Introduction

Forests cover 67 % of Japan (Fujimori 2006). Coniferous plantations account for 41 % of this forested area and comprise mainly Japanese cedar (Cryptomeria japonica) and Japanese cypress (Chamaecyparis obtusa). Most of these plantation forests were developed after World War II, and were managed before ~1980. However, such management was abandoned during harsh economic conditions experienced by the Japanese forestry industry as a result of increased competition from inexpensive imported timber (Japan Forestry Agency 2010). Recently, Japanese people have started expect these plantation forests to function as water sources, and to provide flood control and carbon fixation rather than timber production (Kuraji 2012). Their function as water supplies is regarded as especially important. Indeed, many local governments have introduced taxes on forest-management practices, for example thinning, to enhance their function as water supply (Imawaka and Sato 2008).

To enhance the function of water supply by forest management we need information about the relationship between forest structure (e.g., diameter at breast height (DBH) and stem density) and canopy transpiration (E C) for two reasons. First, E C is a major component of evapotranspiration and therefore substantially affects runoff (Wilson et al. 2001; Ford et al. 2007; Oishi et al. 2008). Second, forest management can change E C by changing forest structure. Indeed, several studies in other countries have reported different E C for stands with different structures (Cienciala et al. 1997; Zimmermann et al. 2000; Moore et al. 2004; Ewers et al. 2008; Macfarlane et al. 2010). However, results of these studies are not consistent. Vertessy et al. (2001) reported that differences between E C among stands corresponded to differences between stand sapwood area among mountain ash forests in southern Australia, and Irvine et al. (2002, 2004) reported that differences between E C corresponded to the differences between sap flux density per unit sapwood area among ponderosa pine forests in a water-limited semi-arid climate. This variability suggests the relationship between forest structure and E C could differ among regions or species.

Information on the relationship between forest structure and E C for major species in Japan is still quite limited, because very few studies have examined E C or its substitutes (e.g., the sum of E C, soil evaporation, and understory evapotranspiration) for plantation forests with different structures by use of a consistent method. Most other studies examining E C were conducted on a specific stand with no management practice or forest growth (Hattori et al. 1993; Tanaka et al. 1996; Kosugi and Katsuyama 2007), although Morikawa et al. (1986) and Kumagai et al. (2007, 2008) are exceptions. Morikawa et al. (1986) measured E C, on the basis of sap flow measurements, for a Japanese cypress forest before and after thinning. Among forest structural data, stem density differed most before and after thinning (1750 and 1325 trees ha−1, respectively); E C after thinning was 79 % of that before thinning. Kumagai et al. (2007, 2008) measured E C for two Japanese cedar stands of equal age but different mean DBH and stem density (23.8 and 40.3 cm and 1575 and 904 trees ha−1, respectively). The E C for the former stand was 68 % of that for the latter stand. Although Morikawa et al. (1986) and Kumagai et al. (2007, 2008) examined differences between E C for stands with different structures, differences between stand structures are more pronounced for stands with contrasting age; small DBH and high stem density are observed for young stands. Therefore, examination of the differences between E C for stands of different age would be useful for clarifying the relationship between stand structure and E C.

In this research, on the basis of sap flux measurements obtained by use of the thermal dissipation method (Granier 1987; Oren and Pataki 2001; Ford et al. 2010), we measured E C for Japanese cypress stands with contrasting DBH and stem density, as a result of different stand age, and quantified differences between E C different stands. In this method E C is expressed as the product of mean stand sap flux density (J S) and stand sapwood area (A S_stand). We thus quantified differences between J S and in A S_stand for different stands to evaluate differences between E C. We examined differences between reference J S for given meteorological conditions and the responses of J S to meteorological conditions in different stands.

Materials and methods

Site description

The experiments were conducted in the Shintate region of the Kasuya Research Forest, Kyushu University, Japan (33°38′N, 130°31′E; 50–340 m a.s.l.). Mean annual precipitation, recorded for 1995–2005 at a meteorological observatory 15 km from the region, was 1790 mm. Mean annual temperature at the site was approximately 16 °C.

