Plant community functional composition and diversity 1 drive fine-scale variability in carbon cycling in the 2 tundra 3

1. The functional composition and diversity of plant communities are globally applicable predictors of ecosystem functioning. However, exactly how traits influence carbon cycling is yet unclear, as are the implications in a warming world. 2. To study how traits affect carbon cycling in the tundra environment, we built a hierarchical model that included abiotic conditions (summer air and winter soil temperatures, and soil resources), plant community functional composition and diversity (plant size and leaf economics), and carbon cycling (above-ground and soil organic carbon stocks, and photosynthetic and respiratory fluxes). We also simulated warming effects on peak-season ecosystem CO​2​ budget. 3. Plant size was the strongest predictor of all carbon cycling variables except soil carbon stocks. Communities of larger plants were associated with larger CO​2​ fluxes and above-ground carbon stocks. Communities with fast leaf economics had higher rates of photosynthesis and soil respiration, but lower above-ground biomass. 4. Diversities on axes of size and leaf economics affected ecosystem functions differently. Leaf economic diversity increased CO​2​ fluxes and soil organic carbon stocks, while size diversity increased the above-ground carbon stock. The contributions of functional diversity metrics to ecosystem functioning were about as important as average leaf economic traits. 5. Simulations suggested that warmer summer air temperatures increase plant size, while warmer winter soil temperatures increase plant size and accelerate leaf economics. All these changes would enhance CO​2​ uptake during peak growing season. Synthesis: We show how traits mediate the link between abiotic conditions and carbon cycling. Community composition and diversity on the two axes of the global spectrum of plant form and function have clear and separate effects on ecosystem functioning. Warmer temperatures increase plant size and accelerate leaf economics, which leads to faster net assimilation of carbon during peak growing season. More research on soil carbon stocks is needed.

We investigated how plant functional traits influence fine-scale patterns of tundra 26 carbon cycling, and how carbon cycling responds to climate warming. 27 We built a hierarchical model that included abiotic conditions (summer air and winter 28 soil temperatures, and soil resources), plant community functional composition and 29 diversity (plant size and leaf economics), and carbon cycling (above-ground and soil 30 organic carbon stocks, and photosynthetic and respiratory fluxes). We also simulated 31 warming effects on the peak-season CO 2 budget. 32 Plant size was the strongest predictor for most carbon cycling variables. Communities 33 of larger plants were associated with larger CO 2 fluxes and above-ground carbon 34 stocks. Communities with fast leaf economics had higher rates of photosynthesis and 35 soil respiration, but lower soil organic carbon stocks. Leaf economic diversity 36 increased CO 2 fluxes, while size diversity increased the above-ground carbon stock. 37 Simulations suggested that warmer summer air temperatures increase plant size and 38 accelerate leaf economics, while warmer winter soil temperatures increase plant size. 39 Both changes would enhance CO 2 uptake during the peak season. 40 We show that traits act as mediators between abiotic conditions and carbon cycling. controlling vegetation changes and carbon cycling in this complex system to be teased apart 120 (Díaz et al., 2007). To our knowledge, such an approach has not been considered much thus 121 far. A deeper understanding of the proximate (vegetation) and ultimate (environment) drivers 122 of carbon cycling is therefore needed to identify climate change -related feedbacks in Arctic 123 ecosystems. 124

