Seasonal patterns of gross primary production and ecosystem respiration in an alpine meadow ecosystem on the Qinghai-Tibetan Plateau

Authors


Abstract

[1] We measured the net ecosystem CO2 exchange (NEE) in an alpine meadow ecosystem (latitude 37°29′–45′N, longitude 101°12′–23′E, 3250 m above sea level) on the Qinghai-Tibetan Plateau throughout 2002 by the eddy covariance method to examine the carbon dynamics and budget on this unique plateau. Diurnal changes in gross primary production (GPP) and ecosystem respiration (Re) showed that an afternoon increase of NEE was highly associated with an increase of Re. Seasonal changes in GPP corresponded well to changes in the leaf area index and daily photosynthetic photon flux density. The ratio of GPP/Re was high and reached about 2.0 during the peak growing season, which indicates that mainly autotrophic respiration controlled the carbon dynamics of the ecosystem. Seasonal changes in mean GPP and Re showed compensatory behavior as reported for temperate and Mediterranean ecosystems, but those of GPPmax and Remax were poorly synchronized. The alpine ecosystem exhibited lower GPP (575 g C m−2 y−1) than, but net ecosystem production (78.5 g C m−2 y−1) similar to, that of subalpine forest ecosystems. The results suggest that the alpine meadow behaved as a CO2 sink during the 1-year measurement period but apparently sequestered a rather small amount of C in comparison with similar alpine ecosystems.

1. Introduction

[2] Gross primary production (GPP) and ecosystem respiration (Re) determine the carbon budget of an ecosystem. GPP is mainly affected by leaf area index (LAI), leaf physiological activity, weather, and growing-season length, but Re, which is the sum of autotrophic respiration (Ra) and heterotrophic respiration (Rh), is controlled disparately by plant and soil microbe activities. Temperature is a major controlling factor of Ra, which is the sum of plant growth and maintenance respiration, whereas soil moisture may play a more important role in Rh [Lloyd and Taylor, 1994; Davidson et al., 1998; Xu and Qi, 2001]. Thus the combination of phase and amplitude of variation in all these above components finally determines the temporal patterns of net ecosystem CO2 exchange (NEE) [Randerson et al., 1999; White et al., 1999; Cramer et al., 1999; Falge et al., 2002].

[3] The eddy covariance method makes it possible to measure NEE with precision and contributes to the identification of the characteristics of source/sink activities of various global ecosystems. Valentini et al. [2000] divided daytime NEE data, derived from flux observations from a global tower network, into GPP and Re by extrapolating site-specific exponential relationships between nocturnal soil temperature and CO2 efflux. This approach allows the phase and amplitude of seasonal variations in GPP and Re to be investigated. Falge et al. [2002] recently used the FLUXNET database to research the factors that control seasonal changes in GPP and Re in a wide range of terrestrial ecosystems. Their results showed that the seasonality of GPP is determined by plant life-form (e.g., broad leaves or needle-shaped leaves), but that of Re is affected more by climate, which is often characterized by temperature. Saigusa et al. [2002] showed that the seasonal change in GPP in a cool-temperate deciduous forest is tightly associated with changes in the maximum GPP and light use efficiency. They also reported a high correlation between maximum GPP and LAI. Flanagan et al. [2002] found that GPP was strongly related to LAI and canopy nitrogen content in a northern temperate grassland. It thus seems that GPP and Re are associated with changes in plant phenology and climatic seasonality in lowland ecosystems. Little evidence, however, is available by which we can determine the influence of environmental factors on the seasonal changes in the carbon dynamics of alpine ecosystems, in particular those on the vast Qinghai-Tibetan Plateau.

[4] In alpine meadow ecosystems on the Qinghai-Tibetan Plateau, the growing season is shortened mainly by low temperature, since abundant solar radiation and precipitation during the growing season are likely to favor plant growth during the short growing season. Kato et al. [2004] showed that the ecosystem has four times the potential daily sequestration capacity of a subalpine coniferous forest ecosystem at equivalent altitude and latitude in Colorado, United States [Monson et al., 2002]. Moreover, low temperature in winter restricts the decomposition of litter. This ecosystem, therefore, might be an annual net sink of atmospheric CO2. The soil carbon density of an alpine ecosystem (18.2 kg m−2; Ni [2002]) has been reported to be much higher than those of savanna (5.4 kg m−2; Adams et al. [1990]) or temperate grassland (13.0 kg m−2; Adams et al. [1990]), but similar to that of tundra (22.0 kg m−2; Adams et al. [1990]), where low temperature also shortens the growing season. However, tundra is reported to be at risk of changing from a sink to a source of atmospheric CO2 with global warming [Oechel et al., 1993]. Consequently, detailed information on the relationship between environmental factors and carbon dynamics of the alpine ecosystem on this extensive plateau is necessary in order to assess regional and global climate change.

