The Dynamics of Transpiration to Evapotranspiration Ratio under Wet and Dry Canopy Conditions in a Humid Boreal Forest

Abstract

Humid boreal forests are unique environments characterized by a cold climate, abundant precipitation, and high evapotranspiration. Transpiration ($E_T$), as a component of evapotranspiration (E), behaves differently under wet and dry canopy conditions, yet very few studies have focused on the dynamics of transpiration to evapotranspiration ratio ($E_T/E$) under transient canopy wetness states. This study presents field measurements of $E_T/E$ at the Montmorency Forest, Québec, Canada: a balsam fir boreal forest that receives $\sim 1600$ mm of precipitation annually (continental subarctic climate; Köppen classification subtype Dfc). Half-hourly observations of E and $E_T$ were obtained over two growing seasons using eddy-covariance and sap flow (Granier’s constant thermal dissipation) methods, respectively, under wet and dry canopy conditions. A series of calibration experiments were performed for sap flow, resulting in species-specific calibration coefficients that increased estimates of sap flux density by $34\%\pm 8\%$, compared to Granier’s original coefficients. The uncertainties associated with the scaling of sap flow measurements to stand $E_T$, especially circumferential and spatial variations, were also quantified. From 30 wetting–drying events recorded during the measurement period in summer 2018, variations in $E_T/E$ were analyzed under different stages of canopy wetness. A combination of low evaporative demand and the presence of water on the canopy from the rainfall led to small $E_T/E$. During two growing seasons, the average $E_T/E$ ranged from $35\%\pm 2\%$ to $47\%\pm 3\%$. The change in total precipitation was not the main driver of seasonal $E_T/E$ variation, therefore it is important to analyze the impact of rainfall at half-hourly intervals.

1. Introduction

Boreal forests occupy around a third of the world’s forest biomes [1] and represent the second largest vegetated area behind tropical forests [2]. Given its large size, the boreal forest regulates water fluxes over a vast area and thus impacts global climatology and hydrology [3,4]. Understanding the interactions between this ecosystem and the atmosphere is crucial [3,5], as it is particularly vulnerable to climate change [6]. Among the anticipated changes, the boreal biome is expected to experience a large increase in temperatures [7,8] and a modest increase in precipitation [7]. Although evapotranspiration is more sensitive to changes in precipitation than temperature [9], a possible shift in the geographical distribution of conifer tree species due to climate change could alter the composition of boreal forests (e.g., more deciduous tree species) and result in greater evapotranspiration in the summertime [2].

Among the water exchanges between forest ecosystem and atmosphere, evapotranspiration (E) is certainly the least well characterized, making its study of utmost importance [10,11,12]. In forest ecosystems, E is the transfer of water to the atmosphere through the combination of (i) evaporation from the soil and understory surfaces evapotranspiration ($E_G$), as well as evaporation from wet overstory canopy surfaces ($E_C$), and (ii) vegetation transpiration ($E_T$) [13,14,15]. In this paper, $E_T$ is defined as the overstory canopy transpiration. The ratio of transpiration to evapotranspiration ($E_T/E$) on a global scale depends on the leaf coverage over a given land surface and the growing stage of the forest stands [16]. Generally, $E_T$ represents the largest fraction of total boreal forest E during the growing season [17], except in treed peatlands [18,19] and in dry sparse forests where $E_G$ is a significant proportion of E [18]. Despite their importance, in situ measurements of $E_T/E$ are rather minimal in boreal forests [17,18,19], especially in wet regions of the biome.

A recent study by Isabelle et al. [20] at a humid boreal forest in eastern Canada found the mean cumulative annual E to be $\approx 550$ mm, the highest amongst a total of 15 sites across the boreal biome, despite representing a rather modest fraction of the mean annual precipitation ($\approx 1600$ mm). In humid boreal forests, where tree growth and $E_T$ rate are not limited by water availability, one may expect a relatively large $E_T/E$ ratio. On the other hand, frequent precipitation implies a low vapor pressure deficit and reduced solar radiation, both not favoring E nor $E_T$[21,22]. Further, rain droplets covering the foliage have also been shown to be an important physical factor limiting $E_T$[22]. In brief, we still lack a proper understanding of the evolution of the $E_T/E$ ratio in sites subjected to frequent variations in canopy wetness.

Another key issue relates to the dynamics of both E and $E_T$ under different canopy wetness conditions. Nearly all of the studies on $E_T/E$ of forests biomes reported to date have contrasted wet versus dry canopy conditions only [23,24,25,26], with the exception of Aparecido et al. [22], where the authors have investigated the variations of $E_T$ under different canopy wetness (dry, semi-dry and wet) conditions. Yet the vertical distribution of leaf wetness within a forest canopy is a very important factor for simulating $E_T$, especially in models that use the single big leaf approach [13].

The most common method for estimating tree $E_T$ is to measure water flux in sapwood, commonly referred to as sap flow [27,28,29]. Sap flow monitoring can quantify the whole-plant water use continuously, regardless of the species, canopy, and terrain heterogeneity [30,31,32]. On the basis of the SAPFLUXNET database, the thermal dissipation method [33,34] appears to be by far the most popular technique to measure sap flow [35]. This method monitors the temperature difference between two probes inserted in sapwood to quantify sap flux density ($F_d$) for a given sapwood depth and height [23,28]. Results can then be scaled up to the whole tree and extrapolated to obtain tree-stand $E_T$ [24].

Caution should be taken when estimating sap flow with this approach, as multiple sources of uncertainty exist [36]. First, one should use species-specific calibration factors in the calculation of $F_d$[31,37,38]. Peters et al. [36] found that 90% of thermal dissipation sap flow studies use the calibration coefficients from the original experiment by Granier [33]. However, several studies have reported variability in the empirical coefficients across different tree species (e.g., [31,36,39]). Other uncertainties may arise from sensors being partially inserted into non-conducting heartwood [40], circumferential variations of $F_d$ [41,42,43], raw signal processing to determine maximum temperature differences between two probes during zero flow [29,44], and the upscaling and extrapolation processes from sapwood scale to tree-stand scale [23,45].

