Introduction
In the last decade, the carbon dioxide exchange in grassland ecosystems has
attracted much attention (Baldocchi, 2008; Huang et al., 2008; Hunt et al.,
2004; Jing et al., 2010; Suyker et al., 2003) because grasslands cover
32 % of the global land surface and make a substantial contribution to the
carbon cycle on a global scale (Parton et al., 1995). The annual net
ecosystem exchange (NEE) of grasslands has a large range, from -650 to 160 g C m-2 yr-1, due to climate variability and land use changes
(Gilmanov et al., 2007; Wang et al., 2016a). The climatic factors
controlling CO2 exchange also vary under different climate conditions
(Du and Liu, 2013; Huang et al., 2016; Xu and Baldocchi, 2004). Most
previous studies have focused on low-lying grasslands (Gilmanov et al.,
2010).
Alpine meadows in China are the primary grassland type of the nation and are
mainly distributed in the Qinghai-Tibetan plateau (DAHV and CISNR, 1996; Liu
et al., 2008). The warming trend in high-altitude areas, such as the Tibetan
Plateau and its south-east margin, has been observed to be more pronounced
(Fan et al., 2011; Liu and Chen, 2000). Several studies of CO2 exchange have been carried out
on the Qinghai–Tibetan plateau, where the mean annual
air temperature (Ta) is approximately 0 ∘C (Gu et al., 2003;
Kato et al., 2006; Shi et al., 2006; Zhao et al., 2006). The daily CO2
fluxes of the alpine meadow steppe in Damxung, Tibet were shown to be
jointly affected by Ta and soil moisture (Fu et al., 2009), while the
daily CO2 fluxes of an alpine shrubland at Haibei, Qinghai were found
to be sensitive to Ta (Zhao et al., 2006). On an annual scale, the
measurements at the Haibei alpine meadow revealed that the annual CO2
uptake was increased by the earlier onset of the growing season, which was
caused by higher Ta (Kato et al., 2006). The Lijiang alpine meadow is
located in a much warmer area (the mean annual Ta is
12.7 ∘C). A spring drought event and relatively low soil moisture
were shown to significantly delay the start time of grass germination and
reduce the annual CO2 uptake (Wang et al., 2016b). How the annual
CO2 exchange responds to the mean annual Ta is not clear for
alpine meadow ecosystems.
(a) Daily sum of solar radiation (Sin), (b) daily mean air
temperature (Ta), soil temperature (Ts), (c) vapour pressure
deficit (VPD, 5 cm), (d) soil water content (SWC, 5 cm), daily total
precipitation (PPT) and (e) 16-day average normalized difference vegetation
index (NDVI) from 2012 to 2015.
Previous studies have attributed year-to-year changes in CO2 exchange
to climatic variability (Hui et al., 2003; Xu and Baldocchi, 2004). Fluxes
may directly respond to climatic drivers or be indirectly affected by
functional changes or changes in the flux–climate relationships (Polley et
al., 2008). Statistical models have been used to partition the interannual
variation (IAV) of the CO2 exchange (Hui et al., 2003; Richardson et
al., 2007; Teklemariam et al., 2010). For example, Shao et al. (2014) found
that 77 % of the observed variation in NEE was explained by functional
changes in the moist grassland in USA, while variations in climatic
variables could better explain the IAV of NEE of a meadow in Denmark
(Jensen et al., 2017) and mixed-grass prairies in the semi-arid area of USA
(Polley et al., 2008). The relative importance of the direct and indirect
effects of the climatic variables on the interannual variations in CO2
exchange for alpine meadows in China has not been quantified.
The CO2 exchange between the atmosphere and the Lijiang alpine meadow
was measured using an eddy covariance technique from 2012 to 2015. The
objectives of this study were to (1) examine the seasonal and interannual
variation in NEE, gross primary production (GPP), ecosystem respiration
(RE) and the parameters of ecosystem photosynthesis and RE; (2) investigate
the main environmental controls of the total GPP, RE and NEE on seasonal and
annual scales; and (3) partition the interannual variation in GPP, RE and
NEE into climatic variability and vegetation growth.
