2.1 MPI-ESM 1.2 in transient deglaciation experiments

We use MPI-ESM, the Max Planck Institute for Meteorology Earth System Model (Mauritsen et al., 2019; Mikolajewicz et al., 2018), consisting of the ECHAM6 atmospheric general circulation model, the MPI-OM ocean general circulation model, and the JSBACH land surface model in coarse resolution (T31GR30 ≈ $\mathrm{3.75}{}^{\circ }\times \mathrm{3.75}{}^{\circ }$), to investigate the evolution of atmospheric methane during the deglaciation. The basis for our modelling set-up is model version 1.2, which is the same model version used in the Coupled Model Intercomparison Project Phase 6 (CMIP6; Eyring et al., 2016).

MPI-ESM, similar to many other ESMs, considers the land–sea and glacier masks as well as the river routing as fixed. To determine the transient evolution of climate, these are updated automatically in our model version, as described in Meccia and Mikolajewicz (2018) and Riddick et al. (2018). Briefly, the model evaluates the changes in ice sheet and sea level after every decade. From these, combined with the RTopo-2 topography (Schaffer et al., 2016), it determines a new land–sea mask, bathymetry and orography as well as a new river routing set-up to be used for the next decade in the model experiment.

2.2 Modelling the methane cycle

Based on the recent Global Carbon Project CH₄ budget (Saunois et al., 2016, 2020), we consider methane emissions from wetlands, termites, wildfires and herbivores as relevant during the deglaciation, and we also include emissions from geological sources. We assume the latter to be independent of climate and, therefore, prescribe them, whereas we aim to model all of the former interactively. We neglect anthropogenic methane emissions in the experiments described here, as they are highly uncertain but likely very small for the time period considered here. In terms of methane sinks, the atmospheric sink of methane is most important, and soils are a secondary sink. We include both of them in our model.

2.3 Atmospheric methane sink

As described in Kleinen et al. (2021), the spatiotemporal evolution of the methane abundance in the ECHAM6 atmospheric model is simulated using a methane tracer which undergoes transport and chemical removal, while emissions are calculated using the JSBACH land surface model. The atmospheric sink of methane is calculated using a zonally averaged reactivity field obtained from the comprehensive ECHAM/MESSy Atmospheric Chemistry (EMAC) model (Joeckel et al., 2010). This reactivity field contains, in aggregated form, all atmospheric sink terms considered in the EMAC model (i.e. reaction with OH, tropospheric Cl, stratospheric reactions, etc.).

Following Gromov et al. (2018) and Kleinen et al. (2021), the subsequent updated parameterisation is used to account for variations in the atmospheric oxidative capacity and, therefore, tropospheric CH₄ reactivity $r_{{\mathrm{CH}}_{\mathrm{4}}}$:

where LN is the global emission of lightning nitrogen oxides (NO_x), simulated interactively according to Price and Rind (1992, 1993); M is the CH₄ atmospheric burden; RN and RC are the terrestrial (surface) emissions of reactive nitrogen (NO_x) and carbon compounds given in Tg N yr⁻¹ and Tg C yr⁻¹ for the emission fluxes, respectively; and A is the mean tropospheric reactive aerosol surface area (about 100 Mm² in both present-day and past climates). Fit parameters (α=7.45 Tg N^−p Tg C^−q yr⁻¹, p=0.36, $q=-\mathrm{0.60}$, k_N=0.59, k_C=4.25 and k_A=4.21 Tg C Mm⁻²) are obtained from an ensemble of EMAC simulations covering a broad range of RN, RC, LN and M values under LGM (21 ka), Middle Holocene (6 ka), PI and PD conditions. The fitted $r_{{\mathrm{CH}}_{\mathrm{4}}}$ value is accurate to 7 % at 95 % confidence intervals (CIs). In the MPI-ESM experiments, the natural emission components of RN and RC are obtained from the Model of Emissions of Gases and Aerosols from Nature (MEGAN) model (Guenther et al., 2012) for the biogenic sources and from the SPITFIRE model (Lasslop et al., 2014), with emission factors from Kaiser et al. (2012), for fire emissions. We use the terrestrial NO_x emissions for the RN term, and we use the biogenic CO and isoprene (C₅H₈) fluxes, scaled by a factor of 1.4 to account for secondary biogenic co-emitted compounds, as proxies for the total RC emitted. These scaling factors were derived from the simulated PD total RC emissions. In experiments covering the historical and future periods, anthropogenic emissions of RC and RN would be considered as well (see Kleinen et al., 2021), but they are neglected for the experiments described below.