If sap flow measurements are the norm to estimate $E_T$, the eddy-covariance method is the most direct and well-accepted approach to monitor total E over forest stands (e.g., [23,46,47]). It has been regularly paired with sap flow measurements to estimate $E_T$ and E independently (e.g., [19,23,48]). Comparing the results from tree-based sap flow measurements with eddy-covariance, which has a much larger spatial scale (footprint area ${10}^4$–${10}^6$ m²), can, however, be a challenge [49]. The heterogeneity between trees on which sap flow measurements are being conducted can lead to poor representation of the $E_T$ rate within the eddy-covariance measurement footprint [50].

The main objective of this study is to assess the impact of high precipitation on the dynamics of $E_T/E$ from half-hourly to seasonal time scales in a humid boreal forested site. In order to achieve this goal, we have two specific objectives. The first is to measure the $E_T$ of balsam fir (Abies balsamea (L.) Mill.) trees, notably by calibrating Granier’s approach for this tree species, and then by analyzing the multiple sources of uncertainty in the upscaling process. The second specific objective is to monitor the state of the canopy around rainfall events and link this to the dynamics of E and $E_T$.

2. Materials and Methods

2.1. Study Site

This study was conducted in a region representative of the humid boreal forest, namely Montmorency Forest in eastern Canada (47${}^{\circ }$17${}^{\prime }$18${}^{\prime }$${}^{\prime }$ N; 71${}^{\circ }$10${}^{\prime }$05.4${}^{\prime }$${}^{\prime }$ W) [20]. This region is under the influence of a continental subarctic climate with a short and cool growing season occurring between June and October [51]. The mean annual temperature and precipitation are $0.5 °$C and 1583 mm ($61\%$ rain, $39\%$ snow), respectively [52]. The site is located within the balsam fir-white birch bioclimatic domain, the dominant vegetation being balsam fir (Abies balsamea (L.) Mill.) with sparse occurrences of white birch (Betula papyrifera Marsh), white spruce (Picea glauca (Moench) Voss) and black spruce (Picea mariana Mill.) [53,54].

The measurement sites were located around two eddy-covariance towers in an experimental watershed (Figure 1a) called the “Bassin Expérimental du Ruisseau des Eaux-Volées” (BEREV) [53,54,55,56]. The first tower is surrounded by a young tree stand that developed after an $85\%$ clear cut that occurred in 1993–1994 [51], hence this stand is referred to as the “Juvenile” site (Figure 1b,c). The second tower is located 1.3 km east of the first tower and is surrounded by a younger stand resulting from logging that happened progressively between 2000 and 2010 and is designated as the “Sapling” site (Figure 1d,e). The age difference between trees in Juvenile and Sapling sites allowed us to investigate the dynamics of $E_T/E$ in stands with different characteristics (

Table 1. Characteristics of balsam fir trees inside the 400-m² plot: tree density per hectare (extrapolated from the number of trees in 0.04 ha), canopy height (h), diameter at breast height (DBH), leaf area index (LAI) and sapwood area per unit ground area ($S_T$). Values are mean ± standard deviation.

Plot	Tree Density	h	DBH	LAI	${\mathit{S}}_{\mathit{T}}$
	[Number of Trees per ha]	[m]	[cm]	[m${}^2$ m${}^{-2}$]	[m${}^2$ m${}^{-2}$]
Juvenile
J1	6500	10.2 ± 3.2	10.2 ± 2.5	3.87	0.00253
J2	5000	11.6 ± 3.5	11.4 ± 3.8	3.35	0.00252
J3	6750	9.5 ± 2.7	8.9 ± 2.0	3.55	0.00223
Sapling
S1	9250	6.3 ± 1.2	6.8 ± 1.5	3.07	0.00157
S2	9250	5.6 ± 1.1	5.6 ± 1.2	2.96	0.00154
S3	6750	5.7 ± 1.5	4.6 ± 1.0	2.58	0.00147

). Specifically, the more mature trees at the Juvenile site were taller and had larger stem diameter and higher leaf area index compared to the younger trees of the Sapling site.

Three 400-m² circular plots were established around each flux tower. The plot locations were determined based on the relative flux footprint contribution. The flux footprint area depends on the direction of prevailing winds. Canopy height and DBH were measured for every tree inside the plots, whereas the leaf area index was measured under overcast sky conditions on a 5 m × 5 m grid using a plant canopy analyzer (model LAI-2000, Li-Cor Biosciences, Lincoln, NE, USA).

Measurements of sapwood width ($S_W$) were obtained by the destructive sampling (i.e., felling at a height of 1.4 m) of 15 balsam fir trees with different diameters located outside the measurement plots. The conducting sapwood and inactive xylem were identified from the cut segment of the stem following the method of Coyea et al. [57] to obtain the allometric relationship between $S_W$ or sap wood area ($S_A$) and DBH. These relationships are plotted in Figure 2 (with the associated coefficient of determination R²) and were used to estimate sapwood area per unit of ground area ($S_T$).

2.2. Eddy-Covariance and Micrometeorological Measurements

The Juvenile flux tower is 15-m high (≈1.5 times the mean canopy height) and is equipped with two identical sets consisting of a 3D sonic anemometer and an open path CO₂/H₂O analyzer (IRGASON, Campbell Scientific, Logan, UT, USA), both installed at a height of 14.63 m. The high-frequency sensors are facing opposite directions (${303}^{\circ }$, northwest; and ${118}^{\circ }$, southeast), so that time series from both devices can be combined depending on wind direction to minimize the effect of flow distortion by the tower structure. The Sapling flux tower is 10-m high and is equipped with a similar eddy-covariance system (IRGASON, Campbell Scientific, Logan, UT, USA) installed at a height of 8.5 m (≈1.5 times the mean canopy height). Each tower was also equipped with several meteorological instruments to measure net radiation (CNR4, Kipp & Zonen, Delft, The Netherlands), air temperature, and relative humidity (HC2S3 and HMP45C, Campbell Scientific, Logan, UT, USA). Total precipitation data was measured at a station located $\approx 4$ km north of the study sites and operated by the Québec government [58]. Additional tipping bucket rain gauges (ECRN-100, Decagon, Pullman, WA, USA) were installed on each site during the sap flow measurement campaign and the readings were used to overwrite the data from the govermental weather station during that specific period of time.

