleocene lignite deposits.
 Our modeling suggests that the Paleocene Carbon Isotope Maximum can be accounted for by net burial of 1.25 × 1018 moles C during the mid to late Paleocene. For scale, we note that the modern terrestrial biosphere contains 0.17 × 1018 moles C in both living biomass and soil carbon [Schlesinger, 1997]. Beerling  modeled the evolution of the terrestrial carbon cycle and suggested that the late Paleocene terrestrial biomass and soils contained 0.24 × 1018 moles C, which is 30% larger than today's inventory. The thick Paleocene coal deposits suggest that a significant fraction of this terrestrial carbon stock was buried, rather than remineralized on short time scales. Using the Paleocene Fort Union coal average of 36% carbon [Ellis et al., 1999], 46,000 Gt of coal burial would be required to account for 1.25 × 1018 moles of net C burial. Since >1% of this amount (i.e., 550 Gt) is presently available as recoverable deposits in one member alone of the Paleocene Fort Union Formation, it seems reasonable that coal deposition could account for the Paleocene Carbon Isotope Maximum. Furthermore, Fort Union Paleocene coals are extremely sulfide-poor. Ellis et al.  estimate that the Wyodak-Anderson member averages 0.1% pyrite. This amounts to a molar Corg/Spy ratio of ∼1800. Even if this extreme ratio is not typical of Paleocene coals (e.g., Paleocene Hangu Formation coals of Pakistan have molar C/S ratios of ∼34 [Shah et al., 1993]), it clearly would contribute to globally averaged elevated Corg/Spy ratios during the Paleocene.
3.4. A Paleocene-Eocene “Global Conflagration”?
 The Paleocene-Eocene boundary is marked by a pronounced, well-documented short-term excursion in both carbon and oxygen isotopic records. One explanation for the negative carbon excursion accompanying the Paleocene-Eocene Thermal Maximum is rapid release of methane from gas hydrates on continental slopes [e.g., Dickens et al., 1995]. While an active methane subcycle is certainly an important part of the global carbon cycle [Dickens, 2001], and to a lesser extent, the global sulfur cycle [D'Hondt et al., 2002], here we propose an alternative explanation for the events surrounding the PETM that does not call upon methane at all.
 If the Paleocene was as we have suggested a time of widespread terrestrial organic carbon burial, it is worth considering whether the abrupt negative carbon isotope excursion at the PETM might have a terrestrial, rather than marine, origin. One possibility is the rapid oxidation of terrestrial organic carbon as late Paleocene coal-forming basins waned. Rampant wildfires, in a form of “global conflagration,” as an oxidative weathering mechanism could rapidly return a large amount of isotopically light carbon to the Paleocene-Eocene atmosphere. The role of wildfire in the global carbon cycle has been underappreciated. For example, droughts resulting from the 1997 El Niño event caused dramatic burning of Indonesian peatlands. The magnitude of carbon release from these fires was estimated between 0.7 and 2.1 × 1014 moles C, comparable to annual global carbon uptake by the terrestrial biosphere [Page et al., 2002]. Page et al.  suggested that burning of the top ∼0.5 m of peat covering a small part of Earth's surface (20 Mha) may have been largely responsible for the largest annual increase in atmospheric CO2 in almost 50 years of instrumental records.
Dickens  argued that accounting for the −2.5‰ PETM carbon excursion by isotopically normal (e.g., −22‰) organic carbon would require the release of 0.7 × 1018 moles C in 10,000 years, approximately half of our calculated Paleocene growth of the sedimentary organic carbon reservoir. This amounts to an average carbon release by burning of 0.7 × 1014 moles C/y, which is within the range estimated for Indonesia in 1997, and perhaps not unreasonable given the widespread occurrence of unusually thick peatlands inferred from the Paleocene coal record. Interestingly, Crosdale et al.  noted that some Paleocene Russian coals are unusually rich in inertinite, and inferred that fire played an important role in Paleocene peatlands of the Zeya-Bureya Basin. Unusually high concentrations of macroscopic charcoal have been identified in lignite beds at the Paleocene-Eocene boundary in southern England [Scott, 2000; Collinson, 2001; Collinson et al., 2003] suggesting that wildfire could have contributed to the observed abrupt negative carbon isotope excursion.
 What might cause such sustained wildfires? One possibility is that Paleocene net growth of the sedimentary organic carbon reservoir would have increased atmospheric O2, thus increasing the susceptibility of peatlands to burning [Watson et al., 1978]. Net burial of 1.25 × 1018 moles C would result in an addition of 1.25 × 1018 moles O2 to Earth's atmosphere. However, because the atmospheric O2 reservoir is large (38 × 1018 moles at present), it isn't clear that this increase would be significant enough to increase the susceptibility of peatlands to fire. Furthermore, our model suggests that Paleocene sulfide burial was net negative, i.e., the weathering flux of sedimentary pyrite was larger than the pyrite burial flux. Thus the Paleocene sulfur cycle was likely a net sink of O2 and the net addition of O2 to the Paleocene atmosphere was less than 1.25 × 1018 moles.
