In global land-surface modelling the term “wetland” is commonly used in the sense of a hydrological diagnostic, as opposed to an ecosystem characterization: a grid cell (or sub-grid fraction) is labelled a wetland when the simulated water table reaches or exceeds the soil surface, which creates anaerobic conditions that allow methanogenesis. This definition does not imply any change in vegetation type: often, the vegetation cover is prescribed from boundary condition data sets, in our case the from the Land-Use Harmonization 2 (LUH2) land use dataset . This then remains unchanged, regardless of the water-table status. Consequently, a cell that is classified as a wetland in our model may still be labelled “cropland” in the land use map and retain the same crop vegetation. The term “wetland” therefore refers only to the soil-water state that controls methane production, not to the ecological concept of a wetland ecosystem.

We estimated the wetland area fraction using a diagnostic TOPMODEL approach , where the compound topographic index (CTI), which describes the relation between the (hydrological) upstream area and the local slope for any point, is used to relate the grid-mean water table to the water-table distribution on the sub-grid-scale topography. The wetland fraction is then assumed to be the fraction of the grid cell where the water table is at or above the soil surface.

The CTI index was obtained from the data set presented in . However, in order to avoid unrealistically low CTI values that would generate spurious wetlands, we enforce a minimum CTI value of 5.

The calculation of the wetland area does not directly feed back to the soil hydrology. Instead, is assumed that the part of the soils, which are diagnosed as inundated using this approach, are anaerobic. Thereby the wetland extent has an influence on the carbon cycle, which again influences the hydrology (see Sect. ).

We extended the scheme presented in detail in with two improvements:

The simple piecewise linear estimate of the ice fraction of soil water used in has been replaced by calculations internally in JSBACH's soil hydrology . Thus, the freezing and melting of soil water now also influences both the hydraulic conductivity, the energy calculations of the soil and the soil carbon handling.

The wetland methane production model implementation is adopted from , which is based on the approach by . In this model, the decomposition of soil organic matter (SOM) is assumed to occur under anaerobic conditions in the diagnosed grid cell wetland fraction (Sect. ). Due to the anaerobic conditions, decomposition products are not just CO2, as under aerobic conditions, but both CO2 and CH4. The model then transports CO2, CH4 and O2 vertically through the soil column by considering diffusion, ebullition, and plant-mediated transport through aerenchyma, also determining the oxidation of CH4 in soil layers where sufficient O2 is present.

In laboratory incubations under strictly anaerobic conditions, SOM mineralisation yields a 50 % CO2 : 50 % CH4 split . However, in actual wetland soils this ratio is usually much greater than 1:1, and can vary significantly among different kinds of wetlands . In our model, we therefore assume a significantly lower fraction of CH4 production, which we furthermore assume to be temperature-dependent in a Q10 formulation. We calibrated this to give a 24 % CH4 fraction at 295 K and a Q10=1.2 . This formulation does not distinguish between acetoclastic and hydrogenotrophic methanogenesis, which is the standard approach in most global methane models. Distinguishing acetoclastic from hydrogenotrophic methanogenesis would require explicit representation of substrate availability (acetate vs. H2/CO2) and is not currently implemented. A recent intercomparison reported that 37 out of 40 global wetland-methane models do not separate the two pathways, highlighting that this simplification is common practice, albeit a known source of uncertainty.

Following , we furthermore assume a time- and space-independent reduction of SOM decomposition rate, so that wetland SOM is decomposed with 35 % of the aerobic decomposition rate.

The produced methane, along with CO2 and O2, is transported vertically through the soil column via diffusion, ebullition and plant aerenchyma, with process descriptions and parameters adopted from . Depending on the availability of O2, which can be supplied either via diffusion or via plant aerenchyma, CH4 may be oxidized partly or completely, also parameterized using a Q10 formulation with a Q10 value of 1.9.

Finally, uncertainty ranges for parameters used in the wetland emission model are sufficiently large that the total emissions (but not the spatial distribution) can be calibrated over a range of wetland area estimates. Thus, we used the CO2:CH4 ratio in methane production to calibrate our model to yield global total emissions similar to the estimate.

