1. Introduction

Marine gas hydrates have attracted increasing attention in recent years due to their potential role in climate change (Dickens 2001), continental slope instability (Sultan et al. 2004), and as a future energy resource (Collett 2001). Gas hydrates are formed through the inclusion of small gas molecules into the interstices of an isometric crystal ice lattice under suitable pressure, temperature, and solubility conditions (Kvenvolden 1993). In the marine setting, high pressure and low temperatures within the seafloor allow gas hydrates to form, given sufficient gas and water supplies (Dickens 2001). Terrestrially, gas hydrates may exist in high-latitude continental regions due to cold subsurface temperatures (Collett and Dallimore 2001).

Due to the supply of methane from the anaerobic breakdown of organic material within seafloor sediments, methane structure I hydrates dominate the natural marine methane hydrate reservoir (Sloan 2003). Continental margins appear to contain the majority of the marine methane hydrate inventory, due to increased organic input from terrestrial debris, upwelling (Gornitz and Fung 1994), and gas and fluid flux expulsion from pelagic sediments during active subduction (Hyndman et al. 1993). Several studies have quantified the global extent of marine gas hydrates using various techniques [see Milkov (2004) for a comprehensive review]. Recent published values for the volume of the marine gas hydrate stability zone (GHSZ) and the mass of carbon stored as hydrate within it lie between 5 and 31.2 × 10⁶ km³ and 100–74 400 Gt C (Milkov 2004; Soloviev 2002; Klauda and Sandler 2005), depending on the geographical, sedimentary, and geochemical constraints used in particular models.

The volume of the marine GHSZ is dependent upon external conditions, namely hydrostatic pressure and seafloor temperature, that determine the depth to the base of the hydrate stability zone (BHSZ). Accordingly, several studies have investigated the effect of changes to sea level and seafloor temperature on the methane hydrate reservoir in order to gauge the sensitivity and response of the marine GHSZ to climate change. Xu et al. (2001) modeled potential changes to the flux of methane from the seafloor in response to temperature and pressure perturbations for both the Paleocene/Eocene thermal maximum (PETM) and the present day, using the hydrate model of Xu and Ruppel (1999). Katz et al. (2001) modeled the GHSZ response to PETM warming using model parameters based on Ocean Drilling Program (ODP) site 1051, while Dickens (2003) considered the carbon capacitor behavior of the marine gas hydrate reservoir under changing external conditions. Such studies have generally utilized idealized temperature changes over one ocean–sediment column, thereby ignoring the regional variability of the GHSZ response to realistic climatically driven seafloor temperature perturbations over global bathymetry.

To address this shortcoming, we utilize the intermediate complexity University of Victoria Earth System Climate Model (UVic ESCM) to determine both estimates of the volume of the GHSZ and potential spatial and temporal changes to this volume in response to a range of climate simulations. The outline of the rest of this paper is as follows: in section 2 we provide a brief description of the UVic ESCM as well as an overview of the marine GHSZ model. In section 3 we present results from two sets of climate model experiments. The first set studies the sensitivity of the marine GHSZ to equilibrium seafloor temperature changes, while the second set explores the pulsed response of the marine GHSZ to potential anthropogenically driven climate change. Sections 4 and 5 include a discussion of results and conclusions, respectively.

2. Model description

All experiments are conducted with version 2.7 of the UVic ESCM, detailed in Weaver et al. (2001). The UVic ESCM consists of an ocean general circulation model coupled to a vertically integrated two-dimensional energy–moisture balance atmospheric model, a dynamic–thermodynamic sea ice model, a land surface scheme, and a dynamic global vegetation model (Matthews et al. 2003; Meissner et al. 2003). The ocean is a fully nonlinear 3D GCM with a global resolution of 3.6° (zonal) by 1.8° (meridional), with 19 vertical levels that increase in thickness parabolically with depth from a surface-layer thickness of 50 m to a deep-ocean thickness of 518 m. Mixing in the ocean is accomplished using the vertical diffusivity distribution of Bryan and Lewis (1979) as well as the Gent and McWilliams (1990) parameterization for mixing due to mesoscale eddies. In addition, water columns are instantaneously mixed if the column becomes gravitationally unstable.

