Pathway-dependent fate of permafrost region carbon

Permafrost soils in the high northern latitudes contain a substantial amount of carbon which is not decomposed due to frozen conditions. Climate change will lead to a thawing of at least part of the permafrost, implying that the stored carbon will become accessible to decomposition and be released to the atmosphere. We use a land surface model to quantify the amount of carbon released up until 2300 and determine the net carbon balance of the northern hemisphere permafrost region under climate warming following the RCP scenarios 2.6, 4.5, and 8.5. Here we show for the first time that the net carbon balance of the permafrost region is not just strongly dependent on the overall warming, but also on the CO2 concentration pathway. As a result moderate warming scenarios may counterintuitively lead to lower net carbon emissions from the permafrost region than low warming scenarios.


Introduction
Permafrost soils in the high northern latitudes contain substantial amounts of carbon (C) which are not accessible to decomposition due to frozen conditions. Current estimates put the C content of soils in the northern circumpolar region in the order of 1300 PgC, with a substantial fraction of this carbon stored in perennially frozen ground (Hugelius et al 2014). However, climate change will lead to thaw of at least part of the perennially frozen permafrost (Koven et al 2013), implying that the stored carbon will become accessible to decomposition and potentially be released to the atmosphere.
The potential carbon release from northern hemisphere (NH) permafrost thaw has been investigated in a number of modelling studies (McGuire et al 2010, Koven et al 2011, Schaefer et al 2011, Schneider von Deimling et al 2012, Burke et al 2017. These, as well as review studies (McGuire et al 2010, Schaefer et al 2014, Schuur et al 2015, point to critical factors to be considered. McGuire et al (2018) show that large effects of permafrost thaw are to be expected in the 22nd and 23rd centuries, making studies that limit the time horizon to the 21st century of limited value. Model studies that use model-derived carbon stocks as initial conditions increase uncertainty about permafrost carbon stocks and emissions, as the modelled initial stocks may differ widely from observations (see Schaefer et al 2014 for an overview). Furthermore, large changes in vegetation productivity are to be expected under future climate conditions, pointing to a need to include these, in addition to C release from thawing permafrost soils, in a full C balance of the NH permafrost region (McGuire et al 2010, Koven et al 2011. We therefore aim to improve on the existing studies. We use a land surface model to quantify the amount of carbon released until 2300 and determine the net carbon balance of the NH permafrost region under climate warming following the representative concentration pathway (RCP) scenarios 2.6, 4.5, and 8.5 (Meinshausen et al 2011). We initialise the the frozen carbon stocks from an observation-based data set (Hugelius et al 2013), ensuring that initial conditions are close to observations, similar to Jafarov and Schaefer (2016), but different from Koven et al (2015) in that we use a full land surface model to estimate fluxes. We finally determine the cumulative net C flux to assess the effect of permafrost region C cycle changes on climate.  (Hagemann and Stacke 2015), using 11 soil layers in the experiments described here, with increasing thicknesses from 6.5 cm at the surface to 23.2 m at the bottom, coming to a total soil column depth of 52.14 m. We have also added a representation of permafrost physical processes (Ekici et al 2014), including soil insulation by a surface organic layer, as well as the improved soil carbon model YASSO (Tuomi et al 2008, 2009, 2011, Goll et al 2015 for soil organic matter (SOM) decomposition. Vegetation disturbances from windbreak and fire are included in the model, with windbreak dependent on the wind velocity and fire dependent on litter amount and humidity. In addition herbivory is considered as a fixed fraction of net primary production.
As implemented in JSBACH (Goll et al 2015), YASSO represents SOM as carbon solubility fractions. It distinguishes litter in acid soluble (A), water soluble (W), ethanol soluble (E) and non-soluble (N) fractions, with the pools replicated for above-and belowground. In addition to the litter pools, a humus pool contains carbon that decomposes on long timescales. These pools are replicated for green and woody litter on each vegetation tile. A simplified schematic is shown in the SI, figure S1(a) available online at stacks. iop.org/ERL/13/094001/mmedia. We have extended the YASSO model to represent permafrost carbon stocks. We estimate the thickness of the active layer (AL), the part of the soil column that thaws during the summer season, from the modelled 2 year mean thaw depth, inspired by the definition of permafrost as ground that remains frozen for at least two consecutive years. Carbon cycling in the AL is represented by the YASSO model, while frozen carbon stocks below the AL are prescribed from the Northern Circumpolar Soil Carbon Database (NCSCD), version 2 (Hugelius et al 2013), thereby insuring that the carbon stocks are in line with observations. As the AL thickness increases under climate warming, we transfer carbon from the prescribed frozen carbon stocks into the AL pools described by the YASSO model, as shown schematically in SI figure S1b. The carbon amount transferred is proportional to the change in AL thickness, and the transferred carbon is allocated to the different YASSO carbon pools according to pool proportions in preindustrial (PI) climate. We therefore assume that conditions in preindustrial climate are representative for the conditions under which the permafrost C stocks formed.

