Controls of Evapotranspiration and CO2 Fluxes from Scots Pine by Surface Conductance and Abiotic Factors

Evapotranspiration (E) and CO2 flux (Fc) in the growing season of an unusual dry year were measured continuously over a Scots pine forest in eastern Finland, by eddy covariance techniques. The aims were to gain an understanding of their biological and environmental control processes. As a result, there were obvious diurnal and seasonal changes in E, Fc, surface conductance (gc), and decoupling coefficient (Ω), showing similar trends to those in radiation (PAR) and vapour pressure deficit (δ). The maximum mean daily values (24-h average) for E, Fc, gc, and Ω were 1.78 mmol m−2 s−1, −11.18 µmol m−2 s−1, 6.27 mm s−1, and 0.31, respectively, with seasonal averages of 0.71 mmol m−2 s−1, −4.61 µmol m−2 s−1, 3.3 mm s−1, and 0.16. E and Fc were controlled by combined biological and environmental variables. There was curvilinear dependence of E on gc and Fc on gc. Among the environmental variables, PAR was the most important factor having a positive linear relationship to E and curvilinear relationship to Fc, while vapour pressure deficit was the most important environmental factor affecting gc. Water use efficiency was slightly higher in the dry season, with mean monthly values ranging from 6.67 to 7.48 μmol CO2 (mmol H2O)−1 and a seasonal average of 7.06 μmol CO2 (μmol H2O)−1. Low Ω and its close positive relationship with gc indicate that evapotranspiration was sensitive to surface conductance. Mid summer drought reduced surface conductance and decoupling coefficient, suggesting a more biotic control of evapotranspiration and a physiological acclimation to dry air. Surface conductance remained low and constant under dry condition, supporting that a constant value of surface constant can be used for modelling transpiration under drought condition.


Introduction
The climate in the boreal zone has warmed already in the last century and is predicted to warm significantly further in this century [1]. Climate variability has the potential to affect carbon exchange, evapotranspiration, and other ecophysiological processes in forest ecosystems. Moreover, the extent of the boreal forests, their role in contemporary northern hemisphere climatology and the global carbon cycle, and their sensitivity to climate change are sufficient reasons for better understanding of boreal ecosystematmosphere interactions [2].
To obtain an understanding of environmental and biological controls of evapotranspiration and CO 2 flux has been a central focus of climate change research for decades [3]. Evapotranspiration is an important process that is controlled by the interaction of a number of environmental factors (e.g., solar radiation, air temperature, vapour pressure deficit, and soil water content) and biological processes (e.g., leaf emergence, leaf development, and stomatal conductance) [4][5][6][7][8][9]. It has been found that biological control of evapotranspiration in a forest ecosystem is reflected in the form of changes in surface conductance [10,11]. The transpiration from coniferous forests in the boreal region is largely controlled by canopy conductance [12,13], because boreal forest canopies are aerodynamically rough and well ventilated, so that the effect of aerodynamic conductance is minimized. CO 2 flux is simultaneously regulated by canopy conductance, since both water vapour and CO 2 pass through the stomata. Hence, stomatal conductance is not only the key to the assessment of transpiration and water balance, but also important for estimating carbon flux. The stomata are highly responsive to environmental variables such as high vapour pressure deficits [14], drying soils, and low light, so that all of these may act to regulate stomatal conductance.
Scots pine (Pinus sylvestris L.) is a major tree species in boreal areas, and its response to climate change and the roles it plays in carbon and water cycling are of great interest to ecologists. Understanding of the control of stomatal conductance over transpiration and photosynthesis at ecosystem level for Scots pine is limited relative to a large body of knowledge at leaf level [8]. As the spatial and temporal scale of our measurements and experiments has increased, more concerns have been focused on the understanding of how canopy conductance is involved in regulating carbon and water cycles on the ecosystem scale and over long periods of time [14][15][16]. Many previous reports have pointed to seasonal and interannual variations in the fluxes of carbon and water in forest ecosystems [3,17,18]. A better understanding of the biotic and abiotic control processes of evapotranspiration is necessary for the assessment of local, regional, and global water and carbon budgets as climate change progresses. Studies to quantify the relationship between mass flux and surface conductance at the forest ecosystem level are still needed. Analysis of dry-canopy transpiration measured with the eddy covariance method will provide information on stomatal behaviour and influence on water vapour and CO 2 fluxes.
Our primary objectives were (1) to examine diurnal and seasonal changes in CO 2 flux and evapotranspiration, (2) to provide information on the magnitude and temporal variation of parameters representing the bulk canopy characteristics, and (3) to understand the biotic and abiotic control processes of transpiration of Scots pine ecosystem.