 A more likely explanation may be related to late Paleocene climate change. Page et al.  suggested that it was the unusually long El Niño dry season that caused the 1997 Indonesian fires to spread out of control. A shift toward a drier climate during the late Paleocene could plausibly trigger widespread burning of abundant Paleocene peatlands. Clay mineral assemblages of Tethyan sediments indicate a progressive change from high rainfall in the early Paleocene to a more arid climate in the late Paleocene and early Eocene [Bolle and Adatte, 2001]. The drying trend, inferred from decreasing kaolinite abundance in six Tethyan sedimentary sections, is briefly interrupted in most sequences by a kaolinite-rich layer at the PETM which may reflect a pulse of increased chemical weathering in response to elevated CO2, warmth, and humidity [Gibson et al., 2000; Bolle and Adatte, 2001]. Fossil leaf morphology may also be used as a proxy for Paleocene-Eocene paleoprecipitation [Wolfe, 1993, 1994; Wilf et al., 1998; Wilf, 2000], but interpretations are controversial [Wolfe et al., 1999], and at present there are not enough data to make global generalizations. The evidence from Wyoming is equivocal with respect to our hypothesis: humid conditions apparently prevailed during the late Paleocene, but the long-term trend from the late Paleocene into the early Eocene seems to be warming and drying [Wilf, 2000]. We suggest that abundant, thick peatlands, a trend toward increased aridity, and a major change in atmospheric circulation [Rea et al., 1990] may have triggered a period of increased wildfire that provides an additional mechanism to explain the PETM event.
3.5. Early Eocene Pyrite Burial
 Our calculated pyrite burial flux increases rapidly across the Paleocene-Eocene boundary to a peak in the early to middle Eocene. We interpret the early Eocene maximum in pyrite sulfur burial to be a consequence of several factors. High-latitude (or in general, bottom water source area) warming would result in a generally dysoxic ocean (as in the late Permian [e.g., Hotinski et al., 2000]). This interpretation is supported by the analysis of Kaiho  who used the species distribution of benthic foraminifera to calculate an oxygen index, reflecting the dissolved oxygen content of the deep ocean for the whole Cenozoic. Kaiho's  oxygen index drops dramatically from relatively high values (oxygenated) during the Paleocene to a Cenozoic low at the Paleocene-Eocene boundary, and remains low until the middle Eocene.
 Early Eocene sea level rise increased the global area of flooded continental shelves, where pyrite burial is most important. Early Eocene Corg burial rates were low, while Spy burial rates were high, suggesting the predominance of carbon burial in euxinic environments at this time. Geologic evidence for such environments exists in rocks such as the London Clay, which is an extremely pyrite-rich transgressive shale that was deposited in the early Eocene North Sea Basin [King, 1981; Newell, 2001].
 Modeling the coevolution of the exogenic carbon and sulfur cycles can significantly improve our understanding of the evolution of the global carbon cycle. We interpret stable C and S isotope records as evidence for high Paleocene organic carbon burial rates accompanied by remarkably low pyrite sulfur burial rates. Although we cannot absolutely rule out an increase in marine organic carbon burial, we interpret this as evidence that Paleocene organic carbon burial was dominated by terrestrial environments, where sulfate was in limited supply. Coals of Paleocene age appear to be sufficiently voluminous to account for the net burial of 1.25 × 1018 moles C inferred from isotopic mass balance models. Furthermore, these coals have low sulfur content, consistent with high Corg/Spy burial ratios.
 The δ34S record contains a rapid shift in the early Eocene that we interpret as a pulse of pyrite burial in globally widespread euxinic environments. The calculated magnitude of this pulse is dependent on the mass of the marine sulfate reservoir during the Paleocene-Eocene, which may have been smaller than at present. The further development of proxy records of seawater [SO42−] will be an important complement to δ34S records in reconstructions of the evolution of the exogenic sulfur cycle.
 The inference of a Paleocene carbon cycle dominated by terrestrial organic carbon burial raises the possibility that the changes in the terrestrial C reservoir also contributed to the short-term negative carbon isotope excursion at the Paleocene-Eocene Thermal Maximum. As an alternative or in addition to gas hydrates, we propose that a “global conflagration,” sustained burning of accumulated Paleocene terrestrial organic carbon in response to increased aridity may have contributed to the PETM negative carbon isotope excursion. This hypothesis can be tested by further study of the occurrence of fossil charcoal in Paleocene-Eocene boundary terrestrial sediments.
 ACK, LRK, and MAA acknowledge support from the NASA Astrobiology Institute. LRK also acknowledges the support of the NSF Biocomplexity in the Environment Program. We thank Ellen Thomas for providing a revised age model for the Cenozoic sulfur isotope record. This paper was much improved thanks to thorough reviews by Jerry Dickens and an anonymous referee.