To determine the sensitivity of the GHSZ to climate change, modeled ocean bottom water temperatures are recorded and subsequently applied, grid cell by grid cell, to a 1D thermal diffusive model at 100-yr intervals to simulate temperature perturbations within the sediment column (Fig. 1).

All initial geotherms are linear and correspond to the initial model seafloor temperature field. Implicit in this approach is the assumption that the global marine geothermal gradient is in steady state at the beginning of the model run. This is likely a reasonable assumption for modern climate simulations due to the relatively invariate Holocene climate of the last 10 kyr (Petit et al. 1999). Initial geothermal gradients and sediment thermal diffusivities are prescribed for all grid points with values chosen to be most representative of the global average for continental shelves. The default geothermal gradient used in the following discussion is 40°C km⁻¹, although identical experiments have been performed with a global geothermal gradient of 60°C km⁻¹, which results in shallower BHSZs. As the value of seafloor thermal diffusivity (κ) in marine sediments is not well constrained, a range of plausible values (from 1 to 10 × 10⁻⁷ m² s⁻¹) are carried out for each model integration. This approach produced an array of model results that allowed us to bracket plausible GHSZ response time scales and magnitudes for each experiment.

The BHSZ corresponds to the intersection of the methane hydrate three-phase stability boundary and the geothermal gradient. The hydrate phase stability curve is calculated using the polynomial fit of Brown et al. (1996) to the seawater/methane hydrate stability measurements of Dickens and Quinby-Hunt (1994). This approach assumes that in situ pore water is similar to seawater in composition and marine hydrates are dominantly methane filled. A Newton–Raphson iterative procedure is used to determine the intersection of the geotherm and the phase curve for each time step at each model cell, enabling regional GHSZ variability to be modeled over time.

The GHSZ within the model is confined to within ∼120 km of coastlines to simulate an idealized continental margin, as this region experiences sufficient upwelling and continental runoff to support large-scale microbial methane generation and methane hydrate growth (Gornitz and Fung 1994). The abyssal seafloor is excluded from the modeled GHSZ, as seafloor there typically exhibits methane concentrations that are well below those needed to saturate the pore waters and form methane hydrate (Dickens 2001). To explore the effect of climate change on the actual hydrate inventory, a first-order estimate of the volume of hydrate present within the GHSZ is made using a sediment porosity profile that decreases exponentially from 60% to an e-folding depth of 1.5 km (Davis and Hyndman 1990). Hydrate is assumed to fill 5% of the available porosity at the BHSZ, with occupancy decreasing linearly to 1% of the available pore space at the seafloor. This idealized approach, which assumes that the prescribed hydrate inventory reacts instantaneously to changes in temperature, neglects important physical aspects of the hydrate system such as fluid flow, solubility effects, the presence of dissolved and gaseous methane (i.e., Xu and Ruppel 1999), and a sulfate reduction zone. However, it allows for preliminary estimates of the global average mass of carbon stored within the seafloor (and changes to this mass in response to future climate change) to be made.

Several additional simplifications are implicit in the model setup. Bathymetry is limited to the horizontal and vertical resolutions of the UVic ESCM and therefore provides a coarse representation of the seafloor. This is reflected in the continental margin bathymetry, which is used in the calculation of regional BHSZ depths. While a higher-resolution bathymetric dataset would markedly improve seafloor topography, we use the UVic ESCM bathymetry when calculating BHSZ depths in order to maintain internal consistency between model-derived seafloor depth and temperature [analagous to the approach of Gornitz and Fung (1994)]. The model could hypothetically be site-tuned to fit specific locations where seafloor temperature and depth and sediment chemical and thermal characteristics are well known, but in the interest of global consistency parameters used in the calculation of the hydrate stability zone are held to reasonable globally averaged values. Latent cooling during endothermic hydrate dissociation is ignored. Similar to Xu et al. (2001), we suggest that over long time scales this effect is negligible, and any temperature decrease will relax toward the seafloor-derived temperature signal. It is assumed that in situ gas overpressure resulting from hydrate dissociation also relaxes, over the time scales discussed here, to the hydrostatic pressure gradient. Advective heat transport is ignored as fluid flow is typically slow in marine sediments and heat transfer is conduction dominated (Xu et al. 2001). Sea level is held constant for all experiments, although it is a simple exercise to simulate higher sea levels by increasing the hydrostatic pressure. The effect of changing eustatic sea level on the marine GHSZ, particularly over time scales of less than 10⁴ yr, is likely insignificant compared to the potential temperature change at the seafloor (Buffett and Archer 2004).

