3.1.1 Peatland area

The modern peatland distribution simulated by LPX-Bern (standard run) compares well to the distribution given by PEATMAP (see Fig. 1 and Table 1). LPX and PEATMAP yield a very similar global peatland area, with respective values of 4.37 and 4.23 M km². The same is true for the latitudinal distribution. LPX simulates 3.2 M km² in the high latitudes (>30^∘ N) and 1.15 M km² in the tropics (30^∘ S–30^∘ N), whereas PEATMAP gives 3.18 and 0.99 M km², respectively. This broad-scale agreement between LPX and PEATMAP notably emerges without any tuning of LPX against PEATMAP. These results are similar to previous results using the LPX (Stocker et al., 2014) with slightly larger tropical peatland coverage in the current study. Differences are due to a new model version after data assimilation, LPX v1.4, (Lienert and Joos, 2018) and the additional model changes described in Sect. 2.1.

Minor and major differences in peatland area between LPX and PEATMAP are seen on the local to regional scale (Fig. 1 and Table 1). In the tropics, LPX simulates more peat in South America and Southeast Asia than PEATMAP indicates. Compared to the estimate of Gumbricht et al. (2017), the LPX peatland extent is similar for South America and a factor of 2 smaller for Southeast Asia. The vast peatland complex in the Congo Basin is almost absent in LPX. In the northern mid to high latitudes, LPX seems to underestimate European peatland area by a factor of 2 and slightly overestimates peatland area in northern Asia, mostly west and east of the western Siberian lowland (WSL) peat complex. In North America, LPX simulates more peat in Alaska and Quebec and less in Western Canada than PEATMAP.

Other modeling studies present results from prognostic simulations of Northern Hemisphere peatlands. Kleinen et al. (2012) simulated peatland dynamics and carbon accumulation over the past 8000 years using the coupled climate carbon cycle model CLIMBER2-LPJ. However, no quantitative results in terms of peatland area were reported. Qiu et al. (2019) used the ORCHIDEE-PEAT DGVM (Qiu et al., 2018) to simulate northern (>30^∘ N) peat expansion over the Holocene. Their simulated northern present-day peatland area is, at 3.9 M km², slightly larger than in LPX. They find similar regional discrepancies between simulated and observation-based peat area in North America, northern Europe, and Asia, as described above for LPX. Peat area dynamics in ORCHIDEE-PEAT are also using the TOPMODEL approach following Stocker et al. (2014), with some different expansion criteria. This might indicate that these discrepancies could have their source in the TOPMODEL approach and its limitations. Another major source of uncertainties is in the climate data used to force LPX (see also Sect. 3.2.2). In particular, precipitation data show large discrepancies between available observational products (Sun et al., 2018).

Table 1Peatland area for different regions and latitudinal bands as given by PEATMAP (Xu et al., 2018) for today and peatland area and their carbon stocks as simulated by LPX-Bern for the preindustrial (PI) period and the Last Glacial Maximum (LGM) in the transient “standard” setup simulation from 22 kyr BP to present. The extent of the northern Asia, western Siberian (WS) lowland, and Southeast Asia regions are shown in Fig. 1.

Download Print Version | Download XLSX

3.1.2 Peatland carbon

Total peat carbon estimates are closely linked to the estimates for area and, thus, inherit their uncertainties. Additional assumptions on bulk density and peat depth introduce additional uncertainties. Therefore, the ranges of both carbon estimates and area estimates are large (Gorham, 1991; Turunen et al., 2002; Yu et al., 2010). The research bias allows for more constrained estimates in well-studied regions such as Europe and North America and less constrained estimates in the tropics and northern Asia. Estimates for northern peatlands range from 270 to 604 GtC and were obtained using various methods and area estimates (see Yu, 2012 and Yu et al., 2014 for a review). The modern carbon inventory of northern peatlands simulated by LPX at the end of the transient standard run from the LGM until present is, at 361 GtC, well within this observational range. In the tropics, LPX simulates a peat carbon inventory of 136 GtC which is substantially larger than classical estimates from the literature that range from 50 to 87 GtC (Yu et al., 2010; Page et al., 2011). These, however, also assume a substantially smaller tropical peatland area than LPX or PEATMAP suggest (see Sect. 3.1.1). Including estimates for the newly discovered peat in the Congo Basin, Dargie et al. (2017) estimate a tropical peat inventory of 69.6–129.8 GtC, which is closer to the LPX results. Gumbricht et al. (2017) calculate an even larger area than LPX, and combining their area estimate with the peat properties assumed by Page et al. (2011) would result in a tropical peat inventory of 350 GtC.

