Gas hydrates are the most substantial store of mobile organic carbon on the Earth1, estimated at ~1,800 GtC (ref. 2). Climatic warming causes changes in the pressure and temperature in sediment on continental margins, driving methane hydrate dissociation and the release of methane into the ocean3. The landward limit of the marine gas hydrate stability zone (HSZ) has been a focus of investigation, as hydrate occurs immediately below the seabed and is therefore sensitive to the warming of bottom water3,4,5,6. However, the volume of hydrate that is exposed during a warming event is relatively small7. Also, because the landward limit is located at water depths of 450–700 m, most of the gas would not reach the atmosphere8. Tens to hundreds of kilometres oceanward of this exposed area, greater quantities of methane could dissociate beneath the continental slope and rise7,9. In this vast unexposed region, muddy and hydrate-clogged sediment layers above the methane generally prevent it from venting vertically to the ocean10, although there are exceptional geological circumstances in which methane can migrate through the HSZ11,12,13. Permeable beds within the clinoforms that form the shelf, slope and base of the slope of continental margins could provide a route for methane to migrate up-dip14, below the HSZ and bypassing it, to beyond its landward limit where water depths are lower at the shelf break and shelf (~100–300 m). This could be important given the amount of methane that could be released and because thousands of vents beyond the landward limit have been observed elsewhere15. Also, venting in shallower water increases the chance of flux to the atmosphere8,16.

Seismic observations

The 3D seismic data we use cover an area of 3,520 km2 on the Mauritanian passive margin17 from the shelf break to the base of the slope (330- and 2,100-m water depths, respectively)—a distance of ~64 km (Figs. 13). The data image hydrates and methane hosted in a set of margin-scale clinoforms, with slopes of ~2.0°. These comprise deepwater mudstones with intercalations of sands and silts18. Hydrocarbon boreholes show that the hydrate is within Quaternary-age sediment and that sedimentation rates have averaged at 91 m Myr−1 to 116 m Myr−1 (ref. 18) since the Miocene (Fig. 2). Modern and buried coral mounds19 are located at the shelf break (Fig. 2).

Fig. 1: Map of the seabed offshore Mauritania from the base of the slope to the shelf break and imaging of the pockmarks, showing the locations of all other figures.
figure 1

a, Map of the seabed. Inset: the northwestern African margin. b, R.m.s. seismic amplitude map (Methods) of a reflection at the base of the field of large pockmarks in the area indicated in a.

Fig. 2: Representative seismic dip-line from the base of the slope to the shelf.
figure 2

The line shows the modern BSR (BHSZM), the relict BSR (BHSZR), and reflections 1, 2 and 3 (R1, 2 and 3). TWT ms, two-way travel time in milliseconds. Black arrows mark the stack of high-amplitude reflections. The two boreholes are projected locations. Inset 1: zoom-in of pockmarks 1 and 2. Inset 2: schematic of the configuration of the BHSZR, BHSZM and reflections 1, 2 and 3.

Fig. 3: Representative seismic strike-line showing the modern and relict BSRs and the modelled magnitude of the shallowing of the BHSZ due to warming since the LGM.
figure 3

a, Strike-line intersecting Fig. 2. Yellow and green triangles, respectively, indicate the points of intersection of reflections 2 and 3 with the BHSZM. Yellow and green circles, respectively, indicate the points of intersection of reflections 2 and 3 with the BHSZR. Inset 1: zoom-in of the BHSZR and BHSZM showing the aligned high-amplitude terminations that are used to map the features. Inset 2: schematic of the configuration of the BHSZR, BHSZM and reflections 1, 2 and 3. HARTs, high-amplitude reflection terminations. b, Schematic of the funnel showing the orientations of the dip and strike lines (Fig. 2 and a, respectively) with the stratigraphy above the BHSZM not shown. c, Schematic hydrate phase diagrams for temperature–depth profiles for modern conditions (full yellow line) and those during the LGM (orange dashed line) for the location at the intersection of Fig. 2 and a in a mid-slope position. White boxes show the modelled magnitudes of the changes in the depth of the BHSZ, the temperature at the seabed and sea level.