To model the sensitivity of the GHSZ to seafloor warming, the UVic ESCM was first integrated from a preindustrial steady state to equilibrium for two sets of experiments driven by changes in atmospheric CO₂ over time (Fig. 2). The first set of experiments used equilibrium-CO₂ profiles that bracketed the range of stabilization scenarios (SRES) suggested by Houghton et al. (2001). These were carried out in order to assess the sensitivity of the GHSZ to permanent seafloor warming and estimate the hydrate-derived climate feedback factor f_hyd. In addition, while the long-term CO₂ concentrations were artificially maintained at constant elevated levels in these simulations, they afforded a clear study of the response of the GHSZ to the initial seafloor temperature increases and allowed absolute upper bounds to be placed on the magnitude of the GHSZ response to climate change. In contrast, the second set of experiments was forced by more physically plausible (albeit still highly idealized) atmospheric CO₂ profiles. In these integrations, the GHSZ evolution was forced by seafloor temperature changes resulting from simulated anthropogenic carbon emissions and subsequent millenial-scale oceanic drawdown. Maximum CO₂ concentrations used in this set of experiments also bracketed the levels used in the SRES scenarios, and the time scale of the atmospheric drawdown was based on coupled carbon cycle–ocean model simulations carried out by Ewen et al. (2004).

In all model integrations, atmospheric CO₂ increased from 1850 preindustrial levels (280 ppmv) following the exponential profile of Weaver et al. (2000), based on the two-parameter exponential fit (a and b) of the observed 1850–1990 increase in atmospheric CO₂ and other gases used by Mitchell et al. (1995):

View Expanded

where C_PI = 280 ppmv, a = 7.944 × 10⁻⁸, b = 3.0058, and t is the year. In the equilibrium-CO₂ experiments, CO₂ was capped at years 2000, 2050, and 2100 and then held constant. In the pulsed-CO₂ experiments, CO₂ was capped at year 2000, 2050, or 2100 levels and then decreased over 2000 yr to 20% of the original exponential increase. A final run increased CO₂ to year 2100 then decreased it over 500 yr; this run was carried out in order to explore the effect of varying rates of oceanic CO₂ drawdown on the GHSZ. All UVic ESCM model runs were integrated for 3000 yr or more to allow for ocean thermal equilibration.

After the global average seafloor temperature in all climate model runs reached equilibrium (typically after 3000–4000 yr from year 1850), bottom water temperatures were held constant at their final values and the total temperature record (i.e., UVic ESCM output temperatures from 1850 to the end of the climate model runs and subsequent constant temperature fields corresponding to the last temperature field output from the UVic ESCM) was applied to the GHSZ model using 100-yr time steps, for 40 kyr. The assumption that the seafloor temperature field for the GHSZ model integrations after the end of UVic ESCM output is equal to the last output from the UVic ESCM is justified because the UVic ESCM was always run until the ocean mean temperature reached equilibrium.

3. Results

The prewarming GHSZ exists along continental margins and varies as a function of bathymetry and model-derived seafloor temperature (Fig. 3).

