The “coral reef hypothesis” asserts that carbonate production on newly flooded shelves contributes importantly to the rise of atmospheric carbon dioxide during deglaciation. We seek to constrain the timing and strength of such carbon dioxide flux by re-assessing reef and platform distribution in the world ocean. The pattern of reef growth that emerges suggests that emission of CO2 resulting from carbonate production was important particularly during the late stages of deglaciation. The effect peaked during the early Holocene and presumably contributed to the warming in the climatic optimum.
 The reason for the rapid rise of atmospheric CO2 during deglaciation [Petit et al., 1999] is still unknown, despite considerable work on the subject [e.g., Berger, 1982a, 1982b; Broecker, 1982; Martin, 1990; Keir, 1995; Shackleton, 2000]. Clearly, there is no single mechanism at work but instead several factors are responsible [Archer et al., 2000; Sigman and Boyle, 2000]. The precise timing of global warming, melting of the high-latitude ice sheets, and rise in atmospheric pCO2 is of special interest in this context, but is difficult to reconstruct at the required resolution. The most important information comes from ice cores covering the last 400 kyr, especially from the drilling at the Vostok site [Barnola et al., 1987; Jouzel et al., 1993; Petit et al., 1999], and the crucial data are those allowing an assessment of relative phase of major climatic parameters within the same record. Of course, the sequence of sea level change, or of shallow water carbonate precipitation, which drives CO2 into the atmosphere, cannot be read in the ice record, so that independent assessment is necessary to identify the original trigger for deglaciation. The trigger mechanism for deglaciation is still unknown. Presumably, it is related to the buildup of instability in an isostatically sinking ice sheet, combined with Milankovitch forcing [Berger, 1999].
 Changes in the lysocline (the depth below sea level where strong calcite solution sets in) have been invoked to compare the phase of the ocean's carbon chemistry with that of the atmospheric CO2 [Archer et al., 2000; Sigman and Boyle, 2000]. The argument is that reduced carbonate accumulation on the narrow glacial shelves resulted in a deepening of the lysocline, by increasing the ocean's alkalinity. These changes resulted in the uptake of atmospheric CO2 by the ocean, on a timescale commensurate with deep mixing. A change in ∼25 ppm (atm.) in CO2 is linked, theoretically, to a change in the depth of the lysocline of ∼1 km [Sigman and Boyle, 2000]. The observed amplitude of depth change is somewhat less than 1 km. This weakens the case for changes in shelfal carbonate accumulation as the major factor controlling glacial-interglacial CO2 fluctuations, which span a range of 100 ppm. The coral-reef mechanism is but one factor among many [Archer et al., 2000; Sigman and Boyle, 2000], and apparently none is strongly dominant.
 We suggest here that addition of CO2 to the oceans and atmosphere by the coral reef mechanism nonetheless is a factor to be considered when discussing the rise of CO2 late during deglaciation, particularly when the sea level rise slowed. Such an effect was suggested by a model of the coral reef mechanism since the LGM [Opdyke and Walker, 1992; Walker and Opdyke, 1995]. To explore this effect, we summarize new information regarding the role of carbonate production in providing CO2 to the atmosphere. The timing and strength of these fluxes are as yet poorly known. We are especially interested in the timing of the change of CO2 relative to that of sea level, since sea level drives the shelfal carbonate production. Such production decreases alkalinity and increases pCO2 of the upper ocean, which directly affects the atmosphere. This was demonstrated in several Pacific reefs, where measured CO2 is up to 48 μatm higher in surface water than in the atmosphere [Gattuso et al., 1993; Kawahata et al., 1997, 2000]. In contrast, the immediate effect on the deep ocean and the lysocline, on a timescale of 1000 years or less, is minor due to the slow mixing of shallow and deep ocean waters. We use atmospheric pCO2 and temperature data from Petit et al.'s  measurements and deuterium (D) proxy, respectively. Conventional 14C ages were converted to calendar ages [cf. Bard et al., 1990]. “Production” and “accumulation” refer to shelfal benthic carbonates.