Raw eddy-covariance data were processed using EddyPro^® version 6.0 (Li-Cor Biosciences, Lincoln, NE, USA). Data processing routines followed the standard Fluxnet procedure, except for the coordinate rotation that used a sector-wise planar fit [59]. Notably, periods of rainfall were filtered out since rain droplets can obstruct the lenses of open-path gas analyzers. Gaps in the E time series caused by filtering procedures were filled using marginal distribution sampling as described in Reichstein et al. [60]. Remaining gaps were filled using monthly linear regression with zero-set intercept between E and net radiation.

2.3. Sap Flow Measurement

Sap flux densities were measured using commercially available Granier-type constant thermal dissipation probes (TDP-30, Dynamax, Houston, TX, USA) during two full growing seasons, from 5 July until 18 October in 2017 and 2018, when trees are actively transpiring. The starting date was a bit late due to several technical issues after the installation process at the beginning of April 2017. Each probe consists of a pair of 30-mm long needles (1.2 mm in diameter) installed one above the other. The upper one includes an electric heater and a thermocouple junction referenced to a junction in the lower needle [31]. The two needles were inserted radially into the sapwood and vertically separated by approximately 40 mm as suggested by the manufacturer [61]. The upper needle was heated with a constant voltage of 3 V using a voltage regulator (AVRD, Dynamax, Houston, TX, USA). The temperature difference ($\Delta T$ which is measured as electrical potential difference, in mV) between the two needles was measured every 60 s and recorded as 10-min averages on a CR10X (Campbell Scientific, Logan, UT, USA). $\Delta T$ is related to sap flux density ($F_d$): simply put, increasing sap flow decreases $\Delta T$ by cooling the heated needle [28,33].

Thermal dissipation probes were installed on balsam fir trees having stem diameter similar to that corresponding to the average DBH (Figure 2) at each site. Four adjacent trees were selected in each plot surrounding the Juvenile and Sapling flux tower. For three out of four trees, the needles were inserted on the north side of the tree to avoid heating from solar radiation. The fourth tree of each plot was equipped with probes on both the north and south sides to investigate potential circumferential variation. The probes were installed at a height of 1.4 m and 0.7 m above the ground at the Juvenile and Sapling plots, respectively. They were covered with a reflective insulation coating to prevent exposure to rain and direct sunlight [62,63], and to minimize the influence from ambient thermal gradient [28]. Adequate insultation is crucial to minimize uncertainties related to these three sources. The tree stems were covered from $\approx 15$ cm above the upper probe down to the ground as suggested by Lu et al. [28].

Observed $F_d$ is underestimated if part of the needles extends beyond the conducting sapwood depth, in which case the following correction has to be applied [40]:

$$\Delta T_{sw}=\frac{\Delta T-b\Delta T_{max}}{a}$$

(1)

where $\Delta T_{sw}$ is the temperature difference in the conducting sapwood only [mV], a and b are the fraction of needle in the conducting sapwood and the inactive xylem, respectively (i.e., $a+b=1$), and $\Delta T_{max}$ is the maximum temperature difference between the two needles occurring when the flux is 0 for a given time period. This correction assumes that the thermal properties of inactive xylem are the same as those of the conducting sapwood when $F_d=0$[40]. Values of a and b for each sampled tree were estimated using the aforementioned relationship between DBH and sapwood width (Figure 2).

The $F_d$ [cm³ cm${}^{-2}$ h${}^{-1}$] was calculated using the Granier [33] power-type relationship with the flux index (K):

$$K=\frac{\Delta T_{max}-\Delta T_{sw}}{\Delta T_{sw}}$$

(2)

$$F_d=\alpha K^{\beta }$$

(3)

where $\alpha$ and $\beta$ are calibration factors. Granier [33] found a strong correlation between K and $F_d$ in two different conifer and one broad-leaf tree species, where $\alpha$ = 42.84 cm³ cm${}^{-2}$ h${}^{-1}$ (0.0119 cm³ cm${}^{-2}$ s${}^{-1}$× 3600) and $\beta$ = 1.231 [28,31,36]. Granier [33] stated that, based on the three species he studied, this empirical equation was not species-dependent [28]. However, other studies later found that for a given K, Granier’s factors could underestimate $F_d$ in several conifer tree species [36].

A combination of the environment-dependent method by Oishi et al. [29] and the daily maximum method by Granier [33] was used to estimate $\Delta T_{max}$ in Equation (2). The initial approach to define $\Delta T_{max}$ (daily maximum) was based on the assumption that $F_d=0$ occurs every night, thus leading to the determination of $\Delta T_{max}$ on a daily basis [44]. Unlike the original approach, the environment-dependent method determines $\Delta T_{max}$ using actual environmental conditions by selecting the highest daily $\Delta T$ observed during conditions of low vapor pressure deficit (D). The advantage of the environment-dependent method is its ability to take into account seasonal shifts in $\Delta T_{max}$ and nocturnal water flux. However, this method requires high-humidity conditions (i.e., $D<0.05$ kPa), which were not met on certain days. For these specific days, the daily maximum method was used as recommended by Rabbel et al. [44] for humid environments.

Ultimately, stand transpiration [mm h${}^{-1}$] was calculated as:

$$E_T=\overline{F_d}{S}_T$$

(4)

where $\overline{F_d}$ is the mean sap flux density of sampled trees [converted in L m${}^{-2}$ h${}^{-1}$, which is equivalent to mm h${}^{-1}$] and $S_T$ is the sapwood area per unit of ground area [m${}^2$ m${}^{-2}$] [23,64]. The estimation of $S_T$ can be obtained by applying the allometric relationship between DBH and $S_A$ (Figure 2) to all trees within the plots [64].

2.4. Species-Specific Calibration of the Thermal Dissipation Approach for Sap Flow Measurements

In order to obtain the most accurate measurement of $E_T$, sap flow calibration was conducted in an environmentally-controlled laboratory by comparing gravimetric $F_d$ with measurements of K using constant thermal dissipation probes [36]. Stem segments used for the calibration experiments were harvested from three different balsam fir trees of similar size (diameter $\approx 15$ cm) located within the study site. The tree trunks were cut in length of 25 cm, wrapped in a wet cloth, and stored separately in black plastic bags to prevent dehydration during transportation and storage. A razor blade was used to trim both cut surfaces and to remove the top 2 cm of bark to ensure the water used for calibration only passed through the xylem [31]. Flow was induced from the top to the bottom of the reversed stem segment at a constant pressure-head using a Mariotte-based verification system, as described in Steppe et al. [31].