125
In this study, we examine the relationships between abiotic environmental conditions, plant 126 community functional composition and diversity, and carbon cycling in a tundra landscape 127 using a unique study setting of 87-200 intensively measured plots and coupled Bayesian 128 regression models. To understand carbon cycling comprehensively, we focus not only on CO 2 129 fluxes (gross primary productivity, ecosystem respiration, soil respiration), but also on carbon 130 stocks (above-ground and soil organic carbon stocks; for abbreviations used in materials and 131 methods, and results, see Table 1). Finally, we simulate how summer and winter warming 132 could cause cascading effects on the carbon uptake capacity of the ecosystem, measured as 133 the peak-season CO 2 budget. 134  Height was quantified as the height of the highest leaf on two random ramets within the 227 collar. LDMC was measured from two leaf samples taken from two different ramets. If only 228 one ramet of a species was present within the collar, height was measured only once, and two 229 leaf samples were taken from that one ramet. The leaf samples were taken at the time of 230 community surveys, put in re-sealable plastic bags with moist paper towels, and transported 231 to the lab to be stored at 4 for up to three days, before weighing for fresh mass. probe GMP343 (A1A1N0N0N0A model; CO 2 range 0-1000 ppm, ± 3 ppm + 1%), an air 251 humidity and temperature probe HMP75 (temperature range between -20 and 60°C, relative 252 humidity range 0-100 %; temperature accuracy at 20°C ± 0.2°C, relative humidity accuracy 253 between -20 and + 40°C ± (1.0 + 0.008 * relative humidity reading)), and a measurement 254 indicator MI70 (Vaisala, Vantaa, Finland). In the chamber, CO 2 concentration, air temperature 255 and relative humidity were recorded at 5-s intervals for 90 s. Concentrations of CO 2 were 256 already corrected for atmospheric pressure and relative humidity during the measurements. is the change in CO 2 concentration in time, M is the molecular mass of CO 2 (44.01 287 g mol -1 ), Vmol the molar volume (22.4*10 -3 m 3 mol -1 ) and Vc the volume of the air space in 288 the chamber, and A the area of the collar. The flux unit was µmol m -2 s -1 . 289

Materials and Methods
Carbon stock data 290 We measured the depth of the soil organic and mineral layers from three points on each plot 291 using a metal probe. In the analyses, we used the mean value of the three point measurements 292 to represent the organic and mineral layer depths in each plot. We collected samples of 293 roughly 1 dl from the soil organic and mineral layers with metal soil core cylinders (4 to 6 cm 294 in diameter, 5 to 7 cm height). The organic samples were collected from the top soil, and 295 mineral samples directly below the organic layer. The soil samples were taken ca.
We had the C% and bulk density information for all organic samples, but we had 54 mineral 311 soil samples with C% and 70 mineral samples with bulk density estimates from the study 312 area. We used the median of these samples (3% for C% and 830 kg m -3 for bulk density) for 313 the carbon stock estimates in all mineral layers, but the mineral layer depth varied across the 314 sites. We consider this reliable due to the relatively low variability in mineral soil C% (0.4-315 6.5%) and bulk density (470-1400 kg m -3 ). In some plots, we used soil organic matter content 316 (SOM%) instead of C%. For more details, see Supporting Information, Methods S2. 317 318 Finally, organic and mineral layer stocks were summed together to calculate the total SOC 319 stock. Plots with no soil were excluded from the analysis. 320 321 Above-ground vascular plant biomass was collected between the 1 st and 10 th of August from 322 the collars. Biomass samples were oven-dried at 70°C for 48 h and weighed after drying. 323 Above-ground carbon stocks (AGC) were estimated by multiplying the total biomass by 324 0.475 (Schlesinger, 1991). In the analysis, we refer to these variables as SOC and AGC. We also used this model together with PAR and temperature logger data to simulate plot-340 specific NEE with 10 minute intervals for the 30 day time period between the 8 th of July and 341 7 th of August 2017. Temperature measurements were first interpolated from 2-4 h to 10 342 minute resolution. We summed the predictions to create an estimate of peak-season CO 2 343 budget. 344 345 Soil respiration was normalized to 20 with a Bayesian linear mixed model. The model 346 included a plot-specific random intercept and a linear temperature response. All priors were 347 left as the brms defaults. We refer to this variable hereafter as SR. 348 349 GPP and ER were calculated for all plots from which CO 2 measurements were taken (n = 350 200). Some of the steel collars were displaced between measurements of NEE and SR, thus 351 SR was only measured in 192 plots. Peak-season CO 2 budget was calculated for all plots in 352 which CO 2 fluxes were measured and were also continuously monitored for air temperature 353 (n = 103). GPP, ER, and SR are always shown with positive signs. Positive numbers for NEE 354 and CO 2 budget indicate net CO 2 gain to the ecosystem (i.e. CO 2 sink) and negative numbers 355 indicate net CO 2 loss to the atmosphere (i.e. CO 2 source). 356