[5] In this study, our major objectives were (1) to assess the seasonal pattern of carbon dynamics for an alpine meadow, which is a representative part of a grassland ecosystem of vast area for which little information from global carbon dynamics studies is available, and (2) to clarify the underlying mechanisms involved in environmental control of the carbon dynamics in an alpine ecosystem. Our primary hypothesis was that the alpine meadow ecosystem is a large CO2 sink. The hypothesis was based mainly on the observation that primary production is favored by plentiful solar radiation, sufficient water availability, and favorable temperatures during the growing season, while the long winter, which is characterized by low temperature and dry air, should limit decomposition processes. We tested the hypothesis by examining the ecosystem CO2 flux using eddy covariance measurements.

2. Sites and Methods

2.1. Site Description

[6] The NEE and environmental factors were measured at the Haibei Alpine Meadow Ecosystem Research Station, Northwest Plateau Institute of Biology, Chinese Academy of Sciences (latitude 37°29′–45′N, longitude 101°12′–23′E; 3250 m above sea level (asl)). The annual average temperature and precipitation for 1981–2000 were −1.7°C and 561 mm. The soil is a clay loam, and its average thickness is 65 cm. The surface 5–10 cm of soil, which is classified as a Mat Cry-gelic Cambisol according to the classification system of the Chinese National Soil Survey [Institute of Soil Science and Chinese Academy of Sciences, 2001], is wet and high in organic matter. The study site is grazed by yaks and sheep every winter.

[7] The plant community at the flux measurement site is dominated mainly by three major perennial sedges, Kobresia humilis, K. pygmaea, and K. tibetica (Cyperaceae) [Li and Zhou, 1998]. The plants start to grow in May, when the daily average air temperature exceeds the freezing point. The maximum aboveground biomass had reached 283 g d.w. m−2 on 30 July (day of the year (DOY) 211) in 2002 [Kato et al., 2004], and it varied within the range of 342 ± 50 g d.w. m−2 (average ± s.d.) from 1980 to 1993 during the summer [Li and Zhou, 1998]), which is when most of the annual precipitation is concentrated and air temperature is high. The foliage almost dies out in October [Li and Zhou, 1998]. There are a few very small shrubs in the meadow, so we ignored their biomass because their contribution was relatively insignificant. Kato et al. [2004] also showed that the LAI began to increase from late May and reached a maximum of 3.8 on 16 July 2002 (DOY197) and then decreased slowly until October.

2.2. Eddy Covariance and Meteorological and Soil Measurements

[8] CO2 and H2O flux (evapotranspiration) were measured by the open-path eddy covariance method from 1 January to 31 December 2002. The flux observation site has a fetch of at least 1 km in all directions, except for a 100-m-long clay fence about 1 m high on the west side, 250 m from the observation tower. Further details are described by Gu et al. [2003]. Wind speed and sonic virtual temperature were measured at 2.2 m above the ground with a sonic anemometer (CSAT-3, Campbell Scientific Inc., Logan, Utah, United States). CO2 and water vapor concentrations were also measured at 2.2 m with an open-path infrared gas analyzer (CS7500, Campbell Scientific Inc.). A digital micrologger (CR23X, Campbell Scientific, Inc.) equipped with an analog multiplexer (AM416) was used for sampling and logging data. Wind speed, sonic virtual temperature, and CO2 and H2O concentrations were sampled at a rate of 10 Hz. Their mean, variance, and covariance values were calculated and logged every 15 min. The cross spectrum between vertical wind w′ and CO2 concentration c′ showed that the covariance for the two variables was close to zero when the sampling frequency was lower than 0.002 Hz (8.3 min per cycle). A period of 15 min was thus considered to be sufficient to include the necessary information for our analysis. The collected data were then adjusted by the WPL (Webb, Pearman, and Leuning) density adjustment [Webb et al., 1980]. In this study, three common flux data corrections, that is, coordinate rotation, trend removal, and water vapor correlation, were not enforced on the collected data. However, the effect of not making these corrections on the calculated flux was examined for 10 days in July 2002 by using fluctuation data sampled at the frequency of 10 Hz, and the implicit estimation error in the flux data was evaluated by comparing corrected and uncorrected fluxes in the latent heat and CO2 flux calculations. The regression line slopes show small differences, within 4%, between corrected and uncorrected fluxes. This indicates that the bias due to these corrections is likely to be negligible in the study. The closure of the surface energy budget was examined by performing a linear regression between the sum of eddy fluxes (H + λE) and the available energy (Rn − G) throughout the year in 2002: (H + λE) = 0.70 × (Rn − G) − 0.24; r2 = 0.89, where H and λE are the sensible and latent heat fluxes, respectively, Rn is the net radiation, and G is soil heat flux, and where all the flux values are daily averages (MJ m−2). Micrometeorological measurements were taken at the same site. Net radiation, photosynthetic photon flux density (PPFD), air temperature and humidity, wind speed, soil heat flux, soil temperature, soil water content, soil surface temperature, and rainfall were logged every 15 min. Details are reported by Kato et al. [2004].