Before the start of the calibration procedure, each stem segment was covered by a plastic sheet to prevent dehydration and was maintained flow-less for 10 h to establish zero-flow conditions. K was measured by three constant thermal dissipation probes which were installed at the same height on three different sides of the stem segment ($0^{\circ }$, ${90}^{\circ }$, and ${180}^{\circ }$). Flow was induced in the stem segment during a 2-h period to saturate the sapwood and stabilize probe readings before the start of the measurements. Calibration was then performed at constant pressure heads of 2.5, 5, 10, 15, and 25 cm for 45 min each time. Water dripping out at the bottom of the cut stem segment was weighted using an electronic balance (PM2500, Mettler Toledo, Toledo, OH, USA) and logged at a 1-min frequency to determine sap flow. The calibration curve was obtained by fitting a power function between K and gravimetric $F_d$.

2.5. Monitoring of Canopy Wetness

Wet and dry canopy conditions were determined using leaf wetness sensors (PHYTOS 31, METER Group, Inc., Pullman, WA, USA) from 5 July until 18 October in 2018. Leaf wetness sensors (LWS) monitor fluctuations in the dielectric constant of the sensor’s upper surface to retrieve wetness [65]. A total of three LWSs were installed in plot J1 at three heights above the ground surface (2, 4, and 6 m) on branches $\sim 15$ cm from the tree trunk. While the number of LWS used in this study was rather limited and their shape was not optimal to represent needles, they provided, based on our results (including visual assessment), a reasonable approximation of the general state of the canopy wetness. The LWSs were mounted at a 45${}^{\circ }$ angle to simulate the typical position of the foliage and to prevent the accumulation of drops as recommended by the manufacturer [65]. The final output data is in raw counts (1 raw count = 1/$0.733$ mV measured using a datalogger with 3000 mV excitation), which ranged from 435 raw counts (dry) up to 1100 raw counts (saturated), and was stored on an EM50 Data Logger (METER Group, Inc., Pullman, WA, USA) at one minute intervals. Values $<445$ raw counts were considered as indicative of dry conditions and values $>445$ raw counts, of wet conditions.

Around a rainfall event, the canopy will undergo a “wetting–drying event” ($t_{wd}$), which consists of a wetting phase ($t_w$) and a drying phase ($t_d$). Data analysis included LWSs readings starting 30 min before the wetting phase ($t_{w-1}$) up to 30 min after the drying phase has ended ($t_{d+1}$). Based on the state of LWSs, the canopy wetness conditions were divided into four levels of wetness ($W_L$) as described in Figure 3. If all LWSs report dry conditions, the canopy is considered dry; if one of the LWSs report wet conditions, then the canopy is slightly wet; if two of the LWSs report wet conditions, then the canopy is fairly wet; and if all LWSs report wet conditions, then the canopy is considered wet.

3. Results

3.1. Determination of $E_T$

3.1.1. Calibration of Sap Flow Measurements for Balsam Fir

The calibration process yielded 45 data points from each of the three stem segments (three sap flow sensors with a mean K value at each of the five constant pressure heads). Figure 4 shows the relationship between gravimetric sap flux density ($F_d$) and sap flux index (K). A power-type function, as with the original equation by Granier [33], was used to obtain the calibration curve where $\alpha$ = 54.997 cm³ cm${}^{-2}$ h${}^{-1}$ and $\beta$ = 1.204 (R${}^2$ = 0.89). The coefficient of determination was slightly below that obtained by Granier [33] or other more recent sap flow calibration studies on coniferous-softwood type trees (e.g., [36,63]) where R²≥ 0.95, but was still deemed satisfactory. Note that this is the first time that calibration coefficients for the thermal dissipation method are being reported for balsam fir trees.

3.1.2. Circumferential and Tree-to-Tree Variations

Other uncertainties related to the scaling process are from $F_d$ variations across the azimuthal direction and between measured trees. Significant differences (p-value $<0.01$) in $F_d$ measured on the north and south sides of the trees were found in all plots during measurement period of 2017 and 2018. Nevertheless, none of the sampled trees (one in each plot) showed higher or lower $F_d$ towards a specific azimuthal direction during the measurement periods (Figure 5a). Moreover, there were no meteorological parameters, nor soil water content fluctuations, from 2017 to 2018, which could explain the increase in circumferential variation. Overall, the coefficients of variation (CV) of $F_d$ measured on north and south sides of the tree ranged from 32% to 47%.

Mean plot $F_d$ values in this study were computed from four measurement trees. The tree-to-tree $F_d$ variations within each site were significantly different (p-value $<0.01$) both in Juvenile and Sapling throughout two measurement periods. The deviation of each tree $F_d$ from the site mean was variable, as represented by 1.5× the interquartile range (IQR) in Figure 5b. The observed $F_d$ values of in Juvenile plots had slightly higher CVs, 52% to 57%, compared to those in Sapling plots, 44%.

3.2. Characteristics of Wetting–Drying Events

As many as 30 wetting–drying events were recorded from 5 June 2018 until 18 October 2018 at Juvenile site. A total of 482 mm of rain fell during this period. The wetting–drying events were characterized by lower net radiation ($R_n$; 59 ± 121 W m${}^{-2}$) and vapor pressure deficit (D; 0.09 ± 0.14 kPa) than during dry canopy conditions, 91 ± 186 W m${}^{-2}$ and 0.41 ± 0.34 kPa, respectively.

The length of each wetting–drying event ($t_{wd}$) was quite variable, ranging from 4 h up to 116.5 h, and was weakly correlated with the total amount of rain (R) during the event (R${}^2=0.46$). However, the relationship between the length of wetting phase ($t_w$) and R was quite strong (R${}^2=0.83$), as one would expect. In most cases, the wetting phase lasted less than 10 h, which was long enough to wet LSW at all three heights (Figure 6a,c). For 25 out of 30 events, rainfall led to a completely wet canopy, whereas in the remaining five cases, the canopy only reached fairly and slightly wet states (see

Table A1. Characteristics of each wetting–drying event described by the total rain (R) during event, length of the event ($t_{wd}$), length of wetting phase ($t_w$), length of drying phase ($t_d$) and contribution of each canopy condition in Juvenile during the measurement period of 2018.