We established two plots of Japanese cypress approximately 50 m apart, called the Sakuta and Hiwada plots (340 m a.s.l.). DBH and stem density for these plots contrasted markedly because of the different stand age (Table 1). The Sakuta and Hiwada plots were located on slopes of approximately 20° facing southwest and south, respectively. Both pyroxenite and amphibolite rocks underlie the plots. The understory in the Hiwada plot was covered with Lauraceae and Theaceae species. The experiments were conducted from April 1, 2009 to March 31, 2010.

Table 1 Stand characteristics of the two plots
Full size table

Measurements

Environmental conditions

Meteorological conditions were measured at a meteorological station installed in a forest opening approximately 500 m from the plots. Solar radiation (R S) was measured by use of a pyranometer (LP PYRA 04; Delta OHM, Padova, Italy), and the air temperature (T a) and relative humidity (RH) were measured by use of a temperature and humidity probe (HMP155; Vaisala, Helsinki, Finland). Data were acquired every 10 s and stored in a data logger (CR10X; Campbell Scientific, Logan, UT, USA) at 10-min intervals. Precipitation (P) was measured by use of a tipping bucket rain gauge (TK-1; Takeda Keiki, Tokyo, Japan). Data were integrated every 10 min then stored in the data logger (CR10X). When meteorological data from the station were missing, the gaps were filled with data from the Ochozu experimental watershed, located approximately 4 km south of the station (Shinohara et al. 2009; Kume et al. 2010).

In addition to meteorological data, volumetric soil moisture content (θ; m3 m−3) was measured by use of a dielectric aquameter sensor (EC-10; Decagon Devices, Pullman, WA, USA) at one point in each plot at 10, 30, and 50 cm below the forest floor. Data were acquired every 30 s and stored in the data logger (CR10X) at 30-min intervals. These measurements were used to represent the average θ between 0 and 50 cm (θ 0–50).

Sap flux

Sap-flux was measured by use of the thermal dissipation method with Granier-type sensors (Granier 1987). Each sensor consists of a pair of probes 20 mm long and 2 mm in diameter. The probes were inserted into the sapwood approximately 150 mm apart vertically. The upper probe, which includes a heater, was supplied with a constant 0.2 W of power. The heat was dissipated into the sapwood and the vertical sap flux surrounding the probe. The temperature difference between the upper and lower probes was measured every 30 s and averaged every 30-min by use of a data logger (CR1000; Campbell Scientific) with a multiplexer (AM16/32; Campbell Scientific). The recorded temperature difference was converted to sap flux density (F d) as reported by Granier (1987). Electricity was supplied to the sensors via photovoltaic generation.

For sap flux measurements we selected ten of 21 trees in the Sakuta plot and all seven trees in the Hiwada plot (Table 2). One or two sensors were inserted in each sampled tree at depths of 0–20 and 20–40 mm to obtain measurements from the entire sapwood depth. For only two trees of the Hiwada plot, sapwood depth was 4.8 and 4.5 cm. Sapwood area of 4–6 cm of the two trees was less significant, compared with total sapwood area (<0.1 %). Thus, we did not measure the sap flux density at this depth. For some trees, the sapwood thickness was less than the probe length. For such trees the correction proposed by Clearwater et al. (1999) has often been used for correction of F d data. However, on the basis of Granier’s assumption that the Granier-type sensor integrated one section of the probe intercepting active F d and the other section in contact with non-conducting xylem, we did not use the correction, in common with previous studies on Japanese cedar (Kumagai et al. 2005a, 2007, 2008; Shinohara et al. 2013).

Table 2 Sensor installation
Full size table

To consider circumferential variation of sap flux (Lu et al. 2000; Tateishi et al. 2008; Tsuruta et al. 2010), we measured sap flux in the outer 20 mm of xylem on the south side for four trees at the Sakuta plot, and on the south, east, and west sides of two trees at the Hiwada plot. Sap flux in the outer 20 mm of xylem was averaged over the two or four circumferential values. For trees for which circumferential variations of F d was not measured, we used F d data measured in a single direction. Although we did not measure circumferential variations of F d for all the trees, previous studies reported that tree-to-tree variation of F d was more significant than circumferential variation of F d in E C estimates (Kume et al. 2012; Shinohara et al. 2013). We analyzed the significance of circumferential variation of F d in this study. First, we compared the coefficient of variation (CV) for circumferential variation of F d and tree-to-tree variation of F d for both the Sakuta and Hiwada plots. In the Sakuta plot, CV of within-tree variation of F d for four trees averaged 9.6 %. CV of tree-to-tree variation of F d for all the 10 trees was 20.2 %. In the Hiwada plot, CV of within-tree variation of F d was 15.1 % on average. CV of tree-to-tree variation of F d for all the seven trees was 19.0 %. For both plots, circumferential variation of F d was less significant than tree-to-tree variation of F d. We also examined the effects of circumferential variation of F d in E C estimates. We estimated E C for the plots by use of the equation (Granier et al. 2000; Ewers et al. 2005; Herbst et al. 2007; Kume et al. 2010):