The relationship between daytime NEE and PAR (a) for August from
2012 to 2015. The NEE and PAR data were averaged with PAR bins of 100 µmol m-2 s-1.
(b) The relationship between NEEsat and NDVI on
a monthly scale.
Observation site and methods
Observation site
The observation site (27∘10′ N, 100∘14′ E, 3560 m a.s.l.) is located at Maoniuping in the Yulong snow mountains, to the north
of Lijiang city on the south-east margin of the Tibetan Plateau, China. The
study area has a plateau monsoon climate, which is influenced by the
south-west and south-east monsoons. There are distinct wet and dry seasons,
with a wet season from June to October. The 30-year mean annual total
precipitation (1981–2010) at Lijiang city (2400 m a.s.l.) is 980.3 mm, and
85 % of the precipitation is concentrated in the wet season. The 30-year
mean annual air temperature (MAT) is 12.6 ∘C (data from the
Lijiang Meteorology Bureau). The dominant species in this alpine meadow are grass
of the genus Kobresia Willd, with a maximum height of 20 cm,
and the shrub Berberis Linn, with a maximum height of
more than 60 cm. The surface is covered by green vegetation, litter and bare
soil. The soil type is a loam, with a dark brown colour, which has a lower
reflectance than the grass canopy (Guo et al., 2009).
Field measurements and normalized difference vegetation index
(NDVI)
The eddy covariance (EC) system was used to measure 3-D wind speed and the
H2O and CO2 concentrations at a height of 2.5 m, with a 10 Hz
frequency. The system consisted of a three-dimensional sonic anemometer
(CSAT3, Campbell Scientific, Logan, UT, USA) and an open-path
CO2/H2O infrared gas analyzer (LI-7500A, LI-COR, Lincoln, NE,
USA). The low response measurements (1/3 Hz frequency) made in this study
were Ta and relative humidity at a height of 2.5 m close to the EC
system (HMP45C, Campbell Scientific). Net radiation (including shortwave and
longwave radiation, CNR4, Kipp&Zonen, Delft, Netherlands) and
photosynthetically active radiation (PAR; LI190SB, LI-COR) were measured at
1.5 m. Soil temperature (109-L, Campbell Scientific) and soil water content
(SWC; CS616, Campbell Scientific) were measured at a depth of 5 cm below
the ground. The precipitation (including solid precipitation in the winter)
was measured using a weighing bucket precipitation gauge (T-200B, Geonor,
Eiksmarka, Norway). All measurements were controlled by a data logger
(CR3000, Campbell Scientific), and the data were stored on a 2-GB CF card.
The average value of daily solar radiation (Sin,
MJ m-2 d-1), the mean annual air temperature (Ta, ∘C), the
mean annual vapour pressure deficit (VPD, kPa), the mean annual soil water
content (SWC, m3 m-3), the total amount of precipitation (PPT,
mm) for the whole year and the wet season, and the maximum value of NDVI for
each year from 2011 to 2015.
Variables
2012
2013
2014
2015
Sin
14.23
14.40
14.44
14.59
Ta
5.93
5.92
6.32
6.16
VPD
0.32
0.30
0.32
0.30
SWC
0.232
0.227
0.232
0.233
PPT (whole year)
1190.4
1066.1
1204.8
1257.4
PPT (wet season)
1086.5
906.1
1092.6
1067.1
NDVImax
0.60
0.68
0.72
0.72
Four points around the flux tower were selected to investigate the
variations in vegetation growth. The 250 × 250 m2 gridded NDVI
data at 16-day intervals (product name: MOD13Q1) for the four points were
obtained from the Moderate Resolution Imaging Spectrometer (MODIS) on the
EOS-1Terra satellite and were averaged to represent the meadow at this
observation site. Observations affected by clouds were removed during this
process, and the gaps were filled linearly.
Flux calculation and quality control
EddyPro software (version 5.1, LI-COR) was used to calculate the half-hourly
CO2 flux based on the 10 Hz raw data. After a spike detection (Vickers
and Mahrt, 1997), the sector-wise planar fit method was used to transform
the coordinate system due to a terrain slope of approximately 10∘
(Wilczak et al., 2001). The CO2 flux was also subjected to a spectral
loss correction (Moore, 1986) and density correction (WPL correction; Webb
et al., 1980).