Site Description
The research was conducted in a 50-year-old pure stand of Scots pine (Pinus sylvestris L.) at Huhus (62u529N, 30u499E, 145 m a.s.l.), eastern Finland. The stand density was 1175 trees ha 21 (ranging from 7.2 cm to 29.5 cm in diameter at breast height), with a mean height of 11.8 m above the ground and a mean diameter at breast height of 11.2 cm. The leaf area index was  [19]. The climate is characterized by a long, cold winter. The mean monthly temperature is lowest in January, 210.4uC, and highest in July, 15.8uC. The average annual precipitation at the site  is 724 mm, of which 38% falls as snow. The mean soil moisture (10 cm depth) was lower than 30% that is regarded as roughly representing drought year [30]. The soil moisture (25 cm depth) was lower than 15% in the mid summer (June 21-August 8) of 2003, thus considering as an unusual drought summer. The ground was covered by small patches of litter (30% of the area) or lichen (65% of the area). The understorey is principally mosses (Dicranum spp, Pleurozium schreberi) and dwarf shrubs (Vaccinium vitisidaea, Calluna vulgaris), so the site represents the Calluna type, on a sandy soil with a low nitrogen supply. The field studies did not involve endangered or protected species and no specific permits were required for the described field studies. The site is flat, and there is a homogeneous underlying surface. The terrain is relatively level, extending at least 2 km around the tower used for the eddy covariance (EC) measurements. It is assumed that zero plane displacement ranges from 7 to 11 m and that the roughness length is between 1 and 2 m. The 80% contribution of the measured flux comes from within 460 m of the upwind area under unstable conditions. This contribution may be from within 1890 m of the upwind area under neutral conditions [20].

