| 23 Jun 2026
Reassessing the Permafrost Carbon Feedback: The Decisive Role of Dynamic Vegetation
Abstract. The permafrost region stores ∼1300 PgC, and its response to warming is a significant uncertainty in future climate projections. We assess how the treatment of vegetation dynamics and CO2 fertilisation influences the permafrost carbon feedback (PCF) by performing experiments with the land-surface model ICON-Land. A vertically explicit implementation of the YASSO soil-carbon scheme resolves depth-dependent carbon pools, as well as temperature and moisture controls on decomposition, while a newly introduced cold-adapted shrub plant-functional type (PFT) enables a more realistic description of Arctic vegetation.
Using CMIP5-derived climate forcing, we performed nine offline experiments from the year 2020 to 2299 CE under RCP 2.6, 4.5 and 8.5. The experiments isolate the effects of (a) climate warming with fixed 2019 CO2 and vegetation, (b) rising CO2 with constant vegetation cover, and (c) the effects of both rising CO2 and a fully dynamic vegetation. All simulations start from a data-constrained pre-industrial permafrost soil carbon inventory.
When vegetation is held static, strong warming (RCP 8.5) alone drives a loss of ∼650 PgC by 2300 CE, turning the permafrost region from a modest sink of carbon into a strong source. Allowing CO2 fertilisation but no vegetation change reduces the loss to ∼250 PgC because enhanced NPP increases litter inputs. In the fully dynamic experiments, shrubification and northward tree expansion dramatically increase aboveground biomass (of up to 199 PgC) and litterfall, limiting soil-C loss to only ~70 PgC. Consequently, total permafrost-region carbon declines by < 100 PgC under RCP 8.5 and even shows a net gain under the low-emission pathways (RCPs 2.6 and 4.5).
These results demonstrate that the sign and magnitude of the PCF are highly sensitive to the representation of vegetation dynamics. Dynamic competition among PFTs, CO2-driven NPP enhancement, and the resulting shifts in carbon accumulation together can offset most of the carbon released by thawing soils. Incorporating realistic Arctic vegetation, especially cold-tolerant shrubs, is therefore decisive for reliable projections of the permafrost carbon feedback and its impact on the global carbon cycle.
General assessment
This study addresses negative land feedbacks that could compensate for warming-induced soil carbon losses in the permafrost region, with a focus on CO2 fertilisation and dynamic vegetation. This is a very important topic since historically, permafrost has been seen as a carbon source caused by soil carbon losses under climate change, overlooking compensating mechanisms such as those assessed in this study. Using the land surface model ICON-Land and the vegetation and biogeochemistry module JSBACH4, the authors show that accounting for vegetation dynamics strongly limits carbon losses in the permafrost region under moderate-to-high warming scenarios, by favouring vegetation settlement and biomass accumulation in the Arctic.
The experimental design is robust and relevant for addressing the scientific questions of the study. It allows to properly separate the effects of climate change, CO2 fertilisation, and vegetation dynamics on the permafrost carbon cycle. This study is a meaningful contribution for land surface modeling and highlights the importance of representing a realistic vegetation dynamics for future permafrost carbon cycle projections.
I also thank the authors for honestly acknowledging the use of AI for different tasks. I think it is very important to explicitly mention if and how AI has been used, which is done here by the authors.
I only have minor comments that I would like the authors to address.
Minor Comments
Introduction
l.8-9 : It is not fully clear whether warming is included in (b) and (c). Maybe clearer as “...the effects of climate warming (a) with fixed 2019 CO2 and vegetation, (b) rising…”.
l.11 : I could not find any reference to the ~650 PgC loss by 2300 in the main text. l.286 states that 349 PgC are lost by 2300 in con85.
l.16 : RCP4.5 is not really a low-emission pathway. Same at l.330.
l.38-39 : ESMs or land surface models ? Please add references of ESMs that do represent permafrost carbon.
l.40 : “significantly alters” → in which direction ?
l.56-58 : Garcia Criado et al. 2025 and Mekonnen et al. 2021 could be cited here.
l.58-59 : Could you add a reference for snow trapping by shrubs ? In contrast, shrubs can also act as thermal bridges and conduct heat in winter (Domine et al. 2022).
l.65 : Maybe precise the proportion of models with dynamic vegetation (3 out of 11 from Arora et al. 2020) and /or cite Schädel et al. 2024 ?
l.68-78 : Varney et al. (2026) could be cited here. They show that accounting for vegetation dynamics limits the sink-to-source transition of permafrost regions.
