Quantifying the role of moss in terrestrial ecosystem carbon dynamics in 1 northern high-latitudes 2

Abstract. Mosses are ubiquitous in northern terrestrial ecosystems, and play an important role in regional carbon, water and energy cycling. Current global land surface models that do not consider mosses may bias the quantification of regional carbon dynamics. Here we incorporate mosses as a new plant functional type into the process-based Terrestrial Ecosystem Model (TEM 5.0), to develop a new model (TEM_Moss). The new model explicitly quantifies the interactions between vascular plants and mosses and their competition for energy, water, and nutrients. Compared to the estimates using TEM 5.0, the new model estimates that the regional terrestrial soils currently store 132.7 Pg more C and will store 157.5 and 179.1 Pg more C under the RCP8.5 and RCP2.6 scenarios, respectively, by the end of the 21st century. Ensemble regional simulations forced with different parameters for the 21st century with TEM_Moss predict that the region will accumulate 161.1±142.1 Pg C under the RCP2.6 scenario and 186.7±166.1 Pg C under the RCP8.5 scenario over the century. Our study highlights the necessity of coupling moss into Earth system models to adequately quantify terrestrial carbon–climate feedbacks in the Arctic.


effects on plant growth due to their regulation of nitrogen availability for vascular plants (Hobbie 86 et al., Gornall et al., 2007). It is gradually being recognized that mosses can have 87 comparable influences on high-latitude ecosystems to vascular plants, due to their large density 88 and essential function in plant competition, soil climate, and carbon and nutrient cycling 89 (Longton, 1988;Lindo and Gonzalez, 2010;Okland, 1995;Pharo and Zartman, 2007). They can 90 on average contribute 20% of aboveground NPP in boreal forests (Turetsky et al., 2010), and 91 their annual NPP may reach as high as 350 g C m -2 in some regions in the Arctic (Pakarinen and 92 Vitt 1973), even exceeding that of vascular plants (Oechel and Collins, 1976; Clarke et al.,  where w m is moss moisture (units: mm), and wmin, wmax, and wopt are related parameters (units: 149 mm) that limit f (wm) to a range of zero to one.  170 * (w m ) = 1 − (w m −w min −w opt,r ) 2 (w m −w min ) * w opt,r +w opt,r 2 (7) 171 where w opt,r (units: mm) denotes the optimal water content for moss respiration.

172
Besides, the carbon in litter production from mosses to soil (L C,m ) is modeled as 173 proportional to moss carbon biomass with a constant ratio (Zhuang et al., 2002): where MOSSC denotes the moss carbon biomass, and cfallm is the corresponding constant 176 proportion.

177
Thus, the change of moss carbon pool (MOSSC) can be modeled as: k n (units: g m -2 ) is the concentration of available soil nitrogen at which nitrogen uptake proceeds 188 at one-half its maximum rate. T is the monthly mean air temperature ( o C), and Am is a unitless 189 parameter ranging from 0 to 1, which represents relative allocation of effort to carbon vs. where wf (units: mm) denotes the moss field capacity.

195
The nitrogen in litter production from mosses to soil (L N,m ) is modeled as proportional to where nfallm is the constant proportion to moss nitrogen biomass (MOSSN).

199
Thus, the changes in moss nitrogen pool (MOSSN) can be modeled as:

201
At the same time, total carbon and nitrogen in litterfall, and total nitrogen uptake from 202 soil available nitrogen are changed due to incorporation of mosses: 204 To simulate moss moisture, we added a moss layer on the soil profile by modifying the WBM 224 ( Figure 2). Similar to soil moisture, moss moisture is also treated as a state variable in the revised 225 WBM, which is modeled as: where the term "percolation" denotes the percolation from moss, which is the sum of rainfall 228 percolation and snowmelt percolation from moss. We assume that there is no runoff from moss 229 layer.

230
Accompanied by the above equation, changes in soil water (SM) is modified as:

236
The newly introduced parameters that are associated with moss activities were documented 237 in Where NEP obs,i and NEP sim,i are the measured and simulated NEP, respectively. k is the number 249 of data pairs for comparison. Fifty independent sets of parameters were converged to minimize the 250 objective function, and finally the optimized parameters were derived as the mean of these 50 sets

264
We verified the TEM_Moss simulated NEP, soil moisture and soil temperature. First, we 265 conducted site-level simulations at six sites that contain level-4 gap-filled monthly NEP data from 266 the AmeriFlux network (Table 3). Site-level monthly gap-filled soil moisture and soil temperature 267 data were organized from the ORNL DAAC Dataset (https://daac.ornl.gov/) to make comparison 268 with model simulations (Table 4 and Table 5). Local climate data including monthly air 269 temperature ( o C), precipitation (mm), and cloudiness (%) were obtained to drive these model 270 simulations.

