Components of ecosystem evaporation in a temperate coniferous rainforest, with canopy transpiration scaled using sapwood density


Author for correspondence: M. M. Barbour Tel: +64 3325 6700 Fax: +64 3325 2418 Email:


  • • Here we develop and test a method to scale sap velocity measurements from individual trees to canopy transpiration (Ec) in a low-productivity, old-growth rainforest dominated by the conifer Dacrydium cupressinum. Further, Ec as a component of the ecosystem water balance is quantified in relation to forest floor evaporation rates and measurements of ecosystem evaporation using eddy covariance (Eeco) in conditions when the canopy was dry and partly wet.
  • • Thermal dissipation probes were used to measure sap velocity of individual trees, and scaled to transpiration at the canopy level by dividing trees into classes based on sapwood density and canopy position (sheltered or exposed).
  • • When compared with ecosystem eddy covariance measurements, Ec accounted for 51% of Eeco on dry days, and 22% of Eeco on wet days.
  • • Low transpiration rates, and significant contributions to Eeco from wet canopy evaporation and understorey transpiration (35%) and forest floor evaporation (25%), were attributable to the unique characteristics of the forest: in particular, high rainfall, low leaf area index, low stomatal conductance and low productivity associated with severe nutrient limitation.


Models of climate change effects on vegetation are critically dependent on a thorough understanding of the exchange of energy, water vapour and carbon dioxide between vegetation and the atmosphere. Transpiration is an important component of this exchange but can be difficult to quantify for ecosystems because of feedback responses and complexity in scaling from the leaf or whole-plant level to the stand (Jarvis & McNaughton, 1986; Wullschleger et al., 2002). Micrometeorological techniques, such as eddy covariance, allow ecosystem measurements of evaporation to be made (‘top down’ approach) while whole-tree transpiration is commonly measured using sap flow sensors (‘bottom up’ approach). Comparison of estimates of the components of water balance using the different techniques requires an appropriate scaling framework.

Canopy transpiration as a component of the ecosystem water balance has been intensively studied in a number of subtropical (Hutley et al., 1997) temperate (Granier et al., 1990; Köstner et al., 1992; Berbigier et al., 1996; Oren et al., 1998; Granier et al., 2000) and boreal forest systems (Cienciala et al., 1998; Kelliher et al., 1998; Zimmermann et al., 2000) but there has been relatively little work done in low-productivity, old-growth forests. Scaling stand transpiration is more complex in old-growth forests, compared with younger forests because of the presence of trees of different size, age and canopy depth (Martin et al., 2001). Canopy transpiration forms the main component of ecosystem evaporation from closed-canopy forests (Denmead, 1984; Kelliher et al., 1986; Baldocchi & Vogel, 1996). However, in low-productivity forests, when leaf area index (L) is < 3 m2 m−2, transpiration from understorey species and evaporation from the forest floor and wet surfaces become much larger components of total ecosystem evaporation (Unsworth et al., 2004). Baldocchi & Meyers (1991) have demonstrated that the frequency of large-scale turbulent eddies and fluctuations in air pressure within the forest canopy are important in determining rates of evaporation from the forest floor, if the forest floor litter layer is wet. Clearly, in high rainfall environments with low leaf area index, forest floor evaporation, understorey transpiration and evaporation from wet surfaces may contribute substantially to ecosystem evaporation, and are regulated by incident irradiance and the degree of turbulent mixing within the canopy.

Ecosystem evaporation (Eeco) may be partitioned into four components: transpiration from canopy (Ec) and understorey (Es) trees, and evaporation from the forest floor (Eg) and wet surfaces (Ew) so that:

image(Eqn 1)

The relative importance of each component on daily and annual timescales varies between sites with differing rainfall and forest structure. In low-rainfall environments, on an annual basis, Ew and Eg are minimal except immediately after rain, while, when rainfall is high and frequent, Ew and Eg can make significant contributions to Eeco (Fujieda et al., 1997; Godoy et al., 1999). In forests with simple structure and sparse understorey vegetation, such as plantation forests, Es may contribute very little to Eeco.