Maximum BHSZ depths occur along deep, cold continental margins and extend to more than 600 m below the seafloor (mbsf) (Fig. 4). Several areas of the continental margin exhibit a total lack of GHSZ due to warm seafloor temperatures and/or shallow depths. Notable in this regard are the Gulf of Mexico, Weddell Sea, Timor Sea, and Hudson Bay. Methane hydrate is known to occur extensively in the Gulf of Mexico (Milkov and Sassen 2002), and the lack of modeled methane hydrate there reflects the coarse bathymetry used in the UVic ESCM, which limits depths within the Gulf of Mexico to no greater than 500 m. As explained in section 2, the application of modeled seafloor temperatures to a high-resolution topography is avoided in the interest of global model consistency, resulting in this regional model–observation conflict. The volume of the global GHSZ in the prewarming model state, using the approach and parameters described in section 2, is 6.7 × 10⁶ km³ and contains 6500 Gt C as methane hydrate; these values are comparable to other published estimates (Milkov 2004). While the purpose of our work was not to develop a new estimate of the global inventory of hydrate-bound methane, it is important that our approach lead to reasonable global values since it is the first-order response of these values to global climate change that we are interested in.

In all equilibrium-CO₂ experiments the globally averaged seafloor temperature increases in response to permanently heightened atmospheric CO₂ as thermal diffusion and water mass convection and advection transport heat downward, with the global average equilibrium seafloor temperature increasing by 0.8°C for Equil2000, 1.2°C for Equil2050, and 2.1°C for Equil2100 after 3000 yr. However, seafloor temperature increases exhibit large spatial variability, with the presence of significantly greater warming along the continental margins (Fig. 5).

The average temperature increases there by 1.0°C for Equil2000, 2.0°C for Equil2050, and 3.9°C for Equil2100 after 3000 yr, with high-latitude shelves exhibiting the largest regional seafloor temperature increase resulting from regionally accelerated warming due to sea ice–albedo feedback effects. Interestingly, for both Equil2050 and Equil2100, the postequilibrium seafloor temperature throughout the abyssal Atlantic and western Indian Oceans has cooled by up to 1°C due to the intrusion of Antarctic Bottom Water under a shallowed North Atlantic Deep Water circulation cell. However, this does not noticeably affect the GHSZ in these regions, which is restricted to shallow shelves above the influence of deep water masses.

The global GHSZ response to the seafloor temperature increase (Figs. 6 and 7) is delayed due to the time required for heat to be transported into the water column and then diffused through the sediments to the BHSZ. However, the global GHSZ begins to decrease relatively soon in all model runs forced by equilibrium-CO₂ profiles. For runs with κ held at 5 × 10⁻⁷ m² s⁻¹ (the median diffusivity value), over half of the total observed GHSZ response occurs within the first 5 kyr after initiation of the atmospheric CO₂ increase, with the initial significant GHSZ loss occurring 200 yr into the experiment (i.e., yr 2050).

Shoaling of the GHSZ in shallow-water mid- to high-latitude regions dominates this early reduction due to large thermal pulses that are able to rapidly reach shallow BHSZ depths. After 40 kyr of model integration, the GHSZ for all the equilibrium-CO₂ experiments in which κ = 5 × 10⁻⁷ m² s⁻¹ has largely adjusted to the postwarming equilibrium state, and the GHSZ volume in the first three experiments has decreased appreciably: Equil2000 by 7%, Equil2050 by 14%, and Equil2100 by 27% (Table 1). Totals of 413, 834, and 1688 Gt C, respectively, are mobilized within the sediments due to destabilization of hydrate (using the hydrate-estimating scheme detailed above). Identical experiments utilizing diffusivities of κ = 1 and 10 × 10⁻⁷ m² s⁻¹ respond similarly to equilibrium-CO₂ forcings but with longer (shorter) response time scales resulting from slower (faster) diffusion of heat within the sediments (Fig. 8).