Previous studies with LPX-Bern reported somewhat different carbon inventories than given here. Stocker et al. (2014) reported 460 and 88 GtC for northern and tropical peatlands, respectively. Differences stem from an updated model version, also resulting in different areas as mentioned in the previous section. Additionally, their carbon stocks were the results of an accelerated spin-up scheme, whereas the pools in this study are filled over a transient run. Spahni et al. (2013) also report northern peatland carbon stocks after a transient LPX run from the LGM, although with prescribed not prognostic peatland area. Their simulation resulted in 365 GtC stored in northern peatlands.

Other model studies with dynamic peatland area reported 317 GtC after an 8 kyr Holocene run (Kleinen et al., 2012) and 463 GtC after a 12 kyr Holocene run (Qiu et al., 2019) in northern peatlands.

3.2.1 Peatland distribution and carbon storage

Under LGM conditions, global simulated peatland area and carbon inventories are reduced compared with the preindustrial period (Table 1). Globally, simulated peatland area and the peat C inventory are 38 % and 45 % smaller at the LGM than during the PI period, respectively. This reduction is dominated by the northern extra-tropics, where peat extent and the C inventory are almost 60 % smaller at the LGM than during the PI period. In contrast, the peat C inventory in the tropics is only about 3 % smaller, and the tropical peat area is even 7 % larger at the LGM than during the PI period. This difference in the tropics is mostly linked to large peatlands simulated on flat exposed continental shelves in Southeast Asia at the LGM, which were subsequently flooded during the deglaciation. Another modeling study by Kaplan (2002) also suggests extensive wetlands on the flat Sunda Shelf, but reconstructions of Indonesian peatlands suggest that vast peat presence in Indonesia during the LGM is unlikely (Dommain et al., 2014). Establishment of now existing inland peatlands seems to be connected to rising sea level (Dommain et al., 2011). In sediment cores from the now submerged Sunda Shelf, there is little evidence of peatlands during the LGM (Hanebuth et al., 2011). Dommain et al. (2014) suggest that the shelf, although with a small topographic gradient, had an effective drainage system with deeply incised river valleys, preventing the formation of large wetlands. Both the hydrological feedback of rising sea level and deep river systems are not represented in LPX and, thus, might limit the models ability to reproduce peat and wetland dynamics in this region correctly.

Simulated peatland coverage in northern mid and high latitudes is smaller and is shifted southwards at the LGM compared with the PI period. Ice sheets covered large parts of Europe and North America during the LGM preventing vegetation and peat from growing. However, peat is also mostly absent in northern Asia and the WSL due to the substantially colder and dryer conditions compared with today. On the other hand, large peatland complexes are simulated along the southern ice sheet margins in North America and in Europe (Fig. 2), in regions where modeled peatlands are mostly absent under current conditions. This even leads to a simulated net increase of peatland area in Europe (+43 %) compared with the present. Verifying the existence of these extensive LGM peatlands that do not exist under present conditions (compare Fig. 1) is difficult, as existing compilations of peat core dates focus almost exclusively on today's existing peatlands (MacDonald et al., 2006; Gorham et al., 2007; Yu et al., 2010; Loisel et al., 2017). In a recent study, Treat et al. (2019) presented a compilation of dated buried peat deposits and simulated peatland area and carbon stocks. Their simulation also suggests large midlatitude peatlands in North America in agreement with our results. Their peat deposits data for the LGM (Fig. 2; dots), together with pollen analyses suggesting the presence of at least some Sphagnum in eastern North America (Halsey et al., 2000; Gajewski et al., 2001), provide plausible evidence for the existence of midlatitude LGM peatlands in North America and Europe. Their extent, however, is probably overestimated in our simulation. Comparisons between the North American LGM hydroclimate in TraCE21k and proxy reconstructions have resulted in a poor skill score, especially in eastern North America (Lora and Ibarra, 2019). Bad performance in this region is shared with all PMIP3 models (Lora and Lora, 2018).