When ecosystem evaporation is measured using eddy covariance, canopy transpiration scaled from sap velocity measurements and forest floor evaporation predicted from lysimeter measurements, the difference between Eeco and Ec +Eg must be Es + Ew. Separating Es and Ew is difficult. However, even in high-rainfall environments, surfaces within the canopy do become fully dry during periods without rain when evaporative demand is high. Under these conditions Ew may be assumed to be negligible and partitioning Eeco can be simplified to:

image(Eqn 2)

By contrast, on days when foliage in the lower canopy remains wet all day, such as on days with low air saturation deficit and little mixing of canopy air, transpiration of understorey vegetation may be assumed to be negligible and:

image(Eqn 3)

Scaling sap velocity to canopy transpiration requires the inclusion of an appropriate biophysical variable (Ćermák, 1989), and a number of approaches have been taken. These include the use of sapwood area (Diawara et al., 1991), basal area (Saugier et al., 1997), species (Wilson et al., 2001), canopy position (Granier, 1987; Kelliher et al., 1992) and sunlit leaf area (Ćermák, 1989). Recent work in a New Zealand old-growth rainforest (Barbour & Whitehead, 2003) argues that none of these five scaling methods is appropriate for estimating transpiration for the dominant canopy species, Dacrydium cupressinum. The wide variability between sap velocity of similar-sized individual trees means that scaling using an average of all measured trees would include large errors, while the absence of a relationship between sap velocity and tree size means that the size class method is also unsuitable. Although the position of an individual tree within the canopy (exposed or sheltered) seemed to determine sap velocity to some extent, wide variability in sap velocity between trees was still apparent.

Demonstration of the theoretically predicted relationship between sap velocity (v) and wood density (ρb) prompted Barbour & Whitehead (2003) to suggest that ρb may be a suitable scalar of water use in this species. Strong positive relationships were observed between (1 − ρb)2 and v for trees in exposed canopy positions, with 94% of variation between trees in average sap velocity measured over 160 d explained by (1 − ρb)2.

The relationship between sap velocity and wood density was predicted by Roderick & Berry (2001), who extended Poiseuille's law of rate of liquid flow through a pipe to describe velocity through the population of pipes forming a stem (v) as:

image(Eqn 4)

pf is the drop in pressure along the stem segment due to friction; Tw is water temperature; ηrel is the relative viscosity of water; As is the cross-sectional area of the stem; l is the length of the ‘pipes’ within the stem; Fp is the volume fraction of available space in the stem occupied by pipes; Np is the number of pipes in the stem; and ɛv is the coefficient of variation of the cross-sectional area of pipes). The derivation of Eqn 4 is described in detail in Roderick & Berry (2001) and Barbour & Whitehead (2003), and identifies the importance of water temperature, wood density and the statistical distribution of pipe diameters to water velocity. Conifers typically have rather uniform pipe diameters, leading Roderick & Berry (2001) to suggest that the (1 + ɛv)2 term has little effect on v, and that at constant Tw and Fp/Np, v should be positively related to (1 − ρb)2, as demonstrated in D. cupressinum by Barbour & Whitehead (2003).

In this paper, we quantify transpiration as a component of ecosystem evaporation in a complex, low-productivity temperate rainforest, dominated by the conifer D. cupressinum with a wide range in tree age and size. Our objective was to develop and apply an appropriate technique to scale whole-tree transpiration to the stand using the relationship between sapwood density and sap velocity, and test the scaling approach using independent measurements. A further objective was to use measurements at the tree and ecosystem scale to partition evaporation into components comprising ecosystem water balance.

Materials and Methods

Site description

Measurements were made in a 50 × 50 m plot of a mixed conifer–broad-leaved forest at Okarito Forest, Westland, New Zealand (43.2° S, 170.3° E, 50 m above sea level). A tower erected at the site allowed access to the foliage to a height of 25 m. The forest is dominated by 100- to 400-yr-old D. cupressinum Lamb. (rimu) trees (73% of the stand basal area, and c. 90% of the canopy foliage area) with a subcanopy of Prumnopitys ferruginea (D. Don) Laubenf. (miro), Weinmannia racemosa L. f. (kamahi), Metrosideros umbellata Cav. (southern rata), Quintinia acutifolia Kirk (Westland quintinia) and a wide diversity of shrubs, ferns and bryophytes forming the understorey and ground vegetation. Dacrydium cupressinum trees occupied both emergent and sheltered positions within the canopy. An annual rainfall of approx. 3400 mm and poor drainage results in a high water table (usually within 200 mm of the ground surface) and an acid humic organic soil. Mean annual temperature is 11.3°C, and daily mean varies just 8.6°C between winter and summer. The air saturation deficit at midday is generally < 1 kPa, but can reach values as high as 2 kPa in summer. Estimated annual net carbon uptake for the trees is low (1.1 kg C m−2 yr−1), with nutrient availability identified as the main limitation (Whitehead et al., 2002).

