Responses of alpine grassland on Qinghai Tibetan plateau to climate warming and permafrost degradation: a modeling perspective

Transcription

1 Environmental Research Letters LETTER OPEN ACCESS Responses of alpine grassland on Qinghai Tibetan plateau to climate warming and permafrost degradation: a modeling perspective To cite this article: Shuhua Yi et al 2014 Environ. Res. Lett View the article online for updates and enhancements. Related content - Variation and control of soil organic carbon and other nutrients in permafrost regions on central Qinghai-Tibetan Plateau Wenjie Liu, Shengyun Chen, Qian Zhao et al. - Storage, patterns, and control of soil organic carbon and nitrogen in the northeastern margin of the Qinghai Tibetan Plateau Wenjie Liu, Shengyun Chen, Xiang Qin et al. - Response characteristics of vegetation and soil environment to permafrost degradation in the upstream regions of the Shule River Basin Shengyun Chen, Wenjie Liu, Xiang Qin et al. Recent citations - The impacts of climate change and human activities on alpine vegetation and permafrost in the Qinghai-Tibet Engineering Corridor Lihui Luo et al - Function-related Drivers of Skull Morphometric Variation and Sexual Size Dimorphism in a Subterranean Rodent, Plateau Zokor (Eospalax baileyi ) Junhu Su et al - Quantifying the streamflow response to frozen ground degradation in the source region of the Yellow River within the Budyko framework Taihua Wang et al This content was downloaded from IP address on 18/06/2018 at 16:30

2 Environmental Research Letters Environ. Res. Lett. 9 (2014) (12pp) doi: / /9/7/ Responses of alpine grassland on Qinghai Tibetan plateau to climate warming and permafrost degradation: a modeling perspective Shuhua Yi 1, Xiaoyun Wang 1,2, Yu Qin 1, Bo Xiang 3 and Yongjian Ding 1 1 State Key Laboratory of Cryospheric Sciences, Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Sciences, 320 Donggang West Road, , Lanzhou, Gansu, People s Republic of China 2 University of Chinese Academy of Sciences, , Beijing, People s Republic of China 3 Chongqing Meteorological Administration, , Chongqing, People s Republic of China yis@lzb.ac.cn Received 22 April 2014, revised 1 July 2014 Accepted for publication 2 July 2014 Published 28 July 2014 Abstract Permafrost plays a critical role in soil hydrology. Thus, the degradation of permafrost under warming climate conditions may affect the alpine grassland ecosystem on the Qinghai Tibetan Plateau. Previous space-for-time studies using plot and basin scales have reached contradictory conclusions. In this study, we applied a process-based ecosystem model (DOS-TEM) with a state-of-the-art permafrost hydrology scheme to examine this issue. Our results showed that 1) the DOS-TEM model could properly simulate the responses of soil thermal and hydrological dynamics and of ecosystem dynamics to climate warming and spatial differences in precipitation; 2) the simulated results were consistent with plot-scale studies showing that warming caused an increase in maximum unfrozen thickness, a reduction in vegetation and soil carbon pools as a whole, and decreases in soil water content, net primary production, and heterotrophic respiration; and 3) the simulated results were also consistent with basin-scale studies showing that the ecosystem responses to warming were different in regions with different combinations of water and energy constraints. Permafrost prevents water from draining into water reservoirs. However, the degradation of permafrost in response to warming is a long-term process that also enhances evapotranspiration. Thus, the degradation of the alpine grassland ecosystem on the Qinghai Tibetan Plateau (releasing carbon) cannot be mainly attributed to the disappearing waterproofing function of permafrost. S Online supplementary data available from stacks.iop.org/erl/9/074014/mmedia Keywords: alpine grassland, permafrost degradation, climate warming, modeling 1. Introduction Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. The Qinghai Tibetan Plateau (QTP) is the highest and largest plateau in the world. Because of its high altitude, permafrost is widespread, covering about half of the total area of the QTP. Climate warming and permafrost degradation have been reported in various studies of the QTP (Wu and Zhang 2010, Jin et al 2011). Due to the low rate of water movement in /14/ $ IOP Publishing Ltd

3 frozen soil (Zhang et al 2010), permafrost degradation changes hydrological regimes (Ye et al 2009). It is hypothesized that permafrost degradation lowers water tables and dries surface soil (Niu et al 2011, Cheng and Wu 2007), and thus can cause degradation of alpine grassland, which is the QTP s major type of vegetation (Zhang et al 1988). There are no long-term ecological study sites in the permafrost regions of the QTP, and short-term experimental warming studies using open-top chambers or infrared heating cannot impact the thermal state of permafrost in the short term. Therefore, most previous studies have used the spacefor-time method to investigate the effects of permafrost degradation on the QTP s alpine grassland ecosystem. The space-for-time method compares the soil and vegetation characteristics of selected plots or belts over a chronosequence of permafrost degradation. Studies at different spatial scales have reached contradictory conclusions. Studies at the plot scale have found that permafrost degradation causes degradation of alpine grassland, including the reverse succession of alpine grassland, (i.e., from alpine swamp meadow to alpine meadow to alpine steppe meadow to alpine steppe (Wang et al 2006, Yang et al 