Historically, many agricultural regions have been drained to remove excess water from fields . In the model, this practice is represented by suppressing the diagnostic wetland fraction on all grid cell fractions that are classified as cropland in the LUH2 land use map. Importantly, this drainage flag only suppresses the wetland diagnostics (i.e., the calculation that would label the cell as anaerobic and allow methane production). It does not modify the underlying soil-hydrology equations (soil moisture, infiltration, runoff) nor the carbon-cycle fluxes beneath the crops. The vegetation type therefore remains the prescribed crop (or pasture) throughout the simulation, no transition to natural wetland vegetation (e.g., emergent macrophytes) occurs. Furthermore, the inclusion of drained crops has the advantage, that methane emissions from rice fields are unequivocally attributed as an anthropogenic source and thus fall out of the scope of this study.

The standard JSBACH model uses the ARNO-scheme to separate rain and melting snow into surface runoff and infiltration into the soil, based on the slope distribution of the grid cell. This scheme ignores that water might pile up in small-scale topographic depressions forming ponds or puddles, from where the water is available for evapo(transpi)ration or later infiltration into the soils. Thus, it may suppress important water and energy feedbacks between surface and atmosphere. It is assumed, that in addition to the micro-topographical effects, this water retention takes place where local soil surface hydraulic conductivity is low. Therefore, the conductivity of these areas is assumed to be equivalent to that of clay, which is the natural soil type with the lowest conductivity. To account for these effects, we implemented the SWR scheme described in detail in , .

It is assumed that the local SWR is a temporary, though possibly repeated, phenomenon compared to the wetlands arising due to saturated soils (Sect. ). The areas inundated by SWR is therefore not counted as wetlands. Due to the lower infiltration in areas with SWR, the soil is comparatively dry, and we assume that these soils stay oxic and therefore do not contribute to the wetland methane production.

The average June–August (JJA) wetland distribution (Fig. a) shows the main wetland areas in the tropical rain forests of South America and Africa, as well as in the Hudson Bay area and central northern Siberia. Smaller, but still important, areas are found in India, eastern North America and larger parts of northern Siberia. In addition, some wetlands are found in central western Europe, south-eastern US, Indonesia and eastern China. Roughly, the majority of the Northern Hemisphere summer wetland area is concentrated in two latitudinal bands: the boreal zone (55–70° N), contributing 34 % of the global area, and the tropics (20° S–15° N) with 38 %. The global JJA wetland area is ≈5.2Mm2, with a slight decrease over time (2 % per century). This trend over the entire experiment period, however, hides the fact that the wetlands are only diminishing during the first part of the experiment, while they are growing again at a rate of 7 % per century after 1980. This increase is mainly driven by changes in the NXT area, though the trend also turns the tropics from negative to positive over the course of the experiment. In fact, the wetland area in the NXT region is slightly larger in the later part of the experiment than in the long-term average, and is thus opposing the global decreasing trend (Fig. a), leading to an increase in the NXT fraction of the global wetland areas from <30% in the 1940s to 33 % in the last decade of the experiment (2005–2014). At the start of the experiment, this fraction was about 31.6 %.

The geographical trend distribution (Fig. b), shows areas of both increasing and decreasing wetlands: Decreasing wetlands are found in the east-central US, eastern Europe, southern Brazil and India/Bangladesh, while increasing wetlands are found in eastern US and western Europe. In general there is a close relation between the development of cropland (and thus drainage, Fig. S3) fraction and the development of the wetlands.

The average simulated wetland area on non-snow-covered areas in “Base” is ≈3.8Mm2 with a pronounced annual cycle (Fig. ), ranging from ≈2.8 to ≈5.4Mm2. The year to year variation on the other hand is small (1σ=0.06Mm2). The annual cycle stems only from the NXT areas, with a maximum in the Northern Hemisphere (NH) summer, June to September, where the northern extratropics are snow free. Though the NXT wetland area in the non-NH-summer increase by up to almost 1Mm2 (November) if no snow masking is applied, the shape of the annual cycle remains. In the annual average, snow is covering ≈0.5Mm2 of wetlands. The lack of an annual cycle in the tropics is to some degree a result of a cancellation of the annual cycles on the difference continents being out of phase (Fig. S2). The rather weak South American wetland annual cycle peaks in April–July and has its minimum in October–January, quite opposite to the somewhat more pronounced annual cycle in Africa, which is the other main tropical wetland region. South East Asia and tropical North America contribute only with very small wetland areas, leaving their influence on the annual cycle diminutive.