2.4 Model forcing and experiments

We forced the model with prescribed orbital parameters from Berger (1978) and greenhouses gases from Köhler et al. (2017). Orbital parameters and greenhouse gas concentrations are supplied to the model as decadal means and are updated every 10 model years. Atmospheric aerosols were prescribed to constant 1850 CE conditions (Kinne et al., 2013), and we considered no anthropogenic land use. Ice-sheet extent was prescribed from the GLAC-1D ice-sheet reconstruction (Tarasov et al., 2012; Briggs et al., 2014; Ivanovic et al., 2016). Ice-sheet extent, bathymetry and topography (Meccia and Mikolajewicz, 2018), and river routing (Riddick et al., 2018) were continuously updated every 10 model years.

We initialised the model from a spin-up experiment at constant 26 ka boundary conditions, running for several millennia. From this model state, we performed a transient model experiment until PI times, with continuously updated ice-sheet extent, bathymetry, topography and river routing, similar to the experiments described by Kapsch et al. (2022). In this experiment, called “base” in the following, the prescribed ice-sheet forcing from the GLAC-1D reconstruction leads to a collapse of the Atlantic Meridional Overturning Circulation (AMOC) early in the Bølling–Allerød (Kapsch et al., 2022), due to excess meltwater entering the North Atlantic, and the Younger Dryas does not occur in this experiment. Thus, we performed a second experiment, called “MWM” (meltwater manipulation), in which we manipulate the meltwater fluxes from the Laurentide ice sheet: starting in 15.2 ka, we prevent meltwater from the Laurentide ice sheet from reaching the ocean, instead storing it. We then release this accumulated meltwater over a period of 1200 years, starting in 12.8 ka, adding it to the Mackenzie River watershed, thereby mimicking the storage of glacial meltwater in proglacial lakes, like Lake Agassiz, and its subsequent release (Murton et al., 2010).

3.1 Transient climate and land carbon changes

At the last glacial maximum, around 20 ka, the global mean temperature is 282 K in our model, which is 4.7 K lower than during the PI period (Fig. 1a). As the ice sheets melt and CO₂ increases, the temperature begins to rise significantly after 18 ka, with the rate of temperature increase declining after 9 ka. Dallmeyer et al. (2022) evaluated model temperatures in the experiment against reconstructions and found a reasonably close match, as illustrated by the comparison against the reconstructions by Shakun et al. (2012) and Osman et al. (2021) in Fig. 1a. While Shakun et al. (2012) reconstructed a smaller temperature change between the LGM and the Holocene than we see in our experiments, Osman et al. (2021) reconstructed a larger temperature change, with our experiment situated between the two. The AMOC (Fig. 1b) is relatively stable at about 20 Sv between the LGM and 14.8 ka. Here, the base and MWM experiments diverge: in the base experiment, meltwater pulse (MWP) 1a, a sea level rise event recorded in Barbados corals (Fairbanks, 1989), leads to a strong freshening of the North Atlantic and a near-complete AMOC shutdown (remaining overturning of 1 Sv) at 14.38 ka, which in turn leads to a drop in global mean temperature by nearly 1 K. Nevertheless, temperature and the AMOC recover quickly. The AMOC continues to be highly variable until 9 ka, with a further near collapse from MWP 1b at 11.4 ka. Global mean temperature, however, is affected significantly less by the AMOC reduction at MWP 1b than at MWP 1a. After 9 ka, the AMOC is relatively stable, although at a slightly smaller overturning of 18.7 Sv (during the PI period). Similar to global mean temperature, the total land carbon stock (Fig. 1c) remains nearly constant at 970 Pg C, which is 718 Pg C less than during the PI period, from the LGM to 18 ka. Subsequently, total land C increases until 10.95 ka. Afterwards, only minor changes in total land carbon stock occur until the PI state, with a total land C stock of 1688 Pg C. The AMOC collapse at 14.38 ka caused by MWP 1a leads to a short decrease in the C stock of about 6.8 %, but the total duration of this excursion is about 800 years and carbon continues to rise after it.