Leaf area index (projected area basis) of the forest is low, just 2.9 m2 m−2, and spatially variable (Whitehead et al., 2002). This suggests that the transmittance of solar irradiance to the understorey trees may be high for significant periods of the day, particularly in the gaps between canopy trees.

Meteorological measurements

Incident irradiance, air temperature, relative humidity, wind speed and rainfall were measured every 10 s above the canopy, and air temperature, relative humidity and wind speed also measured every 10 s at 1.5 m above the ground. Leaf wetness was measured among foliage at 18 m above the ground and on the forest floor with wetness sensors (Model 237; Campbell Scientific, Logan, UT, USA). Average irradiance, temperature, humidity, wind speed, leaf wetness and total rainfall were recorded every 30 min on a data logger (CR10; Campbell Scientific). The average depth to the water table was also recorded every 30 min from a capacitance probe in a stilling well situated 10 m from the base of the tower. For comparison with sap velocity and eddy covariance measurements, incident irradiance is presented as a daily total (QT), temperature and air saturation deficit (Dm) as averages between 11:00 hours and 13:00 hours (NZST), and leaf wetness and depth to the water table as daily averages. To correct for the effect of changing daylength on total daily transpiration, mean daytime D above the canopy, Dzc, and at the forest floor, Dzg, are normalized by the proportion of daylight hours, after Oren et al. (1999). A power failure in December 2001 caused loss of meteorological measurements for 15 d.

Sap velocity measurements

Sap velocity was measured continuously for 1 yr from early spring (October) 2001 using the thermal dissipation technique (Granier, 1985, 1987), and including a correction for the proportion of the probe in contact with nonconducting tissue (Clearwater et al., 1999), as described in Barbour and Whitehead (2003). The probes, consisting of a thermocouple surrounded by a heating wire placed 40 mm above a second thermocouple were either 40 mm long (TDP-40; Dynamax, Houston, TX, USA) for trees with sapwood 40 mm or more thick, or 20 mm long (following Phillips et al., 1996) for trees with narrow sapwood thickness, as determined by measurements of water content made on wood cores. Between 85% and 100% of the length of the probes were in contact with sapwood, so that corrections for nonconducting tissue (Clearwater et al., 1999) were small. Measurements of sap velocity were made every 10 s, and 30-min averages recorded on a data logger (21X; Campbell Scientific). Large-store gel batteries (480 Ah), charged by solar panels mounted on the top of the canopy access tower, supplied power to the probes and loggers. Insufficient energy storage and low solar angles caused a number of power failures during the winter months. However, successful measurements were made for 306 d, or 84% of the year.

Probes were inserted into trees at 2.5 m above ground level in the south-eastern azimuth (the shaded side of the tree), surrounded by closed cell foam and covered by a layer of foil-coated bubble wrap to reduce vertical temperature gradients associated with direct radiation onto the stem and temperature differences between the ground and the air. Natural temperature gradients within the stem were assessed from the temperature difference between the two probes when power to the heating wire was turned off. The temperature difference never deviated beyond ±0.2°C, suggesting that the probes were well insulated from outside temperature gradients and sunlight. A +0.2°C temperature gradient would result in a 7–8% underestimate of peak midday sap velocities, which is considered to be insignificant.

Sap velocity can vary considerably around trees, so bias introduced by placing a single probe on the south-eastern azimuth was assessed by installing additional probes (one to three, depending on basal area) in four trees chosen to span the range in basal area and canopy position of trees in the plot. On average, over 27 d in mid-summer, probes on the south-eastern side of the trees underestimated average sap velocity of all probes by just 6%. This was considered an insignificant error, given an average coefficient of variation in sap velocity between trees of 61%.

