Large carbon cycle sensitivities to climate across a permafrost thaw gradient in subarctic 1 Sweden 2 3

Abstract. Permafrost peatlands store large amounts of carbon potentially vulnerable to
decomposition. However, the fate of that carbon in a changing climate remains
uncertain in models due to complex interactions among hydrological,
biogeochemical, microbial, and plant processes. In this study, we estimated
effects of climate forcing biases present in global climate reanalysis
products on carbon cycle predictions at a thawing permafrost peatland in
subarctic Sweden. The analysis was conducted with a comprehensive
biogeochemical model (ecosys) across a permafrost thaw gradient
encompassing intact permafrost palsa with an ice core and a shallow active
layer, partly thawed bog with a deeper active layer and a variable water
table, and fen with a water table close to the surface, each with distinct
vegetation and microbiota. Using in situ observations to correct
local cold and wet biases found in the Global Soil Wetness Project Phase 3
(GSWP3) climate reanalysis forcing, we demonstrate good model performance by
comparing predicted and observed carbon dioxide (CO2) and methane
(CH4) exchanges, thaw depth, and water table depth. The simulations
driven by the bias-corrected climate suggest that the three peatland types
currently accumulate carbon from the atmosphere, although the bog and fen
sites can have annual positive radiative forcing impacts due to their higher
CH4 emissions. Our simulations indicate that projected precipitation
increases could accelerate CH4 emissions from the palsa area, even
without further degradation of palsa permafrost. The GSWP3 cold and wet
biases for this site significantly alter simulation results and lead to
erroneous active layer depth (ALD) and carbon budget estimates. Biases in
simulated CO2 and CH4 exchanges from biased climate forcing
are as large as those among the thaw stages themselves at a landscape scale
across the examined permafrost thaw gradient. Future studies should thus not
only focus on changes in carbon budget associated with morphological changes
in thawing permafrost, but also recognize the effects of climate forcing
uncertainty on carbon cycling.



Introduction 52
Confidence in future climate projections depends on the accuracy of terrestrial In this study, we extracted the meteorological conditions at the Stordalen Mire 214 from 1901 to 2010 from the GSWP3 climate reanalysis dataset. The 3-hourly products of 215 air temperature, precipitation, solar radiation, wind speed, and specific humidity were 216 interpolated to hourly intervals with cubic spline interpolation to serve as the 217 meteorological inputs used in our model. 218 The The simulation results from CTRL should represent the reliability of applying 268 ecosys at the Stordalen Mire because CTRL is driven by the best local climate 269 description. We first evaluated predicted thaw depth, water table depth, and CO 2 and CH 4 270 exchanges using the CTRL simulation (section 3.2 to 3.4). In the second set of 271 simulations, BIASED-COLD, the biased GSWP3 air temperature data was used, and we 272 corrected only the GSWP3 precipitation. Deviations between CTRL and BIASED-COLD 273 reflect biased air temperature's effects on responses across the thaw gradient. In the third 274 set of simulations, BIASED-WET, we bias-corrected the air temperature extracted from 275 GSWP3, which allows us to quantify the effects of biased precipitation. Finally, we used 276 the meteorological conditions directly extracted from GSWP3 to drive our fourth set of Similarly, the GSWP3 annual total precipitation data correlates well with ANS 305 measurements (r = 0.80) but has a wet bias of 380 mm y -1 between 1913 and 2010 306 ( Figure 2b). An increasing trend in annual total precipitation was recorded in both time 307 series from the early 20 th century to present (0.47 mm y -2 [ANS] and 1.07 mm y -2 308 [GSWP3]), although a decreasing trend was found from 1980 to 2010 (-0.56 mm y -2 309 [ANS] and -2.39 mm y -2 [GSWP3]). 310 The seasonal cycle of the GSWP3 monthly mean air temperature also matches 311 that measured at the ANS, with a very high correlation coefficient (r = 0.99; Figure 3a). 312 The underestimation bias and inter-annual variability of GSWP3 air temperature are 313 greater in winter (maximum underestimate in December, at -4.52 °C with inter-annual 314 variability of 3.53 °C) and smaller in summer (minimum underestimate in July, at -1.52 315 °C with inter-annual variability of 1.65 °C), respectively. 316 The magnitude and inter-annual variability of the GSWP3 monthly mean 317 precipitation are comparable between winter and summer, while the ANS measurements 318 exhibit stronger seasonality with lower magnitudes during winter. Despite the differences 319 found in seasonal patterns, a high correlation coefficient (r = 0.64) was found between the 320 monthly mean precipitation extracted from GSWP3 and the ANS measurements. The 321 overestimation of monthly mean precipitation was greatest in December (43.25 mm 322 month -1 ) and smallest in August (18.75 mm month -1 ). 323 These comparisons suggest that GSPW3 air temperature and precipitation data 324 reasonably capture measured seasonal and long-term trends over past decades, but are

Thaw depth 334
We first evaluated ecosys against observations using bias-corrected climate 335 forcing (i.e., the CTRL simulation). Predicted thaw depth agrees well with measurements 336 varies with peatland type (i.e., relatively slow, moderate, and rapid in the palsa, bog, and 340 fen, respectively). 341 Predicted and observed maximum thaw depth (i.e., ALD) in the intact permafrost 342 palsa was between 45 and 60 cm in September. In the partly thawed bog, the simulated 343 thaw depth is slightly shallower than that observed before August. The simulated bog 344 thaw depth becomes greater than 90 cm by the end of August, which matches the time 345 when measured thaw depth reaches its maximum. The thaw depth becomes greater than 346 accurate observed daily NEE representative of the entire peatland type due to (1) limited 372 daily data points (less than 14% across the study period, Table 1)  show that the magnitude of mean growing season CO 2 uptake is highest in the fen and 419 lowest in the palsa (Table 3). The same rank applies to the magnitude of mean CO 2 420 emissions over the non-growing season, although differences across the thaw gradient are   (Table 1).