Figure 1Overview of climatic and land carbon changes from the LGM to the PI period: (a) global mean temperature change, (b) North Atlantic meridional overturning circulation, and (c) total land carbon storage and atmospheric CO₂. The base experiment is shown in blue, and the MWM experiment is shown in red. Panel (a) also contains the temperature change reconstructed by Shakun et al. (2012) and Osman et al. (2021), while the reconstructed atmospheric CO₂ in panel (c) is from Köhler et al. (2017).

Download

In the MWM experiment, the initial collapse of the AMOC at 14.38 ka is prevented by the manipulation of the meltwater fluxes, and temperature and carbon continue to rise during this period. As we release the accumulated Laurentide ice-sheet meltwater at 12.8 ka, the AMOC collapses, with the consequence of a 1.25 K drop in global mean temperature and a 100 Pg C or 6.9 % decrease in the terrestrial C stock. After we end the meltwater manipulation in 11.6 ka, temperature, the AMOC and land carbon stocks recover quickly.

3.2 Transient changes in methane

In comparison to a stack of ice-core-derived CH₄ concentrations in Antarctica (Köhler et al., 2017), the atmospheric concentration of methane in our experiments is quite reasonable overall (Fig. 2a), although there are significant discrepancies during some periods. In the ice core, methane is nearly constant at ∼370 ppb from the LGM to 18 ka, at which point it starts increasing slowly. At 14.6 ka (the transition into the Bølling–Allerød warm period), CH₄ increases abruptly by ∼150 ppb, decreasing again at 12.8 ka (the transition into the Younger Dryas). At the end of the YD, at about 11.5 ka, CH₄ again increases abruptly and stays roughly constant at ∼690 ppb for the next 2000 years. Finally, ice-core CH₄ slowly decreases by ∼120 ppb during the Early Holocene between 9 and 4.5 ka, recovering after the Middle Holocene to ∼690 ppb during the PI period.

Figure 2Atmospheric concentration of CH₄ over (a) Antarctica and (b) Greenland in the MPI-ESM base and MWM experiments as well as in ice-core data. Antarctic ice-core data are from Köhler et al. (2017), and Greenland ice-core data are from Chappellaz et al. (2013) and Beck et al. (2018). (c) Interpolar gradient of atmospheric CH₄, expressed as the difference in the CH₄ concentration between Greenland and Antarctica.

Download

Modelled atmospheric CH₄ over Antarctica is only slightly lower than reconstructed at 20 ka (370 ppb in the ice core and 350 ppb in the model; Table 1), although the discrepancy is larger at 21 ka. Similar to reconstructions, the atmospheric CH₄ starts rising significantly at 18 ka, but the increase at the beginning of the Bølling–Allerød and the drop in methane during the transition into the Younger Dryas are dissimilar, at least in the base experiment: in terms of atmospheric CH₄, the Bølling–Allerød seems to start earlier, at 16 ka instead of 14.6 ka. At 14.38 ka, near the beginning of the BA in reconstructions, the base experiment displays a significant drop in atmospheric methane, caused by the AMOC shutdown after MWP 1a. For the transition into the Younger Dryas at 12.8 ka, on the other hand, the base experiment continues at high levels of atmospheric methane and does not show the decrease in methane displayed by the ice-core data, as the model shows no climatic transition at this time.

Table 1Simulated and reconstructed CH₄ concentrations (ppb) at the ice-core sites for selected time periods.

NA stands for not available.

Download Print Version | Download XLSX

The MWM experiment differs from the base experiment at the beginning of the Bølling–Allerød, as the atmospheric concentration of methane does not change at the beginning of the BA. Instead, it is very similar to the ice-core data. Furthermore, the atmospheric concentration of methane very rapidly decreases by 25 % at the beginning of the Younger Dryas at 12.8 ka, similar to ice-core data (−29 %). At the end of the Younger Dryas, recovery of atmospheric methane occurs slightly earlier than in the ice-core data, although not as quickly. Therefore, the Younger Dryas in our MWM experiment is quite similar to the ice-core data. For atmospheric methane during the Holocene, however, there is a significant divergence between model results and ice-core data: while ice-core CH₄ slowly decreases towards the mid-Holocene and increases again during the Late Holocene, the model results show constant atmospheric CH₄ throughout the Holocene.