Figure 2Peatland distribution at the Last Glacial Maximum (LGM) as simulated by LPX-Bern in the standard setup (a), agreement (as number of models simulating peat in given grid cell) between LGM time slice simulations run with LPX and forced with different climates anomalies from six PMIP3 models as well as the TraCE21k anomaly (b), and the squared correlation coefficient for a linear regression between physical properties of the different time slice simulations (precipitation minus evapotranspiration, P−E, and growing degree days above 0 ^∘C) and peat fraction in the respective cells. Only cells with significant correlation (p>0.05) are plotted. The colored shading in panel (b) indicates how many time slice simulations show a peat fraction of >0.05 in the respective cell. The color code in panel (c) denotes the dominant predictor in the respective cell. Dots in panels (a) and (b) show buried and still active peat deposits that indicate active peat accumulation during the LGM (24.5–17.5 kyr BP). Peat core data are from Treat et al. (2019).

3.2.2 Uncertainties from climate forcing

The peat distribution as simulated by LPX-Bern for the LGM and the past 20 000 years is subject to many uncertainties. Uncertainties arise from model parameterizations, not only in the peat module but through all components of the model, and are often hard to quantify. Data assimilation, as done recently for the LPX in Lienert and Joos (2018) to constrain model parameters, is an approach to improve model performance in the light of uncertain key parameters. Another source of uncertainty stems from uncertainties in the prescribed forcings. The orbital parameters, atmospheric CO₂ mixing ratio, and land–sea–ice mask for the LGM and their deglacial evolution are all well constrained for the purpose of peat modeling, in contrast to the climate anomalies. Although there are paleoclimate reconstructions for the LGM (Bartlein et al., 2011; Schmittner et al., 2011; Annan and Hargreaves, 2013), 6 k (Bartlein et al., 2011) and the last millennium (Hakim et al., 2016; Tardif et al., 2019), they lack the temporal resolution and/or spatial coverage needed for a global transient simulation from the LGM to present. Climate models can fill these gaps; however, climate anomalies are model dependent, and model performance differs between variables, regions, and the simulated time period (Harrison et al., 2014). These differences in climate models have been shown to propagate large uncertainties into carbon cycle projections (Stocker et al., 2013; Ahlström et al., 2017).

We assess the uncertainty in peatland area and peat carbon stemming from climate forcing uncertainties. Climate anomalies from seven different models are used to force the LPX into seven different LGM states (see Sect. 2.2). This yields a very wide range for global mean inundated area (2.6–3.6 M km²), peat area (1.5–3.4 M km²), and peat carbon (147–347 GtC). Interestingly, simulated wetland and peatland area and the peat C inventory for the 21 kyr BP period are also substantially different between the standard transient simulation using temporally evolving climate anomalies from the TraCE21k simulation compared with the time slice simulation forced with TraCE21k anomalies (Fig. 3, gray star versus gray dot). This highlights both the influence of different methods of input preparation, with slightly different treatment of anomalies and an interannual variability taken from TraCE21k in the transient simulation and from CRU 3.1 for the time slice, as well as the importance of memory effects for a slowly reacting system such as peatlands.

Agreement on the simulated peat extent among the seven simulations differs among regions (Fig. 2b). It tends to be higher in the tropics and East Asia and lower in North America, Europe, and western Siberia. Differences in temperature and precipitation anomalies propagate into differences in the water balance and productivity, partly limited by growing season length, and thus into differences in peat abundance and extent.

