Observation and simulation of evapotranspiration partitioning under wet and dry canopy conditions in a boreal forest of eastern Canada

Observation and simulation of evapotranspiration

partitioning under wet and dry canopy conditions in a

boreal forest of eastern Canada

Thèse

Bram Hadiwijaya

Doctorat en génie des eaux

Philosophiæ doctor (Ph. D.)

Observation and simulation of evapotranspiration

partitioning under wet and dry canopy conditions in a

boreal forest of eastern Canada

Thesis

Bram Hadiwijaya

Under the supervision of:

Daniel Nadeau, advisor Steeve Pepin, co-advisor

Résumé

La forêt boréale représente environ le tiers des biomes forestiers du monde et occupe la deuxième plus grande supercie de végétation derrière les forêts tropicales. Compte tenu de sa grande répartition géographique, la forêt boréale régule les ux d'eau sur de vastes zones et a donc un impact sur la climatologie et l'hydrologie à l'échelle régionale et mondiale. Il est donc crucial de comprendre les interactions entre cet écosystème et l'atmosphère. De nom-breuses études ont porté sur l'évapotranspiration des forêts boréales, mais seule une poignée d'entre elles se sont intéressées à la dynamique du fractionnement de l'évapotranspiration en transpiration du couvert forestier, évaporation de l'eau interceptée et évapotranspiration du sous-étage forestier à une échelle temporelle ne.

L'objectif principal de cette thèse est d'analyser la dynamique de l'évapotranspiration, en particulier la transpiration du couvert forestier et l'évaporation de l'eau interceptée, dans une forêt boréale humide de l'est du Canada. L'approche est basée sur des observations in situ ori-ginales et des sorties de modèles du Canadian Land Surface Scheme (CLASS, tourné en mode `standalone'), à la Forêt Montmorency (47◦₁₇0_{18”N ; 71}◦₁₀0_{05.4”W) de l'Université Laval,}

Québec, Canada. Ce site est classé comme une forêt boréale humide avec un indice d'aridité de 0,57 et des précipitations annuelles moyennes de 1583 mm (60% de pluie, 40% de neige). Cette région est sous l'inuence d'un climat continental subarctique (classication Köppen Dfc), avec une température annuelle moyenne de 0,5◦_{C et une saison de croissance s'étendant}

de juin à octobre. Le dispositif expérimental est composé de deux sites avec des peuplements de sapins baumiers à diérents niveaux de maturité (stade juvenile (site `Juvenile') et stade gaule (site `Sapling')), tous deux équipés d'une tour à ux. Le peuplement le plus mature du site Juvenile a un indice moyen de surface foliaire plus élevé (3,6) que celui du site Sapling (2,9). L'évapotranspiration des peuplements de sapins baumiers a été mesurée par un système de covariance des tourbillons installé sur la tour à ux, tandis que la transpiration du couvert forestier et son bilan hydrique ont été mesurés à l'intérieur de trois parcelles de 400 m2 _situées

à proximité de chaque tour à ux. L'analyse se concentre sur les saisons de croissance de 2017 et 2018. Cette thèse est divisée en trois objectifs spéciques.

Le premier objectif spécique est d'analyser la dynamique de la transpiration du couvert forestier dans des conditions transitoires d'humidité de la canopée. La transpiration a été

obtenue à partir de mesures de ux de sève à l'aide de la méthode de dissipation thermique sur 12 arbres échantillonnés à chaque site. Les incertitudes liées au processus de passage de l'échelle de l'arbre à celle du peuplement ont été quantiées, en particulier grâce à l'étalonnage en laboratoire des capteurs de ux de sève réalisé sur des échantillons de sapin baumier. Les résultats ont montré que la transpiration diminue lorsque la couverture de la canopée est en phase d'humidication, et augmente lorsqu'elle est en phase de séchage. Le rayonnement net, le décit de pression de vapeur et la présence de gouttes de pluie sur les aiguilles des arbres jouent tous un rôle dans la régulation de la transpiration des arbres. À l'échelle saisonnière, la transpiration du couvert forestier a représenté tout au plus 47% de l'évapotranspiration totale. Ces résultats suggèrent que l'évaporation de l'eau interceptée par la canopée est un terme important du bilan hydrique de la canopée.

Ainsi, dans la deuxième partie de la thèse, nous avons analysé la dynamique de l'interception de la canopée autour d'événements pluvieux de l'échelle de la demie-heure jusqu'à la saison de croissance dans son ensemble. An de mesurer le bilan hydrique de la canopée, des dispositifs de mesure de la précipitation et du débit s'étalant le long des troncs ont été déployés, de même qu'un dispositif mesurant la compression d'un tronc d'arbre en continu. Cette nouvelle tech-nique, qui permet d'associer la compression d'un arbre à l'eau interceptée, n'avait jusqu'alors jamais été testée au cours d'une saison de croissance complète. Outre l'estimation des préci-pitations interceptées, les résultats du bilan hydrique de la canopée et le système de suivi de la compression d'un tronc d'arbre ont également été utilisés pour estimer le stockage maximal d'eau dans la canopée. Le stockage d'eau maximum de la canopée estimé par la méthode du bilan hydrique de la canopée s'est révelé en moyenne de 1,6 mm (≈ 0, 49 mm par unité d'indice de surface foliaire) et de 2,2 mm en utilisant le système de suivi de la compression d'un tronc d'arbre. Au cours des deux saisons de croissance étudiées, l'évaporation de l'eau interceptée a représenté entre 16% et 27% de l'évapotranspiration totale.

Enn, pour le troisième objectif spécique de cette thèse, nous nous sommes intéressés à la capacité du modèle CLASS à reproduire les observations du cheminement vertical de l'eau à travers la canopée forestière. Malgré quelques légères diérences entre les observations et les simulations, CLASS s'est avéré capable de simuler avec précision le fractionnement de l'évapotranspiration observée dans des conditions de canopée sèche et humide. Les perfor-mances se sont avérées encore meilleures lorsque la capacité de stockage de la canopée dans le modèle a été supposée égale aux observations. Les principaux écarts entre les observations et les sorties du modèle ont principalement été causés par des diérences de temps de séchage de la canopée observés vs simulés.