Model, experiments and data
l.102-107: Could you please add the units for γi, k, T, P, and κ ?
eq.1, 2 and 3 : Are the soil temperatures and moisture annual means, like in the original version of YASSO (l.106) ? In that case, does this mean that SOC decomposition could be underestimated if the annual mean T is below zero but summer months have a temperature above zero ?
l.136 : How do you define the permafrost regions in which cryoturbation occurs ? Is there a reference or documentation of the representation of cryoturbation in the model ?
l. 149 : Are these disturbances explicitly represented ? Or are they accounted for by a disturbance term/coefficient ?
Section 2.3 : Is vegetation sensitive to ALT deepening ? If so, by which process(es) ? In particular, is the root profile (for water uptake) dynamic ?
l.183-188 : Is the model at equilibrium in 1850 ? In particular, is there a residual drift in soil carbon due to the initialisation from the NCSCDv2 product ?
l.209 : (2) is a strong assumption. Although it is likely that permafrost contains “labile”, or easily decomposable carbon that could be quickly released under warming and thaw, permafrost SOC is also composed of more stable carbon forms. Allocating the whole SOC content to the green litter may overestimate decomposition. For example, explicitly simulated carbon dynamics from the LGM with the ORCHIDEE-MICT-teb model, Xi et al. (2025) found that about 1/3 of the soil carbon was stored in the passive pool in pre-industrial conditions, representing more decomposed and stable SOC. Could you please explain/discuss the choice to allocate all the NCSCDv2 soil carbon to the green litter pools, and how this could affect the soil C dynamics under warming ?
Results
Section 3.1 would benefit from a quantitative comparison of the vegetation distribution against ESACCI-CALU. For instance, mean bias and RMSE over the region of interest for tree, grass and shrub fractions would help supporting the statement that the vegetation is simulated reasonably well (l.240-241).
l.249 and Figure 2 : How is the permafrost region defined ?
l.246-250 : I might have misunderstood, but these numbers do not seem consistent with those at l.213-216. Why are the pre-industrial carbon stocks in the permafrost region lower than the 991 PgC in the continuous permafrost region (l.215) ? Are the 991 PgC from Hugelius et al. (2014) and not from the model (suggested from l.369) ?
l.255-257 : A third reason for SOC underestimation could be a bias in litter inputs from vegetation, especially since the tree fraction is underestimated in some parts of the permafrost region.
l.259-260 : Although it might not be possible to separate the effects of ALT vs organic soil modeling biases, the simulated ALT could be evaluated against CALM site observations (Nelson et al. 2021, Brown et al. 2000), or the ESA-CCI product (Westermann et al. 2024). This would help understanding whether ALT biases are likely to lead to SOC underestimation or not.
l.269-270 : The permafrost region from NCSCDv2 could be added on Fig.2, or in Supplementary, to support this explanation.
l.270-275 : These explanations are redundant with l.256-260. I suggest removing l.256-260 since the same reasons for SOC underestimation are given here.
l.270-272 : Does this mean that the physical soil properties are those of mineral soils in the permafrost region ? In that case, indeed, upper soil insulation and soil water content are likely to be underestimated, possibly leading to biases in soil thermal state and ALT.
l.271 : Does increased drainage always lead to decreased carbon accumulation ? If soil becomes dry enough, could drainage favor accumulation ? In addition, mineral soils also have a higher thermal diffusivity than organic soils, which could lead to higher summer soil temperatures, and increase SOC decomposition (Zhu et al. 2019).
l.272-273 : ALT could be evaluated against observations (see previous comment).
l.273-274 : I agree that simulated SOC storage is expected to be a conservative lower bound estimate. However, its reactivity (or quality) is likely to be an upper bound estimate, since the entire permafrost carbon content has been allocated to green litter (more easily decomposable than woody litter or HUM) during initialisation, in particular for initially frozen carbon that could be exposed to decomposition under warming. Overall, both effects could partially compensate, resulting in relatively low SOC content with high reactivity.
Table 2 : Could you please add a line with pre-industrial stocks in the table ? This would facilitate the comparison between the different simulations.
l.277 : Adding time series of the main climate forcing variables (temperatures and precipitation) would help linking changes in carbon stocks to warming/climate change.
Fig.4a : Although this is not the main message of this study, permafrost soils are already losing carbon in the historical simulation (although vegetation more than compensates the SOC loss). Is this consistent with current observations and the processes represented in the model ?
l.293-294 : “decreases by 374 PgC to 436 PgC in 2299” → “decreases by 374 PgC to reach 436 PgC in 2299” ?
l. 293 : Initial permafrost soil C stocks are 759 PgC (l.279) and decrease to 436 PgC in 2299 in fix85 (l.294). This is a 323 PgC decrease, not 374 PgC.
l.295 : “the increase in biomass...reaching 121 PgC” is imprecise as this mixes the change in vegetation carbon with absolute stocks.
l.296 : “The larger soil C value” → compared to con85 ?