The modern base of the hydrate stability zone (BHSZM) is marked by a bottom simulating reflection (BSR; Fig. 2). It dips at 2° on the slope and at 5° as it shoals to the seabed at its landward limit at ~636 m below sea level17. It crosscuts other reflections17,20 and has the opposite polarity to the seabed17,20 (Fig. 2). It is evidenced as a series of aligned high-amplitude reflection terminations (Fig. 3a, inset 1). Below the HSZ, the hydrate is not stable and there is a free-gas zone (FGZ). Deeper, relict BSRs are commonly parallel to the modern BSR and can be identified using the same criteria21. We selected three reflections (R1, 2 and 3; Figs. 2 and 3) that are generally parallel to each other, have a consistent vertical separation (50–150 m) and crosscut the BHSZM and the succession immediately above and below it (Figs. 2 and 3). These are continuous for ~60 km from ~2,600 m to ~390 m below sea level. A cluster of faults intersects the succession at the base of the slope (Fig. 2). Strike and dip-parallel sections show that the succession has a synform geometry (Fig. 3a, inset 2 and Fig. 3b), the axis of which dips oceanward. The BHSZM has the opposite geometry. It is concave downward over most of the slope. Therefore, the configuration of the strata and BHSZM in a strike-parallel, north–south direction is that of a lens (Fig. 3a, inset 2 and Fig. 3b).

Reflections 2 and 3 include regions of high seismic amplitude that terminate abruptly (marked by yellow and green dots in Figs. 3 and 4a–c). These terminations align, to consistently define a relict BSR21 (BHSZR) that is parallel to the BHSZM, occurring ~50 m beneath it (Fig. 3a). We modelled the position of the BHSZM and BHSZ during the Last Glacial Maximum (LGM)22 (Methods) over the 3D seismic survey and this predicts that they are vertically separated by 43–54 m (Fig. 3c). The mapped BHSZR is located close to the modelled position of the BHSZ during the LGM (Extended Data Fig. 1a,b). The zone between the BHSZR and BHSZM is the hydrate dissociation zone (HDZ). Here, hydrate is no longer stable and methane should have been liberated. The root-mean-square (RMS) amplitude maps (Methods) for each stratal reflection (Fig. 4a–c) show that the lines of intersection for the BHSZR and BHSZM with reflection 1 are curved in an eastward (up-dip) and westward (down-dip) direction (Fig. 4a). The intersections of the BHSZM and reflections 2 and 3 have the planform of a funnel (herein termed ‘the funnel’), with the line of intersection (Methods)23 between the reflections and BHSZR remaining curved in the westward (down-dip) direction. There are also teardrop-shaped trails of higher amplitude (reflection 3) within the funnel (Fig. 4c, inset), as well as dip-parallel bands of higher seismic amplitude (marked Y in Fig. 4b,c). The intersections of reflections 1, 2 and 3 with the BHSZR and BHSZM can be marked by a drop or increase in seismic amplitude (Figs. 3a and 4).

Fig. 4: RMS amplitude maps showing the funnel and pockmarks.
figure 4

ac, RMS seismic amplitude maps (Methods) for reflections 1 (a), 2 (b) and 3 (c), respectively, showing the intersections of the BHSZR and BHSZM, the HDZ, funnel and pockmarks. Inset 1: teardrop-shaped trails of higher seismic amplitude indicate the movement of methane below the BHSZM (ref. 28) within the funnel, supporting the conclusion that long-distance migration of methane has occurred. Yellow and green triangles, respectively, indicate the points of intersection between reflections 2 and 3 and the BHSZM. Yellow and green circles, respectively, indicate the points of intersection between reflections 2 and 3 and the BHSZR. Y, dip-parallel bands of higher seismic amplitude.

Twenty-three large pockmarks occur ~10–15 km landward of the present landward limit of the HSZ (Figs. 1b and 2, inset 1). Their circular planform is very different from the carbonate mounds that form ridges in deeper water19 (Figs. 1b and 2). They are ~50 m above reflection 1, 600–900 m wide and 20–50 m deep. Some are buried by up to 80 m of sediment (Fig. 2, inset 1), and others have an expression at the present-day seabed (Extended Data Fig. 2). They occur at a water depth of ~330 m above a series of high-amplitude black-red reflections (marked by black arrows in Fig. 2). There is a spatial coincidence between their northern limit and that of the funnel (Fig. 4b,c). The southern limit of the funnel and the pockmarks is uncertain, as the 3D seismic coverage ends. Like other examples of pockmarks24,25, there is no evidence beneath them for vertical pathways (for example, pipes) connecting to reflections 1, 2 and 3. They are much larger than the pockmarks at the landward limit of the marine hydrate17.