2. Carbonate Production and CO2 Rise
 Benthic biota produce most carbonate in tropical-subtropical (warm-water) reefs and on platforms [Milliman, 1993; Milliman and Droxler, 1996]. Those intervals during the deglaciation when reefs and platforms were situated in the highly productive 0–10 m depth for a prolonged time, are likely to show the strongest impact on CO2 emission. A global inventory of production is exceedingly difficult. We concentrate, therefore, on the patterns of distribution of the world's tropical-subtropical reefs and platforms (isolated banks and continent-attached platforms, including the deeper shelves). On the basis of their depth distribution and sea level records [Montaggioni, 2000; Vecsei, 2000, 2003a, 2003b], we recognize four different periods of carbonate production since the LGM, with their appropriate effect on CO2 emission (Figure 1, bottom).
 During early deglaciation (LGM to 14 kyr), melting of ice caps dominated the fast sea level rise from approximately −125 to −70 m [Fairbanks, 1989; Bard et al., 1990]. Reefs were rare and short-lived (e.g., Mayotte slope, Indian Ocean [Dullo et al., 1998]). The few flooded platforms were small, except temporarily the western Indian shelf [Vora et al., 1996; Purnachandra Rao et al., 2003]. Thus production during this early stage of deglaciation was low and the contribution to the initial strong CO2 rise was small. Of course, the flooding of one or more carbonate shelves could have resulted in minor pulses of CO2 emission, nevertheless.
 A few reefs are known from island and continental slopes from the early part of the middle deglaciation period (14–10 kyr), when sea level rose fast (e.g., Antilles [D'Anglejan and Mountjoy, 1973]; Florida [Lighty, 1977; Toscano and Lundberg, 1998]; Barbados [Macintyre et al., 1991]; Australian Great Barrier Reef [Harris and Davies, 1989]). Continuous strong reef growth, as on Tahiti [Bard et al., 1996; Cabioch et al., 1999] and on the Yucatan shelf [Macintyre et al., 1977], apparently was rare. The ages of such long-lived reefs were determined in sediment cores. The slow sea level rise during the Younger Dryas cold spell (13–11.6 kyr) presumably added to the reef terraces near −70 m [Dullo et al., 1998]. During the later part of this period, Caribbean reefs grew along today's submerged shelf edges [Macintyre, 1972]. The sea level reached the “depth window” containing most platforms in approximately −70 m just after 14 kyr before present [Bard et al., 1990; Vecsei, 2003a]. These platforms remained in the highly productive upper photic zone for maximally a few kyr before becoming submerged (“drowned”) below that depth. This production is documented by the thick sediments shed from the platform summits onto the slopes (e.g., Pedro Bank, Caribbean [Glaser and Droxler, 1991]; Sudanese Red Sea [Emmermann et al., 1999]). In addition, continent-attached platforms were subject to siltation and strong nutrient input for a few kyr after flooding [Hopley, 1984], which interfered with coral production, although not necessarily with production from calcifying algae [Hallock and Schlager, 1986]. In summary, from 14 to 10 kyr, production from reefs probably was strongly episodic, and increased with time. The resulting emission contributed to the sustained strong CO2 rise and warming after the Younger Dryas.
 However, shallow-water carbonate production apparently was strongest during late deglaciation (10–6 kyr). In the Indopacific, the site of major reef growth [Spalding et al., 2001], available dates suggest increasing production from ∼10 kyr. The last 2 kyr of Pacific sea level rise induced vigorous reef growth on shallow submerged reef-flats and fore-reefs, referred to by Montaggioni  as the reef growth-stabilization event [see also Hopley, 1982; Davies et al., 1985; Ryan et al., 2001]. On the basis of the very high production on modern shallow fore-reefs (around 10 kg m2 a−1 [Smith, 1983; Kinsey, 1985]), reef production must have peaked at 9–6 kyr (Figure 1, top), although global reef area was somewhat smaller than today [Kleypas, 1997]. Many Caribbean reefs also started growth at ∼10 kyr [Lighty et al., 1982; Blanchon and Shaw, 1995]. Flooding and subsequent drowning of the platforms continued, so that maximum reefal production may be set at ∼9–6 kyr. A pulse of CO2 emission must be postulated for that time, therefore. Apparently this pulse largely overlaps with the early Holocene climatic optimum.
 In many areas of the Indopacific, the sea level rise came to a halt or fell slightly during the late Holocene (6 kyr to present) [Davies and Montaggioni, 1985; Pirazzoli and Montaggioni, 1988]. As a result, many reefs aggraded to sea level and reef-flat growth declined. In the Caribbean and western Indian Ocean, the rise of sea level progressively slowed, allowing production to reach its maximum in these reefs sometime during the Holocene [Lighty et al., 1982; Davies and Montaggioni, 1985]. The largest and shallowest banks, flooded after ∼6 kyr, have aggraded following strong production (Bahamas [Droxler, 1984]; global [Vecsei, 2003a]). As elsewhere, overall production declined during the late Holocene.
