کویر سبز

Introduction

Transpiration is an important component of the rainforest water balance; it is also closely linked to net photosynthesis (via stomatal conductance), and hence has a major influence on forest growth (Lloyd et al., 1995). In Amazonia, transpiration was found to account for around 40% of precipitation (Shuttleworth, 1988), whereas in Australian coastal and lower montane rainforests McJannet et al. (2007c) report lower figures of 20–25%, indicating that transpiration is still a highly significant fraction of precipitation. In both Amazonian and Australian rainforests, transpiration varies markedly within and between days, largely due to changing weather conditions, primarily atmospheric humidity and radiation ([Shuttleworth, 1988] and [McJannet et al., 2007d]). Canopy structure and physiology can also affect transpiration (Roberts et al., 1990 J.M. Roberts, O.M.R. Cabral and L.F. De Aguiar, Stomatal and boundary layer conductances in an Amazonian terra firme rain forest, Journal of Applied Ecology 27 (1990), pp. 336–353. Full Text via CrossRef | View Record in Scopus | Cited By in Scopus (67)[Roberts et al., 1990], [Roberts et al., 1993] and [McJannet et al., 2007d]), predominantly via the amount of foliage (leaf area index) and the stomatal conductance of the leaves. Forest location can also affect transpiration, for example, McJannet et al. (2007d) found that although annual transpiration of coastal and lower montane rainforests were quite similar (590 mm), upper montane cloud forest transpiration was much lower at 353 mm.

 

There is evidence from the work of (Gilmour, 1975) and (Hutley et al., 1997) and McJannet et al. (2007d) that Australian rainforests may transpire at a lower rate than other rainforests ([Schellekens et al., 2000] and [McJannet et al., 2007d]). However, it is unclear whether this is due to different weather conditions or canopy physiology in the Australian forests. The Penman–Monteith equation (Monteith, 1965) allows the interactive effects of weather and canopy physiology (via canopy conductance) on transpiration to be calculated. When transpiration data are available the Penman–Monteith equation can be used in its ‘inverse’ form to calculate canopy conductance and hence decouple weather and physiological controls on transpiration. This paper uses the extensive rainforest transpiration data recorded at four different locations by [McJannet et al., 2007a] and [McJannet et al., 2007d] in the inverse form of the Penman–Monteith equation to derive hourly values of canopy conductance for each location. These canopy conductances are then used to derive a model of canopy conductance following the approach described by Jarvis (1976). Model predictions of 30 min and daily transpiration are compared with those derived using alternative models that relate transpiration directly to the controlling weather variables.

Methods

Field site characteristics

Four field sites were selected to represent the different forest types, geologies and altitudes found in the north Queensland region. The locations of the field sites are shown in McJannet et al. (2007c) and characteristics of each site are detailed in Table 1. Brief site descriptions are given below.

The lowest altitude site (30 m ASL) is called Oliver Creek (OC), this site is a pristine lowland rainforest which is located in the Daintree National Park. The second site, Upper Barron (UB), is a lower montane rainforest located on the Atherton Tablelands (1050 m ASL). The third site, Mt Lewis (ML1), is also a lower montane rainforest, which is at an altitude of 1100 m ASL. The fourth site is located in upper montane cloud forest near the summit of Bellenden Ker (BK) (1560 m ASL), which is the second highest mountain in Queensland. Details of the different rainforest stand composition are given by (McJannet et al., 2007a) and (McJannet et al., 2007b). The Australian Wet Tropics region experiences distinct wet and dry seasons. The dry season covers the months from June to November, while the wet season generally occurs from December to May. The predominant wind direction in this region is from the south-east.

Transpiration was measured using the heat pulse method to quantify sap flow in sample trees at each site. Strong relationships between the size (diameter at breast height) of trees and their total daily water use enabled scaling from sample tree water use to stand transpiration. Full descriptions of transpiration measurements at each site, including sampling and scaling methods, are given by McJannet et al. (2007d). Weather data used in this paper (rainfall, solar radiation, temperature, humidity and wind speed) were measured using automatic weather stations located at each site. More detailed descriptions of these measurements are given by McJannet et al. (2007a).