Gas migration in the HDZ

The abrupt termination, phase shift and very high amplitude of the black-red reflection terminations below the BHSZM and BHSZR (Figs. 2 and 3) are the result of gas beneath hydrate-clogged sediment that remained after hydrate had dissociated21, in some cases revealing the migration pathways for the gas (Fig. 4b,c). Above the BHSZM (Fig. 3, inset 1), high-amplitude ‘hard’ reflections may indicate strata with higher concentrations of hydrate26. The HDZ can be identified in cross-sections and map views (Figs. 24) and mapped over an area of over 300 km2. As the three reflections are equally spaced vertically and parallel to one another, they show the intersections of the BHSZR and BHSZM and are also a proxy for the three developmental stages as upward resetting of the BHSZ occurred and show how the planform of the HDZ changed (Fig. 4). The amplitude map for the shallowest reflection 1 (Fig. 4a) shows that, due to the configuration of both the BHSZR and BHSZM and the reflections, strata in this location were initially probably not connected within the HDZ, preventing gas from migrating up-dip through permeable beds. In contrast, during the next development stages (Fig. 4b,c) there were contiguous, permeable strata within the HDZ27, creating a pathway for up-dip gas migration. We interpret the teardrop-shaped amplitude anomalies for reflection 3 within the funnel to be caused by the up-dip migration of methane28, and the dip-parallel bands of high amplitudes (marked Y in Fig. 4b,c) as permeable sediments that are still gas-charged—and further evidence for migration pathways17,20,28. There is no evidence that the deepwater carbonates are the result of sustained seep activity. The discovery of the funnel and evidence that, within it, the HDZ is connected from its base of slope to the shelf break and to the 23 large pockmarks (Extended Data Fig. 3) are the substantive new observations.

The modelled position of the BHSZ during the LGM is very close to the BHSZR, so the BHSZR is likely to be a feature that formed during or since the LGM. As there are no faults within the funnel, gas migration would have only been prevented by stratigraphic traps that are beyond seismic resolution20. South of the structural high, bands of high amplitude also occur, showing there could be other migration pathways to the south of the funnel (Extended Data Fig. 3). The BHSZR cannot be mapped in the eastern part of the survey, probably because the strata intersecting it dip upwards away from the hydrate, so there was no trapping configuration allowing gas to pond, which would then be detected by the seismic method.

The co-location of the open eastern end of the funnel and particularly the northern boundary of the funnel and the cluster of pockmarks suggests a causal relationship. Thin deepwater sands or silts would have provided migration routes for released methane from the regions that underwent dissociation20. The stack of high-amplitude black-red reflections beneath the pockmarks may represent gas-charged sands, also charged by a process of up-dip gas migration. Given the sedimentation rates, burial of 80 m could have taken ~ 1.5 Myr (ref. 27). Some pockmarks are at the seabed and may have been active recently (Extended Data Fig. 2). Therefore, venting probably occurred episodically from the mid Quaternary and possibly to the present day. A shift of ~50 m from the BHSZR to BHSZM over the 300 km2 of the HDZ would release ~0.023 GtCH4 (with a global warming potential of 0.6 GtCO2e; Methods). There are a small number of very small pockmarks at the landward limit of the marine hydrate17. However, the dominant focused flow of methane into the ocean occurs after the BHSZ reset upwards, and the methane moved up-dip from a 30-km-wide zone, bypassing the HSZ and venting ~10–15 km beyond the landward limit of the hydrate. In addition to this lateral transport, it is likely that methane from deeper strata beneath the HDZ also contributed to the total volume of venting, but the absence of vertical seismic anomalies below the pockmarks suggests that this is a minor contribution.