Measurements
Half-hour eddy fluxes including CO 2 flux (Fc), latent heat flux (LE), and sensible heat flux (H), were measured continuously at the top of a 34 m mast, about 20 m above the canopy. Wind velocity and virtual temperature were determined with a Solent 3D ultrasonic anemometer (R2 Gill Instruments, Lymington, UK), and CO 2 and water vapour concentration fluctuations with a closed-path dual CO 2 /H 2 O analyzer (IRGA; model LI-6262, LiCor, Lincoln, NE, USA). The air was ducted down from a point close to the anemometer to the Li-6262 by means of a sampling tube of length 42 m and radius 3 mm. The air flow rate in the tube was maintained at a constant rate of 6 litres min 21 by a mass flow controller (Tylan FC2900B, Tylan General, Swindon, UK) on the sample line. The analogue signals from the Li-6262 were passed to the Solent 3D ultrasonic anemometer, which used an onboard analogue-to-digital converter to digitise the non-linearized signals at 10 Hz. The digitised signals from the Li-6262, combined with the wind speed components (u, v, and w) and the speed of sound, from which air temperature may be derived at 21 Hz, were sent to a computer. The data were collected and processed in real time to provide near-continuous measurements. The EdiSol system was used to calculate the raw data on-line over 30-min interval [21]. Further details of the instrumental installation and calibration are given by Kellomä ki and Wang [22].
Simultaneously with the flux measurements, environmental and meteorological variables were measured using a Vaisala weather station (MILOS 500, Vaisala Oy, Helsinki, Finland) as placed at a distance of about 20 m from the eddy covariance mast. Temperature, humidity probes (HMP45D, Vaisala Oy, Helsinki, Finland) and anemometers (WAA15A and WAV15A, Vaisala Oy, Helsinki, Finland) were mounted at 4, 9, 12, and 18 m above the ground along the weather mast to record the weather profile. Canopy temperature was measured with an infrared sensor (IR 4000.4GL, Everest Interscience, Inc. Tucson, USA) and net radiation above the canopy with a combination of an albedometer CM7B (Kipp & Zonen, Delft, Holland) and a CG2 pyrgeometer (Kipp & Zonen, Delft, Holland). Photosynthetically active radiation (PAR) above the canopy was measured with a quantum sensor (LI-190SA), and global radiation at a height of 20 m with a pyranometer (model CM6B/2, Kipp & Zonen, Delft, Holland). Precipitation above and below the forest canopy was measured using 8 rain gauges (RG13, Vaisala Oy, Helsinki, Finland) and bole temperature with copper-constantan thermocouple probes inserted 1 cm into the trunk of each of three trees at heights of 15 cm, 285 cm, and 305 cm above the ground.
Soil heat flux (G) was determined with 4 soil heat flux plates (Radiation Energy Balance System, Seattle, WA, USA) buried 5 cm below the surface in a variety of microenvironments (ranging from mostly sunlit to mostly shaded). The soil volumetric water content (W s ) was monitored at depth of 25 cm below the mineral soil surface with water content reflectometers (CS615, Campbell Scientific, Shepshed, Leics., UK). All of these sensors were sampled at 10 s intervals and the data averaged over 30 min periods using a data logger (21X, Campbell Scientific, Logan, UT, USA).

Post-processing of data
Sonic anemometer measurements were removed when a spiking rate greater than 5 spikes per 30 min interval was observed [23]. These spikes were suspected to occur during periods of heavy rainfall. Flux measurements were also removed when data did not fall within the specified realistic limits and when the nonstationarity ratio was greater than 3.5 [24]. These quality controls resulted in the removal of 1% of the eddy flux measurements. Details of the post-processing of data have been given previously [18]. After data processing, energy balance closure was 70% on the basis of the slope of linear regression of half-hourly heat flux (sum of sensible heat and latent heat flux) against available energy (R 2 = 0.68). For post-processed half-hour data, see files of Data S1 and Data S2.