2013). Further, it also causes the release of soil organic carbon (SOC) (Wang et al 2008) and reduces fractional vegetation cover (FVC), which is one of the most important and intuitive vegetation characteristics of alpine grassland ecosystems (defined as the proportion of vertical projection of green leaf on the ground to the total ground area (Qin et al 2014). In contrast, Yi et al (2011) studied the FVC of a semi-arid basin on the northeast rim of the QTP at the landscape/regional scale using a remote sensing dataset. Their results showed that the FVC increased when permafrost degraded from extreme stable permafrost at a mean annual ground temperature (MAGT) (depth of about 15 m) of < 5 C to substable permafrost ( 3 C < MAGT < 1.5 C) and decreased when permafrost degraded further to a seasonal frozen area (MAGT > 0.5 C). However, in an adjacent basin with more precipitation, FVC increased as permafrost degraded (Zhou et al under review). Zhou et al speculated that warming might benefit alpine grassland growth in permafrost regions. Permafrost degradation is primarily a consequence of climate warming, which can also dry soils by enhancing evapotranspiration. The previously mentioned space-for-time method cannot separate these effects. Numerical modeling is a suitable tool to separate the direct effects of climate warming and those of permafrost degradation on an alpine grassland ecosystem. Several models have been used to simulate the responses of alpine grassland ecosystems to climate change, e.g., CENTURY (Zhang et al 2007), ORCHIDEE (Tan et al 2010, Piao et al 2012), and TEM (Zhuang et al 2010). The results of studies using the TEM model suggested that warming would increase the thawing rate of permafrost, increase soil temperature, and increase the drying of soil, all of which might enhance the future strength of the ecosystem s carbon sink, as the increasing rate of net primary production (NPP) would become higher than that of soil respiration on the QTP. However, few models considered the effects of permafrost on hydrology processes. In this study, we answered the following two questions using a process-based biogeochemical model with a state-ofthe-art scheme for permafrost hydrology processes: 1) how will the alpine grassland ecosystems in the permafrost regions of the QTP respond to climate warming and 2) does the permafrost degradation play an important role in soil water dynamics? 2. Methodology 2.1. Model descriptions and modifications In this study, we used a dynamic organic soil version of the Terrestrial Ecosystem Model (DOS-TEM) to simulate the soil thermal and hydrological processes and the carbon dynamics of alpine grassland on the QTP. The DOS-TEM consists of four modules; however, only the environmental and ecological modules were used in this study (Yi et al 2010). The environmental module simulates the soil s thermal and water dynamics. The simulated soil temperatures were validated using field measurements from sites in the boreal forests of Alaska (Yi et al 2009), alpine grassland on the QTP (Yi et al 2013a, b), and the arctic tundra in Siberia (Yi et al 2013c). The ecological module simulates the carbon and nitrogen fluxes among atmosphere, vegetation, and soil, and the carbon and nitrogen pools of vegetation and soil using the soil and atmosphere environments. Previous versions of DOS-TEM used a scheme similar to the Community Land Model (Version 3) to simulate soil water hydrology, including infiltration, surface runoff, soil water redistribution among layers, and drainage out of the soil (Yi et al 2009). The Community Land Model, Version 4 (CLM4; Oleson et al 2010) included an underlying aquifer below the soil column as a special soil layer. Water table depth (WTD) is calculated based on the amount of water in the reservoir. The total soil water content (including both liquid and ice) is used to calculate hydraulic conductivity and matric potential, which are required for updating the soil water content of each layer. A factor called the soil ice impermeable fraction is then used to account for the effects of ice on water movement. However, Swenson et al (2012) found that only liquid water should be used in the equations to calculate hydraulic properties and provided a new way to calculate the soil ice impermeable fraction. With these modifications, CLM4 simulated better hydrographs at both Yenisey and Lena Rivers with various coverage of permafrost (Swenson et al 2012). In this study, we followed the CLM4 and Swenson et al (2012) by including a water reservoir and calculating the hydraulic properties in the DOS-TEM. With these modifications, the exchange of water between the last soil layer and the aquifer is negligible when the last soil layer is frozen. However, when it is thawed, the exchange rate of water can be large. Soil above the permafrost table is usually wet due to a low rate of water movement in the underlying permafrost. Lateral runoff from the wet soil, which is called the perched zone by Swenson et al (2012), is common when 2