Scaling sap flux of individual trees to stand transpiration

Transpiration was scaled from individual trees to the stand using canopy position and sapwood density classes. Conducting wood density was determined from the outer 14 mm of sapwood, as described by Barbour & Whitehead (2003), for all trees classed as ‘exposed’ in the canopy. Sapwood thickness of exposed trees was determined visually from fresh wood cores, and used to calculate sapwood cross-sectional area. The average total daily sap velocity in the measured trees for each class was multiplied by the ratio of sapwood area to plot ground area (Ag, i.e. As : Ag) for all trees with a basal area greater than 0.007 m2 in each class within the 50 m by 50 m plot. The ratios of As : Ag and numbers of trees for each class are presented in Table 1. As wood cores were not taken for all trees, As was estimated from a least-squares regression of basal area (Ab) on As of all measured trees (Fig. 1). Total transpiration from all trees (Ec, mm d−1) is given by adding together transpiration rates from all trees in all tree classes. Trees were divided into two canopy positional classes: exposed and sheltered, as defined by Barbour & Whitehead (2003). Exposed trees were those for which at least two-thirds of the crown emerged above neighbouring crowns, or those in the absence of overtopping neighbours for at least 25 m in at least 120 azimuthal degrees, and all other trees are classed as sheltered. Exposed trees were divided into five sapwood density classes, as shown in Table 1.

Table 1.  Ratio of sapwood area to plot ground area (As : Ag) of Dacrydium trees in the 50 × 50 m experimental plot for each class used to scale transpiration from the individual tree to the stand
ClassAs : Ag (× 10−3)Number of trees MeasuredPresent in plot
  1. Ab and ρb refer to basal area and wood density, respectively.

Exposed ρb < 440 kg m−30.1281 6
Exposed 440 < ρb < 480 kg m−30.1921 9
Exposed 480 < ρb < 520 kg m−30.208110
Exposed 520 < ρb < 580 kg m−30.208210
Exposed ρb > 580 kg m−30.0642 3
Figure 1.

Relationship between basal area (Ab) and sapwood area (As) for all 38 exposed trees in the experimental plot and the five sheltered trees used for sap flow measurements. The line represented a least-squares regression As = 0.30Ab, r2 = 0.67. Squares, all exposed trees; triangles, measured exposed trees; circles, measured sheltered trees.

Measurements of ecosystem evaporation

Ecosystem evaporation rate was measured by the eddy covariance technique at a height of 36 m (approx. 6 m above the tree canopy). The system comprised a three-axis sonic anemometer (Model Solent R3; Gill Instruments, Lymington, UK) and a closed-path infrared gas analyser (Model Li-6262; LiCor, Lincoln, NE, USA), measuring water vapour concentrations at 10 Hz. The analyser and computer were housed inside a thermally insulated box and regulated to within one degree of the analyser's calibration temperature by a ventilation fan. To avoid condensation in frequent conditions of high humidity and rainfall, all sensitive instruments were housed inside an insulated shed placed 13 m below the air intake. Power was supplied to all instruments from two 1025 Ah battery banks. The batteries were recharged using 10 85 W solar panels controlled by a charge regulator (C40; Trace Engineering, Arlington, WA, USA).

All data were post-processed using edisol software (Moncreiff et al., 1997). Two-angle coordinate rotation was carried out to eliminate instrument tilt and terrain distortions to wind flow. Rates of evaporation were calculated as covariances between the vertical wind speed and water vapour density. The time lag between the wind speed and water vapour concentration measurements was computed for each half hour by finding the maximum covariance and was typically 4 s. Corrections for the spectral deficiencies of the system (tubing length, analyser, in-line filters and sensor separation) were made following Moore (1986) and Moncreiff et al. (1997).

Evaporation from the ground surface (Eg, mm h−1) was estimated from the available energy reaching the forest floor (Qg, in W m−2), using a least squares regression analysis of change in mass of lysimeters with changing Qg at the same study site, as described by DeLucia et al. (2003) where

image(Eqn 5)

with Qg modelled using Beer's law where:

image(Eqn 6)

and Q is incident irradiance above the canopy, k is the extinction coefficient (assumed to be 0.5) and L is the leaf area index.

Days on which eddy covariance, sap flow and meteorological measurements were all made were separated into three classes based on the amount of rainfall during the day and the length of time the leaf wetness sensors recorded that the surfaces were wet. Wet days were defined as those with at least 2 mm of rain, for which the leaf wetness sensor near the forest floor was wet for at least 65% of the day, and the upper leaf wetness sensor was wet for at least part of the day. Dry days were those with < 2 mm of rain, both leaf wetness sensors recorded no wetness over the 24-h period and no significant rain (i.e. not > 2 mm in 24 h) had fallen for at least the previous 3 d. Under these conditions the outer layer of epiphytic mosses were observed to be dry, and Ew was assumed to be zero. Of the 121 d for which eddy covariance, sap flow and meteorological measurements were available, there were 36 wet days and 20 dry days. The 65 d that were neither wet nor dry were classed as intermediate.