Cette thèse fournit un examen détaillé de l'évapotranspiration en forêt boréale humide, en mettant l'accent sur les transitions entre les conditions sèches et humides de la canopée. Les résultats obtenus sont importants pour le développement de modèles hydroclimatiques réalistes et performants au Canada et dans d'autres régions froides du monde.

Abstract

Boreal forests account for around a third of the world's forest biomes and occupy the second largest vegetated area after tropical forests. Given its large geographical distribution, the bo-real forest regulates water uxes over vast areas and thus impacts climatology and hydrology at regional and global scales. Understanding the interactions between this ecosystem and the at-mosphere is therefore crucial. Many studies have investigated the evapotranspiration of boreal forests, but only a handful have focused on the dynamics of evapotranspiration partitioning into overstory transpiration, wet canopy evaporation, and understory evapotranspiration on a ne temporal scale.

The main objective of this thesis is to analyze the dynamics of evapotranspiration partitioning, particularly overstory transpiration and wet canopy evaporation in a humid boreal forest of eastern Canada. The approach is based on eld observations and model outputs from the Canadian Land Surface Scheme (CLASS, run in oine mode), at the Montmorency Forest (47◦₁₇0_{18”N; 71}◦₁₀0_{05.4”W) of Université Laval, Québec, Canada. This site is classied as a}

humid boreal forest with an aridity index of 0.57 and mean annual precipitation of 1583 mm (60% rain, 40% snow). This region is under the inuence of a continental subarctic climate (Köppen classication Dfc), with a mean annual temperature of 0.5◦_{C and the growing season}

stretching from June to October. The experimental setup consists of two sites with balsam r stands at dierent levels of maturity (Juvenile and Sapling), both equipped with eddy covariance ux tower. The more mature stand at the Juvenile site has a higher mean leaf area index (3.6) than the Sapling site (2.9). The evapotranspiration of the balsam r stands was monitored by an eddy covariance system installed on the ux tower whereas the overstory transpiration and canopy water balance were measured inside three 400-m2 _{plots located in}

the vicinity of each ux tower. The analysis focuses on the 2017 and 2018 growing seasons. This thesis is divided into three specic objectives.

The rst specic objective is to analyze the dynamics of overstory transpiration under transient canopy wetness conditions. Overstory transpiration was obtained from sap ow measurements using the thermal dissipation method on 12 sampled trees at each site. Uncertainties arising in the upscaling process from tree to stand scale were quantied, in particular through laboratory calibration of sap ow sensors performed with balsam r trunk samples. The results showed

that transpiration decreases when the canopy cover is in its wetting phase, and increases when in the drying phase. Net radiation, vapor pressure decit, and the presence of rain drops on the tree needles all play a role in regulating tree transpiration. At the seasonal scale, overstory transpiration represented at most 47% of the total evapotranspiration. These results suggest that wet canopy evaporation is a signicant term of the canopy water balance, which is the subject of the second specic objective.

Hence, in the second part of the thesis, we analyzed the dynamics of canopy interception around rainfall events from half-hourly to seasonal scales. In order to measure the canopy water balance, throughfall and stemow measurement devices were deployed, along with the stem compression approach. This new technique, which aims to weight a tree continuously to determine the amount of intercepted water, had never been tested for a full growing season. Apart from estimating the intercepted rainfall, results from canopy water balance and stem compression approach were also used to estimate the canopy maximum water storage. The maximum canopy water storage estimated using the canopy water balance method was on average 1.6 mm (≈ 0.49 mm per unit leaf area index) and 2.2 mm using the stem compression method. During the two growing seasons under study, the wet canopy evaporation contributed between 16% and 27% of the total evapotranspiration.

Finally, for the third specic objective of this thesis, we were interested in the ability of the Canadian land surface scheme (CLASS) model to reproduce the observed circulation of water through the forest canopy. Despite some slight dierences between the observations and simulations, CLASS is able to accurately simulate evapotranspiration partitioning during dry and wet canopy conditions, particularly after adjusting the maximum canopy water storage in line with that observed at the site. The discrepancy between observations and model outputs is mostly a consequence of dierences in canopy drying times. Overall, CLASS is able to simulate ratios of evapotranspiration component to total evapotranspiration similar to those derived from observations at the more mature stand, but overestimates these ratios at the younger stand.

This thesis provides a detailed examination of the evapotranspiration partitioning, with a focus on transitions from dry to wet and from wet to dry canopy conditions. These informations are important for the development of accurate hydrological models in Canada and in other cold regions around the world.

Contents

Résumé ii

Abstract iv

Contents vi

List of Tables viii

List of Figures x

List of Variables xv

List of Abbreviations xviii

Acknowledgements xx Foreword xxii Introduction 1 0.1 Literature review . . . 1 0.2 Research gaps . . . 5 0.3 Objectives . . . 8 Methods 9 0.4 Study site . . . 9 0.5 Field observations . . . 10 0.6 Overstory transpiration . . . 11

0.7 Rainfall interception by the canopy . . . 12

0.8 Understory evapotranspiration . . . 13

0.9 Simulation using a land surface model . . . 14

1 The dynamics of transpiration to evapotranspiration ratio under wet and dry canopy conditions in a humid boreal forest 16 1.1 Résumé . . . 16

1.2 Abstract . . . 17

1.3 Introduction. . . 17

1.4 Materials and Methods. . . 20

1.5 Results. . . 27

1.7 Conclusions . . . 41

2 Observations of canopy storage capacity and wet canopy evaporation in a humid boreal forest 43 2.1 Résumé . . . 43

2.2 Abstract . . . 44

2.3 Introduction. . . 44

2.4 Materials and methods . . . 46

2.5 Results. . . 56

2.6 Discussion . . . 62

2.7 Conclusions . . . 66

3 Can a physically-based land surface model accurately represent evap-otranspiration partitioning? A case study in a humid boreal forest 67 3.1 Résumé . . . 67 3.2 Abstract . . . 68 3.3 Introduction. . . 69 3.4 Methods . . . 72 3.5 Results. . . 80 3.6 Discussion . . . 90 3.7 Conclusions . . . 94 3.8 Acknowledgements . . . 95 Implications 96 3.9 Practical implications . . . 96