Calculation of bulk parameters
Ecosystem surface conductance to latent heat transfer was calculated by inverting the Penman-Monteith equation [25] as where g c is the surface conductance (m s 21 ), D (kPa K 21 ) the slope of saturated specific humidity versus air temperature (T), R a the available radiation energy (W m 22 ) as calculated below, l the latent heat of the vaporisation of water (J kg 21 ), E the measured evapotranspiration (kg m 22 s 21 ), d the vapour pressure deficit (kPa), r a the density of dry air (kg m 23 ), c p the specific heat of air at constant pressure (J kg 21 K 21 ), lE the latent heat flux (W m 22 ), and c the psychrometric constant (0.0665 kPa K 21 ). g a is the aerodynamic conductance (m s 21 ) calculated as [26] where u is the wind speed above the canopy (m s21) as measured by the sonic anemometer and u* is the friction velocity (m s21). The decoupling coefficient (V) was subsequently calculated to describe the sensitivity of evapotranspiration to a change in surface conductance [12]. The values of V ranges from 0 to 1 with the control of evapotranspiration by surface conductance increasing as V approaches 0 [12,27].
The analyses were conducted for days with a dry canopy (recorded precipitation both above and below the canopy was 0 and PAR more than 200 mmol m 22 s 21 ). By excluding data recorded up to 2 days after rain, we minimized inclusion of the evaporation of water that had been intercepted by the canopy [28]. The decoupling coefficient (V) was calculated according to Jarvis and McNaughton [12], The available energy, R a, was calculated as where R n is the net radiation energy as measured by a Vaisala weather station, S t the total energy storage in the above-ground air column, and G the soil heat flux. The total rate of energy storage (S t ) in a column extending from the ground surface to the EC measurement height z was calculated as where the subscripts b, n, H, lE and p denote the rates of change in heat content of the stems and leaves, sensible heat content in the air column, latent heat content in the column, and energy consumed in photosynthesis, respectively. Heat storage in the boles (S b ) was calculated as [29] S b~s where s d is the stand density (0.12 stem m 22 ground area), r b the average bole density (400 kg m 23 ), Dt the sampling period (30 min) , V bi the estimated bole volume, and T bi the bole temperature (K). V bi was derived from 49 sample trees and T bi measured from 3 sample trees. C pb is the average bole specific heat (J kg 21 K 21 ), estimated as where Wb is the average water content on a dry mass basis (0.968 kg kg21), CD the specific heat of dry wood (1150 J kg21 K21), and Cw the specific heat of water (4190 J kg21 K21). Energy storage in the needles was calculated as where W n is the gravimetric water content of the needles on a wet mass basis (55%), r n the average needle density (0.52 kg m 22 land area on a dry mass basis [30]), C n the specific heat of the needles, and DT n /Dt the needle temperature change per half hour. Needle temperature was estimated using the air temperature measured at a height of 12 m within the canopy and assuming that changes in needle temperature would be reflected in the air temperature. Needle specific heat was obtained from where Cc is the specific heat of cellulose (for glucose 1260 J kg21 K21 [31]). Storage fluxes of sensible heat H and latent heat lE within the air space beneath the height of turbulent flux measurement were measured from profiles of temperature and relative humidity. The storage terms were then calculated as where c is the psychrometric constant (0.0665 kPa K 21 ), dT/dt and dq/dt are the changes in air temperature (T) and water vapour density (q ) over the 30-min time period, r a is air density (1.229 kg m 23 ), and c p is the specific heat of air at constant pressure (1012 J kg 21 K 21 ). The energy consumed in the process of photosynthesis, S P , was calculated from ecosystem photosynthesis GEP using the photosynthetic energy conversion factor C (0.469 J mmol 21 ) [32]) as  where GEP is the ecosystem photosynthesis in mmol m 22 s 21 and was calculated by the method described by Zha et al. [33]. Soil heat flux (G, W m 22 ) was a sum of measured values (G s , W m 22 ) by the soil heat plates and heat storage of soil above the soil heat plates as where Z (m) is the depth of the layer from plate to surface, C s the soil's heat capacity estimated by soil water content at Z, T s the soil temperature (average of soil temperature at the depth of 3.5 cm and 2.5 cm below soil surface).
Three typical clear days representing respective spring, summer and autumn were selected to analyze diurnal changes in analyzed variables. Water use efficiency (WUE) was estimated in terms of the reciprocal of the transpiration ratio [34], which is defined as the amount of water transpired by ecosystem divided by the net ecosystem CO 2 exchange (F c ). For symbols used above, refer to Table 1.

Statistical analysis
The significance of the seasonal changes in measured and calculated parameters was analysed by univariate ANOVA, and the degrees to which evapotranspiration, CO 2 flux, and surface conductance were controlled by environmental variables (PAR, d, T, W s , and u) were analysed by stepwise linear regression. The growing season was divided into months for significance test of the seasonal changes. All the statistical analyses were performed on the basis of 24-hour averages using the SPSS 12.0 program for Windows (SPSS Inc., Illinois, USA).

Seasonal changes in energy components
The upper panel of figure 1 showed the daily mean rates of components of total above-ground energy storage (S t ) over the growing season, including heat content of the stems (S b ), leaves (S n ), sensible and latent heat in the air column (S H +S lE ), and energy consumed by photosynthesis (S p ). Among these components, energy use by photosynthesis accounted for the largest portion, being 95% of S t . There was seasonal change in S p , R n , and R a (Fig. 1). The other energy components fluctuated around zero. The energy closure on average was 70% over the growing season. Addition of soil heat flux only to available energy increased energy closure by 3%, and addition of both soil heat flux and S t increased by 4%. Therefore energy storage only accounted for little amount of energy over daily time period and can be neglected in energy balance analysis.