A comparison to Greenland ice cores is much more intricate, as recent investigations show excess CH₄ in some Greenland ice-core records (Lee et al., 2020). This issue has not yet been resolved fully; thus, we do not show a full ice-core CH₄ time series from Greenland. Instead, we focus on data derived from the NEEM (North Greenland Eemian Ice Drilling) ice core (Chappellaz et al., 2013), which were obtained using a continuous-flow measurement technique and are, thus, presumably less susceptible to the generation of excess CH₄, as well as data from the Holocene (Beck et al., 2018), when the dust concentration was much lower. For those times when data are available (earlier than 17.5 ka and 14 ka to the PI period), the comparison for Greenland (Fig. 2b) is of a similar quality to the comparison for Antarctica. Thus, Greenland CH₄ is very similar for 20–18 ka, and the evolution of methane during the Bølling–Allerød to Younger Dryas transition is also very similar in model and ice-core data if one considers the MWM experiment (Table 1). Therefore, our model captures the gradient in methane between Greenland and Antarctica adequately for key periods of the deglaciation. The interpolar gradient, shown here as the difference in the atmospheric CH₄ concentration between Greenland and Antarctica (Fig. 2c) is positive throughout the experiment, with values at the LGM near 20 ppb and PI values of about 55 ppb. Again, the rate of increase is highest between 18 and 10.5 ka, with relatively stable values throughout the Holocene. Finally, the tropical methane concentration (Fig. A1a) remains in between the values for Antarctica and Greenland throughout the experiment, which is in contrast to previous studies that have shown the highest LGM methane concentrations in the tropics (Valdes et al., 2005).

The terrestrial methane fluxes (Fig. 3) largely determine the atmospheric methane concentration. The net CH₄ flux (Fig. 3a) increases from 90 Tg CH₄ yr⁻¹ at 20 ka to 165 Tg CH₄ yr⁻¹ during the PI period, with emissions staying very similar between 20 and 18 ka, increasing more strongly between 18 and 11 ka, and staying nearly constant between 10 ka and the PI period. In the base experiment, emissions decrease strongly as a response to the AMOC collapse after MWP 1a, with emissions recovering quickly as the AMOC resumes, which is a pattern that continued after MWP 1b. In the MWM experiment, however, we see a 32 Tg CH₄ yr⁻¹ reduction in net emissions in response to the AMOC collapse induced by the meltwater manipulation, followed by a quick recovery as the manipulation ceases. Wetland fluxes (Fig. 3b) are the most important component of the net CH₄ flux; thus, their temporal change is rather similar to the net flux, although at a slightly smaller overall magnitude, with wetland emissions of 79 Tg CH₄ yr⁻¹ at 20 ka and 142 Tg CH₄ yr⁻¹ during the PI period, while the emission reduction from the AMOC collapse in the MWM experiment is 32 Tg CH₄ yr⁻¹. Here, the AMOC collapse leads to a significant decrease in NH temperatures around the Atlantic Ocean, in turn leading to decreased evaporation and, thus, decreased precipitation, thereby decreasing wetland areas and methane production in the NH tropics. The non-wetland methane fluxes (Fig. 3c) are of significantly smaller magnitude than the wetland fluxes, with emissions at 20 ka of 7.4 Tg CH₄ yr⁻¹ for herbivores, 2.8 Tg CH₄ yr⁻¹ for termites, 3.2 Tg CH₄ yr⁻¹ for fire emissions and −2.2 Tg CH₄ yr⁻¹ for the methane uptake in upland soils. Towards the PI state, these fluxes increase to 14.5, 6.3, 10.9 and −6.3 Tg CH₄ yr⁻¹ for herbivores, termites, fires and uptake, respectively. Here, the differences between the base and MWM experiments are small; thus, we omit the MWM experiment from Fig. 3c. The dynamics of the uptake flux appear to be different from the other fluxes at first glance, with pronounced differences for the Bølling–Allerød and Younger Dryas at the same times as they appear in proxy records. This is due to the fact that the most important driver of terrestrial methane uptake is the atmospheric concentration of methane (Kleinen et al., 2020), which we prescribed from ice-core data for the uptake flux.

Figure 3Terrestrial methane in the MPI-ESM base and MWM experiments: (a) net flux, (b) wetland emission flux, and (c) non-wetland methane emissions from herbivores, fires, termites and terrestrial methane uptake.