Event	R [mm]	${\mathit{t}}_{\mathbf{wd}}$ [hour]	${\mathit{t}}_{\mathit{w}}/{\mathit{t}}_{\mathbf{wd}}$ [%]	${\mathit{t}}_{\mathit{d}}/{\mathit{t}}_{\mathbf{wd}}$ [%]	${\mathit{t}}_{\mathit{w}}$			${\mathit{t}}_{\mathit{d}}$
					Slightly Wet [%]	Fairly Wet [%]	Wet [%]	Wet [%]	Fairly Wet [%]	Slightly Wet [%]
1	11	17	15	85	-	-	100	100	-	-
2	7.8	35.5	23	77	-	-	100	93	-	7
3	13.2	29	14	86	-	-	100	98	2	-
4	78.2	116.5	27	73	-	-	100	91	8	1
5	4	19.5	5	95	-	-	100	81	19	-
6	16.6	15	27	73	-	-	100	95	5	-
7	3.6	28.5	9	91	-	-	100	75	4	21
8	3.8	10.5	10	90	-	-	100	58	21	21
9	11.2	24.5	18	82	-	11	89	90	5	5
10	6	21	17	83	-	-	100	54	29	17
11	0.8	4	13	88	-	-	100	29	14	57
12	50	20	40	60	-	6	94	83	13	4
13	31.2	30	48	52	-	3	97	81	13	6
14	1	23.5	6	94	-	-	100	82	2	16
15	4	23	15	85	-	-	100	67	5	28
16	8.2	20.5	12	88	-	-	100	81	6	14
17	9.4	26.5	26	74	-	-	100	97	3	-
18	1.2	19.5	5	95	-	-	100	95	-	5
19	9	13.5	22	78	-	-	100	100	-	-
20	5	40	23	78	-	-	100	87	2	11
21	12	16	28	72	-	-	100	65	35	-
22	0.6	25	4	96	50	-	50	56	-	44
23	27.2	25	52	48	-	-	100	79	13	8
24	43.4	47	50	50	-	-	100	51	34	15
25	8.8	18.5	35	65	-	-	100	38	17	46
26	2.4	15	17	83	-	-	100	52	20	28
27	4.6	26	13	87	-	-	100	53	7	40
28	6.8	44.5	10	90	-	11	89	34	1	65
29	69.6	114.5	29	71	-	-	100	71	1	28
30	8.6	10.5	71	29	-	-	100	100	-	-

for details).

On the other hand, the length of the drying phase ($t_d$) and the amount of time required to completely dry all three LWSs ranged from 3 h up to 84.5 h (Figure 6b,d). The LWS at 2 m required longer time to dry compared to LWSs at 4 m and 6 m. We used multiple linear regressions to estimate the length of drying phases ($t_d$ [hour]) with average net radiation ($R_n$ [W m${}^{-2}$]), vapor pressure density (D [kPa]), and wind speed (u [m s${}^{-1}$]) as predictors. The model took the following form:

$$t_d=0.056R_n-21.5D+14.1u$$

(5)

and had an R² value of 0.51 (p-value $=4\times {10}^{-5}$). To accurately estimate the drying phase duration requires another element, which is the amount of rain stored in the canopy. Not all of the rain is intercepted by the forest canopy as a fraction passes through canopy gaps and reaches the forest floor.

3.3. Dynamics of $E_T$ and $E_T/E$ during Wetting–Drying Events

Daily courses of E and $E_T$ during two typical wetting–drying events show that both variables decrease when the forest canopy is progressively wetted by precipitation (Figure 7). Surprisingly, a rainfall accumulation of only 0.4 mm received on 15:30 of 17 August 2018 was sufficient to put the canopy in a wet state and reduce E and $E_T$ by $6\%$ and $7\%$, respectively (Figure 7a). Meanwhile, 30.8 mm of rainfall received over $\sim 14$ h during daytime on 22 August 2018 reduced daily cumulative E and $E_T$ by $86\%$ and $116\%$, respectively, compared to the day after that rain event. Once the canopy started to dry, E gradually increased followed by $E_T$.

Figure 8a shows decreases in $E_T$ under various canopy wetness stages compared to that during dry canopy conditions. For instance, the average $E_T$ was $70\%$ lower under wet canopy conditions during the wetting phase ($t_w$) than during the half-hour before ($t_{w-1}$). Interestingly, under wet canopy conditions, $E_T$ values were slightly higher during the wetting phase (0.007 ± 0.009 mm) than those during the drying phase (0.004 ± 0.008 mm). During the wetting phase, the rain was able to wet all three LWSs resulting in a canopy wetness level categorized as “wet” although the canopy was not fully saturated. On the other hand, at the beginning of a drying phase, the canopy was mostly wet due to rain accumulation during the wetting phase. Once the canopy wetness reached fairly and slightly wet conditions, the mean $E_T$ increased to 0.010 ± 0.016 mm and 0.022 ± 0.028 mm, respectively. Thirty minutes after the canopy became dry ($t_{d+1}$), the average $E_T$ was $70\%$ higher than during the $t_{w-1}$ period.

Under wet canopy conditions, $E_T/E$ decreased in a similar fashion as $E_T$ (Figure 8b). $E_T/E$ declined from $0.44\pm 0.29$ half an hour before rainfall to $0.31\pm 0.32$ under wet canopy conditions in the wetting phase. Once the rain ceased and the canopy was still in fully wet conditions, $E_T/E$ further dropped to $0.16\pm 0.22$. Even if the canopy starts to dry, transitioning from “wet” to “fairly wet”, $E_T/E$ reached its lowest value ($0.14\pm 0.15$). The low $E_T/E$ values observed under fairly wet conditions were due to E increasing at a higher rate than $E_T$. Once the canopy was under slightly wet conditions, $E_T/E$ rose to $0.35\pm 0.31$. While the changes in $E_T/E$ ratio during wetting–drying events were similar to those of $E_T$, the difference of $E_T/E$ between the $t_{w-1}$ and $t_{d+1}$ periods was only $13\%$.