2.3. Re and GPP Calculation

[9] Ecosystem respiration Re was measured directly during nighttime periods with strong turbulence (as the NEE at friction velocity u* > 0.2 m s−1), and was extrapolated to other periods by exponential regression of measured Re against soil temperature at a depth of 5 cm with the Arrhenius equation reported by Lloyd and Taylor [1994]:

equation image

where Re is the nighttime ecosystem respiration rate (μmol CO2 m−2 s−1), Re,Tref is the ecosystem respiration rate (μmol CO2 m−2 s−1) at the reference temperature Tref (K), and Ea is the activation energy (J mol−1). Those latter two parameters are site-specific. R is a gas constant (8.134 J K−1 mol−1), and Tsoil is the soil temperature at a depth of 5 cm. Re,Tref was assigned as R10, the respiration rate at a Tref of 283.16 K (10°C), and evaluated for every month. Ea was evaluated from a regression of all Re data in 2002 against Tsoil as a constant value throughout the year (81,519 J mol−1).

[10] GPP was calculated as the sum of NEP (net ecosystem production as CO2 uptake, i.e., NEE) and Re as follows:

equation image

where all variables have units of μmol m−2 s−1.

2.4. NEP Gap-Filling Methods

[11] When natural and anthropogenic impacts on the global ecosystem carbon budget are compared among biome types, phenology, and stress conditions, we often use the annual sum of net ecosystem CO2 exchange [Valentini et al., 2000; Falge et al., 2002]. However, Falge et al. [2001] found that the average data coverage during a year was only 65%, owing to system failures or data rejection. They thus reviewed several methods of gap filling and applied them to data sets available from the EUROFLUX and AmeriFlux databases. They used mean diurnal variation (MDV), look-up tables (LookUp), and nonlinear regression (Regr.) methods, and investigated the impact of these different gap-filling methods on the annual sum of NEE. In this study, we also used these three methods to fill gaps in order to obtain the annual sums of NEP.

[12] In the MDV method, a missing datum was replaced by the mean for that time period (15 min) from adjacent days. We chose data windows of 7 and 14 days using two different algorithms: (a) an “independent” window and (b) a “gliding” window. In algorithm a, for each subsequent period of data, mean diurnal variations were established to fill gaps within that period. In algorithm b, a window of prescribed size around each gap was used to construct mean diurnal variations for gap filling within that window.

[13] In the LookUp method, look-up tables were created for six bimonthly periods or four seasonal periods ranging from 1 April to 30 May, 1 June to 30 September, 1 October to 30 November, and 1 December to 31 March. For the look-up table, average NEPs were compiled for 27 PPFD classes × 35 Ta (air temperature) classes. PPFD classes consisted of intervals of 100 μmol m−2 s−1 between 0 and 2600 μmol m−2 s−1 with a separate class for PPFD = 0. Similarly, Ta classes were defined as 2°C intervals between −35° and +34°C.

[14] In the Regr. method, regression relationships were established between the NEP components, that is, Re and GPP, and associated controlling factors (temperature and light) for every month. Missing Re values were extrapolated by using exponential regression equations (equation (1)) between measured nighttime Re with strong turbulence (u* > 0.2 m s−1) and soil temperature at a depth of 5 cm. Nighttime eddy covariance flux data under low-turbulence conditions, that is, below the u* threshold [Aubinet et al., 2000] (0.2 m s−1 in this study), were also corrected with the regression equation (equation (1)); this correction was called the “u*-correction.” GPP was extrapolated by rectangular hyperbolic regression of daytime GPP against PPFD with a Michaelis-Menten-type equation [Falge et al., 2001] for every month:

equation image

where α is the initial slope of the light-GPP curve (μmol CO2 (μmol photon)−1) and is equivalent to the quantum yield, and GPPSAT is GPP at light saturation (μmol m−2 s−1). Those two parameters were month-specific.