Permafrost C initialisation
Hugelius et al (2014) estimate a total C stock of ∼1300 PgC for the northern circumpolar region, with 1035±150 PgC in the depth range to 3 m depth. The NCSCD (Hugelius et al 2013) contains estimates of the soil carbon stock in the northern circumpolar region in four depth intervals, 0-30 cm, 0-100 cm, 100-200 cm, and 200-300 cm. We convert the second depth interval to 30-100 cm by subtracting the C stock for 0-30 cm from the one for 0-100 cm and perform a conservative remapping of the carbon stocks from the native 0.5°grid to the T63 (∼1.875°) resolution of JSBACH. After remapping, the NCSCD estimates a total C stock of 1151 PgC to 3 m depth in the northern circumpolar region. Of this total stock, 22 PgC are located in areas covered by the JSBACH glacier mask, mostly on the margins of the Greenland Ice Sheet, and 52 PgC lie outside the land-sea mask employed by JSBACH, mostly along the northern coasts of Canada and Siberia. A further 127 PgC are located in regions, where JSBACH does not model permafrost in preindustrial climate, despite the fact that the permafrost region is represented reasonably well in JSBACH, with a total area with AL depths less than 3 m in preindustrial climate of 20×10 6 km 2 , similar to area estimates from the Brown et al (2002) map. Overall this leaves a total carbon stock of 949 PgC in the permafrost area modelled by JSBACH.
For model initialisation, the AL thickness is diagnosed by determining the 2 year mean thaw depth for the preindustrial state (SI figure S2). The carbon stock in the NCSCD below the diagnosed maximum AL thickness is used to initialise frozen C pools in the model at the start of the historical experiment. Using the depth information in the NCSCD and the AL thickness modelled in JSBACH for preindustrial climate, 258 PgC are located in the AL, while 691 PgC are at depths below the modelled AL thickness and are prescribed as initial conditions for the frozen permafrost carbon.
Comparing the distribution of carbon in the AL for the start of our experiments in 1850 (SI figure S3) JSBACH/YASSO underestimates AL carbon by 95 PgC in total. However, the spatial pattern shows that the largest underestimates are in areas of organic soils (SI figure S7(a)), which are not represented explicitly in JSBACH. In addition the NCSCD was compiled from recent measurements and therefore is not representative of preindustrial conditions, but rather of present-day conditions. Performing the comparison for the end of the historical experiments reduces the bias slightly (not shown), though the spatial pattern remains similar.
Hugelius et al (2014) estimate an uncertainty range for the total NCSCD carbon stock of 1035±150 PgC, an uncertainty of±14.5%. We use their uncertainty estimate for the total permafrost C stock to derive an uncertainty range for the model results, as the uncertainty in frozen C stock translates directly into an uncertainty of possible soil carbon gain through thawing of permafrost.

Experiments
We perform a set of experiments with JSBACH, driven by climate forcings from CMIP5 experiments performed with MPI-ESM (Giorgetta et al 2013). These climate forcings were available for the extended period until 2300 and produced using an earlier version of the same model, allowing a relatively consistent setup of the experiments. We use climate forcings for the historical period 1850-2005, followed by three experiments to 2300 using the scenarios RCP2.6, RCP4.5, and RCP8.5 (Meinshausen et al 2011). We use the scenarios extended to 2300, with CO 2 stabilised at 542.9 ppm in RCP4.5 and at 1 961.58 ppm in RCP8.5, while CO 2 peaks at 442.76 ppm in 2050 for RCP2.6, thereafter decreasing to 327.21 ppm. For completeness land use changes are included in the scenario experiments, though their impact in the permafrost region is negligible due to the very small areas affected. The model is run in a global configuration, though only results for the permafrost region are shown, i.e. the region with AL thicknesses smaller than 3 m (SI figure S2). We consider paired experiments with and without the consideration of permafrost C to determine the role of thawing permafrost carbon.
For the historical period, as well as scenarios RCPs 2.6 and 4.5, we also perform two additional experiments each to differentiate between the effects of changes in CO 2 and climate. In one of these we keep CO 2 fixed at the PI level of 285 ppm but impose the climate forcings from the scenario experiment, while we vary CO 2 according to the scenario but impose a climate forcing from the PI control climate in the second.
Prior to the start of the experiments, we performed a model initialisation under preindustrial conditions, forced by the climate forcings from the PI control experiment with MPI-ESM. The model carbon cycle initialisation consisted of a 300 year spinup of the model physics, followed by a 5000 year spinup of the carbon cycle to a state in equilibrium with climate. Combining the physics and carbon states, the model was integrated for a further 400 years to ensure consistency. In the spinup experiments, permafrost C pools were not considered, these were only initialised at the start of the historical experiment.