Download

We model the atmospheric sink of methane as a function of terrestrial emissions of reactive carbon (RC) as well as the NO_x emissions from the soil/vegetation and lightning (Eq. 2). The temporal changes in RC emissions during the deglaciation (Fig. 4a) are very similar to the changes in CH₄ emissions: at 20 ka, RC emissions are 270 Tg C yr⁻¹, and they start increasing noticeably at 18 ka, reaching a maximum of 500 Tg C yr⁻¹ at 11 ka. They are nearly constant, with a very small increasing trend during the Holocene, with values increasing from 477 Tg C yr⁻¹ at 8 ka to 490 Tg C yr⁻¹ during the PI period. The base experiment also contains interruptions of the increasing trend after MWP 1a and 1b that do not occur in the MWM experiment. The MWM experiment instead displays a significant decrease in RC emissions during the Younger Dryas. Surface NO_x emissions (Fig. 4b) increase much more gradually compared with RC emissions, although at a higher relative rate, with total fluxes of 2.2 and 5.8 Tg N yr⁻¹ at 20 ka and during the PI period, respectively. They are also much less affected by AMOC fluctuations (thus not having as strong a response to MWP 1a and MWP 1b in the base experiment) or the meltwater-induced AMOC collapse in the MWM experiment. As a result, the two experiments do not differ significantly in terms of terrestrial NO_x emissions. This is different for the lightning NO_x emissions (LNO_x), which have a minimum of 1.6 Tg N yr⁻¹ at the LGM and increase to 2.4 Tg N yr⁻¹ at 9 ka, decreasing thereafter to 2.2 Tg N yr⁻¹ during the PI period. Decreases after MWP 1a and MWP 1b are clearly visible in the base experiment; in the MWM experiment, LNO_x decreases from 2.2 Tg N yr⁻¹ during the BA to 1.9 Tg N yr⁻¹ during the YD.

Figure 4Atmospheric sink of CH₄: emissions of (a) reactive carbon (RC), (b) NO_x from soils and lightning, and (c) the resulting atmospheric lifetime (τ) of CH₄. τ increases with increasing RC and decreases with increasing NO_x.

Download

In both experiments, the simulated atmospheric lifetime of CH₄ (Fig. 4c) varies considerably, from 10.4 years at 20 ka to a maximum of 12 years at 12–11 ka, slowly decreasing throughout most of the Holocene, from 11.8 years at 8 ka to 11.6 years at 2 ka, with a departure to 11.3 years at 1 ka, before reaching 11.6 years during the PI period. Prior to about 8 ka, the variations are dominated by pronounced changes in the RC emissions and CH₄ burden. Note that larger RC emissions and a larger CH₄ burden increase the methane lifetime, whereas larger NO_x emissions have the opposite effect. The general tendency towards higher NO_x emissions dampens the RC-driven increase in the methane lifetime. Consequently, the reduction in RC emissions driven by the meltwater events results in a stronger decrease in the methane lifetime, as the total NO_x emissions are much less affected by these events. LNO_x is reduced, but soil NO_x is unaffected and even slightly increasing. Across the latitudes, the mean atmospheric lifetime of CH₄ is longest in the Southern Hemisphere (SH) extratropics, where values range from 19.5 to 22.5 years (Fig. A1b), whereas lifetimes in the NH extratropics are slightly lower, about 1 year shorter than in the SH. The mean atmospheric lifetimes in the tropics, however, are substantially shorter (Fig. A1c), ranging from 7.2 to 8.3 years.

3.3 The Bølling–Allerød and Younger Dryas

The model experiments that we performed do not display the methane concentration signature expected for the onset of the Bølling–Allerød: an abrupt increase in atmospheric methane between 14.6 and 14.5 ka. Instead, we see an abrupt increase in atmospheric methane at 16.2 ka, 1600 years earlier than expected from palaeoclimate reconstructions. The AMOC signature that would be associated with Heinrich event 1 (H1), a near collapse as recorded in Bermuda rise sediments (McManus et al., 2004; Stanford et al., 2011), is also missing in the model experiments. Due to the absence of the H1, the Bølling–Allerød occurs earlier in our model and is initially of smaller magnitude.

Comparing net CH₄ emissions between the late BA at 13.2 ka in the MWM experiment and 20 ka (Fig. 5), increases in emissions are apparent south of the NH ice sheets as well as in Siberia. The most prominent increase in emissions, however, occurs in the NH tropics, with increases in Africa and Asia being especially prominent. Specifically, northern Africa is substantially wetter and greener than either at the LGM or at present, with the substantial CH₄ emissions from the Sahel region being very striking. The increase in precipitation here leads to an expansion of wetlands and vegetation, and while the wetland expansion increases the methane emitting area, the expansion of vegetation enhances soil carbon content, thereby increasing the substrate available for methane generation.