3. Results

3.1. Diurnal Courses of GPP and Re

[15] The monthly averaged diurnal courses of GPP and Re are shown in Figure 1. GPP shows a maximum value of 15.1 μmol m−2 s−1 at 1200 hours in August (Figure 1a). Large amplitudes of GPP occurred in July and August when LAI and biomass reached their maxima, whereas small amplitudes were observed in May, June, and September. Re showed a maximum value of 7.0 μmol m−2 s−1 at 1600 hours in August (Figure 1b), and increased in the afternoon as soil temperature increased.

Figure 1.

Diurnal courses of the (a) hourly mean gross primary production (GPP) and (b) hourly mean ecosystem respiration (Re) in 2002 (May, circles; June, squares; July, diamonds; August, crosses; September, plusses; October–April, triangles).

[16] The relationship between NEE and PPFD in the daytime and that between NEE and Tsoil at nighttime in July are shown in Figures 2a and 2b, respectively. With the increase of PPFD, NEE decreased and showed a significant difference in relation to light between the morning (0600–1300) and afternoon (1400–2000) values. NEE increased exponentially with the increase of Tsoil.

Figure 2.

Relationship between (a) net ecosystem exchange (NEE) and irradiant photosynthetic photon flux density (PPFD) in the daytime (0600–1300 hours, solid circles and solid line; 1400–2000 hours, open squares and broken line) and (b) NEE and 5-cm-depth soil temperature (Tsoil) at nighttime in July 2002. Data points show 15-min-mean values (Figure 2a) and averaged values (Figure 2b) for 2°C intervals ± standard deviation. The regression line indicates a rectangular hyperbolic relationship: NEE = a × b × PPFD/(b + a × PPFD) + Re, where a is quantum yield (μmol CO2 (mol photon)−1), b is the minimum NEE (μmol CO2 m−2 s−1), and Re is ecosystem respiration (μmol CO2 m−2 s−1). Parameters are shown as follows: month (symbol, a, b, Re, r2). 0600–1300 hours (solid circles, −0.0196, −23.0, 1.31, 0.528), 1400–2000 hours (open squares, −0.0193, −17.5, 1.37, 0.520), P < 0.0001. The equation for predicting Re from Tsoil is y = 0.678 exp (0.0897x); r2 = 0.989; P < 0.0001.

3.2. Seasonal Changes of GPP and Re

[17] R10 (the respiration rate at a soil temperature of 10°C) reached maximum of 2.27 μmol m−2 s−1 in August and 1.88 μmol m−2 s−1 in September, but was negative in January and February (Table 1). α was high in May, reached maximum from June to August, and stayed near zero from October to April. GPPSAT increased from May to maximum values in July and August, and stayed near zero from October to April (Table 1).

Table 1. Parameterization of Ecosystem Respiration and Gross Primary Production Regression Curves Using 15-Min Averaged Data in 2002a
MonthReGPP
R10, μmol CO2 m−2 s−1r2α, × 103 μmol CO2 μmol photon−1GPPSAT, μmol CO2 m−2 s−1r2
  • a

    Re, ecosystem respiration; GPP, gross primary production. The regression follows an Arrenius-type exponential relationship for Re (nighttime NEE (u* > 0.2 m s−1)): Re = R10 * exp (Ea/R*(1/283.16 − 1/(Tsoil + 273.16a))), where R10 is the respiration rate at the soil temperature of 10°C (μmol CO2 m−2 s−1), Ea is the active energy (is adopted for only one value for whole year; 81519 J mol−1), R is the gas constant (= 8.134 J K−1 mol−1), and a rectangular hyperbolic relationship for GPP: GPP = α × GPPSAT × PPFD/(GPPSAT + α × PPFD), where α is quantum yield (dimensionless), GPPSAT is the saturated GPP (μmol CO2 m−2 s−1).

Jan.−0.750.0030.86−0.020.004
Feb.−0.030.0040.12−0.010.089
March0.520.0070.36−0.020.076
April0.930.0080.12−0.010.120
May1.310.018139.501.540.004
June1.500.02643.9910.460.137
July1.230.00631.8320.370.489
Aug.2.270.00948.5618.290.446
Sept.1.880.12424.478.200.276
Oct.1.260.440−2.320.100.119
Nov.0.210.000−1.36−0.880.055
Dec.0.440.009−2.33−0.220.007

[18] Seasonal changes in the 15-day moving average values of PPFD and Tsoil, which strongly control GPP and Re, respectively, are shown in Figure 3a. PPFD fluctuated during the growing season, May to September, because of frequent cloudy conditions; maxima were observed on days of the year (DOY) 181, 207, and 232; and minima on DOY166, 190, and 217. Tsoil also showed three maxima and three minima during the growing season, but their dates lagged behind those of the PPFD maxima and minima by 1–13 days: maxima were recorded on DOY189, 213, and 237 and minima on DOY167, 203, and 230. Tsoil decreased gradually after the initial maximum (DOY189).