272
With six site-level calibrated parameters, TEM-Moss is applied to the region pixel by pixel based were analyzed. We denoted that a positive NEP represents a CO2 sink from the atmosphere to 292 terrestrial ecosystems, while a negative value represents a source of CO2 from terrestrial 293 ecosystems to the atmosphere.
In these simulations, for each pixel, we assumed its moss distribution area is the same as 295 the vascular plant distribution. The total carbon uptake/emission of mosses in a pixel are calculated 296 as the multiplication of pixel area with the carbon fluxes such as NEP (units: g C m -2 month -1 ).

297
Moss-related parameters for representative ecosystems are calibrated ( Fig. 4

302
TEM_Moss was able to reproduce the monthly NEP and performed better than TEM 5.0 303 at chosen sites, with larger R-square values and smaller RMSE ( Figure 6, for TEM 5.0 are from 0.62 to 0.75 (Table 6). On the other hand, RMSE for TEM_Moss is lower 309 than that for TEM 5.0 at each site (Table 6). 310 We presented the comparisons between measured and simulated volumetric soil moisture  (Table 7). R-squares for TEM_Moss are substantially higher than that for TEM 5.0 at most 314 chosen sites, except for US-Atq (Table 7). RMSE for TEM_Moss is lower than that for TEM 5.0 315 at each site (Table 7). Similarly, comparisons between measured and simulated soil temperature at 5 cm depth (ST_5) from TEM_Moss and TEM 5.0 indicated that TEM_Moss can reproduce 317 the soil temperature with R-squares ranging from 0.81 at US-Ho1 to 0.91 at US-Bkg, while TEM 318 5.0 reproduces the soil temperature with R-squares ranging from 0.69 at BE-Vie to 0.89 at US-319 Bkg ( Figure 8; Table 8). Although R-squares for both models are relatively high and RMSE for 320 them are relatively low, TEM_Moss still shows higher R-squares and lower RMSE than TEM 321 5.0 (Table 8). indicated that the northern high-latitude region stored 3.05 Pg C yr -1 , which is more than twice as 333 the storage estimated by TEM 5.0 (1.33 Pg C yr -1 , Figure 9). Both models indicated that carbon 334 uptake by the northern ecosystems during the second half of the 20 th century was higher than the 335 first half for most part of the region, and only a small portion of the region lost carbon in last 336 century ( Figure 10).

337
Simulated total NPP by TEM_Moss was 9.6 Pg C yr -1 , ranging from 8.52 Pg C yr -1 to 338 10.65 Pg C yr -1 in the 20 th century, with 1.69 Pg C yr -1 of moss NPP and 7.93 Pg C yr -1 of vascular plants NPP (Figure 9). Moss NPP ranges from 1.23 Pg C yr -1 to 2.14 Pg C yr -1 and the 340 ratio of moss NPP to vascular plants NPP is 0.21 (Figure 9). TEM 5.0 estimated 0.8 Pg C yr -1 341 lower total NPP than TEM_Moss, but 0.87 Pg C yr -1 higher NPP for vascular plants (Figure 9).

342
On the other hand, average heterotrophic respiration in the 20 th century was 7.38 Pg C yr -1 and 343 all years were within about 5% of this value (Figure 9). TEM 5.0 projected 0.53 Pg C yr -1 higher 344 RH than TEM_Moss (7.91 Pg C yr -1 , Figure 9). Overall, TEM_Moss predicted higher total NPP 345 but lower RH, which jointly caused a pronounced difference in NEP between two models.