4 Figure 1. Spatial distribution of meteorological stations (black dots) and the locations of two sites used to calibrate the ecological parameters (red squares). slope is greater than 0. We, therefore, also simulated this runoff following Swenson et al (2012). We resampled the vector map of permafrost used by Li et al (2008; figure 1) and the vegetation type to 0.5. There are three types of permafrost on the QTP: 1) mountainous permafrost on the northern and southern parts of the QTP and 2) continuous and 3) discontinuous permafrost, which occupy the central part of the QTP. The MAGT of continuous permafrost is usually about C, whereas that of discontinuous permafrost is about C. A total of grid cells had greater than 10% coverage of permafrost in alpine grassland. Monthly climate data at 0.5, including m nthly mean air temperature, precipitation, radiation, and vapor pressure from the Climate Research Unit (CRU) dataset, were used to drive the DOS-TEM. Alpine meadow and alpine steppe are the two major alpine grassland types on the QTP. Both vegetation types may exist in the same grid cell. We defined each different combination of climate and vegetation type as a cohort. There were 708 cohorts altogether. Other required input static data, e.g., slope and soil texture data, were also resampled to 0.5 ; they were assumed homogeneous in a grid cell. These data preparations are described in detail in the Supplementary Materials. We used 5.1 m thick soil above 50 m thick bedrock. For the soil in the top 1 m, we determined the soil texture using our spatial dataset. We assumed the texture of the soil between 1 and 5.1 m to be sand, representing the poor water holding capacity of gravel, which is common on the QTP (Yi et al 2013b). As initial conditions, all soil and bedrock layers were assumed to be 1 C and the soil layers were assumed to be saturated. Table 1. Mean NDVI/slope/number of grid cells with combinations of different trends (decrease, no trend, and increase) in the NDVI and simulated VEGC. Overall provides the average or sum of each column or row. NDVI No Decrease trend Increase Overall Decrease 0.09/ 3.53/5 VEGC No trend 0.16/ 2.41/ 10 Increase 0.11/ 1.71/ 19 Overall 0.12/ 2.18/ Model calibration 0.12/ 3.75/ / 2.07/ / 1.56/ / 1.82/ / 0.75/1 0.18/ 0.98/ / 1.21/ / 1.13/ / 3.52/ / 1.83/ / 1.48/ / 1.67/ 482 In this study, we calibrated rate-limiting parameters (e.g., maximum rate of photosynthesis, plant and heterotrophic respiration rate, etc) for the alpine steppe and alpine meadow following the protocols given in McGuire et al (1992). The vegetation carbon (VEGC) and nitrogen pools, SOC, organic nitrogen and inorganic nitrogen, gross primary production (GPP), NPP, and NPP without nitrogen limitation were used for calibration. We used the values at Anduo ( N, 3

5 Table 2. Simulated results of the simulations using the CRU dataset unchanged (No Warming) and with 1, 2, and 3 C increases in air temperature with and without consideration of water exchange between soil and water reservoir (D and ND). Meaning and unit of each variable remained the same as shown in figure 3. The values were averaged over cohorts that changed from permafrost to seasonal frozen area in the No Warming simulation. No Warming +1 C +2 C +3 C D ND D ND D ND D ND MUT WTD VWC20CM VWC5M EET QCHARGE NPP VEGC RH SOC E, 4871 m a.s.l.) and West Mila Mountain ( N, E, 4652 m a.s.l.) to represent alpine steppe and meadow, respectively [see tables 1 and 2 of appendix of Zhuang et al (2010)]. The mean monthly climate data for the period from the grid cells corresponding to these two sites were used for calibration Model validation The mean monthly climate data for the period were first used for an equilibrium run. Then the monthly climate data for the period were used in a cycling way for a spin-up run (120 years). Finally, the monthly climate data for the period were used for a transient run. This entire simulation was designated as CONTROL. The simulated VEGC of each cohort over the period was first averaged by 0.5 grid cell and the cells were weighted according to cohort areas. Then, the average and linear trends were calculated (using the Python Scipy package) and finally compared with those of the combined normalized difference vegetation index (NDVI) dataset of GIMMS and MODIS (see Supplementary Material) Sensitivity tests To investigate the effects of climate warming on permafrost and alpine grassland ecosystems, we performed three simulations in which the air temperature was increased by 1 C (W1), 2 C (W2), and 3 C (W3). We used the same initial conditions as in the CONTROL simulation, but only ran the transient run. We compared three aspects of all simulations: 1) time series of the averages of all variables over all cohorts; 2) spatial patterns of the variables averaged over the period (when a grid contained two vegetation types, the variable was first weight-averaged based on the area of each vegetation type); and 3) we first divided all cohorts into three categories based on the thermal state of each cohort, as determined in the CONTROL simulation (i.e., with permafrost throughout the whole simulation period, without permafrost throughout the whole simulation period, and changed from with permafrost to without permafrost). Then we compared the four simulations on each cohort category. We also performed simulations to investigate the effects of permafrost degradation on soil water content without considering the exchange of water between the water reservoir and the last soil layer, even when it was thawed. We ran these simulations with an air temperature increase of 0, 1, 2, and 3 C and used the same initial conditions as in the CONTROL simulation. These results were then compared with the corresponding simulations with consideration of water exchange between the soil and water reservoir. 3. Results 3.1. Retrospective analysis Comparison with NDVI data. Over the period, the mean of the simulated VEGC in the CONTROL run had a similar spatial pattern to that of the NDVI high in the east and low in the west (figure 2). However, the simulated VEGC was considerably larger than estimates from NDVI south of 31 N. At the grid cell level, 5 of the 482 grid cells had a significantly decreasing trend (p < 0.05) in both the VEGC and NDVI, 117 had no significant trend (p 0.05) in either the VEGC or NDVI, and 80 had a significantly increasing trend (p < 0.05) in both the VEGC and NDVI (table 1). Of the cells, 198 had a significantly increasing trend in the VEGC but no significant trend in the NDVI. These discrepancies occurred mostly to the west of 97 E. In the grid cells with a significantly decreasing trend, no significant trend, or a significantly increasing trend in the NDVI, the average slope decreased 2.18, 1.82, and 1.67, respectively; the average of the NDVI increased 0.12, 0.19, and 0.25, respectively (table 1). The trends in the simulated VEGC were similar to those in the NDVI. 4