The global average seafloor temperature in all pulsed-CO₂ experiments exhibits an initial rapid increase and subsequent gradual decrease to the prescribed CO₂ profiles, with maximum and final equilibrium seafloor temperatures lagging approximately 700 yr behind peak and equilibrium atmospheric CO₂ levels. Maximum temperatures reached are slightly less than those obtained during equilibrium-CO₂ experiments, due to the fact that CO₂ is relaxed to lower levels before the ocean has time to reach full thermal equilibration. Pulse2000 experiences a maximum global temperature increase of 0.7°C before decreasing to final equilibrium values, Pulse2050 warms by a maximum of 1.3°C, Pulse2100 warms by 2.0°C, and Pulse2100f warms by 1.3°C. As with the equilibrium-CO₂ experiments, warming is enhanced along continental margins: Pulse2000, Pulse2050, Pulse2100, and Pulse2100f warm there by up to 0.8°, 1.8°, 3.6°, and 2.4°C, respectively. As expected, the GHSZ in these integrations typically responds by initially decreasing in response to maximum margin seafloor warming before rebounding to levels determined by the final seafloor temperatures (Fig. 9).

However, the magnitude of the initial pulsed GHSZ volume decrease depends strongly on the thermal diffusive time scale,

View Expanded

at average BHSZ depths (237 m for preindustrial conditions) and the length of the transient seafloor thermal pulse. The prescribed value of the thermal diffusivity and the length of the thermal pulse (related to the length of the atmospheric CO₂ pulse) are therefore critical in determining the response of the GHSZ.

For low seafloor thermal diffusivity, τ is longer than the length of the thermal pulse, resulting in a smoother transition between pre- and postpulse GHSZ states (Fig. 10). Particularly, if κ < 5 × 10⁻⁷ m² s⁻¹, the seafloor thermal perturbation is not felt at typical BHSZ depths, and the pulsed response of the GHSZ is negligible.

However, as the value of thermal diffusivity increases, a given thermal perturbation is able to penetrate deeper into the seafloor, resulting in a greater pulsed response of the global GHSZ volume. In particular, the initial pulsed decrease of the GHSZ to pulsed-CO₂ forcing is significant for κ > 5 × 10⁻⁷ m² s⁻¹ (Fig. 10). This suggests that future predictions of the marine methane hydrate response to climate change are strongly dependent on accurate seafloor κ values.

Similarly, comparison of Pulse2100 and Pulse2100f integrations (Fig. 11) shows the effect of altering the rate of CO₂ decrease on the evolution of the GHSZ.

A gradual CO₂ decrease (Pulse2100) results in a longer thermal pulse to the seafloor and a greater response of the GHSZ, while a rapid CO₂ pulse (Pulse2100f) results in a negligible initial response from the GHSZ. As the rate of decrease of atmospheric CO₂ is governed largely by the ocean, future predictions of the rates of oceanic carbon uptake and storage must be well constrained in order to accurately predict the future behavior of the marine GHSZ.

Notable among the pulsed-CO₂ experiments is the fact Pulse2000 represents the lowest possible response of the GHSZ to potential future climate change. It simulates the immediate and complete cessation of anthropogenic CO₂ emissions in the year 2000, and still results in a 2% decrease in the global marine GHSZ volume after 40 kyr.

Important regions are those in which the BHSZ shoals to the seafloor, rendering the entire sediment column unstable with respect to methane hydrate. In these regions, methane that is mobilized during hydrate dissociation is not able to re-form higher up in the sediment column and is therefore most likely to enter the exogenic carbon cycle. Cells within the model that experience total GHSZ loss are therefore binned during each GHSZ model run, and the spatial and temporal distributions of these cells are examined to give an indication of areas and times that are potentially most prone to large-scale methane flux into the ocean or atmosphere.

Using κ = 5 × 10⁻⁷ m² s⁻¹, Equil2000 experiences some regions with total GHSZ loss, particularly in locations within the Okhotsk Sea and off the coast of Britain (Fig. 12a). These cells, totaling approximately 1% of the total continental margin area, occur in regions of shallow bathymetry that exhibit shallow prewarming BHSZs (maximum 88 mbsf). The first cells to experience total GHSZ loss appear after 400 yr of model integration (year 2250), and subsequent cells become completely unstable for the next 900 yr (until year 3150). Equil2050 exhibits a greater area under which complete GHSZ loss occurs with binned cells appearing over 4% of the total margin area (Fig. 12b), including coherent regions around Great Britain, New Zealand, and in the Japan Sea. GHSZs are totally destabilized from depths of up to 190 mbsf over a period beginning 400 yr into the model integration and lasting for 2300 yr. Equil2100 displays the greatest amount of total GHSZ loss, with 9% of the continental margin (Fig. 12c) completely destabilizing from depths of up to 239 mbsl (due to maximum seafloor temperature increases of up to 8°C). Regions of coherent total GHSZ loss that existed in Equil2050 increase in size and numerous additional regions appear along continental margins. The period over which cells experience total GHSZ loss in Equil2100 begins 400 yr into the model integration and lasts for 2900 yr.

