Interactive comment on “ Simulating the effects of temperature and precipitation change on vegetation composition in Arctic tundra ecosystems ” by H . van der Kolk

This is an interesting and relevant studied, which in my opinion deserves to be published in Biogeosciences. The paper presents in a clear and interesting way potential changes of Arctic tundra under warming/precipitation change/permafrost thaw. Especially, addressing 3 factors in combination, i.e. warming, precipitation and permafrost thaw is a relevant contribution to our understanding of tundra change.

(2) The model needs a thorough evaluation in the performance of simulating soil water, evapotranspiration and soil temperature, active layer depth, water table depth for the period the observations are available.This is the basis to convince the readers to believe the efficiency of the model.Particularly, the simulated soil temperature doesn't look correct in the 40, 80 cm.REPLY: Indeed, the simulated soil temperature at 40 and 80 cm depth does not look good in winter.It has been acknowledged in the text that the soil temperature model is based on temperatures in the growing season only, as the below-zero temperature in winter hardly affects the model output (Supplement S1.3.4).We used a regression approach which is based on air temperature and does not take snow characteristics and phase changes into account.Still, realistic temperatures for the upper (most important for N mineralisation) soil layers in summer have been simulated, but there are (large) deviations for winter temperatures (Supplement Figure S1.3).These deviations did not affect the simulated periods of frozen and thawed soil, which were in agreement with the observations (Supplement page 29, lines 6-8).For broader applications of the model it will be necessary to couple the vegetation model to a more advanced soil physical model (page 11, lines 29-30).We added a new supplement section (S4.2) on the evaluation of soil temperature and active layer thickness to the revised manuscript.In the main text, the comparison with the soil temperature time series is mentioned on page 6, lines 14-15.The results of this evaluation have been summarized on page 8, lines 6-13.
New supplement section S4.2 also evaluates the simulated active layer thickness (ALT).Based on measured thawing depths on 20 dates, it appeared that the simulated and measured ALT were strongly correlated (new figure S4.2), but simulated ALT proved to be slightly too deep: average difference between simulated and measured ALT was 14 cm.Next, we assessed the consequences of too deep ALT for the simulated vegetation composition in a simulation where deep soil temperatures (and thus active layer thickness) were manually set back.The result is that too deep simulated ALT is to the advantage of graminoids (new figure S4.3).This can be explained by the fact that our vegetation model includes differential rooting depths for graminoids and shrubs.Graminoids are more capable of acquiring available nutrients in deeper soil layers, as they are deeper-rooted than the dwarf shrubs.The result implies that simulated graminoid biomass may be overestimated as a consequence of overestimated ALT.This is discussed in a new paragraph on page 11, lines 19-30.Although there is uncertainty regarding the simulated ALT and the amount of nutrients becoming available, the pattern of vegetation shifts in response to temperature and precipitation changes remained the same.
Unfortunately we have no time series of observations for soil moisture and water table depth for a thorough comparison with model outcome.The meteo tower at the Kytalyk field station does not measure soil moisture.Soil moisture and water table depth are extremely variable even at a small spatial scale in the tundra landscape.The wet depressions (resembling the wettest vegetation type in the model) are much wetter compared to the slightly elevated shrub patches.Therefore, we choose to simulate different vegetation types along a water movement gradient from relatively well-drained elevated shrub patches to downstream sedge-dominated depressions receiving water from their surroundings.Limited soil moisture measurements in graminoid, mixed and shrub dominated vegetation types (Wang et al. 2016) indicate clear trends that moisture during the growing season is high in graminoid but low in shrub dominated sites, similar as is simulated in the model.As described in the manuscript, the model simulates a build-up of the snow layer in winter, followed by a wet period in spring following snowmelt.In summer, the soil gradually dries as evapotranspiration (derived from measured latent heat flux) generally exceeds precipitation, the extent of drying depending on the amount of precipitation and temperature.This sequence and dependence on This is an interesting and relevant studied, which in my opinion deserves to be published in Biogeosciences.The paper presents in a clear and interesting way potential changes of Arctic tundra under warming/precipitation change/permafrost thaw.Especially, addressing 3 factors in combination, i.e. warming, precipitation and permafrost thaw is a relevant contribution to our understanding of tundra change.
REPLY: We thank the referee for his/her constructive comments.We appreciate that the referee recognizes that our study has the potential to provide interesting insights in the response of tundra vegetation composition to climate change, gradual permafrost thawing and abrupt permafrost thaw.
The paper is well written and the results are clearly presented.As this study represents a modelling approach, I would find it helpful if some modelling related issues could be clarified.In particular, many parameters in the NUCOM-tundra model were defined based on e.g.vegetation composition found in the field, so I was sometimes uncertain what I learned in the paper about mechanisms responsible for changes in the tundra.
REPLY: It is correct that some of the parameter values for the model have been derived from field measurements at our study site in north-eastern Siberian tundra, see further reply to comments below.From our field studies we have gained insights in the functioning of (dwarf) shrubs and graminoids which have been used for the model.For example, the graminoids are deeper-rooted than the shrubs which implies that the graminoids are better competitors for nutrients in the deeper soil layers.The model synthesizes own findings from field studies with knowledge on Arctic tundra ecosystem functioning from literature.Also related to this issue: of course simplifications/assumptions need to be made for a model, especially if access to measured data is limited.However, I asked myself a few times if the simplification were justified.
A few examples.Abstract.L.24.The simulations suggest that shrubs are better light competitors... etc.If I understand the model right, shrubs are good competitors because they were defined as good competitors in the first place.Not that this would be incorrect.But several times I get the impression that findings are not necessarily a result of the model but a result of how the model was set up, which assumptions were made and which data were used to feed the model.Again, this is certainly an issue that can be said for all models.But I think the text needs some rephrasing to be clear about what is indeed a model outcome (e.g. increase of graminoids under wetter conditions) and what is not.To me the text seems to go too far, which mechanisms can actually explained by this model and which cannot.See related comments below.
REPLY: Of course a result of a model depends on how the model was set up, etc.This has the advantage that it is possible to trace back how a certain model outcome has been generated and provides a potential mechanism for the simulated vegetation change.It is correct that the mechanism is not independent of how the model was build and we agree that this should not be overstated.Still, we think that models are a helpful tool for exploring mechanisms of vegetation change.
We disagree that shrubs were defined as good competitors in the first place.Several plant traits (parameters values) are related to competition for light or nutrients.For light competition, leaf area and light extinction are important, which are influenced by parameters as specific leaf area, biomass allocation to leaves, leaf mortality and light extinction coefficient.Similarly, nutrient competition is influenced by parameters as specific root length, root distribution and nutrient requirements.This explanation is added to page 12, lines 17-23.Values for the parameters have been assigned independently, partly based on literature values and partly derived from own measurements.It is the combination of multiple plant traits and environmental conditions that determines which plant type is a good competitor.Based on our field observations, we assumed the (dwarf) shrubs and graminoids to be equally tall in the model.Therefore, the shrubs were not a-priori the better light competitor.We removed the sentence about shrubs being better light competitors (page 1, line 27), as it is not that clear from our simulations anyway.
Questionable assumption?p4 l21.Graminoids and dwarf shrubs are assumed to be equally tall.The authors may have their reasons to do so, but this is not entirely clear to me.Betula nana can grow easily 2.5 m tall (e.g. in parts of Alaska) and arctic graminoids don't.An incorrect assumption here could have a large influence on the results.