6 Figure 2. Spatial distribution of (a) mean of NDVI; (b) mean of simulated VEGC (gc/m 2 ); (c) linear trend of NDVI [red/light blue: increase/ decrease significantly (p < 0.05), and green: no significant change]; and (d) linear trend of simulated VEGC [trend; gc/(m 2 yr)] over the period Temporal dynamics. At the beginning of the CONTROL transient run, 82% of the cohorts contained permafrost. About 18% of the cohorts in the permafrost regions disappeared over the period. There were two periods with obvious air temperature increases; one was between 1935 and 1950 and the other was after 1980 (see Supplementary Material). The maximum unfrozen thickness (MUT), which equates to active layer thickness for cohorts with permafrost and equals the total soil thickness, i.e., 5.1 m for cohorts without permafrost, showed the same variation pattern in all runs (figure 3(a)). It was approximately 2.2 and 2.6 m at the beginning and end of the simulation, respectively. The WTD was about 9.8 m, and only reduced very slightly in the simulation (figure 3(b)). The volumetric water content (VWC) of the shallow soil was high (>0.33), but was reduced during the two periods of increased air temperature (figure 3(c)). Changes in the deep soil VWC showed a similar pattern to that of the WTD (figure 3(d)). In the first warming period, heterotrophic respiration (RH) increased abruptly and remained at a relatively high level (figure 3(j)). Both NPP and VEGC increased at the beginning of these periods, but then decreased slightly. SOC decreased from >5400 gc m 2 to <5200 gc m 2 (figure 3(i)). In the second period of warming, RH, NPP, and VEGC all increased, whereas SOC remained at 5200 gc m Spatial patterns. Over the period, the mean MUTs were greater at the rims of the permafrost regions. In some regions, shallow permafrost disappeared [red grid cells in figure 4(a)]. In grid cells without permafrost, the WTD was small (water table is shallow) in the southwest of the study area, where precipitation was large, but was almost unchanged where the amount of precipitation was small (figure 4(b)). The VWCs of both shallow and deep layers had similar spatial patterns to the MUT (figures 4(c) and (d)). In grid cells without permafrost, the VWC values in the southwest region were much higher than in the northern region. The simulated VEGC, SOC, NPP, and RH averaged over the period generally decreased from southeast to northwest (figures 4(e) (h)). The maximum values of the VEGC, NPP, SOC, and RH were greater than 2200 gc m 2, 350 gc/(m 2 yr), gc m 2, and 400 gc/(m 2 yr), respectively. However, in the northwest regions, the values were less than 200 gc m 2, 50gC/(m 2 yr), 1000 gc m 2, and 50 gc/(m 2 yr), respectively. The net ecosystem production (NEP) had a complex spatial pattern (figure 4(i)), with an average of 4gC/(m 2 yr) Effects of climate warming At the end of the simulations, the number of cohorts with permafrost decreased from 442 to 350, 253, and 157, when air temperature increased by 1, 2, and 3 C, respectively. When all cohorts were averaged, an increase in air temperature caused an increase in MUT from 1.98, 2.53, 3.07 to 3.56 m (figure 3(a)), a decrease in WTD from 9.93, 9.74, 9.41 and 9.11 m [which suggested water recharging into water reservoir and water table became shallower, figure 3(b)], and decreases in VWC at both the shallow and deep soil layers (figures 3(c) and (d)). Evapotranspirations increased from 172, 184, 192 to 199 mm yr 1 (figure 3(e)). The first peak of 5

7 Figure 3. Simulated (a) MUT (m); (b) WTD (m); (c) VWC at 20 cm (-); (d) VWC at 5 m (-); (e) evapotranspiration (EET, mm/yr); (f) water recharge rate into reservoir (QCHARGE, mm/yr); (g) VEGC (gc/m 2 ); (h) NPP [gc/(m 2 yr)]; (i) SOC stock (gc/m 2 ); and (j) RH [gc/(m 2 yr)] of simulations using the unchanged CRU dataset (CONTROL, black) and with 1 C (W1, green), 2 C (W2, blue), and 3 C (W3, red) increases in air temperature. The values were averaged over all cohorts. recharging of water into water reservoir occurred earlier when the air temperature was increased, with a maximum of 10 mm yr 1 when the air temperature was increased by 3 C (figure 3(f)). NPP was slightly decreased due to drier soil condition and caused a slight decrease in VEGC (figures 3(g) and (h)). The amount of SOC, soil temperature, and moisture are three important factors in determining RH. Increase of soil temperature will increase RH. When soil moisture is at half saturation, RH is the largest (if other conditions remain the same) and an increase or decrease of soil moisture cause a decrease of RH. Therefore, at the beginning of the simulations, warming caused a higher soil temperature and smaller soil moisture, both of which facilitated RH (figure 3(i)). The RH of the W3 simulation was about 70 gc/(m 2 yr) (or 37%) greater than that of the CONTROL simulation. SOC is determined by RH and litter fall input, which are proportional to VEGC. SOC decreased abruptly when the air temperature increased due to greater RH and smaller litter fall at the beginning of simulations. With the reduction of SOC, the RH of warming simulations were smaller than that of the CON- TROL simulation after 1935, while litter fall inputs were still less than the CONTROL simulation. Therefore, the SOCs of warming simulations decreased rapidly at the beginning of simulations and then decreased gently. At the end of the 6