Diurnal and seasonal changes
There were obvious diurnal changes in evapotranspiration (E), CO 2 flux (F c , minus represents CO 2 into the canopy), ecosystem surface conductance (g c ), and decoupling coefficient (V), showing a diurnal and seasonal pattern that was marginally more similar to those of PAR than to vapour pressure deficit (d) (Fig. 2). The values of all the variables were low at night (closer to 0 before 3 h and after 18 h) and high during the mid-day period from 8 to 16 h. The magnitudes of E, F c , and g c were lower on September 10 than those on the other two days as a whole, showing a correspondence with PAR. V values were lower on May 26 than on the other two days.
Mean daily values (24-h average) from May to October indicated significant seasonal changes in E, F c , g c , and V (Fig. 3, P,0.0001), with seasonal trends similar to those in the environmental variables PAR, d, and air temperature (T) (Fig. 4). Overall, these physiological variables had high values from June to September, with the exception of low values during drought period from late June to July, when there was little rainfall and the soil water content (W s ) was low (Fig. 4). The g c remained relatively constant during drought period from late June to July (Fig. 3). The mean daily values for E, F c , g c , and V ranged from 0.  namic conductance (g a ) remained relatively constant, without any significant seasonal changes, despite short-term fluctuations (Fig. 3, P = 0.088).

Control of evapotranspiration
The relationship between ecosystem evapotranspiration (E) and biological variables, represented by surface conductance (g c ) during periods when the leaves were dry (PAR .200 mmol m 22 s 21 and total hourly rainfall both above the canopy and under the canopy were 0), and corresponding environmental variables are shown in Fig. 5. E increased curvilinearly with increasing g c and vapour pressure deficit (d) (Fig. 5), with g c and d explaining 33% and 26% of the variation in E, respectively. It was also linearly related to radiation (PAR) and air temperature (T). PAR, d, and T individually explain 42%, 26%, and 19% of the variation in E, respectively. No meaningful relationship between E and soil water content (W s ) or wind speed (u) could be found.
Stepwise linear regression of evapotranspiration against the environmental variables (PAR, d, T, W s , and u) shows that E was controlled by PAR, T, d, and u together (R 2 = 0.48, n = 1410). Net radiation had more effect on E than did the other environmental variables.
When the biological variable g c and environmental factors were considered together as independent variables in the stepwise linear regression while E was taken as a dependent variable, 80% of the variation in E was explained by PAR, g c , d, and T together (R 2 = 0.80, P,0.0005).
In order to examine the daily variations in the sensitivity of E to the variables (PAR, g c , d, and T) and variations in their relationship, we plot the slopes, intercepts and R-square of their regression lines over the growing season. As a result, the values of the slopes for the linear regression between E and g c , T, d, and PAR were higher in the morning, from 8 to 11 h, than in the  afternoon, after 12 h (Fig. 6). Similarly, R-square exhibited higher values in the morning than in the afternoon, from 8 to 11 h. The high values for the intercept occurred after 11 h.

Control of carbon flux
F c (minus represents CO 2 flux into the canopy) increased with increasing surface conductance (g c ) (Fig. 7), showing a curvilinear relationship (R 2 = 0.28). Among the environmental factors (PAR, d, T, W s , and u), PAR was the most important factor controlling F c , which increased curvilinearly with increasing PAR (R 2 = 0.45).
Stepwise linear regression of F c against the environmental variables showed that PAR, d, u, and W s together explain 43% of the variation in F c (R 2 = 0.41, n = 1413). When the biological variable g c and environmental variables are taken together as independent variables in stepwise linear regression, 54% of the variation in F c is explained by PAR, g c , d, u, and W s together. Temperature had the weakest correlation with F c .