Figure 5Change in net land CH₄ emissions between the LGM (20 ka) and BA (13.2 ka) in the MWM experiment. Shaded areas indicate LGM land points, and the continental outline is from 13.2 ka.

As the AMOC collapses due to the induced meltwater release at 12.8 ka, the region around the North Atlantic Ocean becomes substantially colder, with temperature decreases exceeding 5 K in large parts of the northern and eastern Atlantic Ocean basin north of the Equator. Precipitation in the NH tropics also decreases significantly, with decreases of more than 1000 mm yr⁻¹ in the tropical Atlantic and precipitation decreases of more than 400 mm yr⁻¹ in the Sahel region as well as over India and Indonesia.

As a result, methane emissions decrease substantially between the BA at 13.2 ka and the YD at 12.5 ka (Fig. 6). Most significant are methane emission reductions throughout the tropics, largely due to reduced precipitation leading to less methane production, but also in Europe, where conditions are substantially colder and dryer during the YD than during the BA.

Figure 6Change in net land CH₄ emissions between the BA (13.2 ka) and YD (12.5 ka) in the MWM experiment. Shaded areas indicate BA land points, and the continental outline is from 12.5 ka.

3.4 The role of exposed shelf areas

Due to the lower sea level in glacial climate, significant areas of the present-day continental shelf were exposed. In our model, 14.2×10⁶ km² of non-glaciated continental shelf that lies below sea level at present was exposed at 20 ka. Slightly more than half this area (7.8×10⁶ km²) was located in tropical latitudes, predominantly in South-East Asia and Australia. In the NH extratropics, some 5×10⁶ km² was exposed, largely in the Laptev Sea and Bering Strait. These areas changed relatively little until 15 ka, but they started decreasing rapidly afterwards. By 7.8 ka, most of these areas were flooded, and the rising sea level covered the remainder at 4.3 ka. Our model shows the exposed shelf areas to be a significant source of methane due to significant vegetation growth and wetland formation. With emissions of 31 Tg CH₄ yr⁻¹, the global shelf areas emitted a significant fraction of the net methane flux at 20 ka (Fig. 7a). Due to global warming, the shelf emissions increased further to 40 Tg CH₄ yr⁻¹ at 15 ka, but they subsequently declined as sea level rise started to submerge these areas. The bulk of this flux – 20 Tg CH₄ yr⁻¹ at 20 ka – is emitted from the extensive shelf areas in South-East Asia (Sunda Shelf) and north-western Australia (Sahul Shelf), combined in Fig. 7b as South-East Asia. Here, emissions increased slightly between 20 and 14 ka, subsequently dropping to near zero by 10 ka. While the AMOC collapse in the MWM experiment does affect the region, decreasing emissions slightly, the effect is less pronounced here than in other regions. In non-tropical regions, the largest exposed shelf area in the Laptev Sea and Bering Strait had emissions of some 3.2 Tg CH₄ yr⁻¹ at 20 ka, which rose to 6.9 Tg CH₄ yr⁻¹ at 14 ka, falling subsequently to zero at 8 ka when the remaining shelf area was flooded.

Figure 7Net methane emissions from shelf areas flooded in a PI climate: (a) global emissions and (b) emissions from the Laptev Sea and South-East Asia.

Download

There were further regions with some shelf methane emissions, but the aforementioned areas cover the most important source regions at 15 ka (Fig. 8). However, the highest net emissions originate from a grid point in the East China Sea, just east of present-day Shanghai. The bathymetric data used to determine wetland area on the shelves indicate a very flat area at this location, which results in a grid-cell mean inundation of nearly 25 % at 15 ka. Whether this high value accurately reflects conditions at the time when the shelf was exposed or whether it is due to subsequent changes in bathymetry caused by the Yangtze River sediment load is an open question that we are not qualified to answer.

Figure 8Net shelf CH₄ emissions at 15 ka – the emission maximum. Shaded areas indicate land points at 15 ka, and the continental outline is from the PI state.