3.10 Implications for future study . . . 97

General Conclusion 100

A Appendix 103

List of Tables

1.1 Characteristics of balsam r 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 (ST). Values are mean ± standard deviation. . . 21

1.2 Comparison of LAI, ET/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 . . . 40

2.1 Characteristics of balsam r trees inside the 400-m2 _{plots: tree density per}

hectare (extrapolated from the number of trees in 0.04 ha), canopy height (h), diameter at breast height (DBH), and leaf area index (LAI). Values are mean

±standard deviation. . . 47

2.2 Comparison of leaf area index (LAI), maximum canopy water storage (S), max-imum canopy water storage per leaf area index (S/LAI) between this study and several previous studies. Sites are ordered by S/LAI, from the site with the

highest value to the lowest. . . 64

3.1 Characteristics of balsam r trees inside the 400-m2 _{plots for the Juvenile and}

Sapling sites: tree density per hectare (extrapolated from the number of trees in 0.04 ha), canopy height, diameter at breast height, and leaf area index. Values

are the mean ± standard deviation. . . 73

3.2 List of CLASS parameter values. . . 78

3.3 Comparison of CLASS performance when three dierent c values were used to obtain maximum canopy storage (S = c × LAI). The model was evaluated with

the Kling-Gupta eciency metric KGE0 _{to determine its ability to simulate}

evapotranspiration E, understory evapotranspiration EG, overstory

transpira-tion ET (under dierent canopy-wetness conditions), and canopy evaporation

EC. The values used for c were 0.49 mm (Hadiwijaya et al., 2021), 0.20 mm

(default value, Verseghy et al., 1993), and 0.62 mm (only at the Juvenile site during the 2018 measurement period, Hadiwijaya et al., 2021). The results for E are from data without gap-lling periods and with daily mean closure fractions

A.1 Characteristics of each wettingdrying event described by the total rain (R)

during event, length of the event (twd), length of wetting phase (tw), length of

drying phase (td) and contribution of each canopy condition in Juvenile during

the measurement period of 2018. . . 105

A.2 Relationships of half-hourly ET measured using sap ow with E and D

mea-sured from the ux tower, under various canopy conditions and increasing values

of time lag. . . 106

A.3 Rain event length, gross rainfall (PG), maximum canopy storage (S),

propor-tion of direct throughfall (p), gross rainfall required to saturate the canopy

(P0

G), ratio of canopy evaporation to rainfall during saturated canopy

condi-tions ( ¯EC/ ¯R), and mean wind speed (¯u) of each rain events used for

List of Figures

0.1 Watershed hydrologic cycle and stand water balance. Taken from Winkler et al.

(2010).. . . 2

0.2 Evapotranspiration partitioning into overstory transpiration (ET), wet canopy

evaporation (EC), and understory evapotranspiration (EG) under dierent canopy

wetness conditions around rain (PG) events. . . 4

0.3 Location of the Juvenile in Basin 7A and Sapling sites in Basin 7 at BEREV. . 10

0.4 Schematic diagram of water and energy exchanges as represented in CLASS.

Source: Verseghy (2000).. . . 15

1.1 (a) Location of Juvenile and Sapling ux towers in the experimental watershed; (b) location of measurement plots around Juvenile ux tower; (c) trees at the Juvenile site; (d) location of measurement plots around Sapling ux tower; and (e) trees at the Sapling site. Vegetation heights are computed from LiDAR surveys (Source: Ministère des Forêts, de la Faune et des Parcs du Québec)

collected in 2016. . . 21

1.2 Allometric relationships between diameter at breast height (DBH) and sapwood

area (SA), as well as sapwood width (SW) from 15 Abies balsamea trees located

outside the measurement plots. Vertical green and blue lines are the average DBH of trees that were selected for sap ow measurements in Juvenile and Sapling site, respectively, whereas shaded areas represent standard deviations.

Intercepts for linear and polynomial ts were forced to zero. . . 22

1.3 Conceptual translation of half-hourly LWS data to a canopy wetness level (WL)

during a wettingdrying event (twd). Wetting phase (tw) is when the rain (R)

> 0 mm and drying phase (td) is when the rain has ceased but the canopy is

not fully dry. tw−1is 30 min before a wetting phase, tw0is the starting point of

a wetting phase, and td+1 is 30 min after the end of a drying phase. The circles

illustrate the evolution of canopy wetness at dierent heights above ground. . . 26

1.4 Sap ux density (Fd) vs sap ux index (K) of balsam r stem segments (n =

3; each having 3 sap ow sensors × 5 pressure heads) obtained for this study, compared to the commonly used Granier (1985) and other calibration curves (P. elliotti and P. palustris in Bosch et al., 2014; P. abies (a) and P. sylvestris in Lundblad et al., 2001; P. echinata and P. taeda in Sun et al., 2012; P. abies (b) in Peters et al., 2018). The green and blue shaded area is the range of K

1.5 (a) Distribution of sap ux density (Fd) dierences between sensors installed on

south and north sides of the tree; and (b) tree-to-tree variations described by

the deviation of each tree Fdfrom the site mean Fd. The variability is described

by the interquartile range (IQR). Letters on the x-axis refer to tree (T), Juvenile

(J) and Sapling (S), and followed by the tree or plot number. . . 28

1.6 Characteristics of (a) wetting (tw) and (b) drying (td) phases duration described

by histograms of tw and td; as well as the vertical distribution of (c) tw and

(d) td between LWS installed at three dierent heights: 2, 4, and 6 m. The

variability is described by the interquartile range (IQR). . . 30

1.7 Eect of rain events (R) from (a) 15:30 until 23:00 of 17 August 2018 and from (b) 04:30 until 18:00 of 22 August 2018 on half-hourly canopy wetness level