$$ E_{\text{C}} = J_{\text{S}} \frac{{A_{{{\text{S}}\_{\text{stand}}}} }}{{A_{\text{G}} }} $$
(1)

where J S is the mean stand sap flux density, A S_stand is the total sapwood area of the stand, and A G is the ground area. J S was calculated by use of the equation:

$$ J_{\text{S}} = \frac{{\sum\nolimits_{i = 1}^{n} {Q_{{{\text{t}}_{i} }} } }}{{\sum\nolimits_{i = 1}^{n} {A_{{{\text{S}}\_{\text{tree}}_{i} }} } }} $$
(2)

where Q t is the whole-tree transpiration, A S_tree is the tree sapwood area, and n is the number of trees sampled. Q t was calculated as the sum of the product of F d, measured at different directions, and the depths and sapwood area at each xylem band.

We calculated E C (E Ccal) assuming circumferential measurements of F d for the entire tree and compared E Ccal with measured E C. In the E Ccal computation, we calculated correction coefficients of circumferential variations in F d which calculate the mean F d for two or four directions from a single direction (i.e., north direction) F d by use of the observed data, and applied the coefficients to the other trees. As a result, relative errors between E C and E Ccal were less than 1 % for both plots. This result indicates that circumferential variations in F d were less significant in E C estimates.

Sapwood area

Sapwood thickness (mm) of each sample tree was measured by use of a ruler on a core extracted with a 5-mm increment borer approximately 1.3 m above the ground, and assessed as the mean from two orthogonal measurements. Distinct color differences were used to identify the boundary between sapwood and heartwood. The individual sapwood area (A S_tree) was obtained from the difference between the stem cross-sectional area beneath the bark and the heartwood area, assuming that the stem cross-sections were circular. These measurements were conducted on 21 trees and 7 trees in the Sakuta and Hiwada plots, respectively.

Methods of analysis

First, we quantified the difference between E C for the Sakata and Hiwada plots. Second, we evaluated contributions of the differences between A S_stand and J S to the differences between E C for the plots on the basis of Eq. (1). Third, we evaluated contributions of the differences between the reference values of J S and the responses of J S to the meteorological variables to the differences between J S. Rearranging Eq. (1) and assuming the simplified Penman–Monteith equation (Jarvis and McNaughton 1986; Oren et al. 2001; Ford et al. 2010), J S is expressed as:

$$ J_{\text{S}} = \frac{{A_{\text{G}} }}{{A_{{{\text{S\_stand}}}} }}E_{\text{C}} = \frac{{A_{\text{G}} }}{{A_{{{\text{S\_stand}}}} }}G_{\text{C}} \cdot \frac{D}{{K_{\text{G}} }} $$
(3)

where G C is canopy conductance, D is the vapor pressure deficit, and K G is a unit adjustment constant (124.3 kPa m3 kg−1), which takes into account the effects of temperature on the psychometric constant, the latent heat of vaporization, and the specific heat and density of air (Phillips and Oren 1998). Although K G depends on air temperature, the dependence is weak and, therefore, we used the K G value when the air temperature is 20 °C. We expressed G C as (Oren et al. 2001; Ewers et al. 2008; Kumagai et al. 2008):

$$ G_{\text{C}} = G_{\text{Cref}} \cdot f_{1} (D) \cdot f_{2} (R_{\text{s}} ) $$
(4)