Stationary and integral turbulence characteristics tests were used for flux
quality control (Foken and Wichura, 1996). When u* was less than 0.1 m s-1, the CO2 flux was dependent on u* and was discarded. Because
there was a coniferous forest approximately 350 m to the north of the site,
an analytical footprint model was used to determine whether the half-hourly
CO2 flux was influenced by the forest and needed to be removed (Kormann
and Meixner, 2001).
(a) Relationship between RE and Tsoil in 2012;
(b) relationship between RE and Tsoil for the wet season from 2012 to 2015; RE
and Tsoil were averaged with Tsoil bins of 1 ∘C.
The daily mean NEE, GPP and RE from 2012 to 2015.
The ecosystem photosynthesis parameters using Eq. (1)
(NEEsat: µmol m-2 s-1, α: µmol m-2 s-1, R2) and
NDVI for each month during the wet seasons from 2012
to 2015. The regression was based on the average values of NEEdaytime
and PAR with PAR bins of 100 µmol m-2 s-1. NEEsat(a)
represents the mean value and the standard deviation, NEEsat(b)
and NEEsat(c) represent the maximum and minimum values of NEEsat
for each month.
Month
NEEsata
NEEsatb
NEEsatc
α
REbulk
June
-11.59 ± 2.45
-9.69
-15.08
-0.037 ± 0.009
3.59 ± 0.52
July
-19.67 ± 1.54
-17.46
-21
-0.050 ± 0.009
3.75 ± 0.83
August
-20.14 ± 3.52
-15.43
-23.75
-0.055 ± 0.016
4.15 ± 0.74
September
-16.44 ± 4.56
-11.43
-21.44
-0.051 ± 0.017
3.70 ± 1.04
October
-9.36 ± 1.62
-7.08
-10.9
-0.031 ± 0.005
2.45 ± 0.37
After quality control, approximately 70 % of the CO2 fluxes were
subjected to further analysis. Linear interpolation was used to fill flux
gaps of less than 2 h. To fill gaps of longer than 2 h, marginal
distribution sampling, an improved look-up table method, was used (Falge
et al., 2001; Lloyd and Taylor, 1994).
Data analysis
Using the homogeneity-of-slopes (HOS) model (Hui et al.,
2003), the control of the CO2 exchanges (NEE, GPP and RE) was
statistically partitioned into four components: the interannual variation of
environmental variables (SSi), the seasonal variation of environmental
variables (SSs), variations of biological variables (SSf, NDVI in
this study) and random error (SSe, resulting from measurement and
analysis random error). To identify the significant control variables, a
multiple stepwise regression analysis of the CO2 exchanges with
environmental variables was conducted using SPSS 12.0 for Windows (SPSS
Inc., Chicago, IL, USA). The environmental variables that were significantly
correlated with fluxes were submitted for further HOS analysis, while the
others were excluded from the analysis. To minimize errors, the daily NEE,
GPP and RE were excluded from regressions if more than 50 % of the data
points in the daytime (Rn > 5 W m-2) were missing. More
details of the HOS model are provided in Hui et al. (2003).
The relationship between daytime NEE (NEEdaytime) and PAR was described
by the Michaelis–Menten model (Falge et al., 2001):
NEEdaytime=αNEEsatPARαPAR+NEEsat+REbulk,
where NEEsat is the NEE at the saturated light level, α is the
apparent quantum yield (µmol CO2 µmol-1 photons), and
REbulk is the bulk estimated RE.
The Van 't Hoff equation was used to evaluate the relationship between the
night-time NEE (NEEnighttime, µmol CO2 m-2 s-1) and
soil temperature at a depth of 5 cm (Ts, ∘C; Aires et al.,
2008):
NEEnighttime=aexp(bTs),
where a and b are the regression parameters. The temperature sensitivity
coefficient (Q10) of RE was determined using the following equation.
Q10=exp(10b)
The partitioning of NEE into GPP and RE was based on the assumption that the
sensitivity of RE to soil temperature was the same during the day and at
night (Falge et al., 2001). The regression parameters derived from the
night-time data were extrapolated to the daytime to calculate the daytime RE
and the daily RE. The daily GPP was calculated as follows:
GPP=RE-NEE.