A statistical analysis of the differences in climatic drivers and simulated peat area reveals regionally different mechanisms (Fig. 2c). Temperature, precipitation, precipitation minus evapotranspiration (P−E), and growing degree days over 0 ^∘C (GDD₀) are considered as climatic predictor variables for the peat fraction within a grid cell. We correlated, for each grid cell, the seven climatological mean values of a selected predictor with the modeled peat fraction from each of the seven time slice simulations. P−E and GDD₀ show significant correlations (p<0.05) in more grid cells than precipitation and temperature, respectively. Both moisture balance and GDD₀ have been shown to be among the most important predictors of northern peat initiation and carbon accumulation in the past (Morris et al., 2018; Charman et al., 2013). In LPX the water balance, influenced by P−E, and the carbon balance, influenced by temperature and growing season length, define thresholds on peatland existence and size. In eastern Europe, differences in peat extent between the seven LGM time slice simulations are mostly driven by differences in local precipitation anomalies driving P−E. The same is true in the tropics, with MRI-CGCM3 and IPSL-CM5A-LR being the driest models with the least tropical peatlands and TraCE21k and COSMOS-ASO being the wettest with the largest tropical peatlands. However, in parts of central South America, temperature is the dominant predictor signaling a fragile carbon balance, where peat presence in some of the model climates is possible because of cooler conditions and, thus, reduced respiration. In the south of North America, moisture balance, with contributions of both P and E, is the dominant determinant of the inter-model differences. The time slice forced with TraCE21k climate shows the peatland distribution in North America more shifted to the east compared with most other PIMP3 forcings alongside warmer and wetter conditions (see also Lora and Lora, 2018). The peatland extent in the north of North America is sensitive to temperature differences with longer growing season allowing for increased productivity and, therefore, peat formation. In the MRI-CGCM3 time slice, temperature anomalies with respect to the preindustrial period are lowest and peat is subsequently shifted northwards compared with other time slices. Similar is true for northern and East Asia where lower temperature anomalies allow for more peatlands. Large areas in central Europe, East Asia, and South America show differences in peatland extent induced by differences in climate forcings, but no significant correlations between peat fraction and predictor variables are found. This might be the result of nonlinear interactions and threshold behaviors not captured by our linear regression approach. Taken together, these findings demonstrate a strong sensitivity of simulated peat extent and the C inventory to the prescribed climate fields and a strong dependence of the results on the choice of climate model output used to force LPX. In other words, caution is warranted when interpreting model results for times and regions in which proxy records or observations are sparse and have limited power to constrain the actual climate conditions. This holds not only in the context of this study but also for global peat and carbon cycle model studies in general.

Figure 3Spread of the global annual mean inundated area (meanIA), peatland area, and peatland carbon during the LGM (21 kyr BP) in time slices forced with climate anomalies from seven different climate models (see Sect. 2.2). The box, the solid line, and the dashed line indicate the interquartile range, the median, and the mean from the seven simulations, respectively. The star indicates the values of the transient simulation for the same time period using climate anomalies from TraCE21k.

Download

3.3.1 22–17.43 kyr BP

Global changes in peatland area and carbon before the onset of the Heinrich Stadial 1 (HS1) are small, due to the relatively small changes in the main drivers. There is initial carbon loss in some regions of the tropics, due to some grid cells still approaching equilibrium after the spin-up (Fig. A1). North America already sees an accelerating carbon accumulation with unchanging area before the HS1, driven mostly by increasing temperature. Carbon and area also increase in Europe with large temperature-driven fluctuations

Figure 6Drivers of the change in peatland area from LGM to present. Colors indicate the most important driver, and colored shading shows the contribution of the respective driver on a scale from 0 (no contribution) to 1 (only contributor).