Total annual rainfall from October 2001 was 3482 mm, and was evenly distributed throughout the year. Maximum incident irradiance occurred in mid- to late January, coinciding with maxima in midday temperature and air saturation deficit (Fig. 2). The depth to the water table was generally within 100–200 mm of the ground surface, but fell below 300 mm three times over a short period in late summer (Feb), and occasionally rose to < 50 mm from the surface from early winter (mid-June). The leaf wetness sensor at 18 m above the ground showed that the canopy was dry all day (average leaf wetness = 0%) for 44% of days throughout the year, and the sensor never remained wet all day (average daily leaf wetness was never 100%, even on days with very heavy rain). By contrast, the wetness sensor at the forest floor was dry all day for just 23% of days, and was wet for most of the day (average leaf wetness > 75%) for 49% of days.

Figure 2.

Meteorological measurements over the year from 7 October 2001: daily rainfall (Pd), incoming irradiance (QT), average midday temperature (Ta) and air saturation deficit (Dm) measured 30 m above the ground, and average daily leaf wetness (Lw).

As expected, air temperature and air saturation deficit, measured both above the canopy and near the forest floor, increased with increasing total incoming irradiance. Air temperature near the forest floor tended to be slightly lower (−0.8°C, on average) than that above the canopy, but the difference in temperature between the lower and upper sensors varied between −5.9 and +1.5°C. Daylength-normalized air saturation deficits measured at the two heights were positively related (r2 = 0.87), with Dzgc. 42% of Dzc, on average. Average midday wind speed above the canopy varied between 0.28 m s−1 and 5.53 m s−1, while average midday wind speed 1.5 m above the ground was about one-tenth of that above the canopy. Although turbulence was not measured within the canopy, comparison of average half-hourly wind speed above the canopy (u) with average half-hourly frictional velocity above the canopy (u*, from analysis of data from the sonic anemometer) revealed a linear relationship between the two (u* = 0.272u − 0.096, r2 = 0.77). Given that half-hourly average wind speed above the canopy and 1.5 m from the ground (ug) were also related (ug = 0.088u − 0.009, r2 = 0.52), u and ug are likely to be good indicators of the degree of mixing within the canopy.

Wood density of exposed trees varied between 383 kg m−3 and 607 kg m−3, with a density between 500 kg m−3 and 520 kg m−3 being the most frequently observed (Fig. 3). The numbers of trees used for sap velocity measurements were evenly distributed along this range in sapwood density, such that one or two measured trees were within each wood density class.

Figure 3.

Frequency distribution of sapwood density for the 38 trees in exposed canopy positions within the experimental plot. Asterisks identify trees used for sap flow measurements.

Annual average sap velocity of each exposed tree was found to be positively related to (1 − ρb)2 (where ρb is wood density in g cm−3), which supports theoretical predictions (Roderick & Berry, 2001) and earlier data (Barbour & Whitehead, 2003). Sap velocity was also strongly regulated by midday air saturation deficit (Dm), as shown by the increase in the slope of the linear regression between average daily sap velocity and wood density, when days are divided into classes with narrow ranges in average Dm (Fig. 4).

Figure 4.

Relationship between average sap velocity and the square of the difference from unity of wood density ((1 − ρb)2) for exposed trees for days grouped into classes with differing average midday air saturation deficit (Dm). The lines represent least squares regressions for: 0.1 < Dm < 0.2 kPa v = 0.80(1 − ρb)2 − 0.07, r2 = 0.64 (squares); 0.4 < Dm < 0.5 kPa v = 2.73(1 − ρb)2 − 0.25, r2 = 0.66 (circles); 0.7 < Dm < 0.8 kPa v = 5.09(1 − ρb)2 − 0.35, r2 = 0.73 (closed triangles); 1.0 < Dm < 2.0 kPa v = 5.67(1 − ρb)2 − 0.29, r2 = 0.82 (open triangles).

Scaled transpiration for the stand varied from 0 to 1.8 mm d−1 (Fig. 5). As expected, Ec was strongly regulated by the evaporative demand, with 67% of variation in Ec explained by variation in mean air saturation deficit above the canopy during daylight, and normalized by daylength (Dzc) when Dzc < 0.6 kPa (P < 0.0001, Fig. 6).

Figure 5.