Results
Weather conditions and NDVI
The daily integrated solar radiation (Sin) varied from 1.15 to 32.40 MJ m-2 d-1 (Fig. 1a).
The mean Sin in spring (March to May)
was 17.0 to 19.93 MJ m-2 d-1 and was clearly larger than in other
seasons. In the wet season, the mean Sin was 9.99 to 11.05 MJ m-2 d-1.
The MAT was 5.92 to 6.32 ∘C (Table 1). The daily mean Ta
ranged from 0.41 to 14.96 ∘C in the wet season and decreased to a
minimum value of -9.06 ∘C in the winter. In contrast, the soil
temperature never decreased below 0 ∘C, and the maximum value was
16.48 ∘C (Fig. 1b). The vapour pressure deficit (VPD) reached
its maximum value of 1.07 kPa before the wet season (Fig. 1c). The VPD
decreased to near 0 kPa, and the mean VPD for the wet season was 0.125 to
0.166 kPa.
The annual precipitation from 2012 to 2015 ranged from 1066.1 to 1257.4 mm.
The precipitation during the wet season ranged from 906.1 to 1092.6 mm,
accounting for 85 to 91 % of the annual total precipitation (Table 1). The
mean annual SWC had a small interannual variability, from 0.227 to 0.233 m3 m-3. In the wet season, the SWC reached a maximum value of
approximately 0.35 m3 m-3, and the minimum SWC was 0.15 m3 m-3 (Fig. 1d).
The NDVI of this alpine meadow displayed a clear seasonal and interannual
variation (Fig. 1e). The NDVI exceeded 0.4 at the end of April or in late
May, depending on the amount and distribution of precipitation in the spring
(March to May). The maximum NDVI for each year ranged from 0.60 (2012) to
0.72 (2013). In all 4 years, the NDVI decreased to below 0.4 at the end of
October.
The ecosystem respiration parameters using Eqs. (2) and (3) (a: µmol m-2 s-1,
b: Q10, R2) for the wet and dry seasons
from 2012 to 2015. The regression was based on the average values of RE and
Tsoil with Tsoil bins of 1 ∘C.
Season
Year
a
b
Q10
R2
Wet season
2012
0.437
0.125
3.48
0.98
2013
0.374
0.124
3.46
0.94
2014
0.442
0.126
3.51
0.98
2015
0.433
0.123
3.43
0.98
Dry season
2012
0.338
0.081
2.25
0.78
2013
0.202
0.096
2.60
0.74
2014
0.283
0.115
3.15
0.99
2015
0.313
0.104
2.82
0.70
Seasonal and interannual variations in NEEsat, α and
Q10
The daytime NEE and PAR were averaged, with PAR bins of 100 µmol m-2 s-1 to avoid random errors. For each month in the wet
season, the daytime NEE decreased with PAR until a critical PAR was reached.
Above the critical PAR, the daytime NEE increased and the CO2 uptake
was depressed (Fig. 2a). To derive NEEsat and α, the NEE and
PAR data were used only when PAR was below the critical value. NEEsat
showed a clear seasonal variation (Table 2). The mean NEEsat values for
each month showed that NEEsat began to increase in June (-11.59 µmol m-2 s-1)
and reached a maximum in August (-20.14 µmol m-2 s-1). The highest NEEsat during the whole observation
period occurred in August of 2014 (-23.75 µmol m-2 s-1).
NEEsat then declined with grass senescence in September and October.
The NEEsat in October (-9.36 µmol m-2 s-1) was less than
half that in August. The interannual variations in NEEsat were also
large. For example, NEEsat in September 2015 (-21.44 µmol m-2 s-1)
was almost twice that in September 2013 (-11.43 µmol m-2 s-1; Table 2). On a monthly scale, 81 % of the variation in
NEEsat could be explained by the mean NDVI (Fig. 2b). Over this
meadow, NEEsat did not significantly correlate with SWC because the
soil water conditions were always good in the wet season.