3.4. $E_T/E$ at the Seasonal Scale

Figure 9 presents a summary of seasonal values for evapotranspiration (E), transpiration ($E_T$) and transpiration to evapotranspiration ratio ($E_T/E$) measured at Juvenile and Sapling sites from 5 July to 15 October in 2017 and 2018. In 2017, Juvenile and Sapling sites received 509 mm and 462 mm of rainfall, respectively, whereas in 2018 the total rainfall was 482 mm at the Juvenile site and 452 mm at the Sapling site. Despite having received less rainfall in 2018, there was a significant decrease (∼21%) in E at the Juvenile site (from 246 mm in 2017 to 200 mm in 2018). In contrast, E at the Sapling site slightly increased from 211 mm in 2017 to 220 mm in 2018. The evaporative index, i.e., the ratio of evapotranspiration to precipitation ($E/P$) or rain ($E/R$), decreased in Juvenile site and increased in Sapling site between 2017 and 2018. The summary from two years of measurement period shows a quite similar $E/R$ ratio between Juvenile ($0.45$) and Sapling ($0.48$) site.

Transpiration from young balsam fir stands, the dominant vegetation at the study site, was not the major contributor to evapotranspiration. Only $0.42\pm 0.04$ and $0.28\pm 0.01$ of E was attributable to aboveground $E_T$ in Juvenile and Sapling sites, respectively, in 2017. In 2018 the proportion of $E_T$ to E increased by $25\%$ in Juvenile and $40\%$ in Sapling. Overall, the increases in $E_T/E$ between years were almost similar to those of $E/R$ with a gain of $18\%$ at the Juvenile site and $48\%$ at the Sapling site.

4. Discussion

4.1. Sources of Uncertainty in $E_T$

In this study, we measured tree transpiration using the thermal dissipation method and upscaled sap flux density ($F_d$) from sampled trees to stand transpiration ($E_T$). During the measurement and calculation process, we applied a series of corrections, including species-specific calibration to improve the accuracy and minimize the uncertainty in $E_T$. Our calibration using 25-cm long balsam fir stem segments resulted in a significantly higher $F_d$ value for a given K than that provided by Granier’s empirical equation (t-test, $\alpha$ = 0.05, p-value $<0.001$).

Previous studies have also shown that Granier’s [33] calibration coefficients may underestimate $F_d$ for several evergreen conifer species such as Pinus elliottii, Pinus palustris, Picea abies, and Pinus sylvestris [36,39,63]. Compared with these studies, our calibration curve resulted in $F_d$ values that were $5\%$–$23\%$ higher than those reported by Bosch et al. [39] and Peters et al. [36] for Pinus elliottii, Pinus palustris, and Picea abies at K values ranging between 0 and 0.4 (similar to that observed during the calibration process). Overall, our calibration coefficients increased $F_d$ values estimated using Granier’s coefficients by $34\pm 8\%$. This finding is similar to results by Lundblad et al. [63], who reported a $40\%$ increase in $E_T$ when using species-specific calibration factors for Picea abies and Pinus sylvestris trees.

Another uncertainty from sap flow measurements is the location of sensors installed on the tree trunk. This study was conducted in the Northern Hemisphere where the southern part of the tree canopy receives more sunlight, and hence is expected to transpire more than the shaded counterpart [66]. However, we found no systematic variation in $F_d$ between sensors installed on north versus south side of the tree. While this might be related to height differences among individual neighbor trees, thus creating a shading effect on measured trees, many studies have also found non-systematic circumferential variations in $F_d$ (e.g., [67,68,69]), yet the reason behind these variations has not been adequately identified [68,70]. These results suggest that the water lost through transpiration on one side of the tree crown might not come from the xylem on the same side. Indeed, for most types of conifers, the hydraulic transport network is composed of tracheids with bordered pits, which allow water to move easily in the tangential direction [71]. Nevertheless, Saveyn et al. [43] emphasized the importance of installing sap flow probes at multiple points around the stem circumference to reduce errors and obtain a more precise scaling of tree $E_T$.

Variations in $F_d$ were also found in a measurement plot from one tree to the other. We could not pinpoint any factor responsible for the tree-to-tree $F_d$ variation. However, several previous studies (e.g., [30,72,73]) showed that the variation in $F_d$ between trees was closely related to tree size which was affected by competition among trees for water and sunlight in the stand. The uncertainty caused by spatial variability between plots can be reduced by sampling more trees or plots [74]. The standard errors of the mean are then computed to obtain a value of uncertainty in the cumulative $E_T$ using summation in quadrature, commonly known as square root of the sum of squares.

4.2. Dynamics of $E_T$ and E during Wetting–Drying Events

In this study, the wetting–drying events were monitored using leaf wetness sensors (LWS) that had a different shape and much larger surface area than balsam fir needles. Leaf wetness sensors have been used to analyze the effect of canopy wetness on transpiration (e.g., [22,75,76]) due to its ease of use. Although the sensors are clearly not suitable to quantify the exact amount of water stored in the canopy during a wetting–drying event, they are used here to provide a qualitative description of the canopy wetness. We also must remind readers that due to logistical constraints, only one tree was monitored using three LWS installed at three different heights. Despite the aforementioned limitations, the LWS sensors were able to estimate the timing of wetting and drying phase, in line with the decrease and increase of $E_T$ and E related to the canopy wetness during all wetting–drying events. Three completely independent measurement methods (sap flow, eddy-covariance and LWS) were able to provide a rather clear picture of the phenomena taking place around a rainfall event. Nevertheless, to precisely quantify canopy wetness requires specific methods to measure the amount of rain intercepted by the canopy and its evaporation as explained in a study by Rutter et al. [77]. We did not include the method mentioned earlier in this study because of its complexity to estimate the canopy wetness in time series.

The decreases and increases in $E_T$ around wetting–drying events were closely related to net radiation ($R_n$) and vapor pressure deficit (D). The lower $E_T$ rate in the pre-wetting phase compared to that in the post-drying phase after the drying phase is due to the presence of clouds, resulting in low $R_n$, and low D. The $E_T$ rate is regulated by stomatal conductance, and in the case of abundant soil water, it depends on $R_n$ and D. Figure 10a,b show different responses of $E_T$ under dry canopy conditions to D during daytime ($R_n>10$ W m${}^{-2}$) and nighttime ($R_n<10$ W m${}^{-2}$). The relationship between $E_T$ and D was higher during daytime (R${}^2=0.58$) compared to nighttime (R${}^2=0.17$), because on several occasions, D was quite high over night while the sap flow had already reached zero.