Daily canopy transpiration (Ec, scaled from sap velocity measurements) and ecosystem evaporation (Eeco, from eddy covariance measurements) between October 2001 and October 2002. Missing Ec values indicate power failures associated with lower solar irradiance. The eddy covariance system was operational nearly continuously between mid-November and the end of February, after which measurements were made every second or third day, depending on available energy (i.e. incoming irradiance).

Figure 6.

Relationship between scaled canopy transpiration (Ec) and daylength-normalized air saturation deficit measured 30 m above the ground (Dzc). The line represents a least squares regression when Dzc < 0.6 kPa: Ec = 2.48Dzc − 0.04, r2 = 0.67.

Total ecosystem evaporation varied from 0 to 4.4 mm d−1, tending to be highest in mid-January (Fig. 5). Ec was positively related to Eeco, although the slope of the relationship varied between dry, intermediate and wet days. On dry days Ec formed 51% of Eeco (Fig. 7, r2 = 0.69, P < 0.0001), while on wet days the slope of the relationship was lower, with Ec contributing just 22% of Eeco (Fig. 7, r2 = 0.80, P < 0.0001). On intermediate days the data were quite scattered, and Ec contributed 40% of Eeco.

Figure 7.

Relationship between scaled canopy transpiration rate (Ec) and ecosystem evaporation (Eeco), as measured by eddy covariance for dry (solid line, squares), intermediate (dashed line, crosses) and wet days (dotted line, circles). Definitions of classes of days are given in the Materials and Methods. Lines represent least-squares regressions forced through the origin: for dry days Ec = 0.51Eeco, r2 = 0.69; for intermediate days Ec = 0.40Eeco, r2 = 0.44; for wet days Ec = 0.22Eeco, r2 = 0.80.

Division of measurement days into three classes allowed further partitioning of Eeco using Eqns 2 and 3. On dry days, when Ew was assumed to be zero, Es was calculated from Eqn 2. Es on dry days varied between 0.04 mm d−1 and 1.2 mm d−1, while for the same days Ec varied between 0.3 mm d−1 and 1.3 mm d−1. On average, Es was slightly less than half of Ec (Es = 0.46Ec), and formed 22% of Eeco on dry days. On wet days, when Es was assumed to be zero, Ew was estimated from Eqn 3. Ew on wet days varied between 0 and 2.6 mm mm d−1, and on average accounted for 51% of Eeco. Ew and Es could not be separated for the 65 intermediate days, and over these days, Es + Ew varied between 0 and 2.2 mm mm d−1. On average, Ew + Es contributed 32% of Eeco on intermediate days.

Estimated understorey transpiration on dry days tended to be positively related to daylength-normalized air saturation deficit (Dzg) and average midday wind speed at the forest floor (umg), although was only significantly (P < 0.01) related to umg (Table 2). About 95% of variation in Ew on wet days was explained by total daily incident irradiance (QT), and significant relationships between Ew and both Dzg and umg are probably a result of covariance of these variables with QT (Table 2). On intermediate days Es + Ew was also significantly (P < 0.001) related to QT, although just 24% of variation in Es + Ew was explained by QT on these days (Table 2).

Table 2.  Slope (a1), intercept (a2) and r2 values for the linear relationships between understorey transpiration (Es) and wet canopy evaporation (Ew) and daylength-normalized air saturation at the forest floor (Dzg), total daily incoming irradiance (QT) and average midday wind speed at the forest floor (umg)
ClassComponentRegression statistics
a1Dzg a2r2a1QT a2r2umg a1a2r2
  • *, ***

    , P < 0.01 and P < 0.0001, respectively; ns, not significant.

DryEsnsns1.13 0.020.42*
IntermediateEs + Ewns0.02−0.080.24***2.29 0.090.19*

Using measurements of Eeco, scaled Ec, modelled Eg, and estimates of Es, Ew and Es + Ew for dry, wet and intermediate days, respectively, the water balance of the site was calculated over the period of time that meteorological, sap velocity and eddy covariance measurements were made. Table 3 shows that Eeco over the 121 measurement days was 223.9 mm, of which the contributions were Ec (39%), Eg (25%) Es and Ew (35%). Ec contributed a lower percentage to Eeco in November and December because of high rainfall and generally low D and QT, resulting in a higher percentage of Ew during these months. The percentage of Eeco estimated to be from Eg was fairly constant between months.