REPLY:
Based on observations at our study site in north-eastern Siberian tundra we assumed the graminoids (Eriophorum spp.) and dwarf shrubs (Betula nana) to be equally tall, which is a simplification, but not unrealistic.Also in Alaska, in moist tussock tundra as in Toolik Lake, dwarf shrubs and graminoids are equally tall (Heijmans, personal observations).We acknowledge that in some other tundra areas the shrubs, including B. nana, can grow taller than the graminoids, although this may be limited to places with some additional nutrient supply.For a wider application of the vegetation model, i.e. to include transitions to tall shrub vegetation, a variable plant height up to a maximum would be required.We added this discussion to page 12, lines 27-31 Explaining mechanisms?P. 9 l16ff.The authors state that the NUCOM model was developed to assess which mechanisms are responsible for tundra change.I found this statement somewhat questionable because many very important mechanisms remain unknown when assumptions are made for models.The biomass example above is one such example.The issue that rooting depth in a warmer climate is not known is another example, but discussed later in the discussion.It might be helpful if the authors adapt their wording a bit.E.g. that they refer mechanisms to effects of warming vs. precipitation, which is the novel contribution of this paper.

REPLY:
The focus of this modelling study was on exploring potential vegetation shifts in response to changes in temperature, precipitation and permafrost thaw, as is evident from the abstract.We agree that in parts of the main text the wording is a bit misleading with too much emphasis on mechanisms.We went through the whole text and adapted the wording in several places (e.g.page 1, lines 15, 18; page 4, line 1; page 10, lines 8-13).
So I suggest that the authors go through the entire manuscript another time and re-think carefully how to not over-sell their results.Apart from that, to state this again, I like the paper and find it helpful and novel.