Contrary with the study by Cienciala et al. [21] who found a strong correlation between $E_T$ and D under wet canopy conditions, the weak relationships between both parameters in this study suggest that the presence of rain drops on the tree needles do play a role in ‘regulating’ tree $E_T$ rate. A rainfall accumulation of 0.2 mm on dry canopies during the first half-hour of the wetting phase ($R\left(t_{w0}\right)$) was able to reduce $E_T$ on four out of 10 occasions. The reduction rate of $E_T$ caused by the rain was not clear until $R\left(t_{w0}\right)$ was greater than 1 mm (Figure 11). This could provide a first estimation of the amount of rain needed to saturate the canopy. However, forest canopies only intercept a fraction of the total rain, depending on the closeness of the canopies. A specific method is required to estimate the amount of rain that is intercepted by the canopy during rainfall events.

The strong correlation between E and $E_T$ under dry canopy conditions (R${}^2$ = 0.70) and weak correlation under wet canopy conditions (R${}^2$ = 0.23) is in line with the study of Granier et al. [23] (R${}^2$ = 0.85 under dry canopy conditions and R${}^2$ = 0.46 under wet canopy conditions). We did not find any time lag between both parameters under any canopy conditions, nor on the relationship between sap flow measurements and D (for details see

Table A2. Relationships of half-hourly $E_T$ measured using sap flow with E and D measured from the flux tower, under various canopy conditions and increasing values of time lag.

Time Lag [hour]		All	Dry	Wetting-Drying Events
				Wetting Phase	Drying Phase
					Wet	Fairly Wet	Slightly Wet
E [mm 30 min${}^{-1}$] vs. $E_T$ [mm 30 min${}^{-1}$]
R${}^2$	0	0.67	0.82	0.32	0.31	0.46	0.64
	0.5	0.66	0.78	0.25	0.32	0.34	0.66
	1	0.67	0.77	0.22	0.29	0.24	0.63
	1.5	0.65	0.72	0.21	0.28	0.17	0.55
	2	0.61	0.65	0.20	0.22	0.15	0.48
Slope	0	0.59	0.70	0.22	0.11	0.19	0.51
	0.5	0.58	0.69	0.18	0.11	0.16	0.51
	1	0.59	0.68	0.16	0.10	0.13	0.50
	1.5	0.58	0.67	0.15	0.10	0.11	0.46
	2	0.57	0.65	0.14	0.09	0.08	0.43
D [kPa] vs. $E_T$ [mm 30 min${}^{-1}$]
R${}^2$	0	0.54	0.48	0.32	0.22	0.38	0.68
	0.5	0.51	0.45	0.28	0.22	0.26	0.57
	1	0.48	0.41	0.26	0.20	0.18	0.49
	1.5	0.44	0.37	0.25	0.18	0.13	0.44
	2	0.39	0.33	0.24	0.17	0.13	0.41
Slope	0	0.08	0.08	0.12	0.04	0.06	0.11
	0.5	0.08	0.08	0.11	0.04	0.05	0.11
	1	0.08	0.08	0.10	0.03	0.04	0.10
	1.5	0.07	0.07	0.09	0.03	0.03	0.09
	2	0.07	0.07	0.09	0.03	0.03	0.09

). In a previous study by Saugier et al. [78], a 1.5 h lag was found between $E_T$ measured using branch bag at the canopy level and sap flow measurements. This time lag was caused by changes in wood water storage and by sap flow measurement position (1.3 m). The absence of lag (or presence of a lag $<30$ min) between sap flow measurements and E from eddy-covariance measurements implies that the change of water storage happens quickly in young balsam fir trees.

$E_T/E$ itself varies during the wetting–drying events, showing that the rain also reduces E but not in the same proportion as $E_T$. The lower $E_T/E$ during the drying phase, especially in wet and partially wet conditions, indicates that the most dominant E component during these periods was $E_C$ (with the assumption of $E_G$ occupying a relatively constant proportion of E). $E_C$ was noticeable during the wetting phase and became more dominant after the rainfall had ceased.

4.3. Eddy-Covariance during Rainfall

Unlike the sap flow method, eddy-covariance measurements used in this study rely on open-path sensors that cannot record reliable measurements during rainfall [23]. For this reason, this study relies on the gap filling method (marginal distribution sampling) to obtain E during the wetting–drying events. By design, marginal distribution sampling fills the data gaps with E values of periods with similar meteorological conditions, based on net radiation ($R_n$), air temperature ($T_a$), and vapor pressure deficit (D). We evaluated marginal distribution sampling performance by creating 1000 artificial gaps on the half-hourly time series during dry canopy conditions apart from the the existing gaps that have similar meteorological conditions during wetting–drying events. These artificial gaps covered ∼20% of the available data. Marginal distribution sampling only slightly underestimated the actual observation values and performed well (Figure 12). These results greatly improved our confidence in the gap-filling technique used to estimate E during wetting–drying events.

4.4. $E_T/E$ at the Seasonal Scale

The cumulative rainfall at the Juvenile site during the measurement period of 2018 was $5\%$ less than in 2017. The decrease in rainfall was not followed by an increase of net radiation ($R_n$) or vapor pressure deficit (D): on the contrary, the averages of $R_n$ and D decreased by $9\%$ and $12\%$, respectively. This resulted in the reduction of E but increased $E_T$ by $5\%$ compared to 2017. At the Sapling site, $R_n$ and D increased from 2017 to 2018 by $2\%$ and $31\%$, respectively, leading to an increase of $E_T$ by $42\%$. These results suggest that the magnitude of $E_T$ is not only regulated by meteorological conditions, but also influenced by the growth of both young stands. A study by Tyree [79] demonstrated that stand growth is an important influence on the $E_T$ rate.

The age difference between stands at the Juvenile and Sapling sites are especially outlined by the difference in DBH and LAI. The Sapling site younger stands had a $E_T/E$ ratio $29\%$ lower compared to the Juvenile site for both years. The difference in $E_T/E$ between sites was likely related to LAI, which was $\approx 22\%$ lower at the Sapling site. This finding is similar to that of a study by Breda et al. [80], where $E_T$ is not correlated with DBH, but more related to LAI. Furthermore, Granier et al. [81] found that a reduction in LAI was associated with a decrease in $E_T$ in open stands as a result of the reduction of the transpiring canopy surface, and that it was not associated with a decrease of total E. Other studies have also determined that changes in LAI could alter E partitioning in $E_G$ and $E_T$ by regulating the ratio between area covered by the canopy and stands opening [82,83,84,85].