Carbon budget responses to climate 471
Simulations with the four climate forcing datasets (section 2.5) indicate annual 472 mean (from 2003 to 2007) CO 2 sinks and CH 4 sources, except the weak CO 2 emissions 473 simulated in the fen in experiment BIASED-COLD&BIASED-WET due to reduced 474 sedge productivity driven by increased temperature and oxygen stresses (Figure 9a,b). 475 Our results also indicate that differences in annual CO 2 and CH 4 exchanges across the 476 four climate forcing datasets for a single peatland type are as large as those across 477 peatland types for a single climate forcing dataset (Figure 9a,b). These large CO 2 and 478 CH 4 exchanges climate sensitivities demonstrate that the peatland's dynamical responses 479 to climate have stronger effects on the carbon cycle than on ALDs (Figure 8). 480 With bias-corrected precipitation, increased air temperature (CTRL vs. BIASED-481 COLD) leads to stronger CO 2 uptake and greater CH 4 emissions at all the examined 482 peatland types (Figure 9a  NGGB simulated in the palsa is generally negative (i.e., a net sink from the atmosphere) 529 due to lower CH 4 emissions, except for the simulation conducted without any climate bias 530 correction (correcting only air temperature increased CH 4 emissions but not enough to compensate for the significantly higher CO 2 sink). Our results indicate that the simulated 532 NGGB would be biased by 298, -66, and -252 g CO 2 -eq m -2 y -1 in the palsa, bog, and fen, 533 respectively, without proper bias correction for the GSWP3 climate reanalysis dataset 534 (BIASED-COLD&BIASED-WET vs. CTRL). Using the GSWP3 products directly thus 535 effectively eliminates the positive radiative forcing from the expanding bog and fen, 536 while creating a potentially dramatically inaccurate positive radiative forcing from the 537 shrinking palsa. 538 539

Climate sensitivity versus landscape heterogeneity 540
Climate sensitivity and landscape heterogeneity are defined here as variability 541 across the four climate forcing datasets for a single peatland type, and variability across 542 three peatland types with bias-corrected climate (CTRL), respectively. We estimated 543 carbon cycle variability associated with climate sensitivity and landscape heterogeneity to 544 quantify the corresponding uncertainty in our annual carbon cycle assessments from 2003 545 to 2007. Our results indicate that differences in simulated annual mean CO 2 exchanges 546 and NCB from climate sensitivity are greater than those from landscape heterogeneity 547 (Figure 9a,c); i.e., annual CO 2 uptake strength is more sensitive to climate forcing 548 uncertainty than to peatland type representation. In terms of the simulated annual mean 549 CH 4 emissions and NGGB, our results indicate that variability from climate sensitivity is 550 comparable to those from landscape heterogeneity (Figure 9b,d). Therefore, bias-551

Conclusions 570
We evaluated the climate bias in a widely used atmospheric reanalysis product 571 (GSWP3) at our northern Sweden Stordalen Mire site. We then applied a comprehensive 572 biogeochemistry model, ecosys, to estimate the effects of these biases on active layer 573 development and carbon cycling across a thaw gradient at the site. Our results show that 574 ecosys reasonably represented measured hydrological, thermal, and biogeochemical cycle cold and wet biases in the GSWP3 climate reanalysis dataset significantly alter model 577 simulations, leading to biases in simulated Active Layer Depths, Net Carbon Balance, 578 and Net Greenhouse Gas Balance by up to 28.6%, 38 g C m -2 y -1 , and 298 g CO 2 -eq m -2 579 y -1 , respectively. The Net Carbon Balance simulated with bias-corrected climate suggests 580 that all the examined peatland types are currently net carbon sinks from the atmosphere, 581 although the bog and fen sites can have positive radiative forcing impacts due to their 582 higher CH 4 emissions. 583 Our results indicate that the annual means of ALD, CO 2 uptake, and CH 4 584 emissions generally increase along the permafrost thaw gradient at the Stordalen Mire 585 under current climate, consistent with previous studies in this region. Our analysis 586 suggests that palsa, bog, and fen differ strongly in their carbon cycling dynamics and 587 have different responses to climate forcing biases. Differences in simulated CO 2 and CH 4 588 exchanges driven by uncertainty from climate forcing are as large as those from 589 landscape heterogeneity across the examined permafrost thaw gradient. Model 590 simulations demonstrate that the palsa site exhibits the strongest sensitivity to biases in 591 air temperature and precipitation. The wet bias in GSWP3 could erroneously increase 592 predicted CH 4 emissions from the palsa site to a magnitude comparable to emissions 593 currently measured in the bog and fen sites. These results also show that increased 594 precipitation projected for high latitude regions could strongly accelerate CH 4 emissions 595 from the palsa area, even without degradation of palsa into bog and fen. Future studies 596 should thus recognize the effects of climate forcing uncertainty on carbon cycling, in 597 addition to tracking changes in carbon budgets associated with areal changes in 598 permafrost degradation.