Control of surface conductance
Surface conductance (g c ) is controlled by vapour pressure deficit (d), radiation (PAR), air temperature (T), and wind speed together (stepwise linear regression, R 2 = 0.17, n = 809). Among the environmental factors, d had the closest correlation with g c , a better fit being achieved with the exponential equation (Fig. 8, R 2 = 0.27). The g c decreased with increasing d in the regression analysis. The slopes of the regression curves are steep at high PAR, i.e. 0.89 at PAR .900 mmol m 22 s 21 and 0.73 at PAR between 500-900 mmol m 22 s 21 . Surface conductance also affected the decoupling coefficient V, which increased in a non-linear manner with increasing g c ( Fig. 9; P,0.0001).

Water use efficiency
WUE was calculated on the days when leaves were dry (PAR .200 mmol m 22 s 21 and total hourly rainfall both above canopy and under the canopy were 0). The mean monthly values of WUE ranged from 6.67 mmol CO 2 (mmol H 2 O) 21 in May to 7.48 in October, with a mean value of 7.06 (Fig. 10).

Diurnal and seasonal changes of physiological parameters
The trend in diurnal and seasonal changes in the physiological parameters (evapotranspiration E, ecosystem CO 2 flux F c , surface conductance g c , and decoupling coefficient V) was consistent with that in radiation. It was noted that the little rainfall and low soil water content from late June to early July led to lower values of E, g c , and V (Fig. 3, 4). One possible explanation would be that the stomatal pores closed under conditions of a soil water deficiency in order to prevent excessive water loss, thereby leading to a reduction in transpiration. Lower and relatively constant g c further support that stomata was closed under severe drought condition from late June to July. This supports the conclusion by Duursma [35] that a model with constant plant conductance and minimum leaf water potential can accurately predict the decline in daily maximum transpiration rate during drought. F c remained high in the main growing season from June to August. This could be explained by high LAI [18]. Another explanation might be that stomata closed to some degree in response to greatly vapour pressure deficit (d), but photosynthesis proceeded at a slightly  higher WUE and without any metabolic limitation.

Control of evapotranspiration
The present analysis of environmental control over evapotranspiration in a Scots pine ecosystem showed that radiation, water vapour pressure deficit and temperature are the main environmental factors involved (R 2 = 0.48), with radiation the most important. However, since radiation alone explained 42% of the variation in evapotranspiration, it is suggested that ecosystem evapotranspiration cannot be adequately predicted by this alone, despite its importance.
Previous studies have shown that evapotranspiration is controlled by interaction between a number of environmental and biological factors [15,16,36,37]. Biological control can generally be represented by surface conductance [8,10,15], which is thus taken here as a representative of biological variables. The basic explanation for this is the following. The low values for the decoupling coefficient (V) (Fig. 3) imply that the canopy was well coupled to the atmosphere, so that it can be assumed that the calculation of surface conductance does not violate the assumption of the Penmen-Monteith equation in which canopy is treated as a single layer [38]. Although surface conductance calculated from the Penmen-Monteith equation does not explicitly represent the biological parameter known as stomatal conductance, because theoretical and experimental investigations indicate that surface conductance is often related to the weighted integration from individual leaves [39,40]. Surface conductance can therefore reflect biological control of ecosystem transpiration.
The finding that surface conductance and environmental variables including radiation, vapour pressure deficit and air temperature together explained 80% of the variation of evapotranspiration, which is much more than that achieved by the environmental factors either together (48%) or separately (19%-42%), suggests that ecosystem evapotranspiration is codetermined both biologically and environmentally. Surface conductance alone explained 33% of the variation of evapotranspiration in the present case. Results in the literature suggest that transpiration flux from coniferous forests in the boreal region is largely controlled by canopy conductance [12]. This is because boreal forest canopies are aerodynamically rough and well ventilated, which minimizes the effect of aerodynamic conductance. The relatively constant figures obtained here for aerodynamic conductance (g a , Fig. 3) provide further support for this.
The decoupling coefficient V has also been widely used to determine the relative importance of surface conductance and net radiation for changes in evapotranspiration [41]. The low values of V recorded here, which are similar to those in temperate coniferous forests [23], indicate that evapotranspiration is highly sensitive to surface conductance. On a daily basis, V was highest in the morning and lowest in the evening (Fig. 1), indicating tighter control over water loss by plants as the day progressed. The close relationship between V and g c further supported the importance of biological control with regard to changes in evapotranspiration ( Fig. 9; R 2 = 0.51, P,0.0001).
High morning values for the slope and R-square of the linear regression of evapotranspiration against surface conductance and environmental variables at a given time indicate that evapotranspiration is more sensitive to biological and environmental factors in the morning than in the afternoon (Fig. 6). The explanation could be that the wide open stomatal pores lead to a greater water flux into the atmosphere in the morning, and thus to greater transpiration or greater biological control over evapotranspiration than in the mid-day period, when the stomata are less open.