Globally, values of $E_T/E$ in this study appear lower than the summary of several studies in boreal forests by Schlesinger and Jasechko [17], in which $E_T/E=65\%\pm 18\%$. However, in several studies that directly measured $E_T/E$ in boreal and temperate forests we could find (

Table 2. Comparison of LAI, $E_T/E$, annual precipitation (P) and evaporative index ($E/P$) between this study and several previous studies of E partitioning in boreal and temperate forests. Sites are ordered by annual precipitation rate, from the site receiving the most precipitation to the site receiving the less precipitation

Site	Climatic Zone	Vegetation	Study year(s)	LAI	${\mathit{E}}_{\mathit{T}}/\mathit{E}$	Annual P	$\mathit{E}/\mathit{P}$	Reference
				[m${}^2$m${}^{-2}$]		[mm y${}^{-1}$]
Coweeta Basin, US	Temperate	Eastern white pine	2004–2005	9.4–14.2	0.55 ${}^{*}$	2241	0.65 ${}^{*}$	[74]
Kahoku, Japan	Temperate	Japanese cedar,	2007–2008	3.6–5.2	0.43 ${}^{*}$	2138	0.39 ${}^{*}$	[86]
		Japanese cypress
This study (Juvenile)	Boreal	Balsam fir	2017–2018	3.6	0.47 ${}^{**}$	1583	0.45 ${}^{*}$
This study (Sapling)	Boreal	Balsam fir	2017–2018	2.9	0.35 ${}^{**}$	1583	0.48 ${}^{*}$
Walker Branch Watershed, US	Temperate	Mixed forest	1998–1999	6	0.43 ${}^{*}$	1333	0.50 ${}^{*}$	[24]
Duke Forest, US	Temperate	Mixed forest	2002–2005	7	0.56 ${}^{*}$	1146	0.56 ${}^{*}$	[29]
Lägeren, Switzerland	Temperate	Mixed forest	2014–2015	1.7–5.5	0.74 ${}^{*}$	1037	0.87 ${}^{*}$	[87]
Vielsalm, Belgium	Temperate	Mixed forest	2010–2011	4.1–5	0.68 ${}^{**}$	1000	0.35 ${}^{*}$	[48]
Krycklan, Sweden	Boreal	Mixed forest	2016	4.4–5.2	0.44 ${}^{**}$	619	0.86 ${}^{**}$	[88]
Norunda, Sweden	Boreal	Norway spruce	1995	4–5	0.65 ${}^{**}$	527	1.29 ${}^{**}$	[5]
Prince Albert Nat. Park, Canada	Boreal	Trembling aspen	1994	2.3	0.95 ${}^{**}$	463	0.89 ${}^{**}$	[49]
Scotty Creek, Canada	Boreal	Black spruce	2013	0.9–0.3	0.02 ${}^{**}$	390	0.76 ${}^{*}$	[19]

* during a full year; ** during a particular period in a growing season.

), $E_T$ represented less than $50\%$ of E, suggesting that $E_C$ and/or $E_G$ were quite significant. Interestingly, there is no clear relationship between LAI and $E_T/E$. Despite only having a LAI of 2.3, the $E_T$ of a trembling aspen stand in Prince Albert National Park, Canada, represented $95\%$ of E [49]. On the other hand, with LAI value ranging from 9.4 to 14.2, $E_T$ of a Eastern white pine stand in Coweeta Basin, US, only represented $55\%$ of E [74].

Compared to the other existing $E_T/E$ studies in boreal forest, Montmorency Forest has the highest precipitation (P), even higher than several studies in temperate regions we found (Table 2). Despite high precipitation, only $45\%-48\%$ returns back to the atmosphere. Isabelle et al. [20] in a study at the same sites found that E appeared to be capped even in the presence of high precipitation. The excess of P generates runoff or recharge of ground water, indicating that the availability of soil water is probably not a limiting factor for $E_T$. In the absence of limitation from soil water availability, LAI appears to be a better proxy to estimate $E_T/E$ at the seasonal scale.

== layout width = 297 mm, layout height = 210 mm, left = 2.7 cm, right = 2.7 cm, top = 1.8 cm, bottom = 1.5 cm, include head, include foot [LO,RE] 0cm [RO,LE] 0cm

==

5. Conclusions

This study aims to investigate the dynamics of transpiration ($E_T$) to evapotranspiration (E) ratio ($E_T/E$) under wet and dry canopy conditions at two measurement sites in Montmorency Forest, a unique boreal forest with abundant precipitation and high E. Studying the variations of $E_T/E$ across different canopy wetness conditions is essential, especially for model development to simulate E partitioning. Half-hourly E, $E_T$, and leaf wetness status were measured using eddy-covariance, sapflow (thermal dissipation method), and LWS. The thermal dissipation sensors were calibrated using trunk samples from balsam fir trees and led to new calibration coefficients ($\alpha =54.997$ cm${}^3$ cm${}^{-2}$ h${}^{-1}$ and $\beta =1.204$).

The amount of time needed to completely dry the canopy was $22\pm 18$ h, and was influenced by net radiation, vapor pressure deficit, wind speed, and the amount of rain intercepted by the canopy. Apart from the low vapor pressure deficit and net radiation during the wetting–drying phases, the presence of water on balsam fir needles decreased $E_T$. During the wetting–drying events, $E_T/E$ ranged from $14\pm 15\%$ to $47\pm 54\%$, depending on the wetness level of the canopy.

At the seasonal scale, the variation of $E_T/E$ between our two measurement sites was likely related to differences in leaf area index (LAI). Compared to the several studies of E and $E_T$ partitioning in boreal and temperate forests, we found that our study sites were among several sites in which $E_T$ was not the major component of E. Based on those studies, LAI appears to be a better proxy to estimate $E_T/E$, although it is not always the case.

This study focused on the dynamics of $E_T/E$ at different levels of canopy wetness. However, a proper method is required to estimate the amount of water stored on the canopy during wetting–drying events and the drying rate at half-hourly time steps. Future studies should address the time series of E partitioning, especially the transition of $E_T$ and $E_C$ from wet to dry and dry to wet canopy conditions.