346
Both models estimated that soil organic carbon and vegetation carbon were accumulating 347 continuously in the 20 th century ( Figure 11). TEM_Moss indicated that regional SOC and VEGC  during the 21 st century (Figure 12 (a)). The regional sink shows a decreasing pattern in the 2000s 359 and then generally increases over the remaining years of the 21 st century (Figure 12 (a)). For 360 comparison, TEM 5.0 predicted that the average NEP of 0.28 Pg C yr -1 with the range from -1.48 Pg C stored in northern ecosystems is less than the estimation from TEM_Moss in the 21 st 363 century. Besides, TEM 5.0 simulated that the regional NEP showed a decreasing trend and the 364 region fluctuates between sinks and sources during the century (Figure 12 (a)). The spatial 365 patterns from two models also showed differences. TEM_Moss indicated that the region 366 accumulates carbon over this century, while TEM 5.0 simulated that some regions changed from 367 a carbon sink to a source in the second half of the century (Figure 13 (a)). Simulated regional 368 NPP by TEM_Moss ranges from 11.2 to 13.7 Pg C yr -1 with a mean of 12.98 Pg C yr -1 in this 369 century, while average NPP predicted by TEM 5.0 is 1.46 Pg C yr -1 lower than that value (11.52 370 Pg C yr -1 (Figure 12(a)). TEM_Moss simulated NPP has 3.74 Pg C yr -1 from moss and 9.24 Pg C 371 yr -1 from vascular plants, which account for 28.8% and 71.2% of total NPP, respectively ( Figure   372 12(a)). Meanwhile, TEM_Moss estimated that RH is 10.91 Pg C yr -1 , while TEM 5.0 predicted it 373 as 11.24 Pg C yr -1 , which is higher (Figure 12(b)). Both models projected that soil organic carbon 374 and vegetation carbon accumulate in this century but with different magnitudes (Figure 14 (a)).

375
TEM_Moss predicted that regional SOC and VEGC accumulated 84.7 Pg C and 112.6 Pg C, TEM_Moss also predicted an increasing of 9.4 Pg C in MOSSC, accounting for 4.5% of the total 379 carbon uptake in this region (Table 12(a)).

380
Under the RCP 8.5 scenario, TEM_Moss simulated annual NPP of 13.84 Pg C yr -1 with a 381 range from 11.09 to 16.94 Pg C yr -1 , which is 1.31 Pg C yr -1 higher than the projection from 382 TEM 5.0 (Figure 12 (b)). Total NPP estimated by TEM_Moss has 3.84 Pg C yr -1 from moss and 383 10 Pg C yr -1 from vascular plants (Figure 12(b)). Annual RH was 11.28 Pg C yr -1 estimated by 384 TEM_Moss and 11.54 Pg C yr -1 by TEM 5.0, respectively (Figure 12(b)). Consequently, TEM_Moss projected NEP was 2.56 Pg C yr -1 with the inter-annual standard deviation of 0.93 386 Pg C yr -1 in this century (Figure 12(b)). NEP ranges from 0.67 Pg C yr -1 to 4.78 Pg C yr -1 387 estimated with TEM_Moss, while from -1.69 Pg C yr -1 to 2.65 Pg C yr -1 with a mean of 0.99 Pg 388 C yr -1 was estimated by TEM 5.0 (Figure 12(b)). TEM_Moss predicted more carbon uptake of 389 157.5 Pg than TEM 5.0 during the 21 st century. Both models predicted that NEP showed an 390 increasing trend during the 21 st century (Figure 12(b)). Moreover, similar spatial patterns of 391 carbon sinks and sources appeared in the projections from two models (Figure 13(b)). Soil

392
organic carbon and vegetation carbon shows an increasing trend from both models ( Figure   393 14(b)). Regional SOC and VEGC increased by 92.5 Pg C and 153.6 Pg C, respectively by the 394 end of the 21 st century predicted by TEM_Moss. In contrast, the increase of 44.2 Pg C and 54.5 Pg

396
TEM_Moss predicted an increase of 10.1 Pg C in MOSSC (Table 12(b)).  respectively. In the 1990s, the regional sink is projected to be 2.7 and 1.1 Pg C yr -1 by 438 TEM_Moss and TEM 5.0 respectively. Compared with other existing studies, our regional . 471 We conducted ensemble regional simulations with 50 sets of parameters to quantify 472 model uncertainty due to uncertain parameters. The 50 sets of parameters were obtained using 473 the method in Tang and Zhuang (2008). The ensemble means and the inter-simulation standard 474 deviations are used to measure the model uncertainty ( Figure 15). TEM_Moss predicted that the

489
In addition, we lack spatially explicit information of moss distribution in the region, which will 490 lead to a large regional uncertainty of carbon quantification. We assumed that moss area 491 distribution is the same as its associated vegetation distribution. Another limitation is that some