Results from section 3.3.3 are very interesting. Do you know which processes lead to increase in shrub cover in RCP2.6 and RCP4.5, but to a decrease in RCP8.5 to the benefit of trees ? Is it an outcome of the competition model that would favor trees over shrubs in a warmer climate ?
l.314 : +275% is misleading as the reader could think this refers to the comparison between fix85 and dyn85. Maybe precise “+275% compared to present-day vegetation carbon” ? Same at l.318.
l.344 : Nitrogen release from permafrost thaw might also increase the amount of available nutrients for vegetation, and could lead to increased C uptake (e.g. Beerman et al., 2017; Keuper et al., 2012). However, the response of vegetation to increased soil nitrogen remains highly uncertain.
Fig.4b : Do you know why vegetation C peaks around 2125 in dyn26 before decreasing ?
Discussion
l.341 : These FACE experiments are mostly located at mid-latitudes, and to my knowledge, none has been installed in boreal forests or tundra. There may be only little literature on this topic, but do you expect high-latitude vegetation to respond similarly to CO2 fertilisation ?
l.370-375 : Not affecting any carbon to the humus pool can indeed overestimate its decomposability. Could you please explain your choice to allocate initial permafrost carbon to green litter (see previous comment) ?
This study raises questions on the simulation of future permafrost carbon dynamics by Earth system models, as JSBACH4 represents vegetation dynamics but most ESMs do not. Could you please discuss the implications of this study for ESMs that do not represent dynamic vegetation ? Can they be used to assess the permafrost climate feedback (further used to calculate remaining carbon budgets) ?
Will the new developments made in JSBACH (vertically-resolved soil carbon, Arctic shrubs) be included in the MPI ESM for CMIP7 ?
Technical comments
l.313 : typo “fix 85” → “fix85”
l.331 : typo “these” → “These”
References
Arora, V. K. et al.: Carbon–concentration and carbon–climate feedbacks in CMIP6 models and their comparison to CMIP5 models, Biogeosciences, 17, 4173–4222, https://doi.org/10.5194/bg-17-4173-2020, 2020.
Beermann, F. et al. (2017) Permafrost Thaw and Liberation of Inorganic Nitrogen in Eastern Siberia. Permafrost and Periglac. Process., 28: 605–618. doi: 10.1002/ppp.1958.
Brown, J. et al. (2000). The circumpolar active layer monitoring (calm) program: Research designs and initial results. Polar Geography, 24 (3), 166–258. https://doi.org/10.1080/10889370009377698
Domine et al., Permafrost cooled in winter by thermal bridging through snow-covered shrub branches. Nat. Geosci. 15, 554–560 (2022). https://doi.org/10.1038/s41561-022-00979-2
García Criado et al. 2025. “Borealisation of Plant Communities in the Arctic Is Driven by Boreal-Tundra Species.” Ecology Letters28, no. 9: e70209. https://doi.org/10.1111/ele.70209.
Keuper, F. et al. (2012), A frozen feast: thawing permafrost increases plant-available nitrogen in subarctic peatlands. Glob Change Biol, 18: 1998-2007. https://doi.org/10.1111/j.1365-2486.2012.02663.x
Mekonnen Z. A. et al 2021 Environ. Res. Lett. 16 053001
Nelson, F. E. et al. (2021). Cool, CALM, collected: the Circumpolar Active Layer Monitoring program and network. Polar Geography, 44 (3), 155–166. https://doi.org/10.1080/1088937X.2021.1988001
Schädel et al., Earth system models must include permafrost carbon processes. Nat Clim. Chang. 14, 114–116 (2024). https://doi.org/10.1038/s41558-023-01909-9
Varney, R. M. et al. : Northern high latitudes could become a net carbon source below 2 °C global warming, Earth Syst. Dynam., 17, 913–928, https://doi.org/10.5194/esd-17-913-2026, 2026.
Westermann, S. et al. (2024): ESA Permafrost Climate Change Initiative (Permafrost_cci): Permafrost active layer thickness for the Northern Hemisphere, v4.0. NERC EDS Centre for Environmental Data Analysis, 24 April 2024. doi:10.5285/d34330ce3f604e368c06d76de1987ce5. https://dx.doi.org/10.5285/d34330ce3f604e368c06d76de1987ce5
Xi, Y. et al.: Representing high-latitude deep carbon in the pre-industrial state of the ORCHIDEE-MICT land surface model (r8704), Geosci. Model Dev., 18, 6043–6062, https://doi.org/10.5194/gmd-18-6043-2025, 2025.
Zhu D. et al. Controls of soil organic matter on soil thermal dynamics in the northern high latitudes. Nat Commun 10, 3172 (2019). https://doi.org/10.1038/s41467-019-11103-1