Table 3.  Measured and estimated components of the water balance at the rainforest site during the summer
MonthdRainfall (mm)Eeco (mm)Ec (mm)Eg (mm)Es + Ew (mm)Drainage (mm)Percentage of Eeco(Ec)(Eg)(Es + Ew)Number days in class DryIntermediateWet
  1. Ecosystem evaporation (Eeco), canopy transpiration (Ec), forest floor evaporation (Eg), understorey transpiration (Es) and evaporation from wet surfaces within the canopy (Ew).

November 18 403.8 27.2 6.5 7.613.2376.5242848 1 8 9
December 21 211.2 44.315.610.118.6166.9352342 213 6
January 21  28.8 52.622.611.318.8−23.843213611 9 1
February 26 230.3 54.925.212.617.1175.4462331 118 7
March 23 224.5 33.313.7 9.7 9.9191.2412930 5 810
April and May 12  28.7 11.5 5.0 4.6 1.9 17.2444016 0 9 3


Density as a scalar of transpiration

Strong correlations between Ec and Eeco for days with similar rainfall suggest that wood density is an appropriate scalar for water use in this forest. Such high r2 values (0.69 and 0.80 for dry and wet days, respectively) are encouraging, given that sap velocity and eddy covariance measurements were made at quite different scales. Eddy covariance provides an integrated measure of ecosystem evaporation within the upwind footprint (2–5 ha), the size of which depends on wind speed and stability. Scaled transpiration rates estimated using measurements of sap velocity, by contrast, provide integrated estimates of transpiration over the same 50 × 50 m experimental plot (assumed to be typical of the whole forest) at a daily scale. This means that the rates of evaporation measured by the two approaches was never from exactly the same trees (Oren et al., 1998; Unsworth et al., 2004).

Regulation of components of forest evaporation

While sap velocity was strongly related to wood density for trees in exposed positions (as previously reported), transpiration at the stand scale was strongly regulated by air saturation deficit. Ec increased linearly with Dzc up to 0.6 kPa, with 67% of variation in Ec explained by variation in Dzc (Fig. 6). On the 9 d for which Dzc was greater than 0.6 kPa, Ec was unrelated to Dzc, suggesting stomatal regulation of transpiration. Stomatal regulation of transpiration when evaporative demand is relatively low in the absence of root-zone soil water limitation implies hydraulic limitation, possibly linked to the negative relationship between wood density and sap velocity. Hydraulic limitation has been documented in forests elsewhere (Ryan & Yoder, 1997; McDowell et al., 2002) but further work, in particular measurement of water potential gradients and hydraulic conductivity, is needed to quantify the extent of this limitation at our forest site.

On dry days, when wet canopy evaporation was assumed to be negligible, transpiration from the subcanopy and understorey trees was estimated to be as high as 1.2 mm d−1. Es tended to be positively related to both daylength-normalized air saturation deficit and average midday wind speed, both measured near the forest floor, although the relationship was significant only for umg (Table 2). This suggests that, unlike canopy trees, the transpiration rate of vegetation below the canopy was regulated by the extent of mixing of within-canopy air, rather than air saturation deficit of air above the canopy. Increased transport of drier, above-canopy air into the canopy on days with greater development of turbulent eddies would drive higher below-canopy transpiration as the leaf boundary layer would be reduced and the air saturation deficit of within-canopy air would be increased (Jarvis & McNaughton, 1986; Baldocchi & Meyers, 1991).

On wet days, when understorey transpiration could be assumed to be negligible, wet canopy evaporation was found to be rather higher than Ec, and up to 2.6 mm d−1. As expected, Ew was positively related to Dzg, QT and umg. This suggests that, as in other high-rainfall environments, wet canopy evaporation is driven by both advective and radiative energy (Pearce et al., 1980; Asdak et al., 1998).