(WL), evapotranspiration (E), and transpiration (ET). . . 31

1.8 Variations of (a) ET, (b) E and (c) ET/E before wetting phase (tw−1), during

wetting (tw) and drying (td) phases as well as after the canopy became

com-pletely dry (td+1). ET and ET/E variability is described by the interquartile

range (IQR). The solid bold lines are connecting the means of each phase. . . . 32

1.9 Summary of evapotranspiration (E) and transpiration (ET) from 5 July to 18

October in 2017 and 2018. The numbers above E and ET bars are mean ET/E

values ± standard deviation. . . 33

1.10 Relationships of half-hourly data between vapor pressure decit (D) and

tran-spiration (ET) at the Juvenile site under dierent canopy conditions: (a) dry

during daytime (Rn> 10W m−2), (b) dry during nighttime (Rn< 10W m−2),

(c) wet and fairly wet during wetting phases, (d) wet during drying phase, (e)

fairly wet canopy during drying phase, and (f) slightly wet during drying phase. 35

1.11 Reduction of ET in response to canopy humidication due to rain in the rst

half-hour of wetting phases (R(tw0)), described with ratio of ET during

pre-wetting phase (ET(tw−1)) and the rst half-hour of wetting phase (ET(tw0)).

Only ET (tw−1)/ET (tw0)< 1 were used for the analysis. . . 36

1.12 Comparison between marginal distribution sampling to ll articial gaps of E under similar meteorological conditions during wettingdrying events, and the

actual observation from eddy-covariance system (n = 1000). . . 37

2.1 (a) Schematic and (b) photo of throughfall collection system; (c) schematic and (d) photo of stemow collection system; (e) Installation of interceptometer; (f) conceptual schematic diagram of the bending test adapted from Martin et al.

(2013); and (g) horizontal view of the bending test diagram. . . 48

2.2 (a) Example of scatter plot between cumulative gross rainfall (PG,cum) and

cumulative net rainfall (Pn,cum) to determine regression lines of before (blue

line) and after (orange line) canopy saturation. (b) Close up of the plot to show the determination of direct throughfall (p), precipitation required to saturate

the canopy (P0

G), and maximum canopy water storage (S). Data points come

from event 8 (7 September 2017, 00:20 to 18:10) at the Sapling site. . . 49

2.3 Strain (ε) and calculated mean bending ratio (φ) from bending test of (a) 0◦_and

180◦ _{sensors; and (b) 90}◦ _{and 270}◦ _{sensors. (c) Estimation of the modulus of}

elasticity (EM) from the relationship between monitored strain ε and calculated

2.4 Half-hourly cumulative gross precipitation (PG,cum) and net precipitation (Pn,cum)

of each rain event, volume of water on the canopy (Ic(t)) estimated over 30-min

periods using the canopy water balance and the stem compression approach,

and leaf wetness level (WL) measured with leaf wetness sensors at Juvenile site

between 2 August and 10 October 2018. Ic(t) was estimated using the canopy

water balance calculated from evapotranspiration adjusted using the residual

(E(R)) and the Bowen ratio (E(B)) methods. . . 57

2.5 Half-hourly cumulative gross precipitation (PG,cum) and net precipitation (Pn,cum)

of each rain event, amount of water on the canopy (Ic(t)) calculated using the

canopy water balance method, maximum canopy water storage (S) estimated using the double linear regressions method, and partitioning of

evapotranspi-ration estimated using the Bowen ratio method (E(B)) into its components

(understory evapotranspiration (EG), overstory vegetation transpiration (ET),

and canopy evaporation (EC)) at Juvenile (a, c, and e, respectively) and Sapling

(b, d, and f, respectively) sites between 21 August and 14 September 2018. . . 58

2.6 Distribution of maximum canopy water storage (S) estimated using the double linear regressions method during the measurement period of 2017, 2018, and the combination from these measurement periods at Juvenile and Sapling sites; and (b) Comparison of maximum canopy water storage (S) estimated using (i) the double linear regressions method and (ii) data from the stem compression system at Juvenile site for 10 rain events spanning between 2 August and 10

October 2018. The variability is described by the interquartile range (IQR). . . 59

2.7 (a) Total gross precipitation (PG), net precipitation (Pn) components (stemow

(Sf) and throughfall (Tf)), and estimated canopy evaporation (EC,est) at the

Juvenile and Sapling sites, during the measurement periods of 2017 and 2018, along with the average of both periods; and (b) Partitioning of evapotranspi-ration (E) observed using eddy covariance and adjusted using the residual (R) and Bowen ratio (B) methods into its components, which are understory

evap-otranspiration (EG), canopy evaporation (EC) obtained from the intercepted

rain, and stand overstory transpiration (ET) at the Juvenile and Sapling sites

during measurement periods of 2017, 2018, and the average of both years. . . . 61

2.8 Comparison of half-hourly evapotranspiration adjusted using the residual (E(R))

and Bowen ratio (E(B)) methods versus evapotranspiration as a sum of the

components (E = ET + EC+ EG) during dry canopy condition (left column),

wetting phase (middle column), and drying phase (right column) at the Juvenile

(rst two rows) and Sapling (bottom two rows) sites. . . 63

3.1 (a) Theoretical evapotranspiration partitioning under dry canopy conditions, and during the wetting and drying phases; and (b) measurement of

evapotran-spiration and its components in this study. ET is overstory transpiration

esti-mated from upscaled sap ux density (Fd), EC is evaporation from intercepted

rainfall (Ic), EG is the understory evapotranspiration, E is evapotranspiration

(E = ET + EC + EG), PG is gross precipitation, Pn is net precipitation that

3.2 Location of (a) the Montmorency Forest in the North American boreal zone; (b) Juvenile and Sapling ux towers in the BEREV experimental watershed; measurement plots around the (c) Juvenile and (d) Sapling ux towers including the footprint relative contribution curves (orange to dark red lines) obtained

using the model of Kljun et al. (2004, 2015). Aridity index values (P/Ep;