Canopy conductance

Transpiration from a forest is controlled by the weather and the canopy and aerodynamic resistances to water vapour transfer. Canopy conductance, g_c (the reciprocal of canopy resistance) depends mainly on the physiological behaviour of the trees, while the aerodynamic conductance, g_a (the reciprocal of aerodynamic resistance) represents the turbulent mixing conditions in the air above the canopy. By rearranging the Penman–Monteith equation (Monteith, 1965), the canopy conductance (g_c) can be calculated using measured transpiration rates (λE_T) as:

(1)where R_n is the net radiation, VPD the vapour pressure deficit, Δ is the rate of change of saturated vapour pressure with temperature, ρ the density of dry air, c_p the specific heat of air at constant temperature and γ the psychrometric constant. The aerodynamic conductance (g_a) is calculated from wind speed (u_z) following Thom (1975), i.e.

(2)where k is von Kármán’s constant (0.41), z the reference height (2 m above the canopy height, h), d the zero plane displacement (0.75h) and z₀ the roughness length (0.1h). As net radiation (R_n) was not measured directly it was necessary to estimate it from measurements of incoming solar radiation (R_s) as follows:

(3)R_n=R_s(1-α)+L^*where L^* is the net long-wave radiation and α the albedo of the forest canopy, taken as 0.13 ([Monteith and Unsworth, 1990] and [Roberts et al., 2005]). Net long-wave radiation (L^*) was calculated using Eq. (4) that assumes clear skies, a surface emissivity of 1 and surface temperature equal to air temperature (T) (Brutsaert, 1982):

(4)L^*=σT⁴(ε_a(o)-1)where σ is the Stefan–Boltzmann constant and ε_a(o) the atmospheric emissivity for cloudless skies calculated using the Brunt formula (Brunt, 1932).

Atmospheric coupling

To determine the relative influences of radiation and atmospheric demand (via VPD and wind speed) on transpiration, the decoupling coefficient (Ω) of Jarvis and McNaughton (1986) was calculated. This coefficient ranges from 0 to 1, with 0 representing complete coupling between the canopy surface and the atmosphere and 1 representing complete decoupling. When Ω is small, the canopy is strongly coupled to the atmosphere, so that the transpiration rate is highly influenced by canopy conductance, VPD and wind speed (via its effect on aerodynamic conductance). Conversely, as Ω approaches 1, these factors have much less effect on transpiration and it becomes highly dependent on radiation. The formula for calculating Ω is given below:

(5)

Transpiration modelling

Our first transpiration model (Model 1) uses the Penman–Monteith equation with values of canopy conductance, g_c, derived from vapour pressure deficit (VPD) and solar radiation (S_t). To define the form of the dependence of g_c on climatic variables, we followed the original work in this area by Jarvis (1976) and Stewart (1988) who proposed a simple multiplicative formula of the following form:

(6)where g_c,max is the (hypothetical) maximum canopy conductance, which is modified by functions of vapour pressure deficit (VPD), solar radiation (S_t) and soil moisture deficit (θ) that each range between 0 and 1. No explicit temperature dependence was included in Eq. (6) as this did not improve the explanation of variance in g_c. Other studies of Australian forests do not have any explicit temperature dependence (e.g. Whitley et al., 2009) and so temperature effects on gc are propagated indirectly via the strong correlations between temperature and VPD and S_t. Various forms of the functions f(VPD), f(S_t) and f(θ) have been used (e.g. see [Stewart, 1988], [Dolman et al., 1991], [Wright et al., 1996], [Tani et al., 2003], [Kumagai et al., 2004] and [Whitley et al., 2009]). The most common form of f(VPD) is a curvilinear decline and for f(S_t) a curvilinear incline. Furthermore, since McJannet et al., 2007d found no evidence for a dependence of transpiration on soil moisture for three of our sites (OC, UB and BK) for these locations we have therefore chosen the simplest curvilinear form for g_c with no soil moisture dependence as follows:

(7)where a, b and c are parameters whose values are determined by optimization around g_c derived using measured transpiration values in Eq. (1) and the concurrently measured 30 min values of VPD and S_t. The normalisation of VPD by its minimum value VPD_min and S_t by its maximum value S_t,max are required to produce functions that range between 0 and 1. The derivation of the dependence of g_c on VPD and S_t is restricted to dry, sunny day-time periods. Once the values of a, b and c have been determined, g_c can be estimated on all days (daylight hours only) using Eq. (7). These conductance values are then used in the Penman–Monteith equation to derive our first transpiration estimate, Model 1. The performance of Model 1 is compared with models that estimate transpiration directly from weather variables as described below.