3. Coral Reef Hypothesis in Light of the New Data
 On the basis of available data on shelfal carbonate production, the resulting CO2 emission contributed strongly to the rise of atmospheric CO2 during late deglaciation and the early part of postglacial time. The reefs produced at maximal rates in the early Holocene, their growth induced by the decelerating sea level rise during its last 2 kyr. After quasi-stabilization of the sea level, reefs and platforms aggraded and shut off the opportunities for rapid growth.
 A conservative estimate of global production in reefs, and of the subordinate production on isolated banks and continent-attached platforms, demonstrates its overall importance (Table 1). For the calculation of reefal production, Spalding et al.'s  modern reef area is extrapolated to past reefs, averaged for the 0–6, 6–8, and 8–14 kyr periods, using the percent area change modeled by Kleypas . The production of isolated banks is based on their modern area [Vecsei, 2000] and flooding history (A. Vecsei, Carbonate production on isolated banks since 20 k.a.: Climatic implications, submitted to Palaeogeography, Palaeoclimatology, Palaeoecology, 2003). The continent-attached platforms are compiled from literature data and own measurements on sea charts [Vecsei, 2003a]. We infer a doubling of the average reefal production for the main phase of Indopacific reef growth during the early Holocene. This doubling is conservative, because the widespread shallow submerged reefs of that time should have produced around 10 kg m2 yr−1, similar to modern shallow fore-reefs, by contrast to ∼4 kg m2 yr−1 for reefs at sea level [cf. Kinsey, 1985]. The main production phase was not exactly coeval but similarly lasted about 2 kyr in diverse parts of the Pacific. Our estimates of production on the isolated banks and continent-attached platforms are also very conservative. We assume the banks effectively produced during only 1000 years of the 6–8 and 8–12 kyr periods each, and that the platforms effectively produced only during 1000 years of the entire 0–14 kyr period.
Table 1. Estimated Global Warm-Water Benthic Carbonate Production and Resulting CO2 Emission to the Atmosphere
 Several potential sources of additional production are as yet unaccounted for, for example, the less well-known deeper fore-reefs [Kinsey, 1985; Vecsei, 2001], and the shelves in the temperate climate zone. Thus an overestimation of the global production is unlikely. The calculation of C emission takes into account the ocean's buffering factor, which results in 0.6 mole of CO2 being released to the atmosphere for every mole of CaCO3 precipitated [Ware et al., 1992], neglecting the slight increase in this factor during the course of the deglaciation [Frankignoulle et al., 1994]. Thus during the deglaciation and the interglacial since 14 kyr, the total accumulation of carbonate, ∼4200 × 1015 g (Gt), potentially resulted in an emission to the atmosphere of ∼225 Gt C as CO2. The portion of CO2 that stayed in the atmosphere depends on the rate of release and the concomitant rate of uptake by the ocean in high latitudes. Because release was pulsed (at the Younger Dryas stillstand, and particularly during the early Holocene), substantial contributions to the rise of CO2 was facilitated (several tens of ppm).
 The total CO2 emission greatly exceeds the amount of carbon necessary to provide for the actual ∼40 ppm rise from ∼13 kyr to the onset of industrialization, even if the error in our estimate should be large. This aggravates the problems of balancing the carbon cycle. While much of the emitted CO2 would have been taken up by the deep ocean (resulting in a rise of the lysocline), a substantial part may have been consumed by the buildup of the terrestrial biosphere (plants and soils). This process must have assimilated large amounts of carbon, on the order of the amount in the atmosphere itself since the LGM [Sigman and Boyle, 2000]. Much of the sequestration occurred already between the LGM and 11 kyr [Adams and Post, 1999], but buildup of soil, peat, and shelf-carbon continued well into the early Holocene, presumably supported by the CO2 released by carbonate production. In essence, we propose that the growth of reef carbonates subsidized biosphere demands on the CO2 in the atmosphere, during the early Holocene.