At monthly intervals, there were large random errors in the regression
between RE and Tsoil. For example, the R2 for each month of the
wet season in 2012 ranged from 0.04 to 0.12. Thus, in 2012, the data in the
wet and dry season were combined to fit the regression (Fig. 3a). The
Q10 in the wet seasons was similar, at approximately 3.45 (Table 3), which
was in the normal range of previous studies (1.2 to 3.7; Falge et al.,
2001). These values were clearly higher than those for temperate grasslands
(1.99 to 3.07; Wang et al., 2016a), Mediterranean grasslands (1.22 to 2.36;
Aires et al., 2008) and the Haibei alpine meadow (1.50 to 2.27; Kato et
al., 2004). Q10 was lower in the dry season than in the wet season.
The annual total NEE, GPP and RE (g C m-2 yr-1) for each
year from 2012 to 2015.
2012
2013
2014
2015
NEE
-114.2
-158.5
-159.9
-212.6
GPP
522.3
546.5
669.4
661.8
RE
412.1
393.6
515.2
456.7
Seasonal and interannual variation in NEE, GPP and RE
The ecosystem started to absorb CO2 (negative value of NEE) on DOY 165
in 2012, DOY 137 in 2013, DOY 116 in 2014 and DOY 104 in 2015, and then NEE
decreased (Fig. 4). The minimum daily NEE for each year occurred in July
or August (-3.52 on DOY 196 in 2012, -3.35 on DOY 218 in 2013,
-3.43 on DOY 243
in 2014 and -4.16 g C m-2 d-1 on DOY 210 in 2015). NEE
increased significantly in September and became positive on DOY 293 in 2012,
DOY 305 in 2013, DOY 295 in 2014 and DOY 297 in 2015. The maximum difference
in the start time of CO2 uptake was 61 days while the difference in the
end time was 12 days. The CO2 uptake period was much shorter in 2012
(129 days) than in 2013 (169 days), 2014 (180 days) and 2015 (194 days).
The cumulative NEE, GPP and RE from 2012 to 2015.
The daily GPP increase started earlier than CO2 uptake. The seasonal
pattern of daily GPP was similar to that of NEE, although the amplitude of
GPP variations was larger than that of NEE variations. The maximum daily GPP
for each year was 6.02, 5.47, 6.23 and 5.95 g C m-2 d-1 for the
4 years from 2012 to 2015. Compared with the NEE and GPP,
the seasonal variation in RE was smaller during the wet season. In
particular, RE varied only slightly from June to August.
The annual GPP in 2014 and 2015 was clearly higher than in 2012 and 2013, as
indicated by the larger NDVI (Fig. 5; Tables 1, 4). In contrast, the RE in
2014 was the highest of all 4 years because, although the Q10 value
was similar to the other years, it had the highest Ta (Table 1).
Therefore, the annual NEE in 2014 was similar to that in 2013, but lower
than that in 2015, although the GPP was similar in 2014 and 2015. The spring
drought resulted in a significantly lower NDVI in 2012 than in the other
years; consequently, the annual GPP in 2012 was the lowest of all 4
years. The annual NEE for the 4 years followed the order of 2015 < 2014 < 2013 < 2012 (Table 4), which is consistent with the
length of the CO2 uptake period.
Relationships between (a) NEE and Ta, (b) GPP and Ta and
(c) RE and Ta for the wet seasons from 2012 to 2015.
The percentage of the contributions of the seasonal climatic
variation (SSs), interannual climatic variability (SSi), the
ecosystem functional change (SSf) and random error (SSe) to the
interannual variations in NEE, GPP and RE.
SSs
SSi
SSf
SSe
NEE
37.7 %
7.7 %
10.3 %
44.3 %
GPP
48.6 %
9.7 %
10.7 %
31.0 %
RE
48.6 %
15.6 %
21.2 %
14.6 %
The GPPdiff for 2013–2012, 2014–2012 and 2015–2012 during the
periods from March to May, June, from June to July and from August to September.