Permafrost carbon feedback
To determine the feedback of permafrost region C cycle changes on climate, one would ideally perform experiments similar to the ones we have made with a full ESM, including an oceanic carbon cycle. As this was not possible for the present study, we determine the permafrost carbon feedback in a simplified way. We use the airborne fraction of emitted CO 2 ( f A ) for the three scenarios from Jones et al (2013) to convert the difference in the cumulative net C flux between the permafrost and the non-permafrost experiments ΔC net into a change in atmospheric CO 2 concentration ΔCO 2 P using D =D´´f CO C 2.12 GtC ppm P net A 2 1 . Using ΔCO 2 and a climate sensitivity α of 3 K, we determine the change in global mean temperature due to the changes in the permafrost region carbon balance . Using ΔT and ΔCO 2 , the change in global mean temperature (from the CMIP5 experiments without considering permafrost) and the change in atmospheric CO 2 concentration (from the RCP scenario), we determine the ratio ΔT P /ΔT as the temperature gain and the ratio ΔCO 2 P /ΔCO 2 as the CO 2 gain due to the permafrost carbon feedback.

Permafrost carbon thaw
Climatic changes in the historical and scenario experiments, especially increasing temperatures, lead to an increase in the depth of the AL, the soil layer thawing during the summer season. Thereby carbon contained in soil layers that were perennially frozen under preindustrial conditions becomes available to active carbon cycling. The initial frozen carbon stock of 691±100 PgC decreases during the historical period to 630 PgC in 2005 (figure 1(a)), with further changes depending on the scenario assumed (table 1). The smallest decline occurs in RCP2.6, where stocks are reduced by 135±20 PgC in 2075, with only small changes afterwards, while the entire stock of frozen C becomes available for decomposition in RCP8.5, with 65% of the carbon thawed by 2100 and the remainder thawing during the 22nd century. Finally, RCP4.5 leads to a reduction in frozen C to 325±47 PgC by 2300, leaving nearly half the PI frozen carbon stock in frozen state. As permafrost carbon thaws, C stocks in the AL increase ( figure 1(b)) leading to enhanced heterotrophic respiration (R h ) in comparison to experiments without permafrost C, as more carbon is available for decomposition (SI figure S6).

Net carbon flux
For an evaluation of the impacts of permafrost thaw on climate, however, we need to assess changes in the net carbon flux from the land surface. The net C flux is comprised of carbon uptake through net primary productivity (NPP) and carbon release through heterotrophic respiration (R h ), as well as further carbon fluxes from land use change (C LUC ), fire (D f ), windthrow (D w ) and herbivory (D h ). These further fluxes are very small in comparison to NPP and R h , and they generally show a release of carbon. As initial analysis showed no strong impact of these fluxes, we include them in the R h flux for the following discussion, noting that their importance may be more pronounced on local to regional scales (Genet et al 2018). The full net C flux therefore is defined as LUC , and in the following we discuss the combined C release flux R h +D f +D w +D h +C LUC , which we imprecisely call R h as this is the major component.
As atmospheric CO 2 and temperature rise, NPP (with positive NPP indicating a carbon uptake by vegetation) in the northern high latitudes increases due to two factors: warmer climatic conditions lead to longer  growing seasons, enhancing plant growth, and more CO 2 leads to higher carbon assimilation in photosynthesis, as C assimilation is dependent on the leaf surface CO 2 concentration in the Farquhar et al (1980) photosynthesis model employed in JSBACH. The impact of climate and CO 2 changes on NPP is of similar magnitude, with a substantial synergy between the two factors (supplementary information and SI figure  S9). As a result, we obtain an increase in NPP over the historical period, as well as in all scenario experiments (table 1 and SI figure S6(a)). R h (with positive R h indicating a release of carbon from the land surface) also grows in all experiments, due to temperature-related increases in respiration rate, as well as AL carbon accumulation (table 1 and SI figure S6(b)). If permafrost C is considered, the latter factor is strongly enhanced due to the transfer of carbon from formerly frozen soil layers into the AL. Therefore the increase in R h is substantially stronger if permafrost C is considered. Details of carbon pool and flux changes are given in table 1, as well as in the supplementary information. Without considering permafrost C the climatically relevant net carbon flux -R NPP h is generally positive, indicating a net uptake of carbon by the land surface. When considering permafrost C, however, the net C flux is decreased markedly, turning negative in most cases (table 1).
There is a well-established relationship between cumulative carbon emissions and global temperature change (Matthews et al 2009, Zickfeld et al 2012, Gillett et al 2013. Consequently we focus on the cumulative net C flux from the NH permafrost region in order to assess the climate feedback of NH permafrost carbon thaw (figure 2). For RCP2.6 with permafrost C the cumulative net flux in 2300 is −24 PgC (79 PgC without permafrost C), while it is 9.6 PgC (185 PgC without permafrost C) for RCP4.5, and −104 PgC (285 PgC without permafrost C) for RCP8.5. Accounting for permafrost carbon thus reduces the cumulative net C uptake to 2300 by 103±14.9 PgC, 175±25.4 PgC, or 389±56.4 PgC, respectively, for RCPs 2.6, 4.5, and 8.5.
With these results for the changes in the cumulative net carbon balance, we can estimate the permafrost carbon feedback gain from the difference in the cumulative net C flux ('C loss' in table 2). The method described in section 2.4 allows us to estimate the feedback of permafrost region C cycle changes on climate, determining the additional CO 2 in the atmosphere, as well as the additional warming. The CO 2 and  temperature gains for 2300 are largest for RCP2.6, with a CO 2 gain of 18% and a temperature gain of 9%, while they are substantially smaller (7% for CO 2 and 2% for temperature) for RCP8.5. The impact of permafrost region carbon cycle changes therefore is much more relevant to climate change under the low-emission RCP2.6 than under the higher emission scenarios.