Ep is potential evapotranspiration and P is precipitation) were obtained from

Trabucco and Zomer (2019) and vegetation heights were computed from LiDAR surveys (Source: Ministère des Forêts, de la Faune et des Parcs du Québec)

conducted in 2016. . . 73

3.3 Simplied diagram of evapotranspiration partitioning during dry canopy

con-ditions, and during the wetting and drying phases by CLASS, where PG is the

gross precipitation, p is the fraction of gross rainfall that directly reaches the

ground through canopy gaps, E is total evapotranspiration, ET is

transpira-tion EG is bare soil evaporation, Ev is vegetation evapotranspiration, Ic is the

amount of water on the canopy, S is the maximum canopy storage, Pn is net

precipitation, and θs is volumetric soil water content. . . 79

3.4 Scatter plots and mean daily cycles of half-hourly evapotranspiration observed

by eddy covariance systems (EObs) and model simulations (ESim) at the

Juve-nile (a and b, respectively) and Sapling (c and d, respectively) sites during the measurement periods in 2017 and 2018. Presented data are without gap-lling periods and with daily mean closure fractions between 0.8 and 1.2. Solid lines and shaded areas in the diurnal cycle plots represent averages and standard

deviations, respectively. . . 81

3.5 Comparison of daily evapotranspiration observed by eddy covariance systems

(EObs) and model simulations (ESim) during measurement periods (5 July

-18 October in 2017 and 20-18) excluding gap-lled (black points) and including gap-lled (blue circles) data, both with closure fractions between 0.8 and 1.2 at the (a) Juvenile and (b) Sapling sites. The proportion of non-gap-lled data in daily sums of E when gap-lled values were included in the calculations were

53 ± 26% and 55 ± 6% at the Juvenile and Sapling sites, respectively.. . . 82

3.6 Scatter plots and mean daily cycles of half-hourly understory evapotranspiration

observed with an eddy covariance system (EG,Obs) and simulated with CLASS

(EG,Sim). Presented data are at the Juvenile site during growing seasons when

the sub-canopy eddy covariance system was operational (between July 5th _and

October 18th _{in 2016 and 2017) during periods without gap-lled data and with}

friction velocity > 0.1 m s−1 _{(a and b); and with the estimates of E}

G (EG,Est)

during the measurement periods in 2017 and 2018 at the Juvenile (c and d) and

Sapling (e and f) sites. . . 83

3.7 Scatter plots and mean daily cycles of half-hourly overstory transpiration

ob-served using sap ow measurements (ET,Obs) and CLASS simulations (ET ,Sim)

at the (a) Juvenile and (b) Sapling sites during dry canopy (left column), wet-ting (middle column), and drying (right column) phase conditions over the two measurement periods (2017 and 2018). Solid lines are averages and shaded

3.8 Cumulative gross precipitation (PG,cum), observed net precipitation (Pn,cum(Obs))

using throughfalls and stemows, and simulated water reaching the ground (Pn,cum(Sim)) at the Juvenile (top row) and Sapling (bottom row) sites

dur-ing measurement periods in 2017 (a and c, respectively) and 2018 (b and d,

respectively). . . 85

3.9 Comparison of water stored on the canopy (Ic) observed using the stem

compres-sion approach, estimated using the canopy water balance method, and simulated with CLASS at the (a) Juvenile and (c) Sapling sites. The (b) leaf wetness level

(WL) was measured using leaf wetness sensors at the Juvenile site during the

measurement period in 2018. . . 86

3.10 Scatter plots and mean daily cycles of half-hourly canopy evaporation calculated

using the canopy water balance method (EC,W b) and simulated with CLASS

(EC,Sim), as well as the mean diurnal cycle of gross precipitation (PG) at the

Juvenile (a and b) and Sapling (c and d) sites. . . 87

3.11 Comparison between observed and simulated results of total evapotranspiration (E) and of the partitioning into three components, understory

evapotranspira-tion (EG), overstory transpiration (ET), and canopy evaporation (EC), during

the measurement periods in 2017 and 2018, as well as averages of the two mea-surement periods at the Juvenile and Sapling sites. Note that total evapotran-spiration observations were adjusted for energy budget closure using the Bowen ratio method. Error bars represent standard variation between measurement

plots in each study site. . . 89

3.12 Normalized hysteresis curves between half-hourly transpiration (ET/ET ,max)

and vapor pressure decit (D/Dmax) values for observations and CLASS

sim-ulations during dry canopy conditions at the Juvenile (a and b) and Sapling (c and d) sites, as presented in Pappas et al. (2018). The variables are normalized with their daily maximum. Shades of blue represent hour of the day. Values are averages from the 2017 and 2018 study periods and bars represent standard

deviation. . . 92

A.1 (a) Location of Montmorency forest within boreal zone of Canada and the loca-tion of Juvenile and Sapling ux towers in the BEREV experimental watershed; (b) location of measurement plots around the Juvenile ux tower; and (c) lo-cation of measurement plots around the Sapling ux tower. Vegetation heights are computed from LiDAR surveys (Source: Ministère des Forêts, de la Faune

et des Parcs du Québec) collected in 2016 . . . 103

A.2 Timeseries of gross precipitation (PG) and leaf wetness status WL around the

rain events used to estimate the maximum canopy storage (S) at the Juvenile

List of Variables

Variable Description Units

A trunk sectional area m2

Ac vertically projected crown area m2

c coecient to calculate maximum canopy storage mm

d distance between bending where the force is applied and

where the strain is measured m

D vapor pressure decit Pa or kPa

DT tree diameter m

E evapotranspiration mm

E(B) evapotranspiration adjusted using the Bowen ratio

method mm

E(R) evapotranspiration adjusted using the residual method mm

EC wet canopy evaporation mm

EG understory evapotranspiration mm

EM Young's modulus of elasticity GPa

Eres dierence of total evapotranspiration and sum of

over-story transpiration and underover-story evapotranspiration mm

ET overstory transpiration mm

Ev evapotranspiration from the vegetated surface mm

F force N

Fb force applied for the bending thest N

Fd sap ux density cm3 cm−2 h−1

G ground heat ux W m−2

h height m

H sensible heat ux W m−2

Ic intercepted rainfall mm

˙

Ic(t) change in water volume stored on the canopy mm

K sap ux index

p proportion of rain falling through the canopy without

being intercepted mm

PG gross precipitation mm

P_G0 amount of rain required to saturate the canopy mm PG,cum cumulative gross precipitation mm