where G Cref is the reference value of G C at D = 1 kPa, and f 1(D) and f 2(R s) are functions describing the response of G C to D and R s, respectively. Here, we did not include functions describing the responses of G C to soil moisture and air temperature (Pataki and Oren 2003; Matsumoto et al. 2008), because there were no clear relationships between soil moisture, air temperature, and G C in the Sakuta and Hiwada plots. We assumed f 1(D) and f 2(R S) as (Granier et al. 2000; Ewers et al. 2008; Komatsu et al. 2012):

$$ f_{1} (D) = 1 - a\ln (D) $$
(5)
$$ f_{2} (R_{\text{s}} ) = \hbox{min} \,(R_{\text{S}} /b,\,1.0) $$
(6)

where a and b are empirical constants. If we define J Sref as \( J_{\text{Sref}} = \frac{{A_{\text{G}} }}{{A_{{{\text{S\_stand}}}} }} \cdot \frac{{G_{{{\text{Cre}}f}} }}{{K_{\text{G}} }} \) and \( J_{\text{Sresp}} = f_{1} (D) \cdot f_{2} (R_{\text{S}} ) \), J S is written as:

$$ J_{\text{S}} = J_{\text{Sref}} \times J_{\text{Sresp}} \times D $$
(7)

where J Sref and J Sresp are, respectively, the reference values of J S at D = 1 kPa and the response of J S to the meteorological conditions.

To evaluate the contributions of differences between J Sref and J Sresp to differences between J S and, in turn, to that between E C in the two plots, we performed the following analysis. We determined J Sref, a, and b for each plot on the basis of G C and meteorological data for the period between April 1 2009 and August 31 2009, when missing data were relatively few. G C was calculated as the daily average conductance using mean daytime T a, D, and E C, summed over 24 h but divided by daytime hours (0600–1800) to exclude the large variation caused by biases introduced under conditions of low sap flow and low D (Phillips and Oren 1998). Daily T a and D were obtained by averaging T a and D over daytime hours by assuming that this is the period in which T a and D affect transpiration and sap flow. E C, however, was summed over 24 h but divided by daytime hours, because this accounted for all water uptake driven by D over the entire day, also including nighttime recharge. We removed the G C data obtained on rainy days and when D < 0.5 kPa, because F d data was subject to noise under these conditions (Phillips and Oren 1998). All G C data under conditions with high irradiance (daytime R s > 400 W m−2) in the Sakuta and Hiwada plots were plotted against mean daytime D to determine the empirical constant a. Note that the threshold value was used to remove the lower scatter of G C data sufficiently. The upper ranges of the data were calculated as the mean plus one standard deviation of all G C values obtained, by use of boundary line analysis (Schäfer et al. 2000; Ewers et al. 2005; Herbst et al. 2008). Equation 5 was fitted to the upper ranges. We first regressed the relationship between D and G C, because D is a usually main force driving transpiration in coniferous forests (Monson and Baldocchi 2014) and could be a more determinant of transpiration than the empirical constant b. To determine the empirical constant b, we then examined the relationship between R s and observed G C divided by f 1(D) calculated with D as input.

We then calculated E C for the three hypothetical scenarios with input of meteorological data:

  1. 1

    J Sref, a, and b are the values for the Sakuta plot;

  2. 2

    J Sref is the value for the Sakuta plot, but a and b are the values for the Hiwada plot; and

  3. 3

    J Sref is the value for the Hiwada plot, but a and b are the values for the Sakuta plot.

The difference between E C for Scenarios I and II indicates the contribution of the difference between J Sresp to that between J S, and, in turn, to that between E C. The difference between E C for Scenarios I and III indicates the contribution of the difference between J Sref to that between J S, and, in turn, to that between E C.

Results

Daily meteorological data for the study period are shown in Fig. 1. Total P was 1224 mm. The range of θ in the 0–50 cm soil layer (θ 0–50 cm) was 0.17–0.24 and 0.12–0.22 for the Sakuta and Hiwada plots, respectively. T a tended to increase during the study period. No very clear trends were observed for R s and D during the period.