3.5 Regional distribution of methane fluxes over time

During the Early Holocene (at 9 ka), the total methane emissions were very similar to the PI state. However, the spatial distribution differed significantly (Fig. 9). Due to the recent deglaciation of North America, with some parts of the Laurentide ice sheet still remaining, CH₄ emissions from North America were strongly reduced, whereas emissions from northern Siberia were enhanced in comparison with the PI period, due to enhanced vegetation growth (and thus soil carbon) from increased solar radiation during the boreal summer season. Furthermore, emissions from South America were reduced. The most striking difference to the PI state, however, was the change in NH tropical Africa. At present, northern Africa is not an important source of methane. Large parts of the continent are rather dry and, thus, do not produce significant amounts of methane. During the Early Holocene, however, the North African monsoon was considerably stronger, leading to significantly enhanced precipitation across the Sahel region, reaching into the southern Sahara (Fig. 10) (Dallmeyer et al., 2020, 2021). As a result, vegetation and wetlands in the Sahel expanded and methane emissions were substantially stronger than during the PI period.

Figure 9Net CH₄ emission change from 9 ka to the PI period. The continental outline is from 9 ka.

Figure 10Precipitation change from 9 ka to the PI period. The continental outline is from 9 ka.

Figure 11Net CH₄ flux over time by (a) latitude band and (b) continent for the NH tropics (0–30^∘ N).

Download

Looking at the time series of the regional distribution of methane fluxes since the LGM in our experiments, the SH tropics had the largest net emissions at the LGM, with the NH tropics and extratropics each having slightly lower emissions. As emissions increased during the deglaciation, their share from the NH tropics and extratropics increased faster than that from the SH tropics (Fig. 11a). Emissions in the NH extratropics increased markedly between 18 and 11 ka, rising from 33 to 60 Tg CH₄ yr⁻¹ 7000 years later, with little change afterwards. Emissions in the NH tropics, on the other hand, were strongly affected by the AMOC perturbation events, after MWP 1a in the base experiment and at 12.8 ka in the MWM experiment. Here, emissions also started to increase from 30 Tg CH₄ yr⁻¹ at 18 ka, and they continued to rise until they reached 62 Tg CH₄ yr⁻¹ at 9 ka. In the base experiment, this increase was interrupted after MWP 1a; here, CH₄ emissions dropped from 46 to 27 Tg CH₄ yr⁻¹ during the period from 15 to 14.4 ka. In the MWM experiment, on the other hand, emissions in the NH tropics strongly reacted to the imposed Younger Dryas event, with emissions dropping from 55 to 31 Tg CH₄ yr⁻¹ between 12.8 and 12.5 ka. Investigating the changes in the NH tropics by continent (Fig. 11b), it is clear that the bulk of the long-term changes in methane fluxes from this region are from NH tropical Africa, with African CH₄ emissions increasing from 5 Tg CH₄ yr⁻¹ at 20 ka to some 26 Tg CH₄ yr⁻¹ between 13 and 8.5 ka and then subsequently decreasing gradually to 8 Tg CH₄ yr⁻¹ after 2 ka. In comparison to these fluxes from NH Africa, fluxes from NH tropical Asia and America change very little, except for the rapid changes in fluxes from NH tropical Asia connected to AMOC changes after MWP 1a in the base experiment and during the Younger Dryas in the MWM experiment.

Finally, focusing on the Holocene changes after 8 ka, we see a decreasing trend in emissions from the NH tropics (Fig. 11a), whereas emissions from the SH tropics increase. The latter increase is due to a generally increasing trend in precipitation over South America (Fig. 10), leading to increasing emissions from the Amazon region, while the former is due to a decreasing trend in precipitation over NH tropical regions, mainly in Africa, but also over northern India. These precipitation trends are ultimately caused by orbital changes, with the change in precession leading to a decrease in radiation reaching NH tropical areas and an increase in radiation reaching SH tropical areas, causing changes in land–sea temperature contrasts and, thus, changes in monsoon circulation and related precipitation (Dallmeyer et al., 2021). Therefore, we see a very similar trend to Singarayer et al. (2011), who reported decreasing emissions from the NH tropics and increasing emissions from the SH over the last 8 ka. Singarayer et al. (2011) were thus able to explain the Holocene trend in reconstructed CH₄, a decrease from the Early to Middle Holocene followed by an increase from the Middle to Late Holocene. In our model, however, these trends more or less cancel each other out, leading to very small changes in overall methane emissions, despite the substantial changes in regional fluxes.