8 Figure 4. Spatial distributions of simulated mean (a) MUT (m); (b) WTD (m); (c) VWC at 20 cm (-); (d) VWC at 5 m (-); (e) VEGC (gc/m 2 ); (f) NPP [gc/(m 2 yr)]; (g) SOC stock (gc/m 2 ); (h) RH [gc/(m 2 yr)]; and (i) NEP [gc/(m 2 yr)] over the period of the CONTROL simulation. simulations, the SOCs of warming simulations were about 900, 1400, and 1900 gc m 2 (or 17, 26, and 35%) smaller than the CONTROL simulation when the air temperature was increased by 1, 2 and 3 C, respectively (figure 3(j)). Warming caused larger MUTs and smaller WTDs (shallower water table) in the southern area, drying both shallow and deep soil layers. For example, figures 5(a) (d) show the differences between the W3 and the control simulations. The soil in the shallow layer was obviously drier (VWC was reduced by 0.3) in the eastern part of the continuous permafrost region (figure 5(c)). Similarly, the VEGC, NPP, and SOC had the most obvious decreases in this region (figures 5(e) (g)). For most of the grid cells, the NEP of the W3 simulation was smaller than in the CONTROL simulation (figure 5(i)). When averaged over the entire simulation period ( ), air temperature increase caused the most obvious increase in MUT within cohorts that had permafrost throughout the entire CONTROL simulation period (hereafter COHORT-P) (figure 6(a)). MUTs were the largest, but showed no changes during the warming periods in cohorts that did not have permafrost in the top 5.1 m of soil (hereafter COHORT-S). WTD showed the most obvious decrease (water table became shallow) in cohorts that had permafrost at the beginning but not at the end of the CONTROL simulation (hereafter COHORT-P2S) (figure 6(b)). VWCs of both the shallow and deep soil layers showed clear decreases in COHORT-P and COHORT-P2S and were the smallest in COHORT-S (figures 6(c) and (d)). VEGC, NPP, and RH all showed slight increases in COHORT-P, but showed slight decreases in COHORT-P2S and COHORT-S when the air temperature was increased by 1 or 2 C. The values of these variables were the largest in COHORT-P2S (figures 6(e) (h)). The SOC and NEP of the three cases all clearly decreased when the air temperature increased (figures 6(g) and (i)); SOC and NEP reduction were the highest in COHORT-P2S Effects of permafrost degradation The sensitivity tests with and without the exchange of water between the soil and the water reservoir had the most obvious effects over cohorts that changed from permafrost at the beginning to a seasonal frozen area at the end of the CON- TROL simulation. The simulations did not consider the exchange of water between the lowest soil layer, and the water reservoir was designated with NODRAIIN. For example, W3-NODRAIN refers to the simulation with a 3 C increase in air temperature and no exchange of water. 7

9 Figure 5. Same data as figure 4, but representations of the differences between simulations with 3 C warming and the control simulation averaged over the period. Generally, the effects of shutting down the exchange of water become greater for simulations with increasing air temperature, which was caused by earlier occurrence of recharging water into water reservoir (table 2). NODRAIN simulations did not cause changes of soil thermal state, e.g., MUT (table 2). As expected, NODRAIN simulations had no recharge into the water reservoir and caused a constant value of WTD, which was greater than simulations that considered water exchange (table 2). NODRAIN simulations caused obvious increases of deep soil water content, especially when the air temperature was increased; however, the changes of shallow soil water content and evapotranspiration were very small. NODRAIN simulations caused very small increases of NPP and VEGC. For example, the differences of NPP and VEGC between the W3- NODRAIN and W3 simulations were both <1%. Those between the W3 and W2/W1 simulations were 5.3%/11.1% and 5.0%/11.0%. A slight increase of soil water content in the NODRAIN simulations caused slight increases of RH and SOC. For example, the differences of RH and SOC between the W3-NODRAIN and W3 simulations were <4 and 7%, respectively. Those between the W3 and W2/W1 simulations were 3.0%/6.0% and 5.9%/14.1%. 4. Discussion 4.1. Comparisons with other studies The simulated results for NPP, VEGC, SOC using the DOS- TEM had spatial patterns similar to those reported in other studies a general decrease from the southeast to the northwest (figure 4; Piao et al 2006, Yang et al 2008, Yang et al 2009, Tan et al 2010). Over the period, the SOC simulated in this study varied slightly, as in Zhang et al (2007), with variations of less than <0.5%. However, the simulation in Zhuang et al (2010) showed an obvious increase in SOC of 4.5% over the same period. As the models used in Zhuang et al (2010) and in this study used the same ecological processes and dataset, the discrepancy might be related to the different simulations of the hydrological processes. The VWCs of the shallow soil layers in the two studies showed similar temporal patterns, but the relative variation in VWC in Zhuang et al (2010) was about 5.3% (between and 0.536), whereas in this study it was 21% (between 0.33 and 0.40). The average simulated annual NPP for the entire study area was approximately 199 gc/(m 2 yr). The annual NPP estimated by Piao et al (2006) was about 122 gc/(m 2 yr) between 1982 and 1999 for alpine grassland on the QTP. The 8

