Eddy covariance and biometric measurements show that a savanna ecosystem in Southwest China is a carbon sink

Savanna ecosystems play a crucial role in the global carbon cycle. However, there is a gap in our understanding of carbon fluxes in the savanna ecosystems of Southeast Asia. In this study, the eddy covariance technique (EC) and the biometric-based method (BM) were used to determine carbon exchange in a savanna ecosystem in Southwest China. The BM-based net ecosystem production (NEP) was 0.96 tC ha−1 yr−1. The EC-based estimates of the average annual gross primary productivity (GPP), ecosystem respiration (Reco), and net ecosystem carbon exchange (NEE) were 6.84, 5.54, and −1.30 tC ha−1 yr−1, respectively, from May 2013 to December 2015, indicating that this savanna ecosystem acted as an appreciable carbon sink. The ecosystem was more efficient during the wet season than the dry season, so that it represented a small carbon sink of 0.16 tC ha−1 yr−1 in the dry season and a considerable carbon sink of 1.14 tC ha−1 yr−1 in the wet season. However, it is noteworthy that the carbon sink capacity may decline in the future under rising temperatures and decreasing rainfall. Consequently, further studies should assess how environmental factors and climate change will influence carbon-water fluxes.

Ecosystem carbon budget. Biometric-based NEP and carbon use efficiency. The inventoried biomass, litterfall (for details on the seasonal and annual variations, see Supplementary Fig. S1), and measured R h data gave values for ΔB, L p , and R h of 1.97, 2.14, and 3.16 tC ha −1 yr −1 between 2014 and 2015, respectively, and the biometric-based NEP was estimated from equation (1) to be ~0.96 tC ha −1 yr −1 in 2015 (Table 1). Carbon use efficiency (CUE), which reflects the capacity of forests to absorb CO 2 from the atmosphere and fix it in terrestrial biomass and the influence of autotrophic respiration on GPP in forests, is defined as the ratio of NPP to GPP, giving a value of 0.60 (Table 1).
Eddy covariance carbon exchange and its variations. Using 32 months (May 2013 to Dec 2015) of data, the amplitude of the averaged daily NEE ranged from −1.28 to 1.57 gC m −2 d −1 (the largest net carbon release was on 22 January) in the dry season, and −3.03 (maximum net carbon uptake was on 30 August) to 0.65 gC m −2 d −1 during the wet season (Fig. 2). The peak values of mean gross ecosystem productivity (GEP) and R eco were both observed in August: GEP (GEP = −GPP) reached a peak of −6.09 gC m −2 d −1 on 30 August, and the maximum value of R eco was 3.37 gC m −2 d −1 on 17 August. Strong seasonality in NEE (NEE = −NEP) was observed at the study site (Fig. 2). The site is almost carbon neutral (i.e., the total sum of CO 2 absorbed by photosynthesis is nearly equal to that released by ecosystem respiration) in the dry season (−0.16 tC ha −1 ), but appreciable CO 2 uptake was observed in the wet season (−1.14 tC ha −1 ) during the study period. A comparison of seasonal values of GEP and R eco shows large variations between the dry season (−2.12 tC ha −1 and 1.96 tC ha −1 , respectively) and the wet season (−4.72 tC ha −1 and 3.58 tC ha −1 , respectively). Overall, the respiration rate of the ecosystem in the wet season is approximately 1.8 times that in the dry season, whereas the photosynthesis rate during the wet season is ~2.2 times that during the dry season, i.e., the wet season GEP was ~70% of the total annual GEP. Thus, the savanna ecosystem served as a carbon sink in our study period and the average annual sum of the NEE was −1.30 tC ha −1 yr −1 .  Representative diurnal patterns of carbon fluxes. Averaged over the year, the savanna ecosystem absorbed and fixed CO 2 for ~9.5 hours per day (07:30-17:00). The savanna ecosystem becomes a carbon sink as the daily global radiation increases, and the average net maximum assimilations (i.e., NEE values) at 13:00 in the dry season, wet season, and annually were approximately 2.2, 4.6, and 3.4 μmol m −2 s −1 , respectively. The rate of daytime CO 2 fixation in the wet season is about twice that of the dry season (Fig. 3a). R eco and GPP of the savanna ecosystem increased with increasing photosynthetically active radiation (PAR) and temperature after sunrise, reached their peaks at 16:00 (R eco ) and 13:00 (GPP), and then decreased until sunrise the next day (Fig. 3b,c). The values of NEE, R eco , and GPP in the dry season are less than half those in the wet season (Fig. 3).