The dip of a graded continental slope is ~2° (ref. 29). The dip of the BHSZ is at least double this at its landward limit4 (Fig. 5). This angular discordance is common to many margins4,15,30 (Extended Data Fig. 4). Also, in layered sedimentary rocks, horizontal permeability is higher than vertical31. Therefore as the BHSZ resets upwards due to climatic warming, instead of migrating along the dipping BHSZ to its landward limit4,15, some methane should migrate up-dip along strata to beyond the landward limit of the hydrate (Fig. 5). This may also be occurring on other continental margins that share some of the characteristics of this one (Extended Data Fig. 4). This is the case, for example, on the US Atlantic Continental margin, where thousands of pockmarks have been reported on the upper slope and outer shelf, again beyond the landward limit of the HSZ32. The phenomenon has been documented on smaller scales elsewhere33,34, but the process could be much more important and transport methane further, from the base of the slope to vents tens of kilometres beyond where hydrates are typically found. Even on more structurally complex margins, methane bypasses the HSZ and vents beyond the landward limit, at similar distances to those documented here15,30. Rather than one contiguous gas-charged bed, critical gas pressurization would have caused seal breach, allowing gas to migrate progressively up-dip (Fig. 5)20,32. Methane from hydrate is not considered to have had a role in Late Quaternary climatic change35, but the occurrence of a field of large pockmarks and their connection to distal hydrate dissociation indicates that hydrate dissociation at least contributed to increased methane venting into the ocean and led to ocean acidification36. Previously, only ~3.5% of hydrates in the exposed zone were considered a potential climate driver2. However, some of the ~96.5% further down the slope can also contribute to methane venting in response to ocean warming, and this requires a reassessment of their role in climate change and potential incorporation in climate models.

Fig. 5: Schematic cross-section.
figure 5

Cross-section illustrating that the HDZ caused by a period of future climatic warming will be crosscut by continental margin clinoforms allowing methane to migrate to the shelf edge and vent beyond the landward limit of marine hydrate.


Seismic data and interpretation

The 3D seismic reflection data have a bin spacing of 12.5 × 25 m and a vertical resolution of ~8 m. A positive increase in acoustic impedance is represented as red in the seismic sections. The data are in the time domain. We use a velocity of 1,700 ms−1 for sediment and 1,500 ms−1 for seawater to calculate sediment thickness and the depth to the seabed, respectively. An RMS amplitude map is a map of the square root of the arithmetic mean of the squares of the values of amplitude for the reflection (Figs. 1b and 4). Changes in the seismic amplitude of a reflection that crosscuts the modern and past positions of the BHSZ can be used to delineate where resetting of the BHSZ has occurred23. Average sedimentation rates from the two hydrocarbon exploration wells were used to estimate the age of the pockmarks.

Modelling the past position of the BHSZ

The modelled positions of BHSZM and BHSZR for seismic inline 12398 were calculated by using an empirical expression for the gas hydrate phase boundary, assuming pure methane and a seawater salinity of 35‰. A geothermal gradient of 42 °C km−1 from the seafloor to a depth of 10 km below that results in the modelled BSR matching with the observed BSR. Sea-level changes since the LGM were extracted as a time series37, and the sea level during the LGM can be calculated by adding the changes to present-day sea level22. The present-day bottom-water temperature (2.39–9.08 °C) was extracted from the World Ocean Atlas around the Cap Timiris Canyon within the area 18°–22° N, 21°–17° W (ref. 38). Bottom-water temperature during the LGM was estimated by adding the past bottom-water temperature anomalies39 to the present-day vertical temperature profile37, and the CCSM4 climate model was used to test for a match with the observed relict BSR22,39. To model the modern position of the BHSZM, we used the same inputs but with the modern seabed temperature and sea level. Changes in the temperature profile of the sediment were calculated vertically using a 1D uniform and constant heat diffusivity of 10−6 m2 s−1 and a constant geothermal gradient, with a boundary condition 10 km below the seafloor.

Estimating the weight of carbon released through the pockmarks

We measured the HDZ area within the funnel. Its thickness was modelled across the 3D survey as 43–54 m. We rounded this to 50 m. We used a porosity of 50% and a hydrate saturation of 5% (ref. 2). A volume of 1 m3 of gas hydrate contains 164 m3 of methane under standard conditions40 and we multiplied the methane weight by 0.75 to calculate the carbon weight. We multiplied the methane released by 25 to obtain the CO2-equivalent global warming potential.