Daily total transpiration models have already been reported for all four rainforest sites by McJannet et al. (2007d). There are two separate functions available; the solar functions are linear with the form:

(8)E_T=aS_t+bwhere E_T is in mm h⁻¹ and the fitted parameters a and b vary with site. The VPD function is curved with the form:

(9)E_T=aVPD^band again the fitted parameters a and b vary with location. Eqs. (8) and (9) can only be used separately, so in the current paper the above equations were improved by fitting a combined VPD and S_t function with the form:

(10)Eq. (10) was optimised using data with two timescales. The first, Model 2, used 30 min values of measured transpiration, VPD and S_t. The second, Model 3, used daily total (24 h) transpiration data, and in this case it was also necessary to use 24 h average VPD and solar radiation. For Model 2 the optimisation was performed using a sub-set of 30 min periods (i.e. those when there was transpiration data available from all the monitored trees on a given site; 12% of all the data) and transpiration then predicted for all 30 min periods. In Model 3 the optimisation was carried out on half of the days available (odd number days) and the prediction of transpiration made for all days.

The potential influence of soil moisture on transpiration was investigated at the one site (ML1) where during the 2002 dry season (an El Niño year) McJannet et al., 2007d recorded transpiration rates that were lower that those in the following year with no El Niño. To do this a soil moisture function was added to Eq. (10), i.e.

(11)where the soil moisture function, f(θ) has the following form:

(12)



where θ_d is the soil moisture deficit below which there is no effect on transpiration and θ_e is the soil moisture deficit above which transpiration is zero. When the soil moisture deficit is between θ_d and θ_e the function f(θ) varies linearly between 1 and zero. In order to optimise Eq. (11), referred to as Model 4, it was necessary to derive daily values of soil moisture deficit at ML1. This was done by calculating a running water balance for the site where θ was estimated as follows:

(13)θ_t=θ_t-1+(P_g+P_c)_t-I_t-E_T,t-E_s,twhere the current daily value of the soil moisture deficit, θ_t, is given by its value on the day before, θ_t−1, plus any rainfall (P_g) and cloud interception (P_c), less interception (I), transpiration (E_T) and soil evaporation (E_s). The values of P_g, P_c, I, E_T and E_s for the ML1 site are those measured and reported by [McJannet et al., 2007b] and [McJannet et al., 2007c]. To start the estimation of θ_t it is necessary to know its initial value the day before the start of the simulation period (i.e. 8 August 2002). This initial value was estimated by running a water balance simulation for the 12 month period before this start date. We were able to demonstrate that the value of θ_t on the 8 August 2002 was independent of the choice of the value of θ_t 12 months earlier, since the soil moisture deficit is reset to zero during the 2001/2002 wet season.

Results and discussion

Weather overview

Measurements at different rainforest sites took place over varying time periods (Table 1) between August 2002 and June 2005. Rainfall during the 2004 and 2005 wet seasons was slightly below average, however, the 2002 and 2003 wet seasons were the driest on record at many climate stations in the region. Lower than average rainfall in these years was the result of the development of strong El Niño conditions during 2002. Precipitation characteristics during the measurement period at each site are shown in Table 1. Normalised annual precipitation equivalent is calculated from average total precipitation in each month. It should be noted that the normalised annual precipitation calculated for each site in Table 1 is affected by the timing of the study; sites measured during the dry conditions of 2002 and 2003 had much lower precipitation inputs compared to those encompassing 2004 and 2005. Long-term rainfall records for Bureau of Meteorology rain gauges near OC and BK show annual averages of 3952 mm and 8100 mm, respectively. During our measurement period, annual rainfall input at OC was only 63% (2484 mm) of long-term annual average, thus reflecting the El Niño conditions experienced. Average annual water input to the rainforest at BK (7471 mm) includes cloud interception (2168 mm; see McJannet et al., 2007c). At this site rainfall alone was only 60% of the long term average, a result of below average rainfall during the 2004/2005 wet season.