Control of CO 2 flux
In our work, surface conductance and environmental variables (radiation, vapour pressure deficit, wind speed, and soil water content) together explained 54% of the variation in CO 2 flux F c in a linear manner as compared with 41% in the case of the four environmental variables together and 45% for PAR alone, indicating a dominant control of canopy conductance over F c . The main reason for this is that both water vapour and CO 2 pass through the stomata. Hence, the stomata of the canopy are not only the key to the assessment of transpiration and water balance, but are also important for the estimation of carbon flux.
We did not find any meaningful relationship between F c and any other environmental factors except for PAR, which was linearly related to CO 2 flux (Fig. 7). A close relationship between PAR and F c has been found previously [33]. The low value for Rsquare in the regression between PAR and CO 2 flux is because that the regression applies only to the daytime, when PAR is more than 200 mmol m 22 s 21 from May to October. We thought that the control of CO 2 flux by radiation would decrease when radiation remained higher than 200 mmol m 22 s 21 and was no longer a limited factor for photosynthesis in the Scots pine ecosystem.

Control of surface conductance
The finding that surface conductance decreased with increasing vapour pressure deficit agrees with previous reports [14,42] in which the surface and canopy conductance decline exponentially with increasing vapour pressure deficit in a variety of leaf, wholetree and stand-level studies. The values for the slopes in Fig. 8 imply that the sensitivity of surface conductance to changes in vapour pressure deficit is high under conditions of more pronounced radiation.
Vapour pressure deficit was the most important environmental factor affecting the surface conductance, explaining much more of its variation than temperature, radiation, or soil water potential, as also reported by Gunderson et al. [43]. Although the mechanism of the stomatal response to vapour pressure deficit is unknown, some studies suggest that this response occurs as a feedback related to transpiration and water loss from the leaf rather than as a direct response to humidity [41,45]. We noted that there was no relationship between surface conductance and any other environmental factors except for vapour pressure deficit, the possible reason being that surface conductance is mostly regulated by the plant itself when environmental conditions are favourable. Whatever the explanation is, plants regulate their conductance in order to optimize photosynthesis while minimizing water loss through their leaves [44,13,46]. WUE was high in June and July when rainfall and soil water content were low (Fig. 10), indicating the role of the stomata in minimizing water loss and maximizing CO 2 flux in dry season.

Conclusions
Obvious seasonal changes in evapotranspiration and CO 2 flux were partially driven by radiation, vapour pressure deficit, and temperature. Radiation had more influence on both evapotranspiration and CO 2 flux than did the other environmental factors. Soil water deficiency in mid summer led to a lower and constant surface conductance during drought and a lower decoupling coefficient, thereby reducing transpiration. Vapour pressure deficit was the most important factor affecting surface conductance which was more sensitive to vapour pressure deficit under conditions of high radiation. Higher WUE in dry condition indicated an acclimation of plant to water deficiency.