3.3.2 17.43–11.65 kyr BP

Three main features characterize the peat area evolution over the last glacial termination: (i) a northward shift in the distribution of northern extra-tropical peatlands, including peat expansion in northern Asia; (ii) dipole-like north–south shifts in tropical South America, associated with north–south shifts of the rain belts of the intertropical convergence zone (ITCZ); and (iii) flooding of peatlands on continental shelves, mostly in Southeast Asia, due to rising sea levels.

The last termination represents the transition of the climate system from the last glacial to the current interglacial, accompanied by large warming, ocean circulation changes, and an increase in atmospheric CO₂ (Monnin et al., 2001; Shakun and Carlson, 2010; Ritz et al., 2013). The termination is divided into the Heinrich Stadial 1 (HS1, 17.43–14.63 kyr BP) Northern Hemisphere (NH) cold period, the Bølling–Allerød (BA, 14.63–12.85 kyr BP) NH warm period, and the Younger Dryas (YD, 12.85–11.65 kyr BP) NH cold period (Rasmussen et al., 2014). These NH cold–warm swings are associated with a large-scale reorganization of ocean circulation, which is thought to have been provoked by freshwater release from ice sheet melting leading to changes in the ocean heat transport (Stocker and Johnsen, 2003). With changing low- to high-latitude temperature gradients, the ITCZ shifted and with it the high precipitation zones in the tropics (McGee et al., 2014; Shi and Yan, 2019; Cao et al., 2019). These climate dynamics are well captured by the transient TraCE21k simulation (Liu et al., 2009).

The responses of peatlands in LPX to these climatic changes are drastic. Large shifts in peatland area start to set in at the onset of the HS1 and increase into the BA. During the BA, peatlands show the fastest gross positive and negative area changes throughout the simulation (see Fig. 4b). In the northern mid and high latitudes, peatlands shift north and eastward (see Figs. 5, A2). Peatlands disappear in midlatitude North America and Europe, and new peatlands emerge at higher latitudes and in cold continental regions of Asia. These new peatlands include the large peat complex in the western Siberian lowland (WSL) region. Some of the peatlands established in northern Europe during HS1 vanish again during the BA. The described changes are driven by the TraCE21k climate which shows a substantial warming and wetting of the Northern Hemisphere beginning during the HS1. Temperature is the dominant driver for peat loss and expansion in Europe and North America. The loss of old peat is especially abrupt in North America (see Fig. A1b). Here, precipitation decreases and temperature increases abruptly over southwestern North America at 13 870 BP. Both changes decrease the water balance given by P−E which leads to a decrease in potential peat area and, thus, loss of the previous extensive peat complexes. As this abrupt climate change occurs at a discontinuity of the trace boundary conditions (changes in ice sheet configuration and freshwater forcing), the speed of this change is probably drastically overestimated. In northern Asia both temperature and precipitation drive the peatland expansion. This expansion sees a pronounced halt during the YD where the Northern Hemisphere climate briefly returns to more glacial conditions (see Fig. A2e).

In the tropics, the area and carbon changes are mostly driven by precipitation changes. Largest changes are simulated in South America, where precipitation patterns respond to changes in ITCZ position (see Fig. 5). During the HS1 and the YD where the Atlantic meridional overturning circulation (AMOC) is in a reduced state, peatland area shifts to the south following the southward shift of the ITCZ. During the BA the AMOC is strong and precipitation and peatland area shift back north. In Africa half the peatland area is lost during the BA mostly driven by drying. In Southeast Asia peatlands are lost over the whole termination due to precipitation changes and the onset of sea level rise, which starts to flood the large continental shelves at about 16 kyr BP.

The shifts in peatland distribution result in a similar global peatland area at the beginning of the Holocene compared with the LGM and at the onset of the HS1. However, much less carbon is stored in active peatlands at the beginning of the Holocene than during the LGM and at the onset of the HS1. Thus, the carbon density per unit area is much lower for the newly established peatlands than for the lost LGM peatlands.