Figure 3.

Seasonal changes in (a) daily mean photosynthetic photon flux density (PPFD; solid line) and 5-cm-depth soil temperature (Tsoil; broken line); (b) daily mean gross primary production (GPP; solid line), ecosystem respiration (Re; broken line), and ratio between GPP and Re (dotted line); and (c) daily maximum GPP (solid line) and Re (broken line) in 2002. Data represent the daily means of a 15-day moving average of the GPP/Re ratio. DOY, day of the year.

[19] Seasonal changes in the 15-day moving average values of GPP and Re are shown in Figure 3b. GPP started to increase on 1 May (DOY121), reaching a first maximum of 5.90 μmol m−2 s−1 on 7 July (DOY188), a second maximum of 6.46 μmol m−2 s−1 on 2 August (DOY214), and a third maximum of 5.71 μmol m−2 s−1 on 19 August (DOY231). After the third maximum, GPP decreased gradually to 1.0 μmol m−2 s−1 on 28 September (DOY271). Re started to increase at the beginning of March, reaching a first maximum of 3.48 μmol m−2 s−1 on 12 July (DOY193), a second maximum of 5.40 μmol m−2 s−1 on 8 August (DOY220), and a third maximum of 4.45 μmol m−2 s−1 on 24 August (DOY236). Thus the seasonal changes in GPP show that the plant growing season lasted from May to September in this alpine meadow ecosystem, but did not correspond completely to those in PPFD, although the seasonal changes in Re showed good agreement with those in Tsoil.

[20] Figure 3b also shows seasonal changes in the ratio GPP:Re. The ratio was usually >1 from May to September, indicating that this ecosystem is a carbon sink during the growing season. The maximum value of the ratio was 2.07 on 24 July (DOY205).

[21] Figure 3c shows seasonal changes in the 15-day moving average values of GPPmax and Remax, which are the daily maximum values of GPP and Re. GPPmax started to increase rapidly on 1 May (DOY121) and remained >20 μmol m−2 s−1 for 2 months beginning 2 July (DOY183), reaching a maximum of 23.6 μmol m−2 s−1 on 31 July (DOY212). After 20 August (DOY232), GPPmax decreased rapidly from 20.4 μmol m−2 s−1 to 1.0 μmol m−2 s−1 on 10 October (DOY283). Remax started to increase on 1 May (DOY121), reached a first maximum of 8.10 μmol m−2 s−1 on 1 July (DOY182), and its annual maximum of 12.0 μmol m−2 s−1 on 25 August (DOY237); it then decreased rapidly to 1.0 μmol m−2 s−1 on 25 October (DOY298). Thus the seasonal changes in GPPmax correspond well to those in GPP, whereas the annual maximum of Remax lagged 2 weeks behind that of Re.

[22] To clarify the effect of LAI on CO2 assimilation potential and resource use efficiency, we examined the relationships between LAI and GPPmax, radiation use efficiency by GPP (RUEGPP), and water use efficiency of GPP (WUEGPP, Figure 4). RUEGPP is defined as the ratio of photosynthesis to irradiance (GPP/PPFD), and WUEGPP as that of photosynthesis to canopy evapotranspiration (GPP/evapotranspiration).

Figure 4.

Leaf area index (LAI) controls on (a) maximum gross primary production (GPPmax), (b) radiation use efficiency (RUEGPP), and (c) water use efficiency (WUEGPP) of the GPP. The vertical axis data show the daily mean value of fewer than 4 days backward and forward from the LAI sampling day. GPPmax data when the PPFD was below 1500 μmol m−2 s−1 were eliminated from the average. Error bars show the standard deviation (horizontal axis: n = 5, vertical axis: n = 7). In the equations for predicting GPPmax, RUEGPP, and WUEGPP from LAI, standard errors are 1.56046, 0.912992, 0.176842, respectively; P < 0.0001, 0.0001, and 0.05.