Fig. 1
figure 1

Time-series of environmental factors: a daily precipitation (P), b volumetric soil moisture content of the 0–50 cm soil layer (θ 0–50 cm), c mean daytime air temperature (T a), d mean daytime solar radiation (R s), and e mean daytime vapor pressure deficit (D)

Full size image

A S_tree for the Sakuta and Hiwada plots ranged from 70.4 to 134.1 cm2 (mean 97.0 cm2) and from 288.0 to 700.5 cm2 (mean 482.7 cm2), respectively (Fig. 2a), and A S_tree was related to DBH for both plots (Fig. 2b). A S_stand for the Sakuta and Hiwada plots were 20.4 and 15.2 m2 ha−1, respectively. Q t for the Hiwada plot was larger than that for the Sakuta plot. J S and E C for the Sakuta and Hiwada plots were estimated on the basis of these values. J S and E C for the Sakuta plot were both higher than those for the Hiwada plot throughout the study period (Fig. 3). The mean J S (±SD) for the measurement period was 0.6 ± 0.3 m3 m−2 day−1 for the Sakuta plot and 0.4 ± 0.2 m3 m−2 day−1 for the Hiwada plot. The mean E C (±SD) for the measurement period was 1.3 ± 0.7 mm day−1 for the Sakuta plot and 0.7 ± 0.4 mm day−1 for the Hiwada plot.

Fig. 2
figure 2

a Histogram of individual sapwood area (A S_tree) for the Sakuta (black bars) and Hiwada (gray bars) plots. b Relationships between diameter at breast height (DBH) and A S_tree for the Sakuta (open circles) and Hiwada (closed circles) plots. The solid lines are regression lines: A S_tree = 12.5 × DBH − 71.8 (R 2 = 0.70) for the Sakuta plot, A S_tree = 16.3 × DBH − 280.1 (R 2 = 0.83) for the Hiwada plot

Full size image
Fig. 3
figure 3

Time-series of a sap flux density (J S) and b canopy transpiration (E C)

Full size image

Figure 4 shows comparisons of E C, J S, and A S_stand for the Sakuta and Hiwada plots. E C, mean J S, and A S_stand for the Hiwada plot were 53, 69, and 75 %, respectively, of those for the Sakuta plot (Fig. 4a–c). Thus, differences between both J S and A S_stand for the stands contributed to the different E C.

Fig. 4
figure 4

Comparison of a canopy transpiration (E C) during the study period, b sap flux density (J S), and c stand sapwood area (A S_stand) for the Sakuta and Hiwada plots

Full size image

Figure 5 shows the relationships between D and G C for the Sakuta and Hiwada plots. G C was negatively correlated with D for the both plots. Under conditions with high irradiance (daytime, R s > 400 W m−2), the correlation coefficient was 0.79 for the Sakuta plot and 0.78 for the Hiwada plot. Figure 6 shows the relationships between R s and observed G C divided by f 1(D) for the Sakuta and Hiwada plots. Values tended to be approximately 1 when R s exceeded approximately 400 W m−2. The model term in Eqs. (4), (5), and (6) were determined as J Sref = 1.9, a = 0.67, and b = 368.7 for the Sakuta plot and J Sref = 1.2, a = 0.72, and b = 388.8 for the Hiwada plot, by use of a nonlinear least-squares technique. E C predicted by use of these values was in good agreement with observed E C in the Sakuta and Hiwada plots (Fig. 7): y = 0.99x (R 2 = 0.75) for the Sakuta plot, y = 1.06x (R 2 = 0.84) for the Hiwada plot. The root-mean-squared error was 0.30 mm day−1 for the Sakuta plot and 0.16 mm day−1 for the Hiwada plot. Figure 8 shows the time series of predicted E C in the three scenarios described above. The mean E C for Scenarios 1, 2, and 3 during the study period were 1.2, 1.3, and 0.7 mm day−1, respectively. E C for Scenarios 1 and 2 were almost the same (Fig. 9). E C for Scenario 3 was markedly lower than those for Scenarios 1 and 2 because of the low J Sref value for the Hiwada plot. This indicates the difference between J S for the two plots was primarily caused by J Sref whereas the difference between J Sresp contributed little to the difference between J S.