Introduction
Tundra ecosystems in the Arctic are shaped by strong interactions between biological, hydrological and climatological factors (Hinzman et al., 2005).An important feature of the Arctic ecosystem is permafrost, which is the soil that is persistently frozen for at least two years.For the near future, global climate models project a further increase in temperature and precipitation with pronounced changes in the Arctic region (Johannessen et al., 2004;Vavrus et al., 2012;IPCC, 2014).
Rising temperatures potentially result in increased seasonal thawing of the permafrost and thus in an increased thickness of the active layer (the soil layer that thaws during the growing season) and drawback of near surface permafrost (Anisimov et al., 1997;Lawrence and Slater, 2005;Zhang et al., 2005).Given the large soil organic carbon stocks in permafrost (Zimov et al., 2006a;Tarnocai et al., 2009), there is a major concern that these stocks get decomposed and subsequently released as CO2 and CH4 to the atmosphere (Dutta et al., 2006;McGuire et al., 2009).An increase in greenhouse gas release, including CO2 and CH4 (Christensen et al., 2004;Walter et al., 2006;Nauta et al., 2015;Schuur et al., 2015), from thawing permafrost could stimulate further warming and result in a positive feedback loop between temperature increase and greenhouse gas emission (Zimov et al., 2006b;Schuur et al., 2008;MacDougall et al., 2012;van Huissteden and Dolman, 2012;Schuur et al., 2015).
Climate change in the Arctic region affects tundra vegetation composition.The northernmost tundra is dominated by mosses and lichens due to the extreme low summer temperatures.Southwards, with increasing summer temperatures, graminoids and dwarf shrubs increase in abundance (Walker et al., 2005).Climate change influences the tundra vegetation in multiple ways.Warming experiments in tundra ecosystems showed an increase of graminoids and deciduous shrubs in response to raised temperatures, while mosses and lichens and the overall species diversity decreased (Walker et al., 2006).Shrubs have been observed to expand with ongoing temperature increase presumably due to the increased availability of nutrients in the warmer soil (Tape et al., 2006;Myers-Smith et al., 2011).Several tree and shrub species, including dwarf birches (Betula glandulosa and Betula nana), willows (Salix spp.), juniper (Juniperus nana) and green alder (Alnus viridis), have expanded and increased in abundance in the Arctic as a response to climatic warming (Sturm et al., 2001;Tape et al., 2006;Hallinger et al., 2010;Elmendorf et al., 2012).However, besides climate, other factors as herbivory, soil moisture and soil nutrient availability affect shrub growth as well and it is therefore complex to predict the expansion of shrubs in the Arctic region (Myers-Smith et al., 2011, 2015).Increased abundance of shrubs might have important consequences for permafrost feedbacks.For example, an increase of low shrubs might slow down permafrost thaw as a result of the shadow they cast on the soil (Blok et al., 2010).However, tall shrubs may increase atmospheric heating and permafrost thawing due to their lower albedo (Bonfils et al., 2012) Another mechanism by which climate change affects tundra vegetation is abrupt thaw resulting in local collapse of the permafrost.Thawing of underground ice masses results in a collapse of the ground, by which water filled thermokarst ponds are formed.These ponds are first colonized by sedges and later by mosses (Jorgenson et al., 2006).Due to thaw pond formation, changes in northern permafrost landscapes from dry birch forests or shrub dominated vegetation towards ponds or wetlands dominated by graminoids may occur with climatic change in both discontinuous and continuous permafrost regions (Jorgenson et al., 2001;Turetsky et al., 2002;Christensen et al., 2004;Jorgenson et al., 2006;Nauta et al., 2015).In the continuous permafrost region, permafrost collapse can change the vegetation from shrub dominated towards a wet graminoid dominated stage within less than one decade.The thermokarst ponds might stimulate further soil collapse and consequently drastically alter hydrological and soil processes, also in adjacent areas (Osterkamp et al., 2009;Nauta et al., 2015;Schuur et al., 2015).
The mechanismsfactors that drive the observed tundra vegetation composition changes, especially shrub expansion, and their consequences are not yet well understood.Increased air temperatures lead to higher soil temperatures, especially in the shallow layers, and may thus promote microbial activity, thereby increasing nutrient availability in the soil.In tundra vegetation, shrubs are hypothesised to be best able to respond to such increased nutrient availability (Tape et al., 2006).It is, however, not known to what extent this and other shrub growth factors, including precipitation, growing season length and disturbances (Myers-Smith et al., 2011), will affect the competition between different plant functional types in the Arctic.As shrub expansion has important implications for land surface albedo and consequently climate feedbacks (Chapin et al., 2005;Pearson et al., 2013), it is crucial that we understand and are able to predict further vegetation change in the tundra.
One approach to better understand the interactions between plants in tundra landscapes is to use a dynamic vegetation model to analyse developments in vegetation composition in response to climatic changes.Tundra vegetation models that aim to predict the impacts of climate change on vegetation-substrate interactions should at least include the most important plant functional types, competition and permafrost feedbacks.Although several tundra vegetation models exist, these models do not take into account hydrological feedbacks, the formation of thaw ponds and vegetation-permafrost feedbacks or do not include mosses as separate plant functional type (e.g.Epstein et al., 2000;Wolf et al., 2008;Euskirchen et al. 2009).
Therefore, in this study, the effects of climate change, including both temperature and precipitation change, on plant competition and vegetation composition were studied by developing a new model named NUCOM-tundra based on earlier NUCOM (NUtrient and COMpetition) models for other ecosystems (Berendse, 1994a(Berendse, , 1994b;;van Oene et al., 1999;Heijmans et al., 2008Heijmans et al., , 2013)).The new tundra vegetation model includes mosses, graminoids, dwarf shrubs, hydrological and soil processes, and permafrost dynamics.In this study, we simulated tundra vegetation changes under different climate change scenarios in order to (1) analyse the effects of future temperature and precipitation scenarios on tundra vegetation composition, (2) identify important mechanisms that drive these vegetation changes, and (32) explore the impacts of thaw pond formation due to local permafrost collapse.