3.3. Annual Sums of Net Ecosystem Carbon Dynamics

[23] Table 2 shows annual sums of GPP, Re, and NEP, and daily GPP and NEP calculated as annual GPP and NEP divided by the number of days when NEP > 0. Missing data were filled by the Regr. method for all parameters, and by the MDV and LookUp methods for NEP. Gap-filling methods reduced the proportion of missing data from 33.3% to 4%. The remaining missing data were almost all during the coldest part of winter, from 1 to 13 January, when there is no biological activity at all, so the missing data would not affect annual sums of fluxes. NEP was calculated from the following equation for the alpine meadow.

equation image

where ΔWabove and ΔWbelow are the increments of aboveground and belowground plant biomass, respectively, Wg is the plant biomass consumed by animals, and Wl is the total carbon in litter.

Table 2. Ecosystem Carbon Dynamics in 2002 in Haibei, Qinghai, Chinaa
MethodSpecificationAverage PeriodGPP, g C m−2Re, g C m−2NEP, g C m−2Days (NEP > 0), daysDaily GPP, g C m−2Daily NEP, g C m−2
  • a

    “Day (NEP > 0)” means the number of days when daily NEP > 0 (length of growing season). GPP, Re, and NEP mean the periodic sums of gross primary production, ecosystem respiration, and net ecosystem production, respectively. Daily GPP and daily NEP were calculated by dividing GPP, Re, and NEP by the length of the growing season.

Regr.with u*-correction1 month575.1496.678.51603.590.49
Regr.without u*-correction1 month527.4413.5113.91643.220.69
 
MDVindependent window7 days  147.2   
MDVindependent window14 days  145.2   
MDVgliding window7 days  149.5   
MDVgliding window14 days  141.8   
 
LookUpTsoil2 months  122.3   
LookUpTsoilseasonal  96.9   
LookUpTair2 months  142.3   
LookUpTairseasonal  113.8   

[24] From equation (4), we estimated the belowground biomass, if we neglect the terms Wg and Wl, as follows:

equation image

[25] The annual NEP was estimated as 78.5 g C m−2 y−1 by the eddy covariance method, and ΔWabove was determined by multiplying the maximum aboveground biomass of 283 g d.w. m−2 [Kato et al., 2004] by a conversion factor of 0.476 (to get carbon content), which was derived by measuring the dry biomass carbon content using an NC analyzer (Sumigraph NC-900, Sumika Chemical Analysis Service Ltd., Osaka, Japan). The ΔWbelow was then calculated to be −56.2 g C m−2 y−1 (= NEP − ΔWabove = 78.5 − 134.7).

4. Discussion

4.1. Controls on Diurnal Changes in the CO2 Exchange

[26] Re increased significantly in the afternoon (Figure 1). In an alpine meadow, the soil surface temperature is significantly high in the midafternoon, which could enhance the CO2 efflux either from the soil or from plant maintenance respiration. Decreasing NEE was closely related to the increase in PPFD (Figure 2), but NEE was suppressed in the afternoon [Kato et al., 2004].

[27] We assumed that the daytime relationship between Re and soil temperature was the same as that at night. It is, however, possible that Re during the daytime is higher owing to the increase in maintenance respiration and photorespiration during the daytime. However, these results provide evidence that respiration by plant biomass and soil microbes greatly decreases the net daytime CO2 uptake in alpine meadow ecosystems, as hypothesized by Kato et al. [2004].

4.2. Controls on Seasonal Changes in GPP and Re

[28] Daily GPPmax showed a pattern of seasonal variation similar to the daily mean GPP. The GPPmax, an index of photosynthesis potential, was about three times the daily mean GPP (Figures 3b and 3c). GPPmax was positively related to LAI (Figure 4a), as shown also by Saigusa et al. [2002] and Flanagan et al. [2002]. RUE and WUE also tended to increase as LAI increased (Figures 4b and 4c). These results suggest that LAI determines the ecosystem capacity for assimilation and resource requirements.

[29] The daily mean Re and Remax showed similar seasonal patterns in that their seasonal variations were associated more closely with the seasonal pattern of soil temperature than with that of PPFD (DOY180-230, Figures 3a–3c). Remax, however, increased even though Tsoil decreased during the same period, as seen by changes in R10 (Table 1). In general, seasonal changes in respiratory processes are controlled by climatic factors more strongly than by biological factors [Falge et al., 2002]. However, Remax seemed to be tightly associated with aboveground and belowground biomass in the alpine meadow [Kato et al., 2004; Li and Zhou, 1998].