4. Contribution to the Holocene Climatic Optimum
 Strong mid-summer insolation in the climatically sensitive high northern latitudes, due to the precession of Earth's orbit, is considered the main cause of the Holocene climatic optimum [Berger, 1978]. This warm phase largely occurred during the same time as the highest CO2 levels in the gas from the Vostok cores, suggesting a contribution from CO2-induced warming. The origin of this additional CO2 is as yet unknown. On the basis of the reef and platform ages above, the shelfal production peak likely made a major contribution to the CO2 during the younger part of the maximum. This high CO2 helped maintain high temperatures during the Holocene in the face of orbitally induced cooling after the climatic optimum [Hay et al., 1997]. Other CO2 sources presumably would have been more important during the early part of the maximum.
 One might argue that the ages measured in cores, used to establish initiation of carbonate accumulation, are systematically too young to support a link to the early Holocene climatic optimum. However, the reason for this observation is probably that most drill holes do not cross the spots where the benthos initially colonized the reef substrates, from where the reefs expanded. This applies to all reefs that had an irregular topography before they reached sea level. Modeling suggests that in reefs with high colonization rates, the mode of the age offset is 1–2 kyr, and increases to 1–3 kyr in slowly colonized reefs [Blakeway, 2003]. Considering this age offset, the maximum of reef growth in the Indopacific is likely to be approximately coeval with the CO2 maximum in the Vostok cores.
5. Late Pleistocene Climate Forcing
 Comparison of the temperature, sea level, and CO2 records shows similar effects for the last four deglaciation periods. Remarkably, concordance emerges when the deuterium (D) stratigraphy for the last 400 kyr from Petit et al.'s  Vostok ice core data is stacked internally in such a fashion as to make the −4°C value (temperature anomaly) for the last four terminations congruent (Figure 2). The warming shown in the stacked D profile has a two-step nature, with a distinct terrace midway through the warming (the Younger Dryas cold spell during the last deglaciation). This terrace denotes a time of significant carbonate production, and also marks the level before which production was unimportant. Variability of temperature increased greatly when peak warming was achieved, i.e., around the time of maximal production. Also, there was great variation, between the different cases, regarding the rate at which cooling resumed after peak warmth.
 The two-step nature of warming seen in the stacked record corresponds well to the times before and after the familiar two-peaked meltwater input during the last deglaciation, which flank the Younger Dryas [Fairbanks, 1989; Bard et al., 1990]. This was interpreted as a change from melting “vulnerable” marine-based ice early during deglaciation to melting “stable” land-based ice which needed to be heated in situ [Berger and Jansen, 1995]. Such a pattern would seem to be applicable in general within the late Quaternary, rather than just for the last deglaciation. The D stratigraphy closely follows the sea level curve based on simple Milankovitch modeling [Berger and Jansen, 1995], which in turn lags Milankovitch forcing as expected. The close match supports the suggestion that the D stratigraphy at Vostok is in phase with the δ18O curve of planktonic foraminifers in the western equatorial Pacific [Berger et al., 1989].
 The CO2 is in phase with the D change, if Petit et al.'s  Vostok glaciological timeframe is accepted. Except for a hint of somewhat elevated CO2 values early during deglaciation, there is no indication that CO2 drives or follows the warming. It does both, presumably because it is part of a positive feedback loop. The D terrace at mid-deglaciation is also expressed in the CO2 data. Given this similarity in climate-related parameters during the last four deglaciations, we suggest that shallow-water carbonate production was important for increasing CO2 during each of these events, in the late stages of the process.
 Pulses of CO2, centered around 8.5 kyr, are a striking feature during the older three of the four deglaciation events (Figure 2). The origin of these pulses is as yet unexplained, but they are tied to the slowing of sea level rise before its quasi-stabilization. Very strong yet short periods of carbonate production around the times of sea level stabilization, similar to the Holocene “reef growth-stabilization event,” provide a plausible mechanism for their origin. In summary, the available data suggest that the coral reef hypothesis applies to the last four deglaciation events, but with emphasis shifted to the late stages, when the rising sea provides for the most vigorous reef growth and floods the greatest extent of pre-existing carbonate platforms.
 This work was largely carried out while A. V. was a visiting researcher at Scripps Institution of Oceanography. We thank B. O. Opdyke, S. V. Smith, and an anonymous reviewer. Work was supported by the German Bundesministerium für Bildung und Forschung through a stipend of the Deutsche Akademie der Naturforscher Leopoldina (BMBF-LPD 9701-9).