10 Figure 6. Simulated (a) MUT (m); (b) WTD (m); (c) VWC at 20 cm (-); (d) VWC at 5 m (-); (e) VEGC (1000 gc m 2 ); (f) NPP [100 gc/ (m 2 yr)]; (g) SOC (1000 gc m 2 ); (h) RH ([100 gc/(m 2 yr)]; and (i) NEP [gc/(m 2 yr)] of simulations with the unchanged CRU dataset (C), and with 1 C (W1), 2 C (W2), and 3 C(W3) increases in air temperature. The values were averaged over the period for the grid cells with permafrost (P), without permafrost (S), and with vanishing permafrost (P2S) during the CONTROL simulation period. values for the simulations of Zhuang et al (2010) using TEM and Tan et al (2010) using ORCHIDEE were about 321 and 233 gc/(m 2 yr), respectively, for grassland on the QTP. The mean of the simulated SOC was about 5200 gc m 2, which is less than the value of 6500 gc m 2 estimated by Yang et al (2008) using a remote sensing dataset and the value of 8600 gc m 2 simulated by Tan et al (2010). This study s underestimation may be due to 1) the restriction to alpine grassland in permafrost regions and 2) the fact that Yang et al (2008) estimated SOC down to 1 m, whereas the SOC used in our calibration was estimated with maximum root depth. Existing studies typically dig soil pits down to 40 cm on the QTP (e.g., Qin et al 2014). The simulated NEP was about 4 gc/(m 2 yr), indicating a weak carbon sink in the permafrost regions. The NEP simulated by Piao et al (2012) over the same period on the QTP was about 17.6 gc/(m 2 yr). None of the current studies on ecosystem modeling considered the soil carbon pool in deep soil, e.g., >1 m, or in permafrost soil on the QTP, mainly for two reasons. First, as mentioned earlier, most of the sampling works only take samples of the topsoil ( 40 cm); few works take samples with depth greater than 1 m. Therefore, it is unrealistic to consider the deep soil pool and predict its change. Second, we currently do not know how or how quickly the soil carbon of deep soil accumulates. Nevertheless, future work should consider deep soil carbon. We speculated the estimated NEP might then become closer to zero or even negative values Effects of warming and permafrost degradation on soil moisture Soil moisture plays an important role in the carbon dynamics of alpine grassland ecosystems. For example, in a study by Baumann et al (2009), soil moisture explained 64 and 60% of the variations in SOC and soil total nitrogen stocks, respectively. Further, Yang et al (2009) found that soil moisture interacting with climatic variables could explain 54% of the variation in SOC. Therefore, it is important to simulate the spatial pattern, temporal dynamics, and vertical distributions of soil water content on the QTP. Except for the work of the TEM model (Zhuang et al 2010), few current studies on ecosystem modeling about alpine grassland of the QTP mention soil water. Further, none considered the effects of permafrost degradation on soil water. Soil moisture in shallow layers is mainly controlled by precipitation, surface runoff 9

11 and infiltration, evapotranspiration, and the redistribution of water in soil layers. Increasing air temperatures enhance evapotranspiration. For the entire study area, VWC in the shallow layer was reduced during the two identified periods of increasing air temperature (figure 3). The effects of climate warming on soil moisture were also seen in the warming simulations (figure 6). However, there were large spatial heterogeneities in shallow soil moisture over the study area. For example, in the center of the continuous permafrost regions, VWC was high, whereas in the Qilian Mountain region (the northeastern part of the study area), VWC was much lower (figure 4(c)). Soil moisture in deep layers is also controlled by the redistribution of water in soil layers and by the exchange of water between the soil and water reservoir when the soil is not frozen. Permafrost degradation is a slow process. Before permafrost disappears and water can drain into a water reservoir, soil already becomes drier because of enhanced evapotranspiration on the topsoil and water redistribution within the soil. Therefore, permafrost degradation does affect soil water content, especially deep soil water content. However, its effects were smaller than those of enhanced evapotranspiration Responses of alpine grassland ecosystems to warming Our modeling result was consistent with previous plot-scale studies that showed warming degrades permafrost, dries surface soil, reduces productivity, and releases carbon from both vegetation and soil carbon pools (Wang et al 2008, Yang et al 2013). Our study was also consistent with studies using remote sensing dataset at the basin scale, which have shown that energy and water are the two major factors controlling the growth of alpine grassland ecosystems. For example, Yi et al (2011) found that the FVC of alpine grassland in extreme stable permafrost regions and in seasonal frozen regions were small. Vegetation growth in the former was constrained by energy; an increase in air temperature would extend the length of growing season, accelerate nutrient cycling, and benefit the growth of alpine grassland. However, in the latter, it was constrained by water. They found that alpine grassland in transit permafrost regions with optimum energy and water conditions had the highest FVC. Similarly, in our study s CONTROL simulation, the highest NPP and VEGC were in the cohorts that had permafrost at the beginning of the simulation