Pn net precipitation mm

Pn,cum cumulative net precipitation mm

qa specic humidity of air kg kg−1

Variable Description Units Q streamow mm r correlation coecient rS runo mm ra aerodynamic resistance s m−1 rc canopy resistance s m−1

rs,min minimum stomatal resistance s m−1

R rainfall mm

Rn net radiation W m−2

S maximum canopy storage mm

SA sapwood area cm2

Sf stemow mm

SW sapwood width cm

ST sapwood area per unit ground area m2 m−2

t time h

td drying phase h

td+1 30 minutes after a drying phase h

tw wetting phase h

tw−1 30 minutes before a wetting phase h

twd wetting-drying phase h

Ta air temperature ◦C or K

Tc canopy temperature ◦C or K

Tf throughfall mm

u wind speed m s−1

WL leaf wetness level

X fraction of canopy covered by water

y distance between the neutral axis and the strain

mea-surement m

αr standard variation ratio

α calibration coecient for Granier's equation cm3 cm−2 h−1 β calibration coecient for Granier's equation

βr bias ratio

γr variability ratio

γw specic weight of water N m−3

∆Q energy storage W m−2

∆QB energy storage in the biomass W m−2

∆QG energy storage in the top soil layer W m−2

∆QH energy storage in the air as sensible heat W m−2

∆QλE energy storage in the air as latent heat W m−2

∆T temperature dierence between sap ow probes mV

∆Tmax maximum temperature dierence between sap ow

probes mV

∆Tsw temperature dierence between sap ow probes in the

conducting sapwood mV

∆t time step 30 min

ε compression strain m m−1

θs soil water content m3 m−3

Variable Description Units

λE latent heat ux W m−2

µ mean

ρa air density kg m−3

σ standard deviation

σb bending stress at the outer viber Pa

List of Abbreviations

Abbreviation Meaning

AMSL Above sea mean level

APES Atmosphere-Plant Exchange Simulator

BEREV Bassin Expérimental du Ruisseau des Eaux-Volées BOREAS Boreal Ecosystem Atmosphere Study

CF Closure fraction

CLASS Canadian LAnd Surface Scheme CLSM Catchment Land Surface Model

CV Coecient of variation

DBH Diameter at breast height [cm]

EC Eddy covariance

EVAP Hydrological modelling with an energy balance approach GCM General Circulation Model

IQR Interquartile range

KGE0 _{Modied Kling-Gupta model eciency}

LAI Leaf area index

LSM Land Surface Model

LWS Leaf wetness sensor

NOAH National Centers for Environmental Prediction - Oregon State University - Air force - NWS Hydrology Lab VIC Variable Inltration Capacity

NOPEX NOrthern hemisphere climate Processes land surface EXperiment

If you can not stand the fatigue of study, you will feel the poignant of stupidity.

Acknowledgements

First and foremost, praise and thanks to Allah, the Almighty, for the blessings throughout my research to complete this study.

I would like to express my deepest and sincerest gratitude to my thesis advisors: Professor Daniel F. Nadeau and Professor Steeve Pepin. They are outstanding mentors and educators, as well as perfect motivators. All these opportunities they oered have been invaluable expe-riences for me, especially their introduction to the hospitality of Québec and its beautiful fall and winter. I am very grateful to be able work with them. I also would like to thank Professor François Anctil for his support. Without the three of you, it would be impossible for me to travel to several cities in North America and Europe, and meet dozens of formidable people. Aside from those professors, I also learned a lot from my senior Pierre-Erik Isabelle. Thank you, Pez, for your knowledge and for being a helpful co-author of my articles. I also must then thank all the colleagues who were involved in the eld work and providing valuable advice on the model simulation as well as statistical analyses. One person who stood out is Annie-Claude Parent. Thank you, Annie-Claude for your hard work during those tough times. I would also like to thank: Laurie Mignault, Achut Parajuli, Simon Lachapelle, Adrien Pierre, Judith Fournier, Kelly Proteau, Benoit Brault, Alicia Talbot-Lanciault, Gonzalo Leonardini Quelca, Habiba Kallel, Benjamin Bouchard, Marco Alves, and Antoine Thiboult. I also want to thank Jean-Daniel Sylvain, Guillaume Drolet, and Luc Papillon from Direction de la Recherche Forestière (DRF) from the Ministère des Forêts, de la Faune et des Parcs (MFFP) for their help in providing trunk samples for sap ow calibration.

Special thanks to the members of the jury for this thesis and the doctoral examinations at the beginning of my doctoral studies: Georgianne W. Moore, Oliver Sonnentag, Audrey Maheu, and Brian Morse. I would like to thank the Natural Sciences and Engineering Research Council of Canada (NSERC), Ouranos Consortium, Hydro-Québec, Environment and Climate Change Canada, and Ministère de l'Environnement et de la Lutte contre les Changements Climatiques du Québec (MELCC) for funding this doctorate study.

Special appreciation to my wife Ajeng Mahestika and my daughter Radinka Hadiwijaya for their relentless love. I also want to thank my father Abdul Latief Abadi and mother Dewi

Anggrahini (rest in peace, Mom), my in-laws Mudji Santosa and Istitimijati, and my brother Bayu, my sister Putri, my brother and sisters in-law Nurin, Debby (and Arif), Okto (and Dinda) for their support from Indonesia.

Lastly, I want to thank Jean-Pierre Caliman, Daud Dharsono, and Franky O. Widjaja of PT. SMART Tbk., for keep supporting me and my family during my stay in Québec.