[30] Seasonal changes in mean GPP and Re showed a temporal variation pattern similar to that seen in temperate and Mediterranean ecosystems [Falge et al., 2002] (Figure 3b)]. In contrast, the seasonal variations in GPPmax and Remax were poorly synchronized, which is consistent with what has been reported for boreal coniferous forests [Falge et al., 2002]. The annual maximum of GPPmax occurred about 20 days before that of Remax. During these 20 days, LAI and biomass reached their maximum and remained relatively constant. The major reason for the lag in the occurrence of the annual Remax behind that of the annual maximum of GPPmax may be a combination of environmental controls. The annual maximum of GPPmax occurred when both PPFD and temperature had reached their annual maximum, whereas the annual maximum Remax was observed after a large decrease in PPFD accompanied by a smaller decrease in temperature, which may have resulted in a high-maintenance respiration because of high biomass and temperature.

[31] The ratio of Ra (autotrophic respiration) to GPP was recently suggested to be conserved (e.g., GPP − Ra/GPP = 0.47 ± 0.04 SD on the basis of data from 12 evergreen and deciduous forests; Waring et al. [1998]). The data set is not available to allow us to determine the contribution of Ra in the present study. However, the high Re and GPP observed from July to August suggest a high contribution of autotrophic metabolisms to the ecosystem CO2 flux. Since the changes in Re and GPP were well associated with changes in the aboveground biomass, which is the major contributor of autotrophic metabolisms in the ecosystem, we can further assume that the contribution of Ra is also relatively high during this period. In terms of the other component of respiration, Rh (heterotrophic respiration), we noticed that the ratio GPP/Re was high in this alpine ecosystem. When GPP considerably exceeds Re, as it did here, the accumulation of litter can deprive the ecosystem of free nutrients. Considering the close link between soil organic matter decomposition and nutrient cycling, this system may exhibit negative feedback in growth and CO2 assimilation or be susceptible to disturbance [Shulze et al., 1999; Amiro, 2001]. We consider the belowground litter fall Lbelow to be the key factor controlling the contributions of Ra and Rh.

4.3. Ecosystem Carbon Assimilation Ability

[32] We compared the GPP of an alpine meadow with those of other ecosystems. The annual GPP of 575 g C m−2 was lower than that of a boreal coniferous forest (723–959 g C m−2 y−1) or a Colorado subalpine coniferous forest (831 g C m−2 y−1) at similar elevations (3050 m asl), and much lower than that of a tropical forest (3249 g C m−2 y−1), but it was within the range of GPPs reported for temperate ecosystems, including forests and grasslands (542–1924 g C m−2 y−1; average, 1262 g C m−2 y−1; Falge et al. [2002]). The daily GPP of the study site (3.59 g C m−2 d−1, Table 2) was similar to those of boreal evergreen forest and Colorado subalpine coniferous forest (4.6 and 4.4 g C m−2 d−1, respectively), although it was slightly lower than those of temperate coniferous forest and C3 crops and grassland (5.7–6.9 g C m−2 d−1). Thus, although our alpine meadow ecosystem had a daily CO2 assimilation equal to that of a Colorado subalpine forest ecosystem, it had a lower annual GPP because of the short growing period.

[33] In comparison with the total annual NEP of other ecosystems reported by Falge et al. [2002], that of our study site (78.5 g C m−2 y−1, Table 2), gap-filled by the Regr. method using the u*-correction, was close to that of the Colorado subalpine coniferous forest (71 g C m−2 y−1), although it was substantially lower than those of grassland (231.3 g C m−2 y−1) and boreal (121.4 g C m−2 y−1) ecosystems. The daily NEP of the study site (0.49 g C m−2 d−1, Table 2) was similar to that of the Colorado subalpine forest (0.38 g C m−2 d−1). Although our alpine meadow ecosystem has a lower annual GPP than the subalpine forest ecosystem, it has a comparable annual NEP. We suppose that not only low temperature but also small biomass suppresses ecosystem respiration; as a result, this ecosystem may sequester a substantial amount of C.

[34] Current studies show no consistent relationships between grazing and carbon budgets in grassland ecosystems [Derner et al., 1997; Reeder and Schuman, 2002]. Cao et al. [2004] suggested that high grazing intensity in a heavily grazed area near our study site reduced both aboveground and belowground biomass, and decreased the soil CO2 efflux and the net ecosystem CO2 fixation. Because our study site was winter pasture, there was no grazing during the summer. Grazing seems to have no direct immediate effects on CO2 dynamics during the growing season. However, it should have an impact on litter decomposition and soil structure, which affect soil respiration and growth indirectly. Further studies are needed to clarify the effects of grazing on CO2 assimilation in an alpine meadow ecosystem.