Global wetland methane emissions averaged over the entire experiment period amount to 166.2Tg(CH4)yr-1, of which 35.2Tg(CH4)yr-1, or about 21 %, are emitted in the NXT area. The rest stems almost exclusively from the tropical rain forest areas in South America, Africa and Indonesia, together with India (Table  and Fig. a). The relation between the boreal and tropical wetland latitudinal bands is thus quite different with respect to methane emissions: The boreal band contributes only 8 % while the tropical band emits 67 % of global wetland methane. report a similar distribution of wetland methane emissions between the boreal and tropical regions using a more microbiologically detailed methane model.

For the period 2000–2012 the global average wetland methane emission is 182.7 Tg(CH4)yr-1, in close agreement with the multi-model mean presented in , who report 185Tg(CH4)yr-1 for 2003–2012, with a 1σ standard deviation of 21Tg(CH4)yr-1.

However, our emissions are more scattered, most pronounced in the Eurasian boreal areas. Higher methane emissions are found in western Europe and southeastern US and lower emissions are found in South East Asia, eastern Africa and the Hudson Bay area. These differences in patterns are quite similar to the multi-model mean of wetland fractions presented in .

The emissions are increasing over time with a trend over the entire experiment of about 20 Tg(CH4)yr-1 (8 % per century). After around 1980, the trend is about four times as large compared to the entire experiment (Table , Fig. d). The global increase is mainly driven by increases in the large production areas in South America and Africa as well as Western Europe (Fig. b). Decreasing emissions are found in the Central US.

To obtain the standard deviation of the time series, we detrended it by separately subtracting the linear regressions in the time periods 1855–1980 and 1981–2014. The results show very little effect of the exact point of partitioning the regression periods and are virtually identical between the two periods. The standard deviation of the annual average emissions thus obtained is ≈3.4Tg(CH4)yr-1.

The emission fraction (the fraction of produced methane that is not oxidised in the soil and thus reaches the atmosphere) is 44.5 % and constant over the experiment period. The emission fraction is geographically very unevenly distributed in complex patterns, where the fractions range from near 0 to almost 80 %. In general, the major methane emitting areas have moderate emission fractions of 30 %–40 %, while hardly any methane escapes from dry areas.

Disabling SWR lowers the average wetland area by about 0.1Mm2 (Fig. a), with the entire reduction taking place in the NXT area. This mainly happens in the period July–October, and the reduction is almost twice the annual mean. Geographically more specific, the reduction is confined to a narrowing stripe reaching from northeastern Europe into central Siberia and the Mackenzie River basin (Fig. a).

The areas of reduced wetlands are within large regions where the precipitation is much lower in “DrySurf” than in “Base” (Fig. b). In parts of central Siberia the precipitation reduction is larger than 30 %. Most of the areas with large and consistent precipitation differences are in the areas of the NH jet stream. This indicates a significant shift in the magnitude of moisture recycling through the atmosphere, which is supported by a large decrease of evapotranspiration in “DrySurf” compared to “Base” (not shown).

In “DrySurf” the NXT wetland area decreases by about 7 % per century and thus 5 times faster than in “Base” (Fig. c), whereas the tropical wetlands change a bit less in “DrySurf” than in “Base”. The geographical patterns of changing trends between “Base” and “DrySurf” coincide almost exactly with those of the absolute wetland area difference. The average wetland methane emissions (Fig. c) from “DrySurf” behave rather similarly to the wetland area: Slightly lower emissions (≈2.2Tg(CH4)yr-1) from the NXT than in “Base” are partly compensated by an even smaller increase (≈0.8Tg(CH4)yr-1) from the Tropics.

In the NXT, the decreasing wetland area trend dominates over the generally increasing trend in methane emissions when looking at the entire experiment: A trend of -3 % per century opposes the tropical trend of +8 % per century. The result is that the global trend is slightly below that of “Base”. Since 1980 this picture is reversed: NXT emissions increase by 42 % per century compared to 30 % in the tropics.

In AMIP experiments with prescribed CO2, there is no direct feedback from the terrestrial carbon cycle to the atmosphere. Thus, it could be expected that the wetlands in “ConstCO2” would be identical to those in “Base”. This seems to be confirmed by the average wetland area (Fig. a), where the two experiments are indeed very close. The trends over the 1855–2014 period, however, are somewhat different, with slightly smaller trends all over the world. In the recent period, the NXT region and the tropics behave very differently: In the tropics the wetland area stagnates with hardly any trend, while the positive NXT trend in “ConstCO2” exceeds that of “Base”. These changes arise as a feedback through changed water-use efficiency of the plants at different CO2 levels, changing the transpiration. Thus, both water and energy fluxes to the atmosphere are changed, which then trigger different atmospheric responses. Although the global numbers of evapotranspiration are very similar in the two experiments, both the differences in the evapotranspiration itself and its trend show complex regional patterns (Fig. S5).