Brief model description
NUCOM-tundra has been developed to simulate long-term dynamics of vegetation composition in tundra landscapes for analysis of vegetation-permafrost-carbon feedbacks in relation to climate change and includes nitrogen and carbon cycling, permafrost and water dynamics (Fig. 1).An extensive model description and all equations are provided in Supplement Section S1.
NUCOM-tundra simulates the dynamics of three Plant Functional Types (PFTs), moss, graminoids (e.g.Eriophorum vaginatum) and deciduous dwarf shrubs (e.g.Betula nana), on a local scale at the decadal timescale using a daily time step.NUCOM-tundra represents tundra landscapes, which are an alternation of shrub dominated, graminoid dominated and mixed vegetation types.The model is based on an area of 1 m 2 .The biomass and nitrogen content of the vascular plant PFTs is separated into fine roots, woody plant parts (for shrubs) or rhizomes (for graminoids), and leaves.Mosses form a layer on top of the soil surface, with a thickness up to 4.5 cm.The soil profile is divided into an organic top soil layer with a variable height and 10 deeper mineral soil layers each with a thickness of 10 cm.The fine roots of vascular plants are distributed throughout the active soil profile, with graminoids rooting deeper in comparison to dwarf shrubs (Shaver and Cutler, 1979;Nadelhoffer et al., 1996;Iversen et al., 2015).The thickness of the active layer depends on the soil temperature profile (Supplement Section S1).The thickness of the organic layer, which consists solely of moss, leaf, stem and root litter, generally increases over time, depending on the balance between litter input and litter decomposition.The mineral soil layers contain initial soil organic carbon and nitrogen pools as well.During the simulations, only fine root litter is added to the mineral soil layers and decomposed there.The other litter types become part of the soil organic layer.Decomposition rates depend on soil temperature and differ among PFTs and plant organs (leaves, stems and fine roots).
Plant growth (net primary production) is determined by temperature, light availability, nutrient availability and moisture conditions (Supplement Section S1).In the model, there are temperature thresholds for plant growth thus excluding growth during the winter season.Graminoids and dwarf shrubs compete for the incoming light based on their leaf area.It is assumed that graminoids and dwarf shrubs are equally tall.The moisture content in the upper 10 cm of the soil can strongly reduce PFT growth as both graminoids and dwarf shrubs have an optimum growth only in part of the range of possible soil moisture conditions.Dwarf shrubs prefer drier conditions, and cannot grow if the soil is completely water saturated.The graminoids in the model do not grow well under dry conditions, but can grow on water-saturated soils.
Mosses acquire nitrogen by nitrogen fixation from the atmosphere and can absorb available nitrogen from the upper cm of the soil profile.Vascular plant nitrogen uptake is determined by the fine root length of both graminoids and dwarf shrubs present in each layer and the amount of nutrients available.At the start of the growing season, when the air temperature is above the threshold for growth but the soil is still mostly frozen, the plants can use nitrogen from an internal storage, which is filled by nitrogen reallocated from senescing leaves and roots.Dying plant material enters the soil organic carbon and nitrogen litter pools.Soil organic carbon is lost by microbial decomposition, whereas the mineralised nitrogen from soil organic nitrogen becomes available for plant uptake.The rate of decomposition and nitrogen mineralisation depends on soil temperature.Subsequently, pPart of thise pool of available nitrogen can be lost by denitrification under high soil moisture conditions (Supplement Section S1).
A simple hydrological module is included in NUCOM-tundra, which simulates the volumetric water content of the organic and mineral soil layers (Supplement Section S1).Water from snowmelt, rainfall and inflow from a neighbouring vegetation type (Fig. 2) fills up the pore space in the soil layers.Evapotranspiration, surface runoff, and lateral drainage out of the moss and organic layer lower the water content.The lateral drainage includes transport of dissolved nitrogen.The hydrological processes follow a seasonal pattern.A snow layer accumulates during the winter season, and snowmelt occurs with positive air temperatures at the start of the growing season.During this period, the shallow active layer becomes water saturated and the excess of water runs off over the soil surface.During the growing season, evapotranspiration generally exceeds precipitation, resulting in gradual drying of the soil.

Parameter values and model input
Parameter values for plant properties such as root characteristics, mortality, reallocation and decomposition have been derived from literature or have been calibrated using field data collected at the Chokurdakh Scientific Tundra Station, located 70° 49′ N, 147° 29′ E, altitude 10m (Supplement Section S2).The Chokurdakh Scientific Tundra Station, also known as Kytalyk field station, is located in the lowlands of the Indigirka river, north-eastern Siberia, which is in the continuous permafrost zone and the Low Arctic climate zone.The vegetation is classified as tussock-sedge dwarf-shrub moss tundra (vegetation type G4 on Circumpolar Arctic Vegetation Map; Walker et al., 2005).The parameter values are provided in

Vegetation types
We defined five initial vegetation types representing a gradient from relatively dry to wet sites for the climate change simulations in NUCOM-tundra.These initial vegetation types represent five landscape positions, ranging from relatively well-drained shrub patches to waterlogged graminoid-dominated wetland (Fig. 2).The first vegetation type represents dry shrub-dominated vegetation.Vegetation types 2, 3 and 4 represent moist mixed vegetation from relatively dry to wet sites.
Vegetation type 5 represents a wet graminoid-dominated vegetation downslope which receives water from the neighbouring cell.In vegetation type 1, only outflow of both surface water and water in the organic layer occur, whereas moist sites are characterized by having both water inflow and water outflow.The water flow into a downslope landscape position is the actual water outflow from the adjacent upslope located simulation cell.This water movement over the gradient includes transport of dissolved nitrogen from the elevated shrub patch to the sedge-dominated depression.

Comparison with field data
The performance of several model parts was evaluated by comparing outputs with data from the Kytalyk field station and Chokurdakh Tundra Station and Weather Station.The simulated timing of snow accumulation and snowmelt was compared with snow height data from the Chokurdakh Weather Station between 1944 and 2008.Time series of soil temperature at different depths and measurements of thawing depth were used to evaluate the soil temperature calculations.Furthermore, simulated PFT total biomass, biomass partitioning among leaves, woody parts, rhizomes and fine roots and vertical root distribution were compared with field data.Field biomass collections were done for both graminoids and dwarf shrubs in graminoid dominated, mixed and dwarf shrub dominated vegetation sites at the tundra field site in north-eastern Siberia (Wang et al., 2016).Roots were collected over different depths, enabling to compare the field data with simulated root distribution over depth.These field data were compared with simulations that were run with climate input from the Chokurdakh Weather Station.A detailed description of these comparisons of simulated and field data is provided in Supplement Section S4.

Climate scenario simulations
For all vegetation types a period of 40 years was simulated to initialize the model, using precipitation and temperature data from the Chokurdakh weather station from 1 August 1954 till 31 July 1994 as input.This period excludes the most pronounced warming that took place over the last decades.Annual precipitation was, however, variable with both increasing and declining trends in the 40-year time period (Supplement Section S2).After the initialisation phase, 16 different climate change scenarios were run for all five vegetation types for the period 1 August 1994 -31 July 2074.Inspired by the RCP emission scenarios over the 21 st century, we combined four temperature increases with four precipitation increases to simulate climatic changes over this 80-year period.The combinations include temperature increases of 0, 2.5, 5 or 8 ⁰ C and precipitation increaseschanges of 0, 15, 30 or 45 % increase overmm per 100 years.In the Arctic region, an increase of 2.5 ⁰ C and 15mm% precipitation increase is expected under emission scenario RCP2.6 and an increase of 8 ⁰ C and 45 mm% precipitation increase are expected over the 21 st century under emission scenario RCP8.5 (IPCC, 2014).A scenario with 5 ⁰ C and 30 mm% precipitation changeincrease is regarded here as an 'Intermediate' scenario.
The weather data series over the period 1 August 1954 -31 July 1994 from the Chokurdakh weather station was used twice to create a baseline (no climate change) for the climate scenario simulations between 1994 and 2074.For all climate change scenarios, temperature and precipitation were gradually increased over this 80 year simulation period.For precipitation, only the intensity of rainfall was increased and not the number of days at which rainfall occurred.
To evaluate the effects of climate change on the vegetation composition, we compared the biomass of moss, graminoids and dwarf shrubs on 31 July, averaged over the last 10 years of the simulation (corresponding with 2065-2074).Furthermore, several factors that influenced plant growth and competition were evaluated using the simulations for moist mixed vegetation sites (vegetation type 3) in which all three PFTs co-occur.The following factors were evaluated for the last 10 years of the simulations: 1) Growing season length: The average number of days per year at which the temperature threshold for vascular plant growth was exceeded; 2) Soil moisture: Average soil moisture content in the upper 10cm of the soil profile during the growing season; 3) Light competition: The average percentage of light intercepted by graminoids and shrubs during the growing season; 4) Moisture conditions: Percentage of time in the growing season with optimal soil moisture content for growth of graminoid and dwarf shrub; 5) Nutrient limitation: Limitation of growth of graminoid and dwarf shrub due to insufficient available nutrients in the growing season, expressed as percentage of potential growth realised.

Thaw pond simulations
Thaw pond initiation was simulated for dry and moist vegetation types (vegetation types 1-4) under three different climate change scenarios: no change, RCP2.6 and RCP8.5.Thaw pond collapse was simulated by imposing a sudden alternation of water flows.Upon a collapse event, water inflow, including surface runoff and lateral drainage through the soil organic layer, doubled whereas water outflow stopped.To evaluate the effects of thaw pond collapse on downslope vegetation sites, simulations were performed in series that included all five vegetation types (Fig. 3).Thaw pond formation was initiated at a fixed time step halfway the simulation (corresponding to 31 July 2034).The change in vegetation composition after collapse was evaluated by determining the abundance of graminoids (in %) in the vascular plant community biomass in the last 10 summers of the simulation (31 July in the years 2065-2074), when the vegetation is at its peak biomass.

Comparison with field data
Simulated snow height, soil temperature, active layer thickness, total PFT biomass, biomass partitioning and vertical root distribution were compared with observations from Chokurdakh Scientific Tundra Station or Weather Station (detailed description in Supplement Section S4).Simulated timing of accumulation and melt of the snow layer between 1944 and 2008 was well in agreement with observations of snow depth at the Chokurdakh Weather Station.Simulated shallow soil temperatures were in agreement with observations from the Kytalyk meteo tower, but simulated deep soil temperatures were higher than observed, resulting in simulated thawing depths that were on average 14 cm higher than observed thawing depths (Fig. S4.2).Furthermore,This deeper active layer thickness resulted in an overestimation of graminoid biomass, as was demonstrated by simulations with lowered active layer thickness (Figure S4.3).However, the patterns of changing graminoid biomass among the main climate change scenarios remained when the active layer thickness was lowered, indicating that the uncertainty about the simulated deep soil temperatures doid not significantly affect the vegetation composition changes for the climate change scenario simulations.NUCOM-tundra simulated total biomass for graminoids and dwarf shrubs within the range of biomass values obtained for shrub, mixed and graminoid vegetation types at the field site.Simulated biomass partitioning over leaves, stems and fine roots was comparable to the partitioning observed in the field with high total biomass, although the simulated biomass partitioning was rather constant and did not change with increasing total biomass, as observed in the field.Fine root vertical distribution patterns were similar between observed and simulated data as NUCOM-tundra took into account the different rooting patterns of graminoids and dwarf shrubs, and variations in active layer thickness.