Monthly patterns of daytime NEE light response parameters. Regarding monthly variations in daytime NEE light response parameters (Fig. 4), the apparent quantum yield (α) (Fig. 4a), maximum net photosynthetic rate (P max ) (Fig. 4b) and dark respiration of the ecosystem (R d ) (Fig. 4c) showed similar trends of monthly variation over the studied savanna ecosystem. The maximum and minimum α, P max , and R d values were observed in August and April, respectively. In general, α, P max , and R d values in the wet season (May-October) were higher than those during the dry season (November-April).
Seasonal daytime NEE responses to photosynthetically active radiation (PAR). Carbon sequestration ability increased with increases in PAR irrespective of dry season or wet season (Fig. 5). However, the ecosystem showed higher (3.7 times) light transformation efficiency (photosynthetic capacity) in the wet season (0.0306) (Fig. 5b) than in the dry season (0.0083) (Fig. 5a), implying that most of the NEE (carbon sequestration amount) accumulated during the wet season (May-October).
Responses of NEE to temperature, monthly rainfall, RH, and VPD. To explore the responses of net ecosystem carbon exchange (NEE, kg C ha −1 month −1 ) to environmental factors, a quadratic regression model was applied to make quantitative predictions of monthly NEE to monthly mean air temperature (T air ), relative humidity (RH), vapor pressure deficit (VPD) and total monthly rainfall (Fig. 6). According to the regression results, carbon sink capacity increased with increasing T air , but decreased rapidly when T air was higher than 24.7 °C (Fig. 6a); a similar trend was observed between NEE and VPD, in that carbon sequestration capacity decreased when VPD was higher than 13.7 hPa (Fig. 6d). Further, the carbon sink capacity of our study area decreased with decreasing RH and monthly rainfall (Fig. 6b,c). Discussion Annual carbon exchange. Both the biometric method (BM) and the eddy covariance (EC) method were applied to quantify carbon exchange at the present study site (Fig. 7), although the two methods are different in terms of their spatial scales, temporal resolutions, the assumptions made by both techniques, as well as their advantages and flaws. EC is not only a less disturbing or non-destructive way to investigate carbon exchange, but also provides a dataset with high spatial resolution. This dataset covers time scales ranging from seconds to years 43 . In addition, EC can usually cover a larger spatial scale than BM and has better spatial representativeness 44,45 . However, EC assumes that the underlying surface should be horizontally homogenous 46,47 . This assumption is extremely difficult to satisfy, because forest ecosystems are comprised of heterogeneous canopy and terrain features. Therefore, as a conventional approach, it is necessary to apply BM simultaneously with EC at our study site, although there are some inherent flaws (including the extensive field work required, the indirect nature of the measurements, the method is more destructive, etc.). Furthermore, it is necessary to use the BM method to study  ecosystem carbon exchange while tracking the contribution to NEP from each carbon pool in order to calculate forest carbon use efficiency. The results obtained using the BM (0.96 tC ha −1 yr −1 ) ( Table 1) and EC (1.30 tC ha −1 yr −1 ) ( Fig. 2) indicate that the site is an appreciable carbon sink, although under the control of high mean annual temperature (24.0 °C), high maximum mean monthly temperature (maximum MMT; 29.2 °C), but low mean annual rainfall (786.6 mm) (Fig. 8). There is a difference between the results of the biometric and eddy covariance methods that cannot be ignored. The discrepancy between them did not sufficiently indicate that the "lost" carbon amount of 0.34 tC ha −1 yr −1 (0.34 = 1.30-0.96) was fully stored as organic soil matter. The explanation for the discrepancy may be as follows. 1) The time periods covered by the EC observations (May 2013 to December 2015) differed from those of the biometric measurements (November 2013 to November 2015) 48 , which appears to be the most likely reason for the discrepancy 49 ; 2) the allometric equations 50 were not site-specific; 3) the flux footprint and the inventory plot were not exactly identical; 4) there is a time lag between tree growth derived from the BM and ecosystem photosynthesis determined from EC 44 ; and 5) NEP measured by BM is usually lower than EC results under the conditions of well-developed turbulence 44,48,51 . In summary, temporal mismatch, the allometric equations, and the inventory are the three main reasons for the discrepancy between BM and EC results. Nevertheless, it is conceivable that the savanna ecosystem could be treated as a carbon sink regardless of which method is used. The EC result was more reasonable, although there are some uncertainties caused by lower turbulence on calm nights, advection, and possible cold air drainage of CO 2 43,52-55 . Therefore, it is important to consider the plausibility of the eddy flux of net ecosystem carbon exchange at the study site.