3.3.3 11.65–0 kyr BP

Modeled peatlands in the Holocene show a continuous net expansion in the northern extra-tropics, with newly forming peatlands more than balancing the loss of peatlands elsewhere. The Holocene experienced relative stability in climate and CO₂ levels compared with the termination. The early to mid Holocene was likely characterized by warmer summer temperatures than the preindustrial period with a larger seasonality in the Northern Hemisphere (Marcott et al., 2013; Liu et al., 2014; Samartin et al., 2017). Ice sheet retreat and sea level rise lagged behind the deglacial temperature increase and was mostly completed at about 7 kyr BP (Peltier, 2004). Locally, new land keeps emerging to this day due to isostatic rebound. This effect is especially pronounced in the Hudson Bay lowlands, where new land emerges at a rate of up to 12 mm yr⁻¹ (Henton et al., 2006).

For northern peatlands, positive area changes are consistently larger than negative changes throughout most of the Holocene (see Fig. 4a). This leads to a large continuous area expansion. Old peatland area is also simulated to increase continuously during the Holocene (Fig. 4b), showing that the parallel positive and negative changes are more than mere fluctuations of existing peat but that there is actual continuous peatland loss and growth. Net area increase picks up at about 9 kyr BP, decreases in the late Holocene, and turns into a net area reduction in the last millennium. This late Holocene slowdown and reversal of net peatland area growth is most pronounced in northern Asia where increasing negative changes start to balance and eventually offset the still large positive changes (Figs. A1 and A2). Here, both negative and positive dynamics are driven by temperature and precipitation. The early, fast expansion in northern Asia is offset by a temperature-driven net area loss in Europe which is recovered partly towards the late Holocene. Net area increase in North America is delayed by continued loss of midlatitude peat and the slow retreat of the Laurentide Ice Sheet, which limits the establishment of new peatlands. Today's peatlands in North America start to establish after about 9 kyr BP with most of today's peatlands forming between 7 and 2 kyr BP. Carbon stocks follow these regional trends but with larger relative increases especially towards the late Holocene. As the timescale for building up carbon pools is generally much longer than the timescale of potential area changes, fluctuations in area, mostly by young peatlands, are smoothed in the carbon stocks (see, e.g., Fig. A1b).

The tropical peatland area is simulated to stay relatively stable throughout the Holocene, with positive changes balancing negative changes. Southeast Asia sees a reduction in area in the early Holocene due to continued sea level rise and a subsequent gradual recovery of integrated peat area driven by precipitation and nonlinear effects. Peat area in South America increases slightly over the Holocene with mostly precipitation-driven fluctuations in between. On the other hand, fluctuations in region-integrated peat carbon stocks are largely absent in South America, as carbon, with changing area, is shifted between peat and old peat pools. Peat carbon stocks in South America show a large relative increase following a near-linear path. Africa sees an increase in area at about 10 kyr BP driven by precipitation and enabled by high CO₂ concentrations. The new area gradually degrades again until 3 kyr BP with another peak at 0.5 kyr BP.

3.3.4 Model versus reconstructions

The study of peatland initiation, life cycle, dynamics, and responses to external forcing has been focused on today's existing and active peatlands. This work includes large compilations of peat core basal dates (MacDonald et al., 2006; Gorham et al., 2007; Yu et al., 2010) which are used to reconstruct initiation dates and lateral expansion (Yu et al., 2010; Korhola et al., 2010; Dommain et al., 2014; Loisel et al., 2017) of the sampled active peatlands. This approach, however, does not include earlier peatlands that dried out, were buried or flooded, or otherwise ceased to be active accumulating peatlands. Treat et al. (2019) presented a first compilation of dated buried peat layers, but the small sample size make quantitative reconstructions difficult. Thus, we limit most of the model–data comparison of the transient behavior to today's existing peatlands. Figures 7 and 8c show modeled initiation date frequency, area, and carbon dynamics of northern peatlands that are still active at present.