Climate scenario simulations: Vegetation changes
In the No climate change scenario, the total vegetation biomass was rather stable throughout the simulation in all vegetation types (Fig. 4A for vegetation type 3, data not shown for other vegetation types).Gradual biomass increases between 1994 and 2074 were simulated under the three climate change scenarios based on the RCP scenarios (Fig. 4B, 4C, 4D).
Considering all combinations of temperature and precipitation scenarios in all vegetation types, shows that the simulated total biomass of the vegetation responded strongly to increased temperature (Fig. 5).The average total (aboveground and belowground) summer biomass between 2065 and 2074 ranged from 1544 1970 g m -2 to 2428 3037 g m -2 with no temperature change, whereas total biomass ranged between 3702 4815 g m -2 to 4483 5402 g m -2 with a gradual increase of 8 ⁰ C over the 21 st century.In contrast, precipitation change did not affect the total vegetation biomass.
The vegetation composition in relatively well-drained sites dominated by shrub vegetation (vegetation type 1) and wet graminoid-dominated vegetation (vegetation type 5) did not change under any of the climate change scenarios (Fig. 5).In mixed vegetation (vegetation type 3), graminoids slightly increased in abundance with the RCP2.6 scenario (2.5 ⁰ C temperature and 15mm 15% precipitation year -1 increase over the 21 st century), intermediate and RCP8.5 scenario (8 ⁰C collapse enhanced shrub expansion in the next grid cell.The thermokarst pond acted as a water sink, due to which water flows into the next grid cell were halted.Consequently, the grid cell adjacent to the thermokarst pond became drier which favoured shrub growth (Fig. 7).Similar patterns for the thaw pond simulations were observed for No change, RCP2.6 and RCP8.5 climate scenarios.

Discussion
The effects of climate change on tundra vegetation are complex since temperature and precipitation drive changes in hydrology, active layer depths, nutrient availability and growing season length, which interact with each other (Serreze et al., 2000;Hinzman et al., 2005).Climate warming might alter vegetation composition by increasing nutrient availability due to faster mineralization rates and increasing active layer depths (Mack et al., 2004).It remains, however, unclear which factors are most decisive with respect to the observed expansion of shrubs at the expense of graminoids (Myers-Smith et al., 2011).NUCOM-tundra washas been developed to explore future Arctic tundra vegetation composition changes in response to climate scenarios, thereby taking changes in soil moisture, thawing depth and nutrient availability into account.and to assess which mechanisms are responsible for these changes.