Fig. 1 shows a comparison of the seasonal variation in measured transpiration (E_T) and potential transpiration at the lowland coastal rainforest at Oliver Creek. was calculated using the Penman–Monteith formula with a canopy conductance of 0.014 m s⁻¹ (i.e. equivalent to a canopy resistance of 70 s m⁻¹, the value associated with closed canopy, well watered vegetation; Shuttleworth and Wallace, 2009). In the middle of the dry season (July–August), is between 1 and 2 times the measured transpiration, but later during the early part of the wet season (November–December), exceeds measured transpiration by a factor of 3–4. Similar differences between potential and actual transpiration were observed at the other three rainforest sites and this implies that these rainforests have canopy conductances that are less than that associated with the potential rate (i.e. 0.014 m s⁻¹). The following analyses explore the reasons for the large differences between potential and actual transpiration by deriving the canopy conductances of all four rainforests and the relationships between their conductances and controlling weather variables.

Canopy conductance

Fig. 2 shows typical diurnal trends in canopy conductance at Oliver Creek along with concurrent variations in transpiration, VPD and solar radiation. Canopy conductance decreased during the early morning and remained fairly constant for the rest of the day at 0.006 m s⁻¹. The error bars indicate that the variation in g_c on sunny days is around ±30%. Fig. 2 also shows the mean diurnal trends in transpiration and it is evident that this reflects the diurnal patterns in solar and VPD, however, it is not possible at this point to determine which of these two variables has most effect on transpiration. Solar radiation reaches very high values (1000 W m⁻²) around midday whereas VPD only peaks at 1200 Pa, reflecting the relatively high humidity conditions that exist at this site. Transpiration increases from 0.05 mm h⁻¹ in the morning and reaches 0.2 mm h⁻¹ around midday.

Fig. 3 shows a comparison of the mean diurnal trends in canopy conductance at all four sites along with concurrent variations in transpiration, solar radiation and VPD. For clarity error bars are not shown in this figure; however they are similar to those for OC shown in Fig. 2. The diurnal patterns of g_c are quite similar at OC, UB and BK where conductances remained fairly constant throughout the day. Midday values of g_c at OC and UB were similar at 0.007 m s⁻¹ whereas BK had the lowest midday g_c values at 0.004 m s⁻¹. Much higher values of g_c were observed at ML1, i.e. 0.012 m s⁻¹ around midday and this site also showed a distinct decrease in g_c during the morning and increase in the late afternoon. Transpiration at BK was much lower than at the other three sites and showed little diurnal variation (Fig. 3b), largely because of the much lower and relatively constant VPD’s at this site (Fig. 3c) and to a lesser extent, the lower solar radiation (Fig. 3d). There were also some differences in the diurnal patterns of transpiration at OC, UB and ML1, for example, E_T reached its maximum value much later in the day at ML1 than it did at OC. Again this pattern reflects the times when VPD’s were highest at these sites. Canopy characteristics such as leaf area index, stem density, tree size distribution and sapwood area also affect the total transpiration from the different forest stands and these factors are discussed in detail by McJannet et al., 2007d. Possible reasons for the differences in canopy conductance observed at different sites are presented later in the modelling section.

The absolute values of g_c observed at the OC and UB sites, 0.006 m s⁻¹ are similar to those observed by Granier et al. (1996) in tropical rainforest in French Guiana (0.006–0.007 m s⁻¹) and higher than those in tropical trees in Panama (0.002–0.004 m s⁻¹; Meinzer et al., 1997). Furthermore, Granier et al. (1996) also found that g_c was fairly constant for much of the day. In contrast, [Shuttleworth, 1988] and [Tani et al., 2003] and Kumagai et al. (2004) reported higher canopy conductances in the Amazonian and Malaysian rainforests, respectively (up to 0.015 m s⁻¹) where the midday values were close to those we observed at our ML1 site (0.012 m s⁻¹). However, the Amazonian and Malaysian rainforests showed a pronounced diurnal variation which is unlike that at ML1, i.e. with g_c peaking in the mid-morning and declining thereafter. The same diurnal pattern was observed in the stomatal conductances of several Amazonian tree species by Roberts et al. (1996), and this behaviour is explained by stomatal conductance declining as VPD increased. We shall show later that there is also a negative relationship between g_c and VPD in the Australian rainforests.