Fig. 5
figure 5

Relationships between vapor pressure deficit (D) and canopy conductance (G C) for the a Sakuta and b Hiwada plots. Only data for solar radiation >400 W m−2 are shown (open circles). The upper ranges (solid circles) were the basis for the logarithmic curve fits (solid lines)

Full size image
Fig. 6
figure 6

Relationships between solar radiation (R s) and canopy conductance (G C) divided by the function expressing the dependency of G C on the vapor pressure deficit, f 1(D), for the a Sakuta and b Hiwada plots. The solid line indicates the regression line

Full size image
Fig. 7
figure 7

Comparison between modeled and observed canopy transpiration (E C) for the a Sakuta and b Hiwada plots. A 1:1 line is also shown

Full size image
Fig. 8
figure 8

Time-series of canopy transpiration (E C) for the three scenarios

Full size image
Fig. 9
figure 9

Comparison of canopy transpiration (E C) during the study period for the three scenarios

Full size image

Discussion and conclusions

A S_tree for the Sakuta and Hiwada plots was comparable with values reported by Kumagai et al. (2005b) and Tsuruta et al. (2011). Kumagai et al. (2005b) and Tsuruta et al. (2011) reported allometric relationships between DBH and A S_tree for 1226 and 109 Japanese cypress trees, respectively. Our A S_stand data were also in the range of those published by Tsuruta et al. (2011), who reported A S_stand data for nine Japanese cypress stands across Japan. Our results also agree with those of Kumagai et al. (2008), who conducted sap-flow measurements along a hill slope in a Japanese cedar forest in Kyusyu. They reported that the maximum E C was approximately 2.0 mm day−1 for the upper slope and 1.5 mm day−1 for the lower slope in the range of high D. Our maximum E C data were comparable with those reported by Kumagai et al. (2008).

E C for the Hiwada plot was 53 % of that for the Sakuta plot. This difference was larger than those reported by Morikawa et al. (1986) and Kumagai et al. (2008) for coniferous plantation forests in Japan (discussed in the “Introduction”). These results agree with the fact that the difference between E C values was caused by both A S_stand and J S in this study, whereas the difference was mainly caused by differences between A S_stand for both Morikawa et al. (1986) and Kumagai et al. (2008). This might occur because stand ages were different for our plots whereas the stand ages for the plots of Morikawa et al. (1986) and Kumagai et al. (2008) were nearly or exactly the same. Several previous studies also reported a lower J S for older stands (Irvine et al. 2002, 2004), which is probably because of the larger hydraulic resistance between the soil and canopy caused by the greater tree height (Ryan and Yoder 1997). The larger hydraulic resistance may reduce J S for a given soil-leaf water potential gradient.

The difference between J S was caused by the difference between J Sref, which indicates G Cref is also a factor determining the difference between E C in the Sakuta and Hiwada plots, because G Cref is calculated from J Sref. These results agree with those of Komatsu (2004). Komatsu (2004) collected E C (or its substitute, for example the sum of E C, soil evaporation, and understory transpiration) data for 26 coniferous forest stands. They found differences between E C were primarily caused by differences between the G Cref (or the reference value of surface conductance), not by the differences between the responses of G C to D and R S. Furthermore, Oren et al. (1999) reported the responses of G C to D were reasonably conservative among 31 forests under conditions of non-limiting light, which also agrees with our results. Oren et al. (1999) also reported that many tree species have a = 0.6, which results from regulation of the minimum leaf water potential (ψ l) to prevent xylem cavitation in the response to D. Although soil moisture in the Hiwada plot was lower than in the Sakuta plot (Fig. 1), the values of a both for the Sakuta and Hiwada plots (0.67 and 0.72, respectively) were within the 95 % confidence intervals of 0.6, which supports Oren’s hypothesis. The value of a is not probably affected by soil moisture conditions.

Previous studies of coniferous plantations in Japan (Morikawa et al. 1986; Kumagai et al. 2007, 2008) reported that differences between E C among stands with different structures were mainly caused by the differences between A S_stand, not J S. If this was always true, we could predict different E C among stands with different structures by use of different structural data. Note that a method exists for estimating A S_stand with input of stem density and DBH (Tsuruta et al. 2011). However, our results were different from those of previous studies: the difference between E C for the stands was caused by both J S (specifically J Sref) and A S_stand. Thus, our results suggest the possibility that different J S among stands should be considered for prediction of differences between E C among stands with different structures. To examine this possibility, we recommend further studies investigating the similarity of and differences between E C and J S among coniferous plantation forests with different structures in Japan.