Water balance for the forest

The maximum rate of canopy transpiration observed in the forest was 1.8 mm d−1, which is considerably lower than maximum Ec for coniferous boreal forest (2.8 mm d−1; Cienciala et al., 1998), and temperate maritime pine forest (3.5 mm d−1; Granier et al., 1990), but higher than old-growth temperate conifer stands (0.4–1.5 mm d−1, mean 0.9 mm d−1; Zimmermann et al., 2000; Irvine et al., 2002; Moore et al., 2004). Leaf area index has been considered to be among the most important factors in determining differences in transpiration between stands (Running & Coughlan, 1988; Granier et al., 1996; Santiago et al., 2000), although recent work comparing young and old-growth stands with similarly high L (ranging between 9.5 m2 m−2 and 12.1 m2 m−2) showed threefold higher maximum canopy transpiration rates in the young stand (Moore et al., 2004). The strong nutrient limitation, resulting in low leaf area index of the forest studied here, clearly contributes to the relatively low value of maximum Ec, although low stomatal conductance (Barbour & Whitehead, 2003) is also important. It is interesting that the dominant trees in the ecosystem had low stomatal conductance and transpiration rates, given ample root zone water. Similarly low canopy transpiration rates were found at comparable daylength-normalized air saturation deficits in a flooded Taxodium distichum forest in North Carolina, USA (Oren et al., 1999). Both our forest and that studied by Oren et al. (1999) have low leaf area index –2.9 m2 m−2 and 2.2 m2 m−2, respectively – and low maximum stomatal conductance –0.30 mol m2 s−1 (Barbour & Whitehead, 2003) and 0.175 mol m2 s−1 (Oren et al., 2001), respectively – presumably as a result of low soil nutrient availability.

As expected, the combination of L and high rainfall in the low-productivity old-growth forest studied here resulted in canopy transpiration contributing considerably less to total ecosystem evaporation than in most other forests studied. The percentage of Eeco partitioned to upper canopy transpiration was just 0.51 and 0.22 on dry and wet days, respectively. This compares with 0.69 and 0.91 in Pinus pinaster forests (Granier et al., 1990 and Berbigier et al., 1996, respectively), 0.77 in a Fagus sylvatica forest (Granier et al., 2000) and 0.69 in a forest dominated by Pinus taeda (Oren et al., 1998). On an annual timescale, Hutley et al. (1997), found that canopy transpiration contributed 41% of total annual evaporation from a subtropical rainforest, a value much closer to that observed in the current work although annual rainfall was considerably (about one-third) lower at Hutley et al.'s (1997) site.

The comparatively low proportion of Eeco contributed by Ec implies that other evaporative components, such as evaporation from the forest floor and wet surfaces within the canopy, are significant in the water balance for this forest. Estimates of transpiration from the subcanopy and understorey layers suggest that these trees and shrubs may make a significant contribution (up to 1.22 mm d−1 and 22% of Eeco, on average) to Eeco, despite forming just 10% of total foliage area. As expected, evaporation from wet surfaces within the canopy made a significant contribution to Eeco (51%) on days for which > 2 mm of rain fell. Epiphytic mosses form thick ‘aerial soils’ (up to 100 mm) on many stems and branches and provide evaporative surfaces within the canopy that would contribute to Ew. Bryophytes, lichen and standing dead wood are a feature of old-growth forests but their contribution to total evaporation has seldom been quantified (Unsworth et al., 2004). The thick, nearly continuous, moss layer on the forest floor and high water table meant that forest floor evaporation also made a significant contribution to Eeco (25% over the summer months). The importance of Es, Eg and Ew contributions, relative to the contribution from Ec, to total ecosystem evaporation reflects the unique attributes of the forest. Compared with most other forests for which partitioning of ecosystem evaporation has been attempted, the high rainfall at our study site resulted in the forest floor and surfaces within the canopy remaining wet for long periods, and the low L meant that irradiance and turbulent eddies penetrated into the canopy resulting in relatively high evaporation rates from the forest floor and wet surfaces, and high understorey transpiration rates.


Scaling of water use from individual trees to the stand in a D. cupressinum-dominated forest was achieved using an approach incorporating differences in sapwood density and canopy position between trees. The scaling technique, based on the theoretically predicted relationship between sap velocity and wood density, may be generally appropriate for coniferous forests, especially where there is a range in tree size and age. By comparing ecosystem evaporation measured using eddy covariance with sap-flow-scaled canopy transpiration and separating data for conditions when the canopy was dry and wet, partitioning of Eeco into components was possible. In this temperate rainforest, with high annual rainfall and low productivity associated with low nutrient supply, transpiration from the tree canopy on dry days accounted for 51% of Eeco, while understorey transpiration accounted for 22%. By contrast, on wet days, canopy transpiration accounted for 22% of Eeco, while evaporation from wet surfaces within the canopy contributed 51%.


This work was funded by the Foundation for Research, Science and Technology, and a Landcare Research Investment Postdoctoral Fellowship to M.M.B. Timberlands West Coast and the Department of Conservation provided the site and support for the work. We thank Dr B. J. Bond and two anonymous reviewers for comments that significantly improved the manuscript.