Foreword

This thesis consists of ve sections, all written in English. It begins with a general introduction, followed by three chapters in the form of scientic articles. Next is a discussion on the implications of our main ndings, and then some general conclusions are presented. I am the lead author responsible for all aspects of the research described in this thesis, including the formulation of the research questions, the design of the eld campaigns and modelling runs, the data analysis, as well as the preparation and submission of manuscripts. My advisor, Prof. Daniel F. Nadeau, and my co-advisor, Prof. Steeve Pepin, contributed to the identication and design of studies in the eld of research, and revised the manuscripts in depth. Dr. Pierre-Erik Isabelle contributed to the eddy covariance and meteorological data processing. He also revised the manuscript thoroughly. It should be noted that these articles presented in Chapter

1 and Chapter 2 have already been published, while the manuscript presented in Chapters3

was still under review at the time of writing.

Chapter

1

Hadiwijaya, B., Pepin, S., Isabelle, P.-E., & Nadeau, D.F. (2020). The dynamics of transpi-ration to evapotranspitranspi-ration ratio under wet and dry canopy conditions in a humid boreal forest. Forests 1(2): 237. doi:10.3390/f11020237

This article was submitted on 31 January 2020, accepted on 19 February 2020 and published in 21 February 2020.

Chapter

2

Hadiwijaya, B., Isabelle, P.-E., Nadeau, D.F., & Pepin, S. (2021). Observations of canopy storage capacity and wet canopy evaporation in a humid boreal forest. Hydrological Processes 35(2): e14021. doi:10.1002/hyp.14021

This article was submitted on 19 June 2020, accepted on 20 December 2020 and published in February 2021.

Chapter

3

Hadiwijaya, B., Isabelle, P.-E., Nadeau, D.F., & Pepin, S. Can a physically-based land surface model accurately represent evapotranspiration partitioning? A case study in a humid boreal forest. Agricultural and Forest Meteorology, under review.

This article was submitted on 2 October 2020. Major revision were requested on 28 December 2020 and a revisied version was re-submitted on 23 February 2021.

Introduction

Canada's boreal zone includes forests, grasslands, and wetlands, and provides economical and social support for 3.7 million people living there (Price et al.,2013). Close to 270 million ha of Canadian boreal regions are covered by forests (Brandt et al.,2013), and primarily consist of cold-tolerant tree species within the genera Abies, Larix, Picea, Pinus, Populus, and Betula

(Brandt et al., 2013; Gauthier et al., 2015). Boreal forests in Canada play an important

role in regulating climate, ood, disease and water availability (Price et al.,2013). However, disturbances arising from a variety of drivers such as climate, re, insects, diseases, all have major impacts on biogeochemical (nutrient, carbon, and water) cycling, productivity, species composition, and ecosystem size (Brandt et al., 2013). Therefore, it is essential to provide reliable projections of the impact of these drivers on the dynamics of boreal forests.

The expected climate change, which includes increases in temperature and variations in pre-cipitation and humidity, will alter the water cycle in forest environments, mainly through evapotranspiration (Price et al., 2013; Helbig et al., 2020). Given the importance of water to the Canadian economy, such as for hydropower generation, household consumption, and agriculture practices, a reliable prediction of the impacts of climate change on the water cycle is crucial to adopt policies and thus reducing the risk for populations (Sandford et al.,2018). Hence, hydrological and climate models should have the ability to reproduce the exchanges of water and energy between the forest and the atmosphere (Lawrence et al.,2007). The quality of these models is therefore dependent on detailed observations of processes in the eld, which are used to support their development, notably through validation studies.

0.1 Literature review

The water cycle applied to a watershed is presented in Figure 0.1, starting with precipitation (PG) as the main input. A fraction of the precipitation falls through the canopy gaps and

reaches the ground, while the remainder is intercepted (Rutter,1967;Rutter et al.,1971;Gash

and Morton, 1978; Gash,1979). According to the well-accepted conceptual framework, once

the intercepted rain saturates the canopy, the excess water is then discharged to the ground from the tree foliage (throughfall) and from trunks (stemow) (Rutter, 1967; Rutter et al.,

the soil depending on its properties and antecedent soil moisture conditions, in which case it is stored as soil water content (θs). If precipitation exceeds the inltration capacity, the surplus

water will either be stored locally in surface depressions or ow over the soil surface following the terrain topography as runo (rS). Both groundwater ow and surface runo contribute

to the streamow (Q) exiting the watersheds.

Figure 0.1: Watershed hydrologic cycle and stand water balance. Taken from Winkler et al.

(2010).

Among the water cycle components, evapotranspiration is certainly the most complex to accurately observe (Longdoz and Granier,2012) and simulate (Launiainen et al.,2019) over forest ecosystems. In view of this, two major land-surface studies on boreal forests have been conducted in North America (BOREAS;Sellers et al.,1995) and in northern Europe (NOPEX;

Halldin et al.,1998) to investigate the exchanges of water and energy between this ecosystem

and the atmosphere. Both studies relied heavily on the eddy covariance technique to monitor evapotranspiration in the eld. In short, this well-established technique links uctuations in water vapor concentration to uctuations in vertical wind speed to obtain the turbulent ux of water vapor. Evapotranspiration is the only process providing a direct link between the water and energy budgets in the form of latent heat ux (λE) (Priestley and Taylor,1972). The energy budget of a control volume reads:

where Rn is net radiation, H is sensible heat ux, G is ground heat ux, and ∆Q is energy

storage inside the control volume. All these terms are in W m−2_{. Evapotranspiration then}

can be calculated by dividing latent heat ux with the latent heat ux of evaporation (λ [MJ kg−1_]).

Hydrological models and land surface models (LSMs) calculate latent heat uxes by taking into account variables describing atmospheric conditions, e.g. air temperature (Ta), specic

humidity (qa), air density (ρa), and wind speed (u) (Qu et al.,1998). One of the land surface

models that has been used regularly to simulate the energy budget of the boreal region is the Canadian land surface scheme (CLASS) (e.g. Bartlett et al., 2000; Wang et al., 2002;

Alves et al.,2020). CLASS is a physically-based LSM that calculates total evapotranspiration

and its components by taking into account the existing interactions between soil, vegetation, and atmosphere (Verseghy, 1991; Verseghy et al., 1993). CLASS was developed to provide a realistic lower boundary conditions in general circulation models (GCM) to assess the impacts of climate change (Verseghy,2000).