The seasonal variation in monthly mean canopy conductance at all four sites is shown in Fig. 4. Conductances tend to be lowest in the months October–March, corresponding to the time when VPD’s are highest (Fig. 4c). Conductances are also much higher at ML1, with OC and UB having broadly similar values and BK the lowest values. Mean daily solar radiation does not vary much during the year at OC (Fig. 4d), so the increase in transpiration in the summer months at this site (Fig. 4b) is largely due to the increase in VPD at this time (Fig. 4c). In contrast to this, at the ML1 site, solar radiation appears to vary more seasonally than VPD, so the increase in transpiration in the summer months at this site has a significant radiation component. The relative influence of radiation and VPD on canopy conductance and transpiration is explored more fully in the following section.

Modelling canopy conductance

From the above results, and previous studies of rainforest transpiration (e.g. [Roberts et al., 1996], [Granier et al., 1996], [Tani et al., 2003] and [Kumagai et al., 2004]), it appears that g_c is well correlated with VPD and it may also be affected by solar radiation (S_t). To explore these relationships in the Australian rainforests we plotted g_c against VPD and S_t, and an example of this for Oliver Creek is shown in Fig. 5. Canopy conductance declines with VPD and the location of this relationship only varies slightly with solar radiation. The weak correlation with S_t is confirmed in Fig. 5b, where there is very large scatter in the data and the fitted regression is very flat. Granier et al. (1996) also found that g_c was highly correlated with VPD and only weakly affected by solar radiation.

The regressions between g_c and VPD and S_t for all four rainforest sites are shown in Fig. 6 and associated parameter values and regression coefficients are listed in Table 2. The individual site data points have been omitted for clarity, but the scatter is similar to that shown in Fig. 5 for OC. The relationships between g_c and VPD at three sites (OC, UB and ML1) are quite similar when VPD is high, however, there are substantial differences in the modelled values of g_c at the three sites when VPD is low (<500 Pa). Tani et al., 2003 compared g_c versus humidity deficit relationships in a Malaysian rainforest with those in the Amazonian rainforest (Dolman et al., 1991) and also found large differences at small humidity deficits. They suggested these differences were either an artefact of morning dew on the canopy or real structural and physiological behavioural differences of the forests. Meinzer et al., 1997 also observed highly variable relationships between g_c and VPD in four different rainforest species in Panama, however, when the data were normalised by the ratio of leaf area to sapwood area a common relationship for all four species emerged. They concluded that this was due to the different species balancing potential transpiration demand (related to leaf area) to tree water transport capacity (related to sapwood area). The lowest canopy conductances in the present study where recorded at Bellenden Ker and this is also the site with the lowest ratio of leaf area to sapwood area (McJannet et al., 2007d). The variation in g_c versus VPD relationships, and hence the absolute values of g_c, found at the different sites in the present study of Australian rainforest may therefore be a reflection of different species mixes with variable leaf area to sapwood area.

Fig. 6b confirms the weakness of the relationships between g_c and S_t at OC and UB and shows that we could find no dependence of g_c on S_t at either ML1 or BK. The lack of dependence of g_c on S_t is reflected in the value of the coupling coefficient Ω at each site. Mean day time values of Ω on sunny days for OC, UB, ML1 and BK were 0.06 (±0.05), 0.03 (±0.02), 0.10 (±0.10) and, 0.02 (±0.03), respectively. These low values are smaller than those reported by Granier et al. (1996) (i.e. 0.2) and Kumagai et al., 2004 (0.1–0.4) and are characteristic of vegetation that is very well coupled with the atmosphere and hence highly influenced by VPD (and wind speed). In contrast, very high values of Ω have been reported for rainforest trees in Panama by Meinzer et al., 1997, 0.7–0.9, and they suggested that this strong decoupling of the canopy from the atmosphere was due to low wind speed (<1 m s⁻¹) and high stomatal conductance. Our Australian rainforest sites are much windier (1–3 m s⁻¹; McJannet et al. (2007c)) which produces very high aerodynamic conductances (450–950 mm s⁻¹) and leads to our low Ω values (see Eq. (5)). At Bellenden Ker the low value of canopy conductance would also lower Ω at this site. This means that in our rainforests transpiration is strongly controlled by canopy conductance which highly responsive to atmospheric humidity.