Vegetation Composition ChangesEffects of increases in temperature and precipitation
Our climate scenario simulations suggest a significant increase in biomass with continuing climate change, and eventually especially increased shrub graminoid abundance under scenarios with strong different magnitudes of climate change, such as the RCP8.5 emission scenario.For the modest and strong emission scenarios (RCP2.6,+2.5 ⁰ C and +15mm % precipitation increase over the 21 st centuryIntermediate and RCP8.5), particularly both shrub and graminoid biomass increased.These simulations are in line with biomass increases in both shrubs and graminoids that have been observed during the past decades in Arctic tundra landscapes as a response to temperature increase (Dormann and Woodin, 2002;Walker et al., 2006;Hudson and Henry, 2009).In addition, sShrub expansion , as predicted by the model for strong climate change scenarios, has been observed in many places in the Arctic (Sturm et al., 2001;Tape et al., 2006;Myers-Smith et al., 2011), but is not explicitly simulated by NUCOM-tundra for the mainRCP-based climate change scenarios which combine temperature and precipitation increases.
As nutrient availability limits plant growth in tundra landscapes (Shaver et al., 2001), a positive effect of warming on nutrient availability is a likely explanation for biomass increase observed in tundra vegetation (Hudson and Henry, 2009).
Climate warming might influence nutrient availability positively by lengthening of the growing season, active layer deepening and increased microbial activity.In our simulations, nutrients were especially limiting for the shrubs, likely due to their shallow rooting systems.Compared to shrubs, graminoids root relatively deep (Wang et al. 2016).As a consequence, active layer deepening is expected to favour especially graminoids.It is, however, unclear how plant root morphology responds to climate warming.An experimental warming study in dry tundra demonstrated that plants do not necessarily root deeper in response to warmer temperatures, but instead may concentrate their main root biomass in the organic layer where most nutrient mineralization takes place (Björk et al., 2007).Nevertheless, it is likely that growing season lengthening and increased microbial mineralization of soil organic matter improve growing conditions for both shrubs and graminoids, as also graminoids have been shown to respond strongly upon fertilization (e.g.Jonasson, 1992).
Strikingly, the climate simulations in this study show that shifts in vegetation composition are not only dependent on temperature change, but are strongly affected by precipitation changes as well.Simulated soil moisture contents decreased with higher temperature and lower precipitation scenarios.Evapotranspiration is an important hydrological process determining soil moisture during the growing season.Throughout the growing season, the top soil layer dries out as evapotranspiration exceeds precipitation during this period, which is in agreement with observations at our study site (Supplement Section S2).Higher summer temperatures increase potential evapotranspiration, and thus lead to drier soils, if precipitation remains unchanged.Consequently, the area of relatively dry sites characterized by dense dwarf shrub vegetation might increase.A similar misbalance of temperature and precipitation change between 1950 and 2002 has been proposed by Riordan et al. (2006) as one of the possible causes for drying of thermokarst ponds in Alaska.A second mechanism for tundra drying with higher temperatures is increased water drainage enabled by gradual deepening of the active layer or permafrost degradation.The latter mechanism is especially important in the discontinuous permafrost zone, where climate change may cause loss of permafrost at thermokarst sites and subsequently lead to increased drainage to adjacent areas (Yoshikawa and Hinzman, 2003).
Whereas NUCOM-tundra predicts that especially graminoids benefit under the main climate change scenarios, other Arctic vegetation models mainly simulate shrub expansion (e.g.Epstein et al., 2000;Euskirchen et al., 2009;Yu et al., 2011).As moisture conditions did not differ much between the combined climate change scenarios (Table 1), the relatively large increase of graminoids in those scenarios is likely enabled by active layer deepening.Nutrients that become available in deeper soil layers are easily accessible by graminoids which are able to root throughout the whole active layer, whereas shrubs rely on the nutrient availability in the top soil layers as their root system is confined to that region.It is important to note that NUCOM-tundra currently uses a simple soil temperature soil module, which calculates soil temperature based on air temperature (Supplement S1.3.4).Especially for deeper soil layers, this module lacks accuracy to precisely predict changes in active layer thickness for different temperature change scenarios, which results in uncertainty in both the depth of the active layer and the amounts of nutrients becoming available.These processes influence graminoid growth more than shrub growth.Coupling NUCOM-tundra to an advanced soil temperature model is needed to simulate the physical soil processes more accurately.The climate simulations in this study show that shifts in vegetation composition are not only dependent on temperature change, but are strongly affected by precipitation changes as well.Simulated soil moisture contents decreased with higher temperature and lower precipitation scenarios.Evapotranspiration is an important hydrological process determining soil moisture during the growing season.Throughout the growing season, the top soil layer dries out as evapotranspiration exceeds precipitation during this period, which is in agreement with observations at our study site (Supplement Section S2).Higher summer temperatures increase potential evapotranspiration, and thus lead to drier soils, if precipitation remains unchanged.Consequently, the area of dry vegetation sites characterized by their dense dwarf shrub coverage might increase.A similar misbalance of temperature and precipitation change between 1950 and 2002 has been proposed by Riordan et al. (2006) as one of the possible causes for drying of thermokarst ponds in Alaska.A second mechanism for tundra drying with higher temperatures is increased water drainage enabled by gradual deepening of the active layer or permafrost degradation.The latter mechanism is especially important in the discontinuous permafrost zone, where climate change may cause loss of permafrost at thermokarst sites and subsequently lead to increased drainage to adjacent areas (Yoshikawa and Hinzman, 2003).
Previously, Tthe effectiveness of shrubs to deal with increased nutrient levels has been proposed as an explanation that shrubs can expand in a warmer climatefor observed shrub expansion in the Arctic (Shaver et al., 2001;Tape et al., 2006).
Betula nana responds to higher nutrient availability by increasing its biomass, which is mainly due to increased secondary stem growth (Shaver et al., 2001).However, Betula nana is also known to respond to increased temperatures and fertilization by growing taller and by producing more shoots and tillers, thereby increasing its ability to compete for light with other species (Chapin and Shaver, 1996;Hobbie et al., 1999;Bret-Harte et al., 2001;Shaver et al., 2001).In NUCOM-tundra, competition for light or nutrients is determined by a combination of Sseveral plant traits (parameters values) are related to competition for light or nutrients.For light competition, leaf area and light extinction are important, which are influenced by parameters as specific leaf area, biomass allocation to leaves, leaf mortality and light extinction coefficient.Similarly, nutrient competition is influenced by parameters as specific root length, root distribution and nutrient requirements.
Although tiller production is not explicitly included in NUCOM-tundra, shrubs had an advantage in the competition for light as they have a higher specific leaf area and higher light extinction coefficient than the graminoids.Yet, this advantage for shrubs was in the combined climate change scenarios overruled by the advantageous ability of graminoids to root deep in the mineral soil layers.Although tiller production is not explicitly included in NUCOM-tundra, shrubs clearly had an advantage in the competition for light as they have a higher specific leaf area and higher light extinction coefficient than the graminoids.Based on our observations at our study site in north-eastern Siberian tundra and at Toolik Lake moist tussock tundra, we assumed the (dwarf) shrubs and graminoids to be equally tall in the model.With warming-induced increases in aboveground biomass for both graminoids and shrubs, the competition for light becomes more important.In parts of the tundra area shrubs, including B. nana, can grow taller than the graminoids.For a wider application of the vegetation model, i.e. to include transitions to tall shrub vegetation, a variable plant height up to a maximum would be required.
The climate simulations in this study show that shifts in vegetation composition are not only dependent on temperature change, but are strongly affected by precipitation changes as well.Simulated soil moisture contents decreased with higher temperature and lower precipitation scenarios.Evapotranspiration is an important hydrological process determining soil moisture during the growing season.Throughout the growing season, the top soil layer dries out as evapotranspiration exceeds precipitation during this period, which is in agreement with observations at our study site (Supplement Section S2).
Higher summer temperatures increase potential evapotranspiration, and thus lead to drier soils, if precipitation remains unchanged.Consequently, the area of dry vegetation sites characterized by their dense dwarf shrub coverage might increase.
A similar misbalance of temperature and precipitation change between 1950 and 2002 has been proposed by Riordan et al. (2006) as one of the possible causes for drying of thermokarst ponds in Alaska.A second mechanism for tundra drying with higher temperatures is increased water drainage enabled by gradual deepening of the active layer or permafrost degradation.
The latter mechanism is especially important in the discontinuous permafrost zone, where climate change may cause loss of permafrost at thermokarst sites and subsequently lead to increased drainage to adjacent areas (Yoshikawa and Hinzman, 2003).
Predicted increases in total biomass and expansion of shrubs are not unique for NUCOM-tundra, but have also been simulated by other Arctic vegetation models (e.g.Epstein et al., 2000;Euskirchen et al., 2009;Yu et al., 2011).OOverall, our results from the climate change scenarios mainly, however, highlight the importance of changes in annual precipitation, as they have a large influence on the water table and thus on the tundra vegetation composition.Precipitation, especially in summer, compensates for evapotranspiration water losses, thereby having a direct positive influence on soil moisture and water table position.Climate models predict considerable precipitation increases for the Arctic region (IPCC, 2014) as the retreat of sea ice in the Arctic Ocean results in strongly increased evaporation and precipitation (Bintanja and Selten 2014).
In agreement with these predictions, significant increases in precipitation over the last decades have been demonstrated for several Arctic weather stations (Hinzman et al., 2005;Urban et al., 2014).Changes in precipitation are, however, less consistent in comparison to climate warming (Urban et al., 2014).In Chokurdakh, the location used for the model simulations, precipitation has not increased along with temperature change between 1981 and 2010 (Temperature +0.0565 ⁰ C year -1 ; Precipitation -1.127 mm year -1 ) (climexp.knmi.nl/atlas;Li et al., 2015).Continuation of the climate trends observed in Chokurdakh between 1981 and 2010 would lead to a scenario that most resembles a temperature increase of 5 ⁰ C and no precipitation change by the end of this century.This scenario would imply that currently mixed vegetation will shift towards shrub dominated vegetation in the near future due to tundra drying.