Figure 7a compares LPX results to a gridded “oldest age” dataset compiled by Loisel et al. (2017), and Fig. 7b and c compare LPX results to two different reconstructions for lateral expansion (Loisel et al., 2017; Korhola et al., 2010) based on similar methods but different underlying peat core datasets (see Sect. 2.3). The two reconstructions for peat expansion agree on a limited pre-Holocene expansion, but they disagree substantially on the timing of fastest expansion during the Holocene (Fig. 7). Both simulated initiation and peat expansion have peaks about 4 kyr earlier than the reconstructions. The model simulates early initiation of today's northern peatlands, already beginning in HS1, and a large expansion during BA. The reconstructions, on the other hand, suggest lateral peat expansion picking up only with the transition into the Holocene. Agreement between model and reconstructions becomes good in the mid to late Holocene.

The early expansion in the model also propagates to the carbon balance for presently active peatlands. The model simulates earlier accumulation extending into the HS1 and slower accumulation during the early Holocene than suggested by net carbon balance (NCB) reconstructions by Yu (2011) (Fig. 7c). The summed simulated carbon increase from the LGM to the PI period in today's northern peatlands amounts to 313 GtC (Fig. 8c).

The early expansion of northern peatlands in the simulation is mostly dominated by peat establishment in western Siberian lowland (WSL) region and northern Asia in general (see Sect. 3.3.2 and Fig. A1). The dominant drivers of this expansion are temperature and precipitation, which, according to TraCE21k, both increase substantially over northern Asia during the HS1 and BA. A similar simulated early expansion into the WSL was reported by Treat et al. (2019), with the coupled CLIMBER2-LPJ setup. Morris et al. (2018) investigated possible climatic drivers for peat initiation in a modeling study using the HadCM3 model. They suggest that the WSL responded to an increase in effective precipitation at about 11.5 kyr BP, instead of the early warming. One source of the model data mismatch could lie in the uncertainties in climate anomalies discussed in Sect. 3.2.2. Especially at high latitudes, climate anomalies can vary greatly between climate models, and model performance at one point in time does not always correspond to performance at another point in time (Harrison et al., 2014). Although the freshwater change in TraCE21k was designed to capture the rapid climate events during the glaciation, the magnitude and timing of regional or even hemispheric changes can still have large biases. To date, TraCE21k is the only available transient GCM simulation, but new simulations under the umbrella of PMIP4 might shed more light on the model dependence of the warming pattern in question (Ivanovic et al., 2016). Another source of the mismatch could lie in the simple representation of peatlands in the model, which might be unsuitable to reproduce specific initiation and expansion pathways, like terrestrialization and fen–bog transition that might have been important controlling factors in that time and region (Kremenetski et al., 2003). One example could be the relative weakness of the initiation criteria on the moisture balance (precipitation over evapotranspiration >1), which is almost always weaker than the indirectly mediated condition on inundation persistence. This might pose a problem, especially in the WSL where moisture balance might have been the driving factor for peat initiation (Morris et al., 2018). Lastly, although the WSL is relatively densely sampled and reconstructions of peat initiation are robust, other areas of northern Asia are vastly under sampled and reconstructions are less reliable.

Throughout the tropics dated buried and active peat cores already show peatland presence during and preceding the LGM. However, peatland extent or evolution towards the presence are not well constrained and are subject to large uncertainties. The small number of available dated tropical peat cores impedes a statistical approach. Applied nevertheless, it indicates a more or less continuous growth of today's peatlands since about 19 kyr BP with the largest expansion rates between 8 and 4 kyr BP (Yu et al., 2010). Dommain et al. (2014) reconstructed the evolution of Indonesian peatlands using a combination of dated cores and a transfer function between depth and age. They argue for a peat expansion much later than that inferred by basal ages alone, with 90 % of today's peat establishing after 7 kyr BP and 60 % after 3 kyr BP. In this study, the dominant control on peatland area was found to be local sea level. Rising sea level during the termination and the early Holocene triggered the establishment of inland peatlands through alterations in moisture availability and the hydrological gradient, and the stabilization of sea level and subsequent sea level regression after 4 kyr BP prompted the establishment of coastal peatlands. In contrast to these reconstructions, the transient simulation shows that 60 % of today's tropical peatland area is already present in the LGM and only small expansion during the recent millennia. The sparsity of the data warrants caution when comparing reconstructions to model results. However, in Southeast Asia, this discrepancy could indicate the importance of the feedback of sea level on local hydrology, which is missing in LPX.