Pathway-dependence
Concentrating on the case with permafrost carbon, we consider how the net carbon balance changes with global warming. Here NPP and R h to first order change linearly with global mean temperature change (figures 3(a) and (b)), implying that changes in these fluxes are independent of the exact greenhouse gas trajectory and mainly dependent on warming. Investigating the cumulative net flux, however, no such linear relationship can be found (figure 3(c)). Instead, the different scenarios clearly fall on different trajectories, showing a dependence of the cumulative net C flux not just on the global mean temperature change, but also on the pathway this was arrived at. This pathwaydependence is due to a combination of factors. First, the net flux, being the difference between NPP and R h , is substantially smaller than the component fluxes, emphasising changes in the component fluxes. Second, the change in NPP is not just dependent on temperature change, but through CO 2 -fertilisation also on the atmospheric CO 2 concentration. As a result NPP keeps increasing until the end of the experiments in RCPs 4.5 and 8.5, while it reaches a maximum in 2113 and decreases subsequently in RCP2.6. Third, warming in RCPs 2.6 and 4.5 is substantially smaller than in RCP8.5, keeping a substantial fraction of the PI permafrost C in frozen state, thereby limiting the increase in R h . The combination of these factors leads to the separation of trajectories observed in our experiments ( figure 3(c)).

Conclusions
We have shown that the mobilisation and partial respiration of thawed permafrost C significantly reduces the net carbon fluxes into the permafrost region. The magnitude of the reduction in net carbon flux is smallest in the RCP2.6 scenario and largest in the RCP8.5 scenario. Taking into account the extent of anthropogenic perturbations to the climate system, however, this ranking is reversed. RCP2.6 assumes anthropogenic emissions of −0.42 PgC/a, i.e. an anthropogenic removal of carbon, for the year 2100. If permafrost C is considered, the net C flux into the permafrost region is decreased by 0.4 PgC/a, implying that the negative emissions would need to be doubled in order to reach the climate target. This ranking is also reflected in the substantially larger permafrost carbon feedback gains in RCP2.6 than in RCPs 4.5 and 8.5. Our results also show the importance of the time horizon considered. While permafrost thaw has limited carbon cycle implications during the 21st century, the relevance becomes substantially larger, if longer time horizons are considered, as part of the permafrost C remains frozen in 2100 even in the high warming RCP8.5 scenario.
Finally, our results demonstrate that the trajectory of the climate system to a final climate state has consequences for the net carbon emissions from the permafrost region. The NH permafrost region becomes a net source of carbon in both the strong warming scenario RCP8.5, and in the climate stabilisation scenario RCP2.6. With the intermediate scenario RCP4.5, however, the NH permafrost region remains a (small) net sink of carbon, despite a stronger warming than in RCP2.6. This is mainly due to the enhanced carbon uptake through CO 2 -fertilisation under the higher CO 2 concentrations and longer growing seasons of RCP4.5. Thus the fate of the permafrost region carbon is not only dependent on the overall warming, but also on the climate trajectory, it is pathway-dependent.