Water intercepted by the forest canopy will eventually evaporate back to the atmosphere (EC, interception loss) after the storm (Rutter et al., 1971; Gash, 1979; Gash et al., 1995).

Water losses to the atmosphere can also occur in the form of transpiration (ET), which is the

evaporation of the moisture inside the plant leaves through stomata (Roberts,1983). Apart from the evaporation at the canopy level, evaporation also takes place at the soil level, in the form of understory evapotranspiration (EG) (Or et al., 2013). These terrestrial water

vapor uxes to the atmosphere, along with open water evaporation, contribute to the total evapotranspiration (E) of a watershed. Thus, E can be expressed as:

E = EC + ET + EG (2)

which is driven by three main factors: (1) water availability; (2) evaporative demand from the surrounding air (i.e. vapor pressure decit and atmospheric turbulent conditions); and (3) energy available to activate the phase change (Winkler et al.,2010).

Each evapotranspiration component has distinct physical characteristics, time of occurrence around rainfall events, and responses to climate feedbacks (Savenije,2004). In any analysis of evapotranspiration, it is thus necessary to consider the cases of water loss from transpiration and wet canopies separately (Oke, 1987). However, there is no easy way of distinguishing between these two processes (Moene and van Dam,2014).

Figure 0.2 presents the occurrence of ET, EC, and EG under dierent canopy wetness

con-ditions around rain events according to the most accepted framework (Bosveld and Bouten,

2003). Under dry canopy conditions, water vapor losses to the atmosphere are dominated by transpiration, for which stomata act as a regulator (Oke,1987). Transpiration is a physical process involving the absorption of soil water by plant roots, inducing movement from the roots to the leaves (Dingman, 2002). The absorbed soil water moves through the xylem, a

collection of vessels inside the stem that represents the hydraulic pipelines for transporting water and nutrients to the canopy (Moene and van Dam, 2014), and eventually evaporates in the substomatal cavities of the leaf (Campbell and Norman, 2012). These processes are regulated by both environmental and plant physiological factors. The environmental factors include solar radiation, vapor pressure decit, aerodynamic conductance, and available soil water content (Wullschleger et al., 2000; Oren and Pataki, 2001; David et al., 2004; Ewers

et al., 2008), while the (morpho)physiological factors typically include tree age, hormones

(Pallardy, 2008), root distribution, leaf geometry, canopy size, and sapwood to heartwood

proportion (Waring and Running,1976).

Figure 0.2: Evapotranspiration partitioning into overstory transpiration (ET), wet canopy

evaporation (EC), and understory evapotranspiration (EG) under dierent canopy wetness

conditions around rain (PG) events.

Around a rain event, the incoming solar radiation and vapor pressure decit are both decreas-ing, resulting in a lower transpiration rate (Granier et al.,2000a;Aparecido et al.,2016). The transpiration rate increases after the rain has ceased in response to increasing solar radiation and vapor pressure decit (Granier et al., 2000a;Aparecido et al., 2016). At the same time, the atmospheric conditions after rain events will favor the evaporation of intercepted water

(Rutter et al.,1971;Gash,1979).

Since transpiration involves water transport from the roots to the leaves/needles, one way to quantify the transpiration rate is by measuring the movement of axial sap ow inside the xylem

(Goldstein et al.,1998). The most popular method to do so is the thermal dissipation method

developed by Granier (1985). This method has been widely used by plant physiologists and forest hydrologists due to its simplicity, and relatively low cost (Clearwater et al., 1999; Lu

et al.,2004). This approach can be used to estimate transpiration from a single tree or a forest plot by sampling trees of dierent diameter classes (Granier,1987). Studies by Granier et al.

(2000a) andAparecido et al.(2016) demonstrated that the sap ow measurement was able to

monitor the decrease in transpiration rate during wet canopy conditions.

To estimate evaporation from the intercepted rainfall, Rijtema(1965) and Rutter (1967) de-veloped a conceptual framework based on the assumption that transpiration and wet canopy evaporation do not take place simultaneously for a given leaf surface. This approach requires independent measurements of amounts of rainfall being intercepted and subsequently avail-able for evaporation by the forest canopy. Rutter et al.(1971) established an analytical model to explain the interception process and to calculate wet canopy evaporation. This model estimates the canopy water balance using hourly rainfall and other relevant meteorological parameters as inputs (Gash and Morton,1978).

According to the Rutter model, the canopy water balance for a given time is:

(1 − p) PG= Tf + Sf + EC+ Ic (3)

where p is the proportion of rain falling through the canopy without being intercepted by the canopy, PGis the gross rainfall, Tf is the throughfall, Sf is the stemow, and Icis the amount

of water stored on the canopy. According to this model, canopy evaporation is calculated using potential evaporation following Monteith (1965).

In LSMs, it is common to see Rutter's model implemented. In CLASS for instance, a slightly modied version of Rutter's model is used to simulate Ic and EC. CLASS calculates p using

user-dened information on the canopy density and net precipitation as a sum of Tf and Sf

(Verseghy,2012). To determine the net precipitation, LSMs typically used a threshold of the

maximum amount of water that can be stored on the canopy. CLASS calculates this threshold using a simple function of leaf area index (Verseghy et al.,1993).

Whereas for transpiration, LSMs usually compute the process using a combination of both environmental and plant physiological factors (Waring and Running, 1976). For instance, CLASS calculates stomatal resistance as a function of environmental conditions such as the incoming solar radiation, vapor pressure decit, soil moisture suction in the rooting zone, and air temperature, combined with physiological factors such as leaf area index and species-specic minimum stomatal resistance (Verseghy et al.,1993).

0.2 Research gaps

0.2.1 Evapotranspiration partitioning in boreal forest

Thus far, evapotranspiration partitioning studies have mostly focused on documenting the transpiration to evapotranspiration (E ) ratio across a variