Seasonal and Inter-Annual Variations in Carbon Dioxide Exchange over an Alpine Grassland in the Eastern Qinghai-Tibetan Plateau

This work analyzed carbon dioxide exchange and its controlling factors over an alpine grassland on the eastern Qinghai-Tibetan Plateau. The main results show that air temperature and photosynthetically active radiation are two dominant factors controlling daily gross primary production. Soil temperature and soil water content are the main factors controlling ecosystem respiration. Canopy photosynthetic activity is also responsible for the variation of daily ecosystem respiration other than environmental factors. No clear correlation between net ecosystem exchange and environmental factors was observed at daily scale. Temperature sensitive coefficient was observed to increase with larger soil water content. High values of temperature sensitive coefficient occurred during the periods when soil water content was high and grass was active. Annual integrated net ecosystem exchange, gross primary production and ecosystem respiration were -191, 1145 and 954 g C m-2 for 2010, and -250, 975 and 725 g C m-2 for 2011, respectively. Thus, this alpine grassland was a moderate carbon sink in both of the two years. Compared to alpine grasslands on the Qinghai-Tibetan Plateau, this alpine grassland demonstrated a much greater potential for carbon sequestration than others. Annual precipitation is a dominant factor controlling the variation of annual net ecosystem exchange over this grassland. The difference in gross primary production between the two years was not caused by the variation in annual precipitation. Instead, air temperature and the length of growing season had an important impact on annual gross primary production. Variation of annual ecosystem respiration was closely related to annual gross primary production and soil water content during the growing season.


Introduction
Terrestrial ecosystems play a crucial role in global carbon balance. Grasslands, as an important part of terrestrial ecosystems, comprise 32% of Earth's natural vegetation [1]. Estimated study site, the multi-year average of annual air temperature is 1.9˚C and the annual mean precipitation is 593 mm with most of the precipitation occurring between May and September. More details on the site have been reported in the previous documents [18,19].

Flux measurement
An eddy covariance (EC) system was used to continuously measure the flux of CO 2 . The EC system was mounted 3.15 m above the soil surface. It consists of a 3D sonic anemometer (CSAT-3, Campbell Scientific, Inc., Logan, UT, USA) and an open path and fast response infrared gas analyzer (LI-7500, LI-COR Biosciences Inc., Lincoln, NE, USA). The separation distance between the two sensors was 0.15 m. An air temperature and relative humidity sensor (HMP-45C, Vaisala, Helsinki, Finland) was also installed at the same height, which was used for correction of flux measurements for density effects due to heat and water vapor transfer (only the mean temperature and humidity from the slow sensor can be used in the correction). Signals from EC instrumentation were recorded at the rate of 10 Hz, and the raw data were stored in a CR3000 data logger (Campbell Scientific, Inc.). The dominant prevailing winds at the site are easterlies and southeasterlies in the summer and northwesterlies in the winter. The fetch is greater than 1.5 km for all directions at the site. Footprint analysis using a model [20] indicated that the peak for the flux footprint was approximately 58 m upwind of the system, with 90% cumulative flux footprint extending to approximately 168 m upwind (the values are calculated using the average of the measurements in 2010).
In order to assess the accuracy of the EC measurements, linear regression analysis between the sum of sensitive heat flux (H) and latent heat flux (LE) versus available energy (net radiation (R n ) minus ground heat flux (G0)) was conducted. During 2010, the intercept, slope and coefficient of determination (r 2 ) were 10.1 W m -2 , 0.80, and 0.89, respectively. In 2011, we attained a relatively low closure degree of the surface energy balance. The intercept, slope and coefficient of determination (r 2 ) were 5.2 W m -2 , 0.77, and 0.79 respectively in 2011. However, the closure degrees for the two years were close to the mean value for 50 site-years (0.79 ± 0.01) across 22 sites in FLUXNET [21].