Hierarchical model of tundra carbon cycling 357
We built a hierarchical model of tundra carbon cycling by combining five submodels. The We used the carbon cycling model to simulate warming effects on ecosystem capacity to 382 sequester carbon via changes in plant community functional composition and diversity. We 383 predicted the peak-season CO 2 budget for median environmental conditions, a winter 384 warming scenario (+1 February soil), a summer warming scenario (+1 July air 385 temperature), and a combination of the two warming scenarios. For the scenarios with 386 summer warming, we ran the simulations with and without direct temperature effects on 387 summer respiration. Direct temperature effects were included in the models by multiplying 388 predicted ER by the exponential of the temperature sensitivity parameter (β Temperature , Eqn S2) 389 from the previously fitted light-response model. We did not take into account residual 390 uncertainty in the predictions. We propagated uncertainty by evenly sampling 500 predictions Models with only trait-related covariates predicted carbon cycling as accurately -or even 405 more accurately -than models with both environmental and trait covariates (Fig. 3). By the 406 principle of parsimony, we therefore focus our analysis on the simpler models. The simulated peak-season CO 2 budget was 1.4 times more sensitive to changes in ER than in 437 GPP (Fig. 4d). Fluctuating light and temperature levels and the nonlinear light response curve 438 did not cause major nonlinearities in the determination of the budget, indicated by the 0.98 R² 439 of the linear model. The expected peak-season CO 2 budget was positively affected by the 440 functional changes brought about by simulated summer and winter warming, even when the 441 direct effects of respiration were taken into account (Fig. 6). The probability of an ecosystem 442 patch being a summer net CO 2 sink grew in all but one scenario. In the simulation with 443 summer warming and a direct temperature effect, the probability of being a sink fell by one 444 percentage point. diversity. One such mechanism is that increased productivity could lead to higher litter inputs 512 from leaves and roots (DeMarco et al., 2014). Since leaf economic diversity did not increase 513 above-ground carbon stocks, our interpretation is that the positive diversity effect on soil 514 organic carbon stocks is probably due to increased root litter production, indicative of higher 515 total investments in acquiring below-ground resources. These observations would be 516 consistent with more efficient partitioning of below-ground resources, which would also 517 explain the positive effects of leaf economic diversity on CO 2 fluxes. Conclusions 560 We found that in a tundra landscape, fine-scale carbon cycling was parsimoniously explained 561 by plant community functional composition and diversity, and that long-term environmental 562 conditions had only indirect effects. Average plant size was the strongest predictor of most 563 carbon cycling variables, but leaf economic diversity mattered the most for soil organic 564 carbon stocks. Plant size increased CO 2 fluxes and above-ground carbon stocks, whereas fast 565 leaf economics were associated with higher CO 2 fluxes and lower above-ground carbon and as strong as key drivers of productivity.    r  n  e  l  d  e  n  s  i  t  y  p  l  o  t  s  o  f  t  h  e  p  o  s  t  e  r  i  o  r  d  i  s  t  r  i  b  u  t  i  o  n  s  o  f  t  r  a  i  t  e  f  f  e  c  t  s  a  b  o  v  e  -g  r  o  u  n  d   864   c  a  r  b  o  n  s  t  o  r  a  g  e  .  T  h  e  l  o  w  e  r  x  -a  x  i  s  i  s  o  n  t  h  e  o  r  i  g  i  n  a  l  s  c  a  l  e  ,  w  h  i  l  e  t  h  e  u  p  p  e  r  a  x  i  s  d  e  p  i  c  t  s  t  h