GPPdiff
Periods
2013–2012
2014–2012
2015–2012
March to May
20.0
63.1 (43 %)
83.3 (60 %)
June
28.7
13.4 (9 %)
23.7 (17 %)
July to August
-8.7
14.2 (10 %)
-18.5 (-13 %)
September to October
-12.0
55.2 (38 %)
48.2 (35 %)
Entire year
24.2
147.1
139.5
Discussion
Partitioning the interannual variation in CO2 exchange
The HOS model was used to partition the interannual variation (IAV) in
CO2 exchange into climatic variability and ecosystem functional change,
which was reflected by the variability of the flux–climate relationship
throughout the years (Hui et al., 2003). During the wet season, the daily NEE, GPP
and RE were mainly related to Ta (Fig. 6). The effect of PAR on NEE
and GPP was very weak, with R values of -0.05 and 0.08, respectively.
Comparison of mean annual temperature (MAT, ∘C), mean
annual precipitation (MAP, mm yr-1), NEE (g C m-2 yr-1), GPP,
RE and RE / GPP between this study and previous grassland studies.
References/
Ecosystem
Latitude
Longitude
Altitude
MAT
MAP
NEE
GPP
RE
RE / GPP
Location
Description
This study/
Alpine meadow/
27∘10′ N
100∘14′ E
3560
6.1
1180
-161 (-213
600 (522
444 (394
0.74 (0.69
Lijiang, China
shrub
to -114)
to 669)
to 515)
to 0.79)
Yu et al. (2006)/
Alpine meadow
30∘51′ N
90∘05′ E
4250
2.1
520
28 (16
167 (144
195 (183
1.16 (1.08
Damxung, China
and 39)
and 190)
and 206)
and 1.27)
Kato et al. (2006)/
Alpine
37∘37′ N
101∘18′ E
3250
-1.0
566
-121 (-193
634 (575
514 (489
0.81 (0.72
Haibei, China
shrub
to -79)
to 681)
to 556)
to 0.86)
Shimoda et al. (2005)/
C3/C4 grassland
36∘06′ N
140∘06′ E
27
13.9
1156
-17 (-78
2365 (2285
2348 (2303
0.99 (0.97
Japan
to 17)
to 2426)
to 2392)
to 1.01)
Aires et al. (2008)/
Mediterranean
38∘28′ N
8∘01′ E
140
15.5
669
-71 (-190
893 (524
822 (573
0.92 (0.85
Portugal
grassland
and 49)
and 1261)
and 1071)
and 1.09)
Jensen et al. (2017)/
Meadow
55∘55′ N
8∘24′ E
0
8.7
809
-156 (-356
1349 (1147
1193 (1069
0.88 (0.75
Denmark
to -18)
to 1570)
to 1406)
to 0.98)
Gilmanov et al. (2007)/
Multiple
–
–
-0.7
3.9
387
-150 (-653
1261 (467
1111 (493
0.90 (0.59
Europe
(19 sites)
to 1770
to 14.6
to 1816
to 171)
to 1874)
to 1622)
to 1.14)
Xu and Baldocchi (2004)/
Mediterranean
38∘24′ N
120∘57′ E
129
16.3
559
-52 (-132
798 (729
747 (735
0.94 (0.85
USA
grassland
and 29)
and 867)
and 758)
and 1.04)
Flanagan et al. (2002)/
Temperate
49∘26′ N
112∘34′ E
951
–
378
-2 (-21
280 (272
278 (267
1.0 (0.93
Canada
grassland
and 18)
and 287)
and 290)
and 1.07)
Seasonal variations in the differences of (a) NEE, (b) GPP and
(c) RE from 2013 to 2012, from 2014 to 2012 and from 2015 to 2012.
Cumulative (a) Ta and (b) GPP and (c) the daily mean SWC from
March to May (DOY60 to 151) for 2012, 2013, 2014 and 2015.
Relationships between (a) NEE and Ta; (b) GPP and Ta;
(c) RE and Ts from July to October at a monthly scale and (d) relationship
between the annual total CO2 exchange fluxes and the mean annual
Ta, which were GPP = -1191Ta2+14 930 Ta-46102,
R2= 0.97 and RE = 276Ta-1235, R2= 0.97.
Relationships between the annual GPP and precipitation (PPT) for
this Lijiang site, the Haibei site (Kato et al., 2006; Fu et al., 2009), the
Damxung site (Fu et al., 2009) and the semi-arid grassland sites (Fu et al.,
2009; Du and Liu, 2013; Yang and Zhou, 2013).
A separate-slopes model was constructed for each year, and the multiple
regression model was based on data from the observational period. Compared
with the multiple regression model, the separate-slopes model substantially
improved the NEE estimation, with R2 increasing from 0.69 in the
multiple regression model to 0.79 in the separate-slopes model. This means
that the separate-slopes model accounted for 10.3 % more variation in the
observed NEE than the multiple regression model, which was attributed to the
functional change (SSf). The other 89.7 % of the variation in the
observed NEE was partitioned to interannual climatic variability (SSi,
7.7 %), seasonal climatic variation (SSs, 37.7 %) and random error
(SSe, 44.3 %; Table 5). Therefore, most of the IAV in NEE, GPP and
RE was attributable to the variation in climatic variables, in particular,
climatic seasonal variation. This is in line with the findings reported for
a Skjern meadow in Denmark and a temperate ombrotrophic bog in Canada (Jensen et
al., 2017; Teklemariam et al., 2010). In contrast, Braswell et al. (1997)
and Shao et al. (2014) found that functional change, rather than the direct
effects of IAV in climate, accounted for more IAV in fluxes. Moreover, the
contributions of different drivers to the IAV in GPP were similar to that of
NEE, while the functional change in RE was twice that of NEE and GPP. The
R2 values for NEE, GPP and RE in the multiple regression model were
0.44, 0.53 and 0.59. It was considered reasonable that the
largest random error was recorded for NEE.
Control of the interannual variation in the CO2 exchange
To examine the interannual variation in CO2 exchange, the cumulative
NEE, GPP and RE in 2013, 2014 and 2015 were compared with the corresponding
values in 2012 (Fig. 7). The cumulative NEEdiff (the difference in
NEE) values for 2014–2012 and 2015–2012 increased rapidly in spring and
autumn. In summer, the differences between 2012, 2014 and 2015 varied
slightly. The cumulative NEEdiff for 2013–2012 increased from April
to early August. These patterns were similar to those for GPPdiff.
However, the annual cumulative GPPdiff (24.3 to 147.2 g C m-2 yr-1) was relatively larger than the annual cumulative NEEdiff
(-44.2 to -98.3 g C m-2 yr-1). The cumulative REdiff
decreased from DOY 1 and then increased in spring. The cumulative
REdiff for 2013–2012 and 2015–2012 reached its maximum at the end of
June, while the cumulative REdiff for 2014–2012 increased throughout
the entire year and was the largest of all the periods considered.
The daily CO2 uptake over this meadow ecosystem has previously been
shown to increase with Ta (Wang et al., 2016). Especially in the spring
(March to May), the temperature affected the vegetation growth and GPP. From
March to May, the cumulative Ta was 592.3, 577.1, 633.1 and
647.6 ∘C in the 4 years from 2012 to 2015.
Consequently, the cumulative GPP in the spring increased in the order of
2015 > 2014 > 2013. The exception was that the spring of
2012 had a higher Ta, but a lower GPP than the spring of 2013. Compared
with the GPP in 2013, the GPP in 2012 increased more significantly due to
the higher Ta from March to April. However, the drought in May 2012
delayed vegetation growth and reduced GPP. The differences in GPPcum for
2013-2012, 2014-2012 and 2015-2012 at the end of May were 20.0, 63.1 and 83.3 g C m-2, representing 82.6, 42.9 and 59.7 % of the differences for
the three periods.
From July to October, the NEE, GPP and RE were all strongly correlated with
Ta on a monthly scale (R2= 0.84, 0.86 and 0.73; Fig. 9a, b, c). The slope of the relationship between GPP and Ta
was much larger than for that between RE and Ta, indicating that when
Ta increased, the alpine meadow ecosystem absorbed more CO2. The
monthly GPP in July and August varied slightly between the 4 years, while
the interannual variability of the GPP in September was the largest because
the monthly means Ta in September for 2012 (8.6 ∘C) and
2013 (8.8 ∘C) were significantly lower than those for
2014 (9.7 ∘C) and 2015 (10.0 ∘C). Consequently, the
differences in GPPcum for 2014-2012 and 2015-2012 from September to
October were 55.2 and 48.2 g C m-2, representing 37.5 and 34.6 % of
the differences for the two periods.
On the annual scale, the annual total NEE decreased with the MAT in 2012,
2013 and 2015 and then increased when the MAT was highest in 2014. The
reason for this was that the annual total RE increased linearly with the MAT
(R2= 0.97), while the relationship between the GPP and MAT was
non-linear (Fig. 9d). The GPP became saturated as the MAT increased. In
contrast, the annual NEE increased with the MAT in the Haibei alpine meadow,
which is in line with previous studies showing that the annual NEE is
comprehensively controlled by the temperature environment (Kato et al.,
2006).
Comparison of annual CO2 exchange with other sites
The annual GPP at the study site was much larger than that reported in
semi-arid grasslands in Tibet and Canada (Flanagan et al., 2002; Yu et al.,
2006), but much lower than that reported in moist grasslands in low-lying
areas in Europe (Table 7). In China, the annual GPP for semi-arid grasslands
and the Haibei alpine meadow increased slightly with annual precipitation
(R2=0.94; Fig. 10). With similar annual precipitation, the annual
GPP of the Damxung site was much lower than the Haibei site, possibly because
of higher elevation. When the annual precipitation increased further, to over
1000 mm yr-1, the annual GPP remained steady (Fig. 10). The annual
GPP for the grassland ecosystems in China was always below 700 g C m-2 yr-1.
In addition to temperature effects, the daily RE was also correlated with
the daily GPP (RE = 0.44GPP + 0.63, R2= 0.82) during the wet season
from 2012 to 2015. Due to the high elevation and low soil temperature in the
summer, the percentage of RE to GPP for this meadow site was lower than for
Mediterranean grasslands (RE = 0.53GPP + 0.72, R2= 0.85, Aires et al., 2008; RE = 0.47GPP + 1.33, R2= 0.85,
Xu and Baldocchi, 2004). The
low level of RE resulted in a similar or even lower annual NEE (mean value:
-161 g C m-2 yr-1) at Lijiang than in moist grasslands with a low
elevation (Table 7). For example, the mean annual NEE for a meadow in
Denmark (annual precipitation: 809 mm) was -156 g C m-2 yr-1,
while the mean annual NEE for a C3/C4 grassland in Japan (annual
precipitation: 1156 mm) was -17 g C m-2 yr-1. The ratio of RE to
GPP ranged from 0.69 to 0.79 over the Lijiang alpine meadow, which was lower
than the Haibei alpine meadow (Table 7). This is why the annual
NEE at the Lijiang site was on average 25 % lower than at the Haibei site.
In general, low RE / GPP ratios occurred in high-altitude and moist areas. The
alpine meadow ecosystem (Lijiang and Haibei) had a lower RE / GPP ratio than
most low-lying grasslands. Compared with semi-arid grasslands (RE / GPP:
approximately 1.0), the RE / GPP ratios reported in moist grasslands are much
lower, e.g. a sown grassland in the Netherlands (0.60) and a natural
grassland in Italy (0.59; Gilmanov et al., 2007).
Conclusions
The 4-year EC data from 2012 to 2015 were used to investigate the
interannual variation in the NEE, GPP and RE. The key parameters for
ecosystem photosynthesis and respiration were determined for the different
seasons of each year. The vegetation growth (NDVI) controlled NEEsat on
a monthly scale, and the interannual variation in Q10 for the wet and
dry seasons was small. The seasonal variation in CO2 exchange was
affected by the seasonal pattern of Ta and the soil moisture in the
spring. In the spring, low Ta and drought events delayed the start time
of CO2 uptake. In the late wet season, the higher Ta in 2014 and
2015 resulted in later grass senescence and CO2 release. The annual NEE
decreased with the length of the CO2 uptake period, but its
relationship with the NDVI was not significant. For this alpine meadow, the
HOS model suggests that most of the IAV in NEE, GPP and RE was attributed to
the seasonal variation in climatic variables. On an annual scale, the annual
RE increased linearly with the MAT, while the annual GPP became saturated
when the MAT increased from 6.16 to 6.32 ∘C. Thus,
the annual NEE decreased and then increased with the MAT. The low RE / GPP
ratio at the study site was responsible for the lower annual NEE compared
with some other grassland ecosystems with a larger GPP.