[35] Annual NEPs estimated using gap-filling methods differed somewhat among methods (78.5–145.9 g C m−2; Table 2). For the Regr. method, which calculates GPP and Re separately, Re with u*-correction was higher by 83.1 g C m−2 y−1 than that calculated without u*-correction. This result, based on the principle that u*-correction replaces nighttime eddy covariance fluxes under more stable conditions with fluxes under higher turbulence, should be expected. The consequent decline in annual NEP due to the u*-correction (35.4 g C m−2 y−1) was lower than that observed in other ecosystems (77.0 g C m−2 y−1; Falge et al. [2001]). Comparison of the total annual NEP estimated using the three gap-filling methods shows that MDV gave a higher sum than the other methods. This result suggests that environmental factors during gap periods may be biased in an extreme climate. The differences among gap-filling methods were smaller than those among various ecosystem types [Falge et al., 2001]. It is difficult to choose which is the most “correct” gap-filling method for most flux measurement studies, in part because of the lack of comparative data. However, when the total annual NEP estimated from gap-filled data is used to evaluate an ecosystem's capacity for CO2 assimilation or for comparison with other ecosystems, the possible inclusion of internal data errors should be considered.

5. Conclusions

[36] This 1-year CO2 flux measurement using the eddy covariance method has fulfilled the need for a long-term CO2 flux study of this vast grassland ecosystem in east Asia. The current observations show that the alpine meadow ecosystem is a weak CO2 sink throughout the year. This result is of particular importance for our understanding of the carbon budget on the world's highest plateau, because the Kobresia humilis meadow observed in the study is typical of about 40% of the grassland area on the plateau. To provide information for further assessment of the impact of climate change on the alpine ecosystem, we examined the factors controlling the carbon dynamics. We concluded that the seasonal changes in GPP and Re associated well with the biological (e.g., LAI and biomass) and meteorological constraints on the alpine ecosystem. The high GPP/Re ratio suggests that Ra also plays an important role in controlling the carbon dynamics in the alpine meadow ecosystem, in particular during the growing season.

[37] Temporal variation patterns, including the phase and amplitude of fluctuations in both GPP and Re, are important for characterizing the carbon dynamics of an ecosystem. Seasonal changes in mean GPP and Re showed compensatory behavior, similar to that reported in temperate and Mediterranean ecosystems. In contrast, seasonal changes in GPPmax and Remax were poorly synchronized, similar to the situation reported for boreal coniferous forests. Since both the total annual and daily GPP and Re fell within the variation range of those for boreal and subalpine ecosystems, we suggest that the alpine meadow ecosystem be classified as a subarctic biome during modeling parameterization.

Notation
α

initial slope of the light-GPP curve, μmol CO2 (μmol photon)−1.

Ea

activation energy, equal to 81,519 J mol−1, J mol−1.

G

annual grazing biomass, g C m−2 y−1.

GPP

gross primary production, μmol CO2 m−2 s−1, g C m−2.

GPPmax

maximum value of diurnal GPP, μmol CO2 m−2 s−1.

GPPSAT

GPP at light saturation, μmol CO2 m−2 s−1.

Labove

annual aboveground litter fall, g C m−2 y−1.

Lbelow

annual belowground litter fall, g C m−2 y−1.

LAI

leaf area index, cm3 cm−3.

NEE

net ecosystem CO2 exchange.

NEP

net ecosystem production, g C m−2.

PPFD

photosynthetic photon flux density, μmol m−2 s−1.

R

gas constant, 8.134 J K−1 mol−1.

Ra

autotrophic respiration.

Re

ecosystem respiration, μmol CO2 m−2 s−1, g C m−2.

Re,Tref

ecosystem respiration rate at Tref, equal to 283.16 K, μmol CO2 m−2 s−1.

Rh

heterotrophic respiration.

Remax

maximum values of diurnal Re, μmol CO2 m−2 s−1.

RUEGPP

radiation use efficiency by GPP, mmol CO2 mol−1 photon.

Ta

air temperature, °C.

Tsoil

soil temperature at a depth of 5 cm, °C.

u*

friction velocity, m s−1.

WUEGPP

water use efficiency by GPP, mmol CO2 mol−1 H2O.

ΔWplant

annual increment of plant biomass, g C m−2 y−1.

ΔWsoil

annual increment of soil+litter biomass, g C m−2 y−1.

Acknowledgments

[38] This study was part of a joint research project of the National Institute for Environmental Studies, Japan, and the Northwest Plateau Institute of Biology, China. It was supported by the Global Environmental Research Program of the Ministry of Environment, Japan (B13), and by a Grant-in-Aid for Scientific Research (grant 13575035) by the Japan Society for the Promotion of Science.

Ancillary