period, but not at the end (figure 6). Increasing the air temperature mitigated the energy constraints on plant growth in cohorts with permafrost during the CONTROL simulation. It caused a slight increase in NPP and VEGC, although shallow soil moisture was reduced remarkably. In contrast, increasing the air temperature exacerbated water constraints on plant growth in cohorts without permafrost in the CONTROL simulation Uncertainties and limitations Atmospheric driving. Zhang et al (2007) and Tan et al (2010) used interpolated meteorological data for the ORCHIDEE and CENTURY models of the QTP. However, the distribution of the meteorological stations on the QTP is very heterogeneous, with a high density in the eastern part and very sparse density in the western part of the QTP. The CRU dataset has wide applications. Zhuang et al (2010) used it to study carbon dynamics on the QTP. However, comparisons between meteorological station data and CRU data for corresponding grid cells were not good (see Supplementary Material). The averages for the simulated VEGCs to the south of 32 N were as high as those in the eastern area, whereas the NDVI only showed such high values in the eastern area (figure 2). Similarly, most of the trends in the simulated VEGCs were positive, whereas those in the NDVI were negative. These discrepancies might be related to the spatial pattern of precipitation in the CRU dataset (see Supplementary Material). For example, there was a high center of precipitation in the southern part of the study area, whereas the meteorological station data showed a decreasing trend from the southeast to northwest (Piao et al 2006). As discussed in section 4.3, the combination of energy and water plays an important role in simulated carbon dynamics. To quantitatively estimate the carbon pools of the alpine grassland ecosystems on the QTP, including their spatial distribution and temporal dynamics, a more realistic climate dataset is required Gravelly soil and disturbances. In addition to soil moisture and temperature, soil texture is another important factor that affects alpine grassland carbon dynamics (Yang et al 2008, Baumann et al 2009). Clay and silt content stabilizes SOC, reduces carbon leaching, and holds soil moisture. The major soil texture used in this study, taken from FAO/IIASA/ISRIC/ISSCAS/JRC was silty clay. However, soil profiles on the QTP contain large amounts of gravels (diameter > 2 mm), which are usually sieved before further processing. Gravel has different thermal and hydrological properties, e.g., porosity, thermal conductivity, water retention curve, hydraulic conductivity, etc, than finergrained soils (Yi et al 2013b). Strong freezing and thawing cycles, overgrazing, and rodent excavating make fine soils vulnerable to wind and water erosion and the accumulation of gravel. For most of the grid cells north of 32 N and west of 88 E, the NDVI data showed no significant trends, whereas the simulated VEGC showed a significant positive trend. The topsoil of the western part of the continuous permafrost of the QTP contains high quantities of gravel and has sparse vegetation cover. Therefore, we speculate that gravel content and erosion processes, which were not considered in this study, might be the reason for the mismatch between NDVI and VEGC. The effects of gravel on soil properties, spatial and vertical distribution, and dynamics due to erosions should be considered in future ecosystem modeling. 10

12 5. Conclusions In this study, we implemented a state-of-the-art permafrost hydrology scheme in an ecosystem model to simulate the responses to climate warming in alpine grassland ecosystems in the permafrost regions of the QTP. The modified model can properly simulate the responses of soil temperature and water content to changes in air temperature and permafrost degradation and to spatial patterns of precipitation. Sensitivity tests showed that increases in air temperature would cause increases in soil temperature and active layer depth and decreases in soil moisture, NPP, and RH. Vegetation and soil carbon pools in the alpine grassland ecosystem in the permafrost regions of the QTP would become smaller. It is worth mentioning that the responses were different in different regions according to their different precipitation regimes and stages of permafrost degradation. When making mitigation and adaptation policies, it is necessary to bear in mind these regional differences. Permafrost prevents water from draining away. However, degradation of permafrost is a slow, long-term process, which is a response to warming, that also enhances evapotranspiration. Degradation of alpine grassland ecosystems (i.e., less vegetation and soil carbon pools) cannot be mainly attributed to the disappearing waterproofing function of permafrost. Acknowledgments We would like to thank Dave McGuire for his helpful suggestions and comments; Huijun Jin for clarifying the classification of permafrost type on the Qinghai Tibetan Plateau; the Data Assimilation and Modeling Center for Tibetan Multi-spheres, Institute of Tibetan Plateau Research, Chinese Academy of Sciences for help with radiation data; and the Environmental and Ecological Science Data Center for West China, National Natural Science Foundation of China ( westdc.westgis.ac.cn) for providing the permafrost, vegetation, and NDVI data. This study was jointly supported by grants from the Major State Basic Research Development Program of China (973 Programme) (no. 2010CB951402), the Strategic Priority Research Program (XDB030303), independent grants from the State Key Laboratory of Cryospheric Sciences (SKLCS-ZZ ), and the Chinese National Natural Science Foundation Commission ( ). References Baumann F, He J S, Schmidt K, Kühn P and Scholten T 2009 Pedogenesis, permafrost, and soil moisture as controlling factors for soil nitrogen and carbon contents across the Tibetan plateau Global Change Biol Cheng G D and Wu T H 2007 Responses of permafrost to climate change and their environmental significance, Qinghai-Tibet plateau J. Geophys. Res. 112 F02S03 Jin H J, Luo D L, Wang S L, Lv L Z and Wu J C 2011 Spatialtemporal variability of permafrost degradation on the Qinghai-Tibetan plateau Sciences in Cold and Arid Regions Li X, Cheng G D, Jin H J, Kang E S, Che T, Jin R, Wu L Z, Nan Z T, Wang J and Shen Y P 2008 Cryospheric change in China Glob. Planet. Change McGuire A D, Melillo J M, Joyce L A, Kicklighter D W, Grace A L, Moore B III and Vorosmarty C J 1992 Interactions between carbon and nitrogen dynamics in estimating net primary productivity for potential vegetation in North America Global Biogeochem. Cy Niu L, Ye B S, Li J and Sheng Y 2011 Effect of permafrost degradation on hydrological processes in typical basins with various permafrost coverage in Western China Sci. China Ser. D-Earth Sci. (EN) Oleson K W et al 2010 Technical Description of Version 4.0 of the Community Land Model (CLM) NCAR University Corporation for Atmospheric Research Piao S L, Fang J Y and He J S 2006 Variations in vegetation net primary production in the Qinghai-Xizang plateau Clim. Change Piao S L, Tan K, Nan H J, Ciais P, Fang J Y, Wang T, Vuichard N and Zhu B 2012 Impacts of climate and CO 2 changes on the vegetation growth and carbon balance of the Qinghai-Tibetan grasslands over the past five decades Global Planet. Change Qin Y, Yi S, Ren S, Li N and Chen J 2014 Responses of typical grasslands in a semi-arid basin on the Qinghai-Tibetan plateau to climate change and disturbances Environ. Earth Sci Swenson S C, Lawrence D M and Lee H 2012 Improved simulation of the terrestrial hydrological cycle in permafrost regions by the community land model J. Adv. Model. Earth Syst. 4 M08002 Tan K, Ciais P, Piao S L, Wu X P, Tang Y H, Vuichard N, Liang S and Fang J Y 2010 Application of the ORCHIDEE global vegetation model to evaluate biomass and soil carbon stocks of Qinghai-Tibetan grasslands Global Biogeochem. Cy. 24 GB1013 Wang G X, Li Y S, Wu Q B and Wang Y B 2006 The relationship between permafrost and vegetation and its effects on cold ecosystem Sci. China Ser. D-Earth Sci (In Chinese with English abstract) Wang G X, Li Y S, Wang Y B and Wu Q B 2008 Effects of permafrost thawing on vegetation and soil carbon pool losses on the Qinghai-Tibet plateau, China Geoderma Wu Q B and Zhang T J 2010 Changes in active layer thickness over the Qinghai-Tibetan Plateau from 1995 to 2007 J. Geophys. Res. 115 D09107 Yang Y H, Fang J Y, Tang Y H, Ji C J, Zheng C Y, He J S and Zhu B 2008 Storage, patterns and controls of soil organic carbon in the Tibetan grasslands Global Change Biol Yang Y H, Fang J Y, Pan Y D and Ji C J 2009 Aboveground biomass in Tibetan grasslands J. Arid Environ Yang Z P, Gao J X, Zhao L, Xu X L and Ouyang H 2013 Linking thaw depth with soil moisture and plant community composition: effects of permafrost degradation on alpine ecosystem on the Qinghai-Tibet plateau Plant Soil Ye B S, Yang D Q, Zhang Z L and Kane D L 2009 Variation of hydrological regime with permafrost coverage over Lena Basin in Siberia J. Geophys. Res. 114 D07102 Yi S H et al 2009 Interactions between soil thermal and hydrological dynamics in the response of Alaska ecosystems to fire disturbance J. Geophys. Res. 114 G02015 Yi S H, McGuire A D, Kasischke E, Harden J, Manies K, Mack M and Turetsky M 2010 A dynamic organic soil 11

13 biogeochemical model for simulating the effects of wildfire on soil environmental conditions and carbon dynamics of black spruce forests J. Geophys. Res. 115 G04015 Yi S H, Zhou Z Y, Ren S L, Xu M, Qin Y, Chen S Y and Ye B S 2011 Effects of permafrost degradation on alpine grassland in a semi-arid basin on the Qinghai-Tibetan plateau Environ. Res. Lett Yi S H, Li N, Xiang B, Wang X, Ye B and McGuire A D 2013a Representing the effects of alpine grassland vegetation cover on the simulation of soil thermal dynamics by ecosystem models applied to the Qinghai-Tibetan Plateau J. Geophys. Res. Biogeosci Yi S H, Chen J, Wu Q and Ding Y 2013b Simulating the role of gravel on the dynamics of active layer and permafrost on the Qinghai-Tibetan plateau The Cryosphere Discuss Yi S H, Wischnewski K, Langer M, Muster S and Boike J 2013c Modeling different freeze/thaw processes in heterogeneous landscapes of the Arctic polygonal tundra using an ecosystem model Geosci. Model Dev. Discuss Zhang J, Wang J T, Chen W, Li B and Zhao K 1988 Vegetation of Xizang (Tibet) (Beijing: Science Press China) in Chinese Zhang Y Q, Tang Y H, Jie J and Yang Y H 2007 Characterizing the dynamics of soil organic carbon in grasslands on the Qinghai- Tibetan plateau Sci. China Ser. D-Earth Sci Zhang Y, Carey S K, Quinton W L, Janowicz J R, Pomeroy J W and Flerchinger G N 2010 Comparison of algorithms and parameterizations for infiltration into organic-covered permafrost soils Hydrol. Earth Syst. Sci Zhou Z Y, Yi S H, Chen J J, Ye B S, Sheng Y and Wang G Responses of alpine grassland to climate warming under different precipitation regimes in the permafrost regions of the Qinghai-Tibetan plateau Arct. Antarct.Alp. Res. under review Zhuang Q, He J, Lu Y, Ji L, Xiao J and Luo T 2010 Carbon dynamics of terrestrial ecosystems on the Tibetan plateau during the 20th century: an analysis with a process-based biogeochemical model Global Ecol. Biogeogr