Is it reasonable that the savanna ecosystem absorbed and fixed ~1.30 tC ha −1 in a year with strong seasonality ? Our answer is positive for the following reasons. 1) We followed strictly the ChinaFLUX procedures for QA/QC (quality assurance and quality control) and post-processing of the data to ensure reliable flux estimates 43 . Furthermore, an online procedure 53 , recommended by FLUXNET and maintained by the Max Planck Institute, was applied for gap filling and partitioning of the flux data with the widely used (particularly for forest systems) friction velocity threshold of 0.2 52,55 . 2) The biometric inventory result (Table 1), which was conducted within the footprint (Fig. 1) is consistent with the EC result, although there is some difference between the values. 3) The Ailaoshan subtropical evergreen broadleaf forest (24°32′N, 101°01′E, 2476 m a.s.l.) acts as a carbon sink of ~9 tC ha −1 yr −1 with little seasonality due to the lower mean annual temperature (MAT) and abundant mean annual rainfall (MAR) 44 . The amount of carbon sequestration at the present study site is just one-seventh of their Seasonal variations in carbon exchange. We have shown that the savanna ecosystem in our study site acts as a carbon sink, that it absorbed approximately 1.30 tC ha −1 yr −1 from the atmosphere by means of photosynthesis, and that this result is reasonable and convincing. We next consider whether the seasonal variation in carbon fluxes is also reasonable. The 32-month (May 2013 to December 2015) averaged NEE of the wet season (−1.14 tC ha −1 yr −1 ) is ~7 times that of the dry season (−0.16 tC ha −1 yr −1 ) (Fig. 2). Interestingly, the dramatic seasonal variations in NEE, GEP, and R eco values in Africa, Australia, and Brazil are also highly consistent with our results 9,15,21,23,25,34,56 . A study of a West African savanna reported a carbon source of 47.72 gC m −2 in the dry season but a carbon sink of −374.49 gC m −2 in the wet season in 2008 23 . Furthermore, in a semi-arid sparse savanna in Demokeya, Sudan, the daily amplitude of NEE in the wet season (−1.8 gC m −2 day −1 ) was 9 times that in the dry season (−0.2 gC m −2 day −1 ) 25 ; the factor reached 20 in a tropical savanna in Australia 15 . Therefore, it is reasonable and not surprising that 87.7% of the NEE is absorbed in the wet season (−1.14 tC ha −1 yr −1 ) while the dry season is nearly carbon neutral in the savanna ecosystem at our study site.
The second question is why four-fifths of the NEE is taken up in the wet season. The reasons are as follows. The daytime NEE responses to photosynthetically active radiation (PAR) tell us that, although the dark respiration of the ecosystem (R d ) increased during the wet season (May-October) (Fig. 4c), both light use efficiency (α) (Fig. 4a) and maximum net photosynthetic rate (P max ) increased in the wet season (Fig. 4b). The wet-to dry-season ratio of α was 3.13 (0.0267 in the wet season and 0.0085 in the dry season), and P max (3.11-13.98 μmol CO 2 m −2 s −1 ) reached its peak (13.98 μmol CO 2 m −2 s −1 ) in August. In addition, GEP and NEE increased rapidly with the coming of the wet season, and peak GEP and NEE were −6.09 and −3.03 gC m −2 d −1 on 30 August (Fig. 2), respectively. Furthermore, the net assimilation of carbon increased dramatically in the wet season compared with the dry season (Fig. 5), and previous studies have also shown higher photosynthesis rates in the wet season than in the dry season 57 . Therefore, the fact that most of the annual NEE accumulated during the wet season in our research area is reasonable and convincing.