The areal percentage of the prescribed continental margin that experiences 100% GHSZ loss under pulsed-CO₂ forcing (Table 2) is only slightly less than that experienced by equilibrium-CO₂ experiments, and the areal distribution of 100% GHSZ loss regions is also similar. With κ = 5 × 10⁻⁷ m² s⁻¹, Pulse2000, Pulse2050, Pulse2100, and Pulse2100f experience 1%, 3%, 8%, and 4% total GHSZ losses, respectively. The fact that the extent and timing of 100% GHSZ loss regions in both pulsed- and equilibrium-CO₂ integrations is similar indicates that the majority of cells that experience 100% GHSZ loss do so during or very soon after the exponential CO₂ increase. These cells are typically shallow and are therefore sensitive to rapid GHSZ loss both because the BHSZ in these cells is initially close to the seafloor and the seafloor above these regions also typically warms the most. In pulsed-CO₂ cases where κ = 5 × 10⁻⁷ m² s⁻¹ or 10 × 10⁻⁷ m² s⁻¹, any 100% GHSZ loss occurs during the initial pulsed decrease of the global GHSZ volume. In experiments with κ = 1 × 10⁻⁷ m² s⁻¹, the pulsed response of the GHSZ is negligible (the thermal skin depth D is less than typical BHSZ depths) and all 100% GHSZ loss occurs during the long-term adjustment to equilibrium seafloor temperatures. Pulse2100f experiences only about half of the areal 100% GHSZ loss as Pulse2100 (Table 2). The difference is due to the lower maximum temperatures reached at the seafloor (particularly along continental margins) because of a quicker atmospheric CO₂ drawdown. This finding indicates that the rate of CO₂ removal from the atmosphere is an important factor in determining the potential future input of carbon into the exogenic carbon cycle from 100% GHSZ loss regions.

To calculate the climate feedback factor f_hyd due to the release of hydrate-sourced carbon to the atmosphere, CH₄ from the GHSZ model cells experiencing total GHSZ loss during equilibrium-CO₂ runs (κ = 5 × 10⁻⁷ m² s⁻¹ and G = 0.04°C m⁻¹) is converted to CO₂ and added instantaneously to the atmosphere as soon as total GHSZ loss at each cell occurs. This assumes that the ocean does not oxidize any outgoing CH₄ nor does it draw down the additional CO₂, and results in a series of pulsed additions of CO₂ that accumulate in the atmosphere (Fig. 13).

If a cell does not experience complete GHSZ loss during the integration, no carbon is added from that cell to the atmosphere; this assumes that any methane mobilized in such situations re-forms as hydrate in the shallowed GHSZ. These runs are labeled Equil2000f, Equil2050f, and Equil2100f, and the equilibrium surface air temperature (SAT) difference between these and Equil2000, Equil2050, and Equil2100, respectively, give an indication of the hydrate-derived feedback effect. The equilibrium SAT difference between Equil2000f and Equil2000 is less than 0.1°C, while Equil2050f warms 0.4°C more than Equil2050 and Equil2100f warms by 0.6°C more than Equil2100; the respective climate feedback factors f_hyd [calculated according to the method of Hansen et al. (1984)] are 1.05, 1.11, and 1.09 (average 1.08).