Meteorology and soil measurements
Meteorological and soil variables were also measured continuously with an array of sensors. Net radiation flux and photosynthetic photon flux density were measured at 1.5 m height with a four-component net radiometer (CNR-1, Kipp and Zonen, Delft, Netherlands) and a quantum sensor (LI-190Sb, LI-COR Biosciences Inc., Lincoln, NE, USA), respectively. Precipitation was measured with a weighing gauge (T200B, Geonor, Norway) at 2 m height. Soil temperature was measured at 1, 3, 5, 10 cm depths and other deeper layers with CS107 temperature probes (Campbell Scientific, Inc.). Volumetric soil water content was measured at 5 and 10 cm depths and other deeper layers with CS616 Time Domain Reflectometer (TDR) probes (Campbell Scientific, Inc.). Soil heat flux was measured using heat flux plates (HPF01, Wohlwend Engineering, Sennwald, Switzerland) buried at 2 and 7 cm below the soil surface and deeper depths. Signals from meteorological and soil sensors were recorded as half-hourly averages with a CR23XTD data logger (Campbell Scientific, Inc.).

Data processing and gap filling
The flux data were calculated off-line using the EddyPro software [22]. The raw data were processed to obtain 30-min averages. The main procedures included spike detection and removal [23], double coordinate rotation [24], sonic air temperature correction [25], frequency response correction [26] and correction for the effect of air density fluctuations [27].
Missing data is unavoidable and universal in continuous field measurements due to instrument malfunction or power failure. For measurements at the current site, missing data were less than 1% in 2010. However, in 2011, 12.7% of the raw data were missed due to the power failure from May 3 to 9 and from October 20 to December 1. In addition, data quality assurance criteria may generate additional gaps in the data sets due to the rejection of unreasonable and/or contaminated data. Measurements taken under the following conditions were rejected: (a) rain events, (b) low quality checked with an overall flag system [28], (c) low turbulent mixing (friction velocity < 0.1 m s -1 ) during nighttime, and (d) negative CO 2 fluxes during the non-growing season. The effects of the above four rejection criteria on data coverage are presented in Table 1. After applying the rejection criteria, data coverage of CO 2 fluxes is 57.9% in 2010 and 48.8% in 2011.
In order to obtain the information on the daily and annual carbon flux data, a gap-filling strategy [29] was adopted to fill in missing and rejected data. The missing data caused by power failure in 2011 were filled via the look-up tables method [30]. For small gaps (< 1 hour), interpolation method was used. Large gaps of daytime missing CO 2 flux data during the growing season were filled by a light-response function [30]: where F c (μmol m -2 s-1 ) is the net flux density of CO 2 , F max (μmol m -2 s-1 ) the maximum CO 2 flux at infinite light, α the apparent quantum yield, Q p (μmol m -2 s-1 ) the incident photosynthetically active radiation, and R eco the respiration from soil and plants.
Nighttime missing data during the growing season and all missing data during the nongrowing season were filled using the exponential relationship [29] between the CO 2 flux during the periods of high turbulent mixing (friction velocity > 0.1 m s -1 ) and soil temperature at the depth of 5cm: where b 0 and b are the two empirical coefficients, T s the soil temperature, from which respiration temperature coefficient (Q 10 ) can be estimated as The exponential relationship was also used to estimate daytime R eco . GPP was estimated by subtracting R eco from NEE. The respiration temperature coefficient (Q 10 ) was evaluated using 5-daytime sliding windows. CO 2 storage term was corrected before gap filling to avoid double counting based on the one point CO 2 concentrations from the open-path IRGA of the eddy covariance system [31].

Results and Discussion
Weather conditions  Table 2. Air temperature during the pre-growing season in 2010 was generally higher than that in 2011. During the growing season, T a was comparable in the two years. Daily maximum T a reached up to 17.1˚C in 2010, which was comparable to that of 15.3˚C in 2011. During the post-growing season, T a in 2010 was lower than in 2011. Variations of soil temperature followed those of air temperature during the non-growing season. However, during the growing season, the average T s in 2010 was slightly lower than that in 2011. Vapor pressure deficit during the non-growing season in 2010 was higher than that in 2011. PAR during different periods had no remarkable differences between the two years.
doi:10.1371/journal.pone.0166837.t002 GPP, R eco and NEE in relation to environmental variables GPP and R eco are simultaneously affected by the environmental variables. Fig 3 shows the response of daily GPP to daily averaged T a , D and daily integrated PAR during the peak growth period (June to August). GPP was positively correlated with T a , D and PAR. Changes in T a accounted for 53% (RMSE = 1.43 g C m -2 ) and 20% (RMSE = 1.41 g C m -2 ) of variability in GPP during 2010 and 2011 peak growth periods, respectively. Linear regression of GPP with D explained only 37% (RMSE = 1.65 g C m -2 ) and 23% (RMSE = 1.38 g C m -2 ) of the variability in GPP during the two periods, respectively. The relatively low GPP in 2011 peak growth period corresponded to lower air temperature and vapor pressure deficit during the period. Light response function of GPP to PAR explained 43% (RMSE = 1.57 g C m -2 ) and 32% (RMSE = 1.30 g C m -2 ) of the variability in GPP during the two periods respectively, indicating that the photosynthesis activity in 2010 was higher than that in 2011. The relatively low air temperature and vapor deficit in 2011 restrained the canopy photosynthesis activity.
Multiple regression of GPP with T a and PAR provided a better fit to the data (data not shown). Changes in T a and PAR together accounted for 71% (RMSE = 1.12 g C m -2 ) and 44% (RMSE = 1.17 g C m -2 ) of the variability in GPP during the two periods, respectively. Regression of GPP on other combinations of environmental variables did not provide a better fit to the data (data not shown). The above results suggest that the air temperature and photosynthetically active radiation are two significant environmental factors controlling the GPP at daily scale for this alpine grassland. Fig 4 shows response of integrated nighttime R eco to T s and θ v during the peak growth period. R eco data were averaged with T s bins of 1˚C and θ v bins of 0.01 m 3 m -3 . R eco was R eco and θ v showed a cubic relationship during the peak growth periods (Fig 4B). Changes in θ v accounted for 59% (RMSE = 0.50 g C m -2 ) and 98% (RMSE = 0.12 g C m -2 ) of the variability in R eco during 2010 and 2011 peak growth periods, respectively. In 2010 (2011), with the increase in θ v , R eco showed an increasing trend when θ v was lower than 0.26 m 3 m -3 (0.28 m 3 m -3 ) and greater than 0.39 m 3 m -3 (0.43 m 3 m -3 ), and a decreasing trend when θ v was within the range of 0.26 to 0.39 m 3 m -3 (0.28 to 0.43 m 3 m -3 ). The positive correlation between R eco and θ v under high soil moisture condition occurred when grass was very active. In general, medium soil moisture condition can promote ecosystem respiration compared to low or high soil moisture conditions, which may limit ecosystem respiration. For example, Xu et al. [33] reported that soil moisture limited ecosystem respiration when soil water content is below a threshold of 0.15 m 3 m -3 over a Mediterranean grassland in California. Yang et al. [34] found that ecosystem respiration and soil water content showed a quadratic relationship with the maximum R eco occurring at the medium soil water content of 0.15 m 3 m -3 over a temperate desert steppe in Inner Mongolia.
An increase in Q 10 was observed in response to increase in soil water content. High Q 10 values occurred during the period when soil water content was high and grass was active. High temperature sensitivity may be caused by the direct effects of temperature on the activities of plant root and microbe and indirect effects related to photosynthetic assimilation and carbon allocation to roots [29,35]. This response had also been found on the research for ecosystem respiration in alpine grassland ecosystems in the Qinghai-Tibetan Plateau [7,13], and in other grasslands such as the Mediterranean annual grassland [29] and northern temperate grassland [36].
Although Q 10 was close in the two years, R eco in 2010 was higher than that in 2011 at the same temperature. It was found that R eco was highly linearly correlated with GPP ( Fig 5). 77% (RMSE = 0.98 g C m -2 )and 52% (RMSE = 0.75 g C m -2 )of the variability in R eco could be explained by the changes in GPP during 2010 and 2011 peak growth periods, respectively. Similar results were also reported in previous studies [29], indicating that R eco was closely related to canopy photosynthetic activity in addition to environmental factors.
There were no clear correlations between NEE and T a , D, PAR and θ v at daily scale. Also no clear correlation between GPP and θ v was observed at daily scale. However, at 30-min scale, NEE was found to be positively correlated with T a and D when T a was lower than 16.5˚C and D was lower than 0.55 kPa. NEE was negatively correlated with T a and D when T a and D were higher than the above two values [18]. NEE and θ v showed a quadratic relationship when θ v was lower than 0.33 m 3 m -3 and the maximum NEE occurred when θ v was 0.25 m 3 m -3 . When θ v was higher than 0.33 m 3 m -3 , NEE was positively correlated with θ v .

Seasonal variations in NEE, GPP and R eco
Seasonal variations in daily NEE, GPP and R eco are shown in Fig 6. R eco was less than 2 g C m -2 d -1 during nun-growing seasons of the two years. Accordingly, NEE was positive during the non-growing seasons. During the growing seasons, GPP, R eco and NEE showed different seasonal variations between the two years. GPP and R eco began to increase gradually since the growing seasons started. When the increment of photosynthesis exceeded that of respiration, NEE was negative. With the start of rainy season, both GPP and R eco increased phenomenally from early May to early July in 2010. However, GPP showed a higher increasing rate than R eco , leading to a sharp increase in NEE. The daily maximum NEE reached up to -4.9 g C m -2 d -1 during early July in 2010. As T a and T s decreased in early July, both GPP and R eco decreased. With the rise of temperature, GPP and R eco then increased again. Canopy photosynthesis and ecosystem respiration peaked in late July. The daily maximum GPP and R eco reached up to 12.9 and 10.7 g C m -2 d -1 respectively in 2010.
In 2011, the daily maximum NEE reached up to 5.2 g C m -2 d -1 during mid-July. GPP showed a sharp increase in early May due to the rapidly rising air temperature and appropriate soil moisture and then decreased in mid-May. With the rising temperature and increasing rainfall from late May, GPP showed a rapid increase and peaked in late June. The daily maximum GPP reached up to 10.3 g C m -2 d -1 in late June. GPP then decreased in early July and late July, and increased in mid-July and early August. The daily maximum GPP in mid-July (10.2 g C m -2 d -1 ) almost reached up to that in late June. R eco increased from Mid-May to Mid-June, and fluctuated till early July. R eco then increased again and reached its peak in mid-July, with the daily maximum R eco reached up to 7.8 g C m -2 d -1 . Both GPP and R eco decreased during the late growing seasons in both years.
Seasonal variations in GPP were consistent with the changes in air temperature in both of the two growing seasons. Changes in T a during early June to mid-July were similar in both seasons. Changes in GPP displayed a similar pattern. However, in late July, T a showed different variations between the two seasons. Both T a and GPP increased and reached their peaks in late July in 2010. In 2011, however, both of them decreased during late July. The decrease in air temperature during late July in 2011 prevented GPP from increasing further, resulting in as large as 2.6 g C m -2 d -1 of difference in daily maximum GPP between the two years.
R eco in 2010 growing season ranged from 0.8 to 10.7 g C m -2 d -1 , as compared to that of 0.8 to 7.8 g C m -2 d -1 during 2011 growing season. The difference in R eco between the two seasons might be caused by the seasonal differences in both soil temperature and soil moisture. T s showed an increasing trend with a rise from 9.3 to 13.9˚C during late July in 2010 growing season, whereas T s decreased from 17.1 to 12.1 o C in 2011. In 2010, θ v was greater than 0.42 m 3 m -3 during the period when the respiration rate was high (Figs 2A and 4B). In contrast, in 2011, θ v was within the range of 0.33 to 0.42 m 3 m -3 during the period when respiration was relatively low (Figs 2B and 4B). As a result, R eco increased during late July in the 2010 growing season but decreased during late July in the 2011 growing season. The divergence led to as much as 2.9 g C m -2 d -1 of difference in daily maximum R eco between the two years. R eco showed sharp increases during three dry periods in the growing season of 2011, resulting in remarkable decreases in NEE. The observation demonstrated that changes in soil moisture played an important role during the process.