Table 2 also shows that overall the conductance model (Model 1) could only explain between 43% and 63% of the variance in the data, with the best fit at UB and the poorest fit at ML1. Both lower and higher explanation of the variance in g_c with VPD and S_t has been reported, e.g. by Kumagai et al., 2004 in a Malaysian rainforest (35–44%); by Harris et al., 2004 for the Amazonian rainforest (32–50%) and Wright et al., 1996 for the Amazonian rainforest (68–83%).

Measured and modelled transpiration

A comparison of the two 30 min time resolution models (Models 1 and 2) for sunny days at Oliver Creek is shown in Fig. 7. At this site the conductance based model (Model 1; r² = 0.69) performed slightly better than the direct E_T model (Model 2; r² = 0.68) with a better distribution of the data about the 1:1 line. Model 1 also gave slightly better predictions of 30 min E_T at UB and was no worse than Model 2 at ML1 (see regression coefficients in Table 3). Model 2 performed better than Model 1 at BK (Table 2). However, the main problem with Model 1 is that it applies to daylight hours only and this becomes apparent when it is used to predict daily total (24 h) transpiration. This is illustrated in Fig. 8 for Oliver Creek and Bellenden Ker (the plots for UB and ML1 are similar to OC). When using Model 1, daily modelled E_T for both sites is consistently lower than when using either Model 2 or Model 3. Model 1 therefore underestimates daily total transpiration because it does not allow for any transpiration during the night, since by using Eq. (7) this model sets g_c = 0 when S_t = 0. This is consistent with the way Jarvis–Stewart conductance models have been used in numerous other studies. In contrast, the direct calculation of transpiration using Model 2 does allow for night time transpiration since when S_t = 0 it uses E_T = a VPD^b. The average amount of night time transpiration is given by the difference between the fitted lines in Fig. 8 and it has a value of 0.5 mm, or an average of 0.04 mm h⁻¹ for a 12 h night. This means that nocturnal transpiration accounts for a substantial amount of the daily total transpiration, i.e. 20% when E_T = 2.5 mm d⁻¹ and 50% when E_T = 1.0 mm d⁻¹. McJannet et al., 2007d highlighted the occurrence of significant sap flow in rainforest trees during the night. On nights with no VPD, very small sap flows were associated with the replenishment of water storage in the trees. However, on nights where VPD remained high (0.5–1.0 kPa) sap flows were 3–12 times higher and McJannet et al., 2007d concluded that this must be due to continued transpiration due to the high VPD. The occurence of nocturnal transpiration has been reported as widespread among a wide range of tree and shrub species, including tropical forest trees (Dawson et al., 2007), with nocturnal rates between 5 and over 40% of daytime maximum rates. Nocturnal sap flow rates of 0.1 mm h⁻¹ or higher have been also reported in Eucalyptus grandis trees in New South Wales, Australia (Benyon, 1999) which cannot be explained by storage replenishment alone.

The flat horizontal grouping of points at low transpiration rates at BK (Fig. 8b) is thought to be an artefact of the difficulty of measuring the extremely high humidities at this site. For these data points the recorded relative humidity is over 98%, so modelled E_T remains fairly constant while measured E_T still varies significantly. In reality, relative humidity (and hence VPD) are probably also varying (albeit in the range very close to 100%), but the sensors used cannot accurately discriminate these very high humidities. A similar, but much less pronounced example of this phenomenon can be seen when using Model 2 at Oliver Creek (Fig. 7b).

The overall performance of transpiration Models 1, 2 and 3 at all four forest sites is given in Table 3. Model 1 has the lowest correlation coefficients and it consistently underestimates cumulative transpiration (over periods from 221 to 427 days) by large amounts (30–70%) because it does not account for nocturnal transpiration.

The statistical fit of Model 2 is not much better overall than Model 1, however, as it does allow for nocturnal transpiration its cumulative total transpiration is within ±20% of the measured cumulative transpiration. The daily time step model, Model 3, gives the best fit to the observed daily total transpiration with correlation coefficients between 0.7 and 0.8. Model 3 also gives the most accurate estimate of cumulative transpiration, usually to within 1% of the measured cumulative transpiration. Whitley et al. (2009) compared a canopy conductance based model (similar to our Model 1) with one where forest (Eucalyptus cerbra and Callitris glaucophylla) transpiration was directly related to environmental variables (similar to our Models 2 and 3) and they also found that the latter model performed better overall.