4. Discussion

The analysis presented here demonstrates the sensitivity of a simple marine methane hydrate stability model to seafloor temperature increases, as simulated by an intermediate-complexity climate model. Even with atmospheric CO₂ concentrations capped at below-present-day levels and allowed to decrease (Pulse2000), the global GHSZ experiences a noticeable decrease in volume with several regions within the model domain destabilizing completely. Any further increase in CO₂ results in greater loss of GHSZ via increased seafloor warming. Of course, projections of future CO₂ concentrations in the atmosphere are prone to uncertainty; the range of CO₂ profiles used in the present study brackets the range of maximum CO₂ concentrations used in the Intergovernmental Panel on Climate Change SRES high-CO₂ stabilization experiments (Houghton et al. 2001). After carbon emitted to the atmosphere equilibrates with the ocean [on a time scale of 10³ years; Archer and Buffett (2005); Archer (2005)], carbon is removed from the exogenic system through carbonate and silicate weathering, a process that takes from 10³ to 10⁵ yr. Therefore, ignoring the chance that anthropogenic atmospheric carbon is rapidly drawn down in the short term (e.g., Pulse2100f), concentrations of atmospheric CO₂ are expected to remain elevated for millennia, resulting in sustained ocean temperature increases similar to those modeled here.

The times during which cells within the model domain experience rapid decreases in GHSZ volume (and total GHSZ loss) are indicative of periods when the response rate of the GHSZ to potential anthropogenic forcing is highest. Decreases in the global GHSZ volume begin to occur after 200 yr (i.e., year 2050) for all model integrations and cells begin to experience total GHSZ loss after approximately 450 yr, even when using atmospheric CO₂ held steady at below-present-day levels. Furthermore, it is apparent that although the global GHSZ does not reach a new equilibrium for 20–40 kyr, the bulk of the response is concentrated in the first 5 kyr and involves GHSZs situated on shallow continental margins that are most prone to total dissociation.

Pulsed-CO₂ experiments provide a useful demonstration of the interplay between the time scale of the seafloor temperature pulse and seafloor thermal diffusivity. For relatively quick thermal perturbations (∼10² yr) at the seafloor, or low thermal diffusivity (κ < 5 × 10⁻⁷ m² s⁻¹), the thermal skin depth D [Eq. (2)] is shallower than the global average BHSZ, and the global GHSZ does not experience an initial decrease in response to the seafloor temperature pulse. However, if the temperature perturbation persists for ∼10³ yr or more, as would be expected from a sustained atmospheric CO₂ increase, a greater percent of the GHSZ will disappear before rebounding in response to postpulse seafloor temperatures. The response time scales for given thermal diffusivities presented here represent a minimum, as the latent heat of dissociation is not taken into account by the GHSZ model.

Areas that are most sensitive to GHSZ destabilization within the model domain are shallow, high-latitude grid cells that correspond to regions of significant sea ice loss, lowered albedo, and increased absorption of incoming radiation at the sea surface. The combination of initially shallow BHSZ depths and rapid, large seafloor temperature increases in these regions makes them the first to experience rapid GHSZ shoaling. This is most noticeable in the Bering and Okhotsk Seas, which display the highest increases in seafloor temperature. We suggest that the Okhotsk Sea in particular is very sensitive to warming and hydrate dissociation as it contains several grid cells that experience rapid and total GHSZ loss, even with CO₂ held at and lowered below year 2000 levels.

Large-scale changes to global ocean circulation play an important role in regulating seafloor temperature changes along deeper continental margins where the seafloor is potentially in contact with global-scale, density-driven water masses. Most noticeably, a shift in the location of North Atlantic Deep Water (NADW) formation results in large regional temperature changes at the seafloor in the Greenland–Iceland–Norwegion Seas as a result of a northeastward migration of the modeled deepwater convection site during periods of elevated CO₂. The seafloor under the vertical component of the new, shallow convection cell experiences dramatic temperature increases and GHSZ loss for all model integrations, while the former region of vertical convection cools significantly. Seafloor temperature change also occurs due to variations in the rate of production of Antarctic Bottom Water (AABW). Equil2050 and Equil2100 in particular experience a sudden increase in AABW production immediately following a rapid drawback of sea ice in the Weddell Sea approximately 900 yr into the climate model integration; this cold, dense water mass invades under the shallowed NADW circulation cell and penetrates north into the Atlantic and west into the Indian Ocean. Any GHSZ there would therefore increase significantly, but due to the specified proximity of the GHSZ to continental margins in these experiments this deep water-mass transition has little effect on the modeled global hydrate inventory. This abyssal cooling effect is also briefly apparent in pulsed-CO₂ experiments, but disappears as AABW production gradually decreases in response to decreasing CO₂ levels.