Cumulative NEE, GPP and R eco
The cumulative NEE, GPP and R eco over the two years are shown in Fig 7. Cumulative NEE, GPP and R eco were -191, 1145 and 954 g C m -2 for 2010, and -250, 975 and 725 g C m -2 for 2011, respectively. According to the cumulative NEE data, this alpine grassland was a moderate carbon sink in both of the two years. Annual integrated values of NEE of alpine grasslands on the Qinghai-Tibetan Plateau reported in previous studies were within the range of 173 to -193 g C m −2 yr −1 [7,8,10,11]. Compared to alpine grasslands on the Qinghai-Tibetan Plateau, this alpine grassland showed a much greater potential for carbon sequestration than others.
The ecosystem gained more carbon in 2010 growing season than that in 2011 growing season via photosynthesis, and produced more CO 2 in 2010 than that in 2011 via respiration regardless in growing season or non-growing season (Table 2). However, the system fixed less carbon in 2010 than that in 2011. It is highly probable that the annual precipitation is a dominant variable controlling the NEE for this grassland. The annual precipitation in 2011 was 13.4% more than that in 2010, and the ecosystem fixed 30.9% more carbon in 2011 than that in 2010.
Many research found that annual grassland productivity is positively correlated with annual precipitation [37,38]. This study, in contrast, found that the difference in GPP between the two years was not caused by the variation in annual precipitation. Though much more precipitation happened in 2011, GPP was 170 g C m -2 less than that in 2010. For present alpine grassland, air temperature had more influence than precipitation in determining the GPP, as described in previous sections. In addition, the lengths of growing seasons also had an important impact on GPP [29]. The length of growing season in 2011 was 14 days shorter than that in 2010, which may be a reason why GPP was lower in 2011. Cumulative R eco was 83% and 74% of GPP in 2010 and 2011, respectively. Lower R eco in 2011 was mainly caused by lower GPP and corresponding soil water content during the growing season. Despite the decrease of GPP and R eco in 2011 compared with the previous year, R eco reduced much more than GPP, resulting in more NEE in 2011.

Conclusions
Alpine grasslands comprise most of the natural vegetation in the Qinghai-Tibetan Plateau. Over the past several decades, the Qinghai-Tibetan Plateau experienced evident climate warming and moistening. Understanding carbon dynamics of alpine grassland ecosystems on the Qinghai-Tibetan Plateau in response to changing environmental conditions is critical to accurately model carbon balance. Carbon dioxide exchange over the alpine grassland on the eastern Qinghai-Tibetan Plateau was measured using the eddy covariance method. The main results show that air temperature and photosynthetically active radiation are dominant factors controlling the GPP at daily scale. Soil temperature and soil water content are main variables controlling R eco . Canopy photosynthetic activity is also responsible for the variation of daily R eco other than environmental factors. No clear correlation between net ecosystem exchange and environmental factors was observed at daily scale. Temperature sensitive coefficient Q 10 was observed to increase with larger soil water content. High Q 10 values occurred during the periods when soil water content was high and grass was active. Annual integrated NEE, GPP and R eco were -191, 1145 and 954 g C m -2 for 2010, and -250, 975 and 725 g C m -2 for 2011, respectively. According to the annual NEE data, this alpine grassland was a moderate carbon sink in both of the two years. Compared to alpine grasslands on the Qinghai-Tibetan Plateau, this alpine grassland showed a much greater potential for carbon sequestration. Annual precipitation was a dominant variable controlling the variation of annual NEE of this grassland. The difference in GPP between the two years was not caused by the variation in annual precipitation. Instead, air temperature had more influences than precipitation in determining the GPP. In addition, the length of growing season also had an important impact on annual GPP. The variation of annual R eco was closely related to annual GPP and soil water content during the growing seasons. Under the background of global climate change, more studies are needed to understand how carbon dynamics of alpine grassland ecosystems in the Qinghai-Tibetan Plateau respond to changing environmental conditions. Supporting Information S1 File. Data set of the Components of ecosystem carbon exchange and controlling environmental variables. T a , air temperature, (˚C); T s , soil temperature (˚C); PPT, precipitation (mm); θ v , volumetric soil water content (m 3 m -3 ); D, vapor pressure deficit (kPa); PAR, photosynthetically active radiation (mol m -2 ); NEE, net ecosystem exchange (g C m -2 ); GPP, gross primary production (g C m -2 ); and R eco , ecosystem respiration (g C m -2 ). (PDF)