Soil moisture effects on transpiration

Fig. 9 shows the time series of measured transpiration and simulated soil moisture deficit (θ) at Mount Lewis during 2002 and 2003. Transpiration during the El Niño period from the beginning of August to mid-November 2002 was particularly low compared with the same period during the following year. It can also be seen that θ towards the end of the 2002 dry season (170 mm) was much larger than in the 2003 dry season (90 mm).

The result of adding soil moisture dependence to Model 3 for the ML1 site is shown in Table 4. Note that the soil moisture function was only applied when VPD exceeded 700 Pa. This limit was introduced because inspection of a plot of E_T against soil moisture deficit showed no discernable relationship at low VPD’s. Furthermore, removing this threshold gave implausible values of the fitted parameters θ_d and θ_e (see below). Table 4 shows that the addition of soil moisture dependence only increased the regression coefficient very marginally and actually decreased the accuracy of the modelled cumulative transpiration. The values of soil moisture deficit where the reduction in transpiration begins (θ_d) and where transpiration stops (θ_e), which are fitted by the optimisation routine, are 61 mm and 280 mm, respectively. The maximum extractable water figure of 280 mm corresponds to a sandy loam soil around 2.5 m deep (Rawls et al., 1996), however, assuming that transpiration begins to be affected by soil moisture when 65% of the available water has been extracted (Shuttleworth, 1993), θ_d would be expected to be 180 mm. From a soil physics perspective therefore, the fitted value of θ_d appears to be too low. When the soil moisture function was applied with no limiting VPD, the fitted values of θ_d and θ_e were more implausible at 10 mm and 460 mm, respectively.

The main reason for introducing the soil moisture function at Mount Lewis was to see if this would explain the low transpiration rates observed during the El Niño period from the beginning of August to mid-November 2002. This question is addressed in Fig. 10 which shows a comparison of the difference between Model 3 and Model 4 transpiration from measured transpiration for this site. Although the inclusion of the soil moisture function marginally improves the overall correspondence between modelled and measured transpiration, Model 4 still overestimates E_T during the El Niño period in 2002. This means that the low values of transpiration at this time cannot be entirely explained by the concurrent soil moisture conditions. One possible explanation is that the amount of foliage (leaf area index) that the forest supported during the El Niño period in 2002 was lower than in more normal rainfall years such as 2003. Rainfall in the 2 months (June and July) before the August to mid-November El Niño period in 2002 was only 123 mm and the average soil moisture deficit was 85 mm, compared to 360 mm rainfall and average soil moisture deficit of 5 mm in the same months in 2003.

Most other studies of tropical rainforests have also found little or no correlation between soil moisture on forest transpiration ([Dolman et al., 1991], [Shuttleworth, 1989], [Tani et al., 2003] and [Fisher et al., 2008]). Further studies by Roberts et al., 1996 report that there was very little evidence that declining soil moisture influenced stomatal conductance at any of their Amazonian rainforest sites. For the same Amazonian forests Wright et al. (1996) modelled surface conductance and found no significant correlation with soil moisture. Likewise McWilliam et al. (1996) found no evidence that the Amazonian rainforest leaf gas exchange (of water vapour and carbon dioxide) was affected by seasonal changes in rainfall and hence soil moisture. More recent studies of the Amazonian forest north of Manaus (Harris et al., 2004) report a 2 month period where soil moisture did appear to reduce canopy conductance and transpiration, currently the only Amazonian site where this has been observed. Fisher et al., 2008 analysed soil hydraulic properties at the Manaus site and another site in eastern Amazonia (Caxiuanã) where there was no correlation between soil moisture and transpiration and concluded that the Manaus response was due to the much lower soil water availability (30–50% of that at Caxiuanã) associated with the higher clay content of the soil at this location.