The hydrate-filling scheme used in these experiments is meant only as a first-order estimate of the global average amount of carbon stored as hydrates. As described previously, a simple methane occupancy gradient neglects features of the hydrate reservoir controlled by fluid flow, solubility effects, the presence of dissolved and gaseous methane, and a sulfate reduction zone. Additionally, the fate of any carbon mobilized during hydrate dissociation may vary widely. While it is implied here that only methane from regions that experience 100% GHSZ loss will escape to the exogenic carbon cycle, it is possible that additional carbon may also be released during overpressure-induced landslides, solubility effects within the GHSZ (Sultan et al. 2004), or through the GHSZ via venting of warm fluids (Wood et al. 2002). Alternately, mobilized hydrate bubbles may remain trapped within the seafloor, negating any potential climate feedback. Such effects are dependent on local subsurface properties and are, therefore, difficult to isolate with the model used here.

5. Conclusions

We find that the volume of the global GHSZ, as modeled with a simple GHSZ model coupled to a climate model of intermediate complexity, decreases in response to potential anthropogenic atmospheric CO₂ emissions. The total GHSZ volume loss depends strongly on the CO₂ forcing applied to the climate model and the rate of loss depends on the sediment thermal diffusivity. The character of the GHSZ response to pulsed-CO₂ scenarios is sensitive to the length of the atmospheric CO₂ pulse and seafloor thermal diffusivity. For experiments using a higher value of seafloor diffusivity or sustained atmospheric CO₂ pulses on the order of 10³ yr, the GHSZ volume experiences a significant initial decrease before rebounding to levels corresponding to final seafloor temperatures. However, for low diffusivity values (κ < 5 × 10⁻⁷ m² s⁻¹) or rapid draw-down of atmospheric CO₂ on the order of 10² yr, the global GHSZ is negligibly affected.

Shallow shelves on continental margins are the first to experience GHSZ loss due to the combination of large seafloor temperature increases and shallow prewarming BHSZ depths. High-latitude Northern Hemisphere shelves exhibit the most dramatic GHSZ decrease as a result of sea ice loss, albedo decrease, and increased energy absorption at the sea surface. Initial coherent GHSZ volume loss occurs within 200 yr of model integration, and the majority of the global GHSZ adjustment to warmer seafloor temperatures occurs within the first 5 kyr of the atmospheric CO₂ increase. Embedded within the global GHSZ volume decrease are regions that exhibit complete GHSZ destabilization. With atmospheric CO₂ held at below-present-day levels, few such regions exist. However, with CO₂ elevated above present-day levels, several coherent regions of total GHSZ loss appear within 3 kyr of model integration, even if the atmospheric CO₂ is allowed to relax to lower levels. These regions display shallow prewarming GHSZs and/or large temperature increases, and are zones in which hydrate-sourced CH₄ is most likely to enter the exogenic carbon cycle. A simple equilibrium feedback analysis shows that the maximum f_hyd resulting from the addition of carbon from these regions to the atmosphere ranges from 1.05 to 1.11, depending on the equilibrium warming scenario.

The results presented here suggest that the marine hydrate reservoir may be affected significantly by future atmospheric greenhouse gas increases, with the GHSZ response varying in timing and intensity as a function of regional seafloor temperature change. Additionally, it is apparent that the global response time scale of the global hydrate reservoir (5–10 kyr) is of the same magnitude as seafloor temperature pulses produced by potential future anthropogenic emissions. This implies that if anthropogenic emissions of CO₂ continue well into this century and oceanic uptake of carbon operates on the order of 10³ yr, the marine methane hydrate reservoir has the potential to contribute to the exogenic carbon cycle for millennia to come.