Keeping the CO2 concentration constant at pre-industrial levels for the carbon model has a huge impact on the trend of the methane emissions (Fig. d). The increasing methane emissions in “Base” are turned into a decrease in the “ConstCO2” and the trend is thus in the opposite direction compared to all other conducted experiments. Only the NXT area has a positive trend in the latter part of the experiment, and this is still smaller than in all other experiments except “ConstVeg” (discussed below). The global carbon litter is – opposite to any other experiment – decreasing after around the year 1900 (Fig. S6). There is still a slight increase in the decomposition of the soil carbon, but only about 5 % of that in the other experiments. That less carbon respires is due to a warmer climate increasing the respiration rate. The reversed trend in the methane emissions in “ConstCO2” reveals, that the trend found in the other experiments is largely due to an enhanced carbon cycle caused by CO2 fertilisation. Furthermore, different effects on the wetland methane emissions from rising temperatures seems to be canceling each other out, at least on the large scale.

Both experiments increase the potential wetland area, but they do so in different ways:

WetCrops: the drainage flag is disabled, so any cropland cell whose water table reaches the surface is diagnosed as a wetland (hydrologically). The vegetation remains the prescribed crop; no ecological conversion to wetland plants occurs.

The significant differences between the “Base” and “DrySurf” experiments highlight the sensitivity of the estimates of wetland extent and thus wetland methane emissions to changes in the soil hydrology. Including SWR dramatically increases the evapotranspiration and thus atmospheric water recycling in some regions. This also has major effects on the fluxes of latent and sensible heat and thus on regional temperature. These feedbacks emphasize the importance of also considering coupled (in this case land-atmosphere) simulations, since such additional water recycling (or lack thereof) is not accounted for in offline, land only, setups. As a result, offline sensitivity studies to changes in, e.g., soil hydrology characterizations may come to wrong conclusions. In similar offline experiments (not shown), forced with the Global Soil Wetness Project Phase 3 (GSWP3) data set , the dominating effect of enabling SWR was to decrease the water infiltration into the soil, thus leading to slightly reduced wetland areas and thus lower wetland methane emissions, completely hiding the changes in moisture recycling dominating the coupled experiments.

However, since model estimates of wetland area do not differ by a very large amount between the “Base” and “DrySurf” experiments, and the increase in precipitation tends to occur in regions, where the difference in precipitation between model and observations is relatively small, there is no clear conclusion, whether inclusion of the SWR parameterisation, as in the “Base” experiment, leads to an improvement in model results in terms of wetland areas and wetland methane emissions.

Compared to the TOPMODEL wetlands, the SWR scheme is another way of allowing super-surface water to form, which can be interpreted as wetlands. Both SWR and TOPMODEL-style wetlands arise in areas with high water availability and since both approaches are based on statistical methods, it is impossible to completely separate the two ways of obtaining surface water bodies.

The correlation between the SWR and TOPMODEL-wetland areas in the entire “Base” experiment is ≈0.3 without trend. The assumption that soils below SWR areas are oxic may thus be too simplistic. Since SWR almost exclusively influences North America north-east of the Mackenzie River, north eastern Europe and north western Siberia (Fig. S7), a (partial) lifting of this assumption would enhance the Arctic and Sub-Arctic wetland methane emissions compared to the tropical ones. By assuming that there is no overlap between the SWR and TOPMODEL wetland areas and that annual methane emissions are linearly dependent on the annual mean area with a wet surface, a rough estimate of the effect of the assumption, that the SWR areas do not contribute to methane production, can be obtained. In this case global wetland methane emissions would increase by 42 Tg(CH4)yr-1 or ≈25 %. This includes an additional 27 Tg(CH4)yr-1 of emissions from the NXT area, increasing the NXT fraction of the global emissions from 21 % to 30 %. Since the assumptions for this estimate are rather unrealistic, these numbers are, however, regarded as the upper extreme of the consequences of excluding the SWR areas from the methane production.

Using an AMIP setup introduces an atmospheric variability, which is different from the observed one. Therefore, comparisons of model results to observations and observation-based model results are only possible in a statistical sense over longer periods. In addition, in AMIP setups the atmospheric model may introduce less well-known biases compared to observational data sets. For this study, the most important known atmospheric bias is an underestimation of the tropical precipitation by about 29 % compared to GSWP3 (Table ). The tropical bias is unevenly distributed between the continents, and is by far strongest over Indonesia and South East Asia, which explains our comparatively low wetland areas and thus methane emissions from this region. The bias distribution agrees with findings of and seems thus to be a general property of the ICON-XPP atmospheric model. Common wetland products based on observations (GIEMS-2 , WAD2M ) find a significant seasonal cycle in the tropical wetland area more or less in phase with the northern boreal wetland areas. That in this study essentially no seasonal cycle is found can be attributed to the severe underestimation of south Asian and Indonesian rainfall and thus diminished wetlands in these regions. Therefore, the weak cycle in the Amazonian region can out-weigh the out-of-phase cycles in Asia and Africa, producing a net-zero cycle for the tropics.

The use of land use maps compared to land use transitions (which are currently not implemented in JSBACH4) is known to underestimate the influence of anthropogenic land use changes on the carbon cycle . Specifically the release of soil carbon is underestimated. Therefore, especially in the tropics, where shifting cultivation is a common agricultural practice, the carbon pools may be overestimated. This could also lead to an overestimate of the methane emissions from these regions. Since farmers generally try to avoid wet areas , we, however, assume the effect on the methane emissions to be small.

Assuming that all cropland areas in the model are drained is a strong simplification, especially due to the large-scale dynamics and development of agricultural practices. Former cropland areas may be abandoned and become either pastures or completely be given back to nature. This, of course, would not mean that drainage systems are removed, and thus the area drained is likely underestimated by the approach of this study. Also, managed forests, which in JSBACH are regarded as natural vegetation, pastures, and areas of urban expansion may be drained to enhance tree growth, making the land more suitable as grassland or to make the ground supportive for technospheric constructions . Again these effects will lead to an underestimate of drained area. On the other hand, not all croplands are drained, either because they are dry enough by nature or because of limited local resources to establish adequate drainage systems. As an example, states that in present-day India, only about 10 % of croplands are drained. This will lead to an overestimate of the drained area, but likely not change the simulated wetlands much, since the extra drained area will largely be too dry to regularly become wetlands, anyway. Though reports a history of several thousand years of anthropogenic drainage, argue that the main wetland area loss due to drainage takes place after 1900 and is still ongoing at a more or less constant rate. Thus, our assumption of drained crops may be most representative for the later part of the simulation period. With the exception of the NXT region in the “ConstVeg” experiment, wetland areas are increasing after 1980 in all of our experiments for both NXT and the tropics. One reason may be different developments in the precipitation and precipitation patterns. GSWP3 and the ICON-XPP atmosphere agree that in the 1980–2014 period, the global (land) precipitation is on average enhanced (≈0.48,0.43 and 0.32mmyr-2 for GSWP3, “Base” and “DrySurf”, respectively). However, the global patterns of increasing precipitation in ICON-XPP correlates better with our simulated wetlands (somewhat better in “Base” than in “DrySurf”), which may thus be enlarged. Nevertheless, it is unlikely that this better pattern correlation alone can explain the trends in wetland area going in different directions in our simulations and observations. It is more likely that we, despite the drainage assumption, largely underestimate the influence of human water management (dams, reservoirs and drainage, also of non-croplands). In total we expect our simulated drained area to be underestimated (and the wetland areas therefore overestimated) in regions with intense and long-lasting anthropogenic influence.

Data-based estimates of global methane emitting areas are scarce, though a few map-based estimates of wetland extent and remote-sensing estimates of surface inundation are available. estimated the global annual maximum extent of lakes and wetlands based on maps, and show a global extent of methane-relevant wetlands of between 4.5 and 7 Mm2, depending on which categories of wetlands are considered relevant for methane. developed a remote-sensing-based data set of surface inundation, estimating an annual mean extent of 2.5 Mm2, after removal of rice-growing areas based on , and a JJA extent of 4.1 Mm2. This estimate includes the areas of lakes, reservoirs, and rivers, though, which are not considered methane-emitting areas in our model. Excluding those areas, estimate a wetland extent of 4.4 Mm2 in the annual mean, with a JJA extent of 5.2 Mm2. Finally, estimate an annual mean extent of 3.5 Mm2 for inundated and saturated wetlands and peatlands, with an annual maximum extent of 7.7 Mm2.

Our model estimate yields between 5.3 Mm2 in the “DrySurf” experiment and 6 Mm2 in “WetCrops”, both for the JJA extent and the 1980–2014 period (Fig. ), thus determining wetland extents generally in line with data-based estimates. However, a large uncertainty in the estimates of wetlands, and especially methane-emitting areas, has to be acknowledged. The TOPMODEL-based approach we are using determines the extent of (super-) saturated soils, and considers those as methane-emitting. This may, however, underestimate the extent, as some categories, for example peatlands, cannot be considered well in this approach.

Furthermore, the ICON-XPP model shows a precipitation bias over the tropics, specifically in South East Asia, likely leading to too low an estimate of wetland extent for this region in our experiments. Also, JSBACH is not equipped with a detailed river flow model capable of simulating temporary flooding of river flood plains as a consequence of heavy precipitation or snow melt events. Since such wetlands tend to be short-term phenomena, it is questionable, however, whether the soils have time to become anaerobic and thus they would likely not contribute significantly to the wetland methane emissions.

Excluding short-term wetlands in dry areas by applying the dynamic criterion in Eq. (), instead of a static mask, allows the model to capture the influence of major climatic shifts. This exclusion, the other hand, might also cause spurious trends in wetland area and thus in wetland methane emissions by changing the wetland-allowance mask, either temporarily or as part of a long-term trend. By selecting the required ratio as low as 0.3 (Eq. ), the bounds of where wetlands are permitted are located far away from where major wetlands are predicted. Thus, the wetland methane emissions and their trends are likely not influenced by the details of this masking.

In the “ConstCO2” experiment the trend in wetland methane emissions is much smaller than in “BestGuess”. Over the entire experiment period it is even negative, following the downward trend in wetland area. This suggests that the main driver of the trend found in all other experiments is an accelerated carbon cycle stemming from CO2 fertilisation. This contradicts the findings of both and the WETCHIMP model ensemble , who attribute their trend to a temperature-driven increase of wetland area and microbial methane production rates.

For the period 2000–2012 the trend in global wetland methane emissions is about 0.4 Tg(CH4)yr-2, of which 75 % stem from the tropics. This is higher than the numbers presented in , who report a multi-model mean of 0.2 Tg(CH4)yr-2, and about 10 % of a trend of 2.2 Tg(CH4)yr-2 from top-down inversions attributed to non-anthropogenic sources. Since the productivity in JSBACH is known to be comparatively sensitive to changes in the atmospheric concentration of CO2, which is the driver of our trend in methane emissions, it appears likely that our predicted wetland emission trend is at the high end.

The basic TOPMODEL assumption defining wetlands as areas with super-saturated soils (i.e., the water level is at least reaching the surface) and using the wetland fraction as a measure of the soil oxygen content are simplifications, which can only serve as a first guess. For instance soils, where the root zone of macrophytes is saturated, may be anoxic. Thus, though the soil column as a whole is not saturated, such areas could be a source of methane, mainly since produced methane may bypass the upper (oxic) soil layers through aerenchyma. On the other hand, soils take time to become anoxic after being inundated. Depending on the state of the soil at inundation, this may take hours to weeks , and the anaerobic area could thus potentially be overestimated by assuming all wetland soils to be anoxic. Further, wetland areas may have very different and highly heterogeneous methane production and emission rates, as shown for Arctic ponds by .

Of course, the tropical precipitation bias reducing our tropical wetland area tends to decrease the importance of the tropical wetland methane emissions. There are, however, also effects which draw in the other direction: In this study, we accounted only for the conversion of contemporary organic matter to methane, while in reality an additional source may be added from the huge amount of ancient organic carbon stored in the thawing permafrost , as well as from degrading peatlands , since we include no explicit peatland model such as HIMMELI . found that incorporating micro-scale ground subsidence following permafrost thaw will increase the wetland fraction and thus methane emissions. Both these effects would enhance the importance of the Arctic regions in the global methane budget. Even though our wetlands are in general smaller, the modeled NXT fraction agrees with the more detailed wetland methane model of .

Offline (stand-alone) land-surface simulations are valuable for isolating the response of wetland hydrology to prescribed climate, but they cannot capture feedbacks such as (i) moisture recycling that modifies precipitation patterns, and (ii) vegetation-mediated changes in surface energy fluxes. Conversely, online coupled simulations may inherit atmospheric biases (e.g., the tropical precipitation deficit in ICON-XPP) that propagate into the wetland response. Therefore, both experimental designs are complementary: offline experiments help to diagnose process sensitivities, while online experiments are required to assess the full climate-wetland feedback loop.