The explanation for the lack of a transpiration response to soil moisture at most of the Amazonian rainforest locations is reported to be either high soil water availability and/or substantial rooting depths of these forests, sufficient to provide an adequate water supply even during the dry season ([Hodnett et al., 1996] and [Fisher et al., 2008]). This is even more likely to be the case in the Australian rainforests where rainfall is greater (2500–7500 mm year⁻¹) and dry seasons shorter than in Amazonia. Alternatively some locations in Australia (e.g. those on the coastal floodplain) may have water tables that are close enough to the surface for tree roots to access water from them even when the overlying soil is relatively dry. For example, in a Melaleuca quinquenervia floodplain forest at Cowley Beach, north Queensland, McJannet (2008) found that the maximum groundwater table depth was only 2.5 m and there was no decline in transpiration when the water table was at this depth. In contrast to rainforests other native Australian forests do show a marked response to soil moisture, for example, the Eucalyptus and Callitris dominated woodlands described by Whitley et al. (2009) in north-west New South Wales. Soil moisture constraints on forest transpiration appear to be more common in temperate forests, e.g. Scots pine ([Sturm et al., 1998] and [Irvine et al., 1998]) and beech (Granier et al., 2000).

Conclusions

This study has shown that transpiration from Australian rainforest is strongly controlled by its surface conductance, which is highly responsive to atmospheric VPD and only weakly influenced by solar radiation. This is because these forests are highly coupled to the atmosphere, due to their very high aerodynamic conductance. This high conductance is a result of exposing tall rough canopies to quite windy conditions. The lack of sensitivity to radiation also implies a low threshold of light saturation of stomatal conductance, which has been observed in other tropical rainforests (e.g. Granier et al., 1996).

The best predictive model of rainforest transpiration is one based on daily average VPD and solar radiation (Model 3). With this model individual daily transpiration rates can be estimated to an average accuracy of between ±15% (at OC) and ±23% (at UB) and cumulative total transpiration over longer periods (220–400 days) can be estimated even more accurately, ±1%. For many purposes, such as working out catchment water balances and potential impacts of land use and climate change, this daily model is entirely adequate and it also has the advantage of only requiring daily mean values of VPD and total daily solar radiation. If sub-daily transpiration rates are required, the best predictive model is that based on the direct regression of E_T on VPD and S_t (Model 2), since this will account for the significant amounts (20–50%) of nocturnal transpiration that the surface conductance based model (Model 1) does not. Whichever model is chosen it is necessary to use parameters that are specific to that site as it is not possible to get accurate predictions of E_T using parameters derived from another location. This is even true for the surface conductance based model and this implies that the physiological factors that influence g_c such as leaf area, stomatal conductance and canopy structure must vary significantly between sites.

We have also found that there is no obvious correlation between soil moisture and surface conductance or transpiration. No evidence for this was found at three locations even though the rainfall for the period that measurements were made was only 60% of the long term average. At the one site where low transpiration rates were observed during a long dry spell these do not correlate with the concurrent low soil moisture conditions. It is possible that low antecedent soil moisture may have reduced leaf area and hence transpiration, however, further observations of leaf area, transpiration and soil moisture would be needed to corroborate this hypothesis.

Finally, we return to the initial observation in this paper that actual transpiration rates are much lower than potential transpiration rates. We have demonstrated that this is not due to weather conditions, but rather to the physiological characteristics of the Australian rainforests, i.e. their surface conductances are generally much lower than that associated with the potential rate (0.014 m s⁻¹). Only one of our rainforest sites has conductance values approaching 0.014 m s⁻¹ and this is the site where E_T comes closest, but not equal to potential. There remains the question of why surface conductances are so low in Australian rainforests. This could be partly due to their low leaf area, 3.3–4.5, compared with values up to twice this in other rainforests (Roberts et al., 2005). Other factors include the possibility of low stomatal conductance in Australian rainforests and canopy structures that do not include many large trees. The absence of large trees may be a consequence of the exposure of these sites to cyclones. On the other hand, low stomatal conductance may be an inherited physiological trait if the Australian rainforests have evolved through periods which were significantly drier than at present.

Acknowledgements

The authors would like to thank, Mark Disher, Peter Fitch, Andrew Ford, Peter Richardson, Trevor Parker, Adam McKeown, Trudi Prideaux, Jenny Holmes and Pepper Brown for their help with field installation and maintenance. Funding for this research was provided by the Cooperative Research Centre for Tropical Rainforest Ecology and Management. We are also grateful for the constructive comments from CSIRO colleagues and the external reviewers.