Climate change and carbon exchange. The savanna ecosystem at our study site acts as a carbon sink of 1.30 tC ha −1 yr −1 in the global carbon cycle, with approximately 88% of this carbon being absorbed during the wet season (May-October), while it is nearly carbon neutral in the dry season (Fig. 2). Carbon sink capacity decreases with increasing T air and VPD and decreasing rainfall and RH (Fig. 6). Therefore, it is important to consider the impacts of future climate changes on carbon exchange in such a savanna ecosystem, as its severe environment may be highly sensitive to changes in rainfall and temperature 5,30 , and many previous studies have revealed that water and temperature have important impacts on savanna ecosystem carbon exchange 5,8,13,29,34,37,[58][59][60] . Observations show that, over the past 36 years (1980-2015), the climate in the present study site has become hotter and drier with increasing T air and VPD, while annual rainfall and RH show decreasing trends (Fig. 9). In addition, there was a significant contradiction between water and heat, with an increasing shortage of rainfall (Fig. 9) and abundance of net radiation 42,61 . Therefore, the carbon sequestration ability of the savanna ecosystem will decrease (Fig. 6) under decreasing rainfall and increasing temperature (Fig. 9). We should, therefore, pay close attention to protecting similar savanna ecosystems and specific research should assess the influence of climate change on carbon and water exchanges.

Conclusions and Prospects
The biometric-based method (BM) and eddy covariance technique (EC) were used to determine carbon exchange over a savanna ecosystem in Southwest China. Our results and the discussion above lead to the following preliminary conclusions.
First, the carbon use efficiency (CUE = NPP/GPP) 62-64 was 0.60 (4.11/6.84), slightly higher than the mean CUE of all forests (0.53), which varies from 0.23 to 0.83 65 . Second, the largest daily net carbon release (22 January) and the maximum carbon sink (30 August) were 1.57 and −3.03 gC m −2 d −1 (equivalent to 1.51 and −2.92 μmol m −2 s −1 ), respectively. Third, the carbon exchange varies dramatically between the dry season (when the savanna is nearly carbon-neutral or a small carbon sink of 0.16 tC ha −1 yr −1 ) and the wet season (when the savanna is an  (Table 2), the mean GPP, R eco , and NEE were 10.13 ± 4.66, 8.78 ± 3.79, and −1.34 ± 1.58 tC ha −1 yr −1 , respectively. Consequently, the carbon sink strength of this savanna was close to the mean carbon sink ability of savannas globally. Note that the carbon sequestration capacity (i.e., the amount of CO 2 that the savanna ecosystem can take up) will decrease in the future under ongoing climate change (Fig. 6) as the climate here becomes hotter and drier than in past decades (Fig. 9). Therefore, it is critical that corresponding policies or management practices should protect similar savanna ecosystems that are subjected to decreasing rainfall amounts and rising temperatures. Further studies, which in turn help protect this area, should be conducted to understand the extent of the influence of climate change and the mechanisms responsible for this influence on energy, carbon, and water fluxes in the region.

Materials and Methods
Experimental site description. Site description. The geographical location of our research site (23°28′25.93″N, 102°10′38.76″E; 553 m a.s.l.) is in Yuanjiang Nature Reserve (YNR) in Yunnan province, Southwest China (Fig. 7a). The slope of the study plot terrain is ~15° and the soil is classified as torrid red earth (dry red soil).
Hot-dry winds dominate the climate due to the Foehn effect and the enclosed nature of the topography 41,66,67 , so the climate here is dry and hot with a high average annual temperature and low average annual rainfall, and there is considerable savanna vegetation spread throughout the area. The phenology shifts dramatically because  Table 2. Comparison of carbon exchange (NEE, GPP, R eco , gC m −2 yr −1 ) in savanna ecosystems worldwide. MAR is mean annual rainfall (mm). In the column labeled T air (°C), MDT is mean daily temperature, MAT is mean annual temperature, and the others are min.MMT/max.MMT (minimum mean monthly temperature to maximum mean monthly temperature). The sites are listed in descending order of NEE (a positive value means the ecosystem is a carbon source, and a negative value indicates a carbon sink that takes up CO 2 from the atmosphere). The values of NEE, GPP, and R eco listed in this table are shown as the mean value ± the standard deviation (sd) over the study period.
of the distinct changes between the dry and wet seasons (Fig. 7b,c). The period of leaf-fall is mainly between the end of the rainy season and the middle of the dry season, and most leaves are shed before the start of the driest month, even though the trees are dry-season deciduous species 68 (Fig. 7b). Vegetation growth is strongest in the middle of the wet season (August) (Fig. 7c). A permanent savanna ecological research plot (1 ha) associated with the Yuanjiang Savanna Ecosystem Research Station (YSERS) of the Xishuangbanna Tropical Botanical Garden of the Chinese Academy of Sciences, was established on a west-facing slope in YNR in 2011, as described in previous studies 41,67 , and the YSERS carried out an investigation of the vegetation in 2012. The savanna vegetation (canopy height of ~8 m) here consists mainly of small trees, shrubs, and herbs. In this community, the dominant trees are Lannea coromandelica, Polyalthia suberosa, Diospyros yunnanensis and similar species. The dominant shrubs are Vitex negundo f. laxipaniculata, Campylotropis delavayi, Woodfordia fruticosa, Euphorbia royleana, Jasminum nudiflorum, Tarenna depauperata, etc. The dominant herbaceous species are Heteropogon contortus and Bothriochloa pertusa 41,67-69 . As an adaptation to the region's high temperature and low rainfall, the leaves of these species are relatively small, with thick cuticle and smooth or waxy leaf surfaces.