Thaw Pond Formation
Formation and expansion of shrub vegetation might be set back by the degradation and collapse of ice wedges in the permafrost.Even in dry vegetation type simulations, the initiation of thermokarst led, as a result of water accumulation in thaw ponds, to complete replacement of shrubs by graminoids.The simple thaw pond module in NUCOM-tundra simulated a peak of moss biomass shortly after ground collapse and before significant colonization of graminoids.After the collapse, the shrubs were dying because of the water-saturated soil, while the graminoids had not yet colonized the new thaw pond, and the mosses took advantage of the increased light availability.At the Siberian field site we observe that mosses mostly drown as well and the first colonizers of thaw ponds are mostly graminoids, for example Eriophorum angustifolium (Jorgenson et al., 2006), although also mosses can be first colonizers.NUCOM-tundra does not (yet) take surface water depth into account for the calculation of plant and bryophyte growth, due to which drowning of mosses could not be simulated.The model is also not able to simulate further vegetation succession in thaw ponds.Most likely, after graminoid and moss establishment, an organic layer starts to accumulate, shrubs recolonize the collapsed sites, after which permafrost aggradation can take place.Water runoff out of collapsed sites is expected to increase when the organic layer accumulates, probably aided by newly developing ice structures which lift the surface.Simulated thermokarst events favoured shrub growth in adjacent grid cells due to increased drainage.Although graminoids become dominant in thaw ponds, ice wedge collapses may provide opportunities for shrub growth on the pond margins due to small scale changes in water flows.Such a mechanism was observed in Alaska, where shrubs became dominant around thermokarst spots as a result of increased drainage from shrub-dominated patches to the new ponds (Osterkamp et al., 2009).
These results further reveal that tundra drying is an important mechanism for shrub expansion in the Arctic.Thermokarst ponds, however, may also act as a heat source and thereby stimulate further permafrost collapse (Li et al., 2014).Overall, permafrost degradation should be recognized as an important potentially graminoid stimulating factor when studying climate change in Arctic landscapes (Jorgenson et al., 2006).Rapid expansion of shrubs might be partly compensated by the formation of graminoid dominated thaw ponds, particularly in poorly drained lowland tundra.
Taking into account the formation of thaw ponds and the subsequent vegetation succession is, however, essential for a full analysis of climate change effects on tundra ecosystem (Van Huissteden & Dolman, 2012).This model feature should therefore receive special attention with further model developments.
requires daily temperature and precipitation data as input for the model simulations.Weather data from Chokurdakh Weather Station (WMO station 21946, located 70˚ 62' N, 147˚ 88' E; altitude 44.0m) for the years 1954 to 1994 obtained through the KNMI climate explorer tool (Klein Tank et al., 2002, http://climexp.knmi.nl)were input to the model.Initial values of the model simulations are provided in Supplement Section S3.The biomass start values were based on field measurements at the Chokurdakh Tundra Station.

Figure 1 . 5 Figure 2 .
Figure 1.Schematic presentation of NUCOM-tundra, including the main processes simulated in organic and mineral soil layers and the effects of moisture and available nitrogen on vegetation.Air temperature influences many of these processes, including active layer depth, plant growth, evapotranspiration, snowmelt, decomposition, mineralisation and denitrification.The mineral soil is divided into 10 layers of 10 cm height each.5

Figure 3 .Figure 4 .Figure 5 . 5 Figure 6 .Figure 7 .
Figure 3. Simulation grid used to simulate thaw pond initiation in four different vegetation types.Water flows at the start of the simulation are as indicated in the diagram.After the local permafrost collapse the water flow from the cell with the thaw pond formation to the following downslope cell is stopped (see Figure 6 for the consequences).

Figure 8 .
Figure 8. Biomass of moss, graminoids and shrubs and organic layer moisture content during a thaw pond formation simulation in an initially dry shrub dominated vegetation type (vegetation type 1) under the No change scenario.The interrupted dotted line indicates the moment of ice wedge collapse.