Figure 7Simulated and reconstructed dynamics of today's existing northern peatlands: (a) peatland initiation frequency (two overlapping histograms), (b) peatland area expansion and (c) expansion rate, and (d) the net carbon balance (NCB) normalized by respective estimates of today's carbon pool. The reconstruction datasets are described in Sect. 2.3. The background colors indicate different time periods, as in Fig. 4

Download

3.3.5 Transient carbon balance of peatland soils and the land biosphere as seen by the atmosphere

In this section, we address how carbon stored in soils of active peatlands and carbon stored in the remains of former peat soils changed over time. Thus, we quantify the overall contribution of peatland soils and peat carbon to the changes in the global carbon inventory of the land biosphere.

When trying to quantify the net effect of peatlands on the atmosphere, looking only at carbon stored in today's active peatlands can be misleading. Former active peatlands have transformed into other landscapes. Organic-rich peat layers may now be buried under mineral soils on land or in coastal ocean sediments (Treat et al., 2019; Kreuzburg et al., 2018). When analyzing the transient carbon balance of global peatlands such “old peat carbon” pools have to be considered.

Figure 8Carbon on global land (soil + litter + vegetation) (a); on active peatland areas, old peatland areas, and on flooded continental shelves (b); and in today's active peatlands (c). The brown line in panel (b) represents global carbon that originated from peatlands, even if it is not part of an active peatland anymore. The background colors indicate different time periods, as in Fig. 4

Download

Figure 8b shows the temporal evolution of carbon stored in soils of active peatlands, old peat soil carbon remaining on former peatlands, and old peat carbon stored on flooded continental shelves. Here, the old peat pools presented exclusively include the carbon from the organic-rich layers of formerly active peatlands that remain after accounting for decomposition over time (see Sect. 2.1). In the PI period, 499, 139, and 22 GtC of peat carbon are stored in their respective pools. The total simulated increase in peat carbon from the LGM to the PI period within these three pools is 351 GtC. This represents the simulated net carbon accumulation of global peat and, thus, the net amount of carbon sequestered from the atmosphere by peat. When only considering carbon stored in active peatlands, we would underestimate the deglacial peat carbon change with a value of 224 GtC. On the other hand, if we only consider the carbon stored in today's active peatlands, the inferred deglacial change amounts to 365 GtC (Fig. 8c), and we would overestimate the net peat accumulation since the LGM. While the latter difference in net peat carbon accumulation between the complete and incomplete accounting scheme appears small for the total deglacial change, the difference can be substantial and relevant for other periods. For example, a particular large difference is identified for the phase of high peatland expansion and loss rates as simulated from the Bølling–Allerød to the Preboreal in our model. Here, the carbon balance is given as 12 GtC when including old peat, versus 102 GtC when only looking at the carbon in today's active peat for the period from 14.6 to 10 kyr BP.

Peatlands contribute about 40 % to the total land biosphere carbon increase of 893 GtC. The result for the total land carbon increase between the LGM and the PI period is in good accordance with a recent estimate, integrating multiple proxy constraints (median: 850 GtC; 450–1250 GtC ± 1 standard deviation) (Jeltsch-Thömmes et al., 2019). The model also simulates the total change of the land biosphere carbon inventory between the beginning of the Holocene and the preindustrial period in reasonable agreement with the reconstruction by Elsig et al. (2009). The simulated temporal evolution, however, is different, with a rapid uptake in the early Holocene in the reconstruction compared with a delayed uptake in the mid to late Holocene in the simulation.