Long-term meteorological conditions and regional climate patterns. Thirty-six years (1980-2015) of meteorological records (Fig. 8) from a weather station located ~20 km northwest of the study site give the monthly variations in relative humidity (RH), water vapor pressure (e), wind speed (WS), rainfall, minimum air temperature (T min ), mean air temperature (T mean ), and maximum air temperature (T max ). Overall, the results show that the wet season RH, e, rainfall, T min , T mean , and T max are larger than the dry season values, but WS is lower in the wet season.
According to the long-term results (Fig. 8b), the mean annual temperature (MAT) is 24.0 ± 0.5 °C, and the monthly average temperatures of the coldest month (January) and the hottest month (June) are 16.9 ± 2.2 °C and 29.2 ± 2.4 °C, respectively. The climate is strongly seasonal; in the wet season (May-October), the climate is dominated by the tropical southern monsoon from the Indian Ocean, which delivers most of the annual rainfall (786.6 ± 153.2 mm). The ratio of wet season rainfall to annual rainfall can reach 81.0%, whereas in the dry season (November-April), the total rainfall is less than 150 mm. There are more than 100 days with temperatures above 35 °C in the YSERS records for 2012-2013 69 . The yearly total number of sunshine hours is 2261.7 61 , the annual average pan evaporation is 2750 mm 41 , and the aridity index (AI) is ~0. 29. These values indicate that the study area belongs to the semi-arid class according to the definition of semi-arid regions (AI = 0.2-0.5) 70 .
Climate change trends in our study area. The results of 36 years (1980-2015) of observations on temporal trends in rainfall, temperature, RH, and VPD ( Fig. 9) showed that the observed declining trend in rainfall (p = 0.6873) and the observed increasing trend in temperature (p = 0.0596) were not statistically significant (Fig. 9a,b). However, both the RH (p = 0.0009) and VPD (p = 0.0017) increased significantly (Fig. 9c,d). Therefore, the climate here is becoming drier and hotter than it previously was, and the opposite trends seen in water supply and heat are becoming exacerbated 42 . Biometric and eddy covariance method for estimate of carbon exchange. Biometric-based NEP estimation. The biometric method is a conventional way of estimating NEP all over the world 49,51,[71][72][73] , and the expression [74][75][76] for NEP as estimated by the biometric method is where NEP is generally defined as the net ecosystem production that represents the balance between GPP and ecosystem respiration (R eco ); NPP is net primary production, with R h the heterotrophic respiration of the ecosystem; ΔB is the biomass increment; L p is above-ground litterfall production; and ΔB t , ΔB s , and ΔB h are the biomass increments of trees, shrubs, and herbs, respectively. To estimate the biomass production, we inventoried 1 ha of vegetation within the footprint of the eddy flux tower in November 2013. All trees with diameter at breast height (DBH) > 2 cm were identified, tagged, measured (in terms of their height and DBH), and mapped. Standard allometric equations for karst vegetation 50 were used to calculate tree biomass from DBH and height, because site-specific allometric equations were not available. Carbon density was derived from the biomass by multiplying by a factor of 0.5 77,78 . In November of 2014 and 2015, we re-measured tree DBH and the heights of trees tagged and measured in 2013 to estimate components of the biomass carbon budget including DBH increment, tree height, recruitment, growth, mortality, and coarse woody debris. Litterfall was captured by 20 litterfall traps (1 m × 1 m) that were randomly located in the 1 ha permanent ecological research plot. The litterfall was collected on the last day of each month and then sorted into leaves, branches, flowers, and fruits. Each component was dried to a constant weight at 65 °C, then weighed and recorded. For the calculation of ΔB s and ΔB h , the harvest method was used to estimate the above-ground and below-ground biomass of shrubs (2 m × 2 m with 3 sets and 4 repeats) and herbs (1 m × 1 m with 5 sets and 3 repeats) near the 1 ha permanent plot.
Root removal by trenching was used to measure R h 76 within the footprint of the flux tower. The volume of root trenching was 100 cm × 100 cm × 40 cm and the volume was wrapped with wire mesh (0.149 mm × 0.149 mm) to prevent the growth of new roots. Two treatments (CK and root trenching) were applied with six replicates in August 2014. Open-top static chambers (60 cm × 32 cm × 30 cm) together with a gas chromatograph (7890D GC, Agilent Co. Produced, USA) were used to measure R h . Sampling was performed (~4 months after root trenching) twice a month from Nov 2014 to Dec 2015 (usually the 15 th and the last day of each month) during 09:00 and 11:00 each day.
Eddy covariance-based NEE estimations. Eddy covariance and meteorological measurement system. Eddy covariance provides a direct and continuous measure of matter and energy fluxes between an ecosystem and the atmosphere 79,80 , and has been applied across the globe to different cover types including forests, farmlands, grasslands, wetlands, tundras, deserts, and aquatic ecosystems to measure energy, carbon, and water exchanges 81 . The EC and meteorological instruments were mounted and oriented in the prevailing wind direction at an angle of 135° from north at a height of 13.9 m on a flux tower that was established near the 1 ha permanent plot in April 2013. The eddy covariance system characteristics and parameters used in this paper are as follows. 1) The EC system consisted of a triaxial sonic anemometer (CSAT3, Campbell Scientific Inc., USA) and a high-frequency open-path CO 2 /H 2 O infrared gas analyzer (Li-7500, Li-Cor Inc., USA) installed at a height of 13.9 m. 2) Measurements of wind speed (A100R, Denbighshire, UK) and direction (W200P, Denbighshire, UK) were made at two heights, and a photosynthetically active radiation (LQS70-10, APOGEE, USA) profile measurement system was also deployed. The sampling frequencies of flux data and meteorological data were 10 Hz and 0.5 Hz, respectively. Control systems were used for the simultaneous acquisition of flux data (model CR1000, Campbell Scientific Inc., Logan, UT, USA) and meteorological data (model CR5000, Campbell Scientific Inc.). All data were collected continuously beginning in May 2013. NEE Calculation. The NEE between the forest ecosystem and atmosphere is the sum of the turbulent eddy flux and the storage flux 79,82,83 as equation (2): where F c is the turbulent eddy flux transported between the EC measurement plane above the forest and the atmosphere, F s indicates the storage flux under the plane of the eddy covariance system (13.9 m in this study), ρ is air density, w is the vertical wind velocity, c represents CO 2 concentration measured by an infrared gas analyzer, the primes denote fluctuations in the target scalar (CO 2 concentration in this case) from the average, and the overbar signifies a time average (30 min in this case). Δc is the variation in CO 2 concentration over a 30 min period at height z r , Δt is the time interval (1800 s in this case) and z r is the height of the plane of the eddy covariance system above the ground (13.9 m in this case). Generally, negative NEE values indicate that the ecosystem fixes CO 2 from the atmosphere by photosynthesis and acts as a carbon sink. Thus, NEE is generally equal to −NEP.
Data processing and carbon flux calculation. Quality assessment and control (QA/QC) are necessary to ensure the reliable processing of flux data before calculating energy, carbon, and water fluxes to account for environmental and meteorological limitations (topography, rain, advection, and low turbulence issues). ChinaFLUX has developed a series of standard methodologies to assess the EC system and to control the quality of flux data. For details of data QA/QC and post-processing procedures used in the present study, see ref. 43 introduce the data processing flow. 1) Three-dimensional coordinate rotation was applied to remove the effects on airflow of instrument tilt or irregularities in the terrain [84][85][86] ; 2) In WPL calibration, flux data were corrected for air density variations arising from the transfer of heat and water vapor 87 ; 3) data recorded in rainy periods were discarded 43 ; 4) storage flux (F s ) was calculated 79,82,88 ; 5) outliers were identified and eliminated 53 , and absolute NEE values > 50 μmol m −2 s −1 (i.e., NEE values larger than 50 or less than −50 μmol m −2 s −1 ) were rejected 44 ; 6) negative nighttime data were rejected; 7) data with friction velocities (u*) < 0.2 were filtered 52,55 ; and 8) gap filling and partitioning were applied to the flux data [52][53][54] using an online procedure that is recommended by FLUXNET and used as standard by EUROFLUX and maintained by the Max Planck Institute (http://www.bgc-jena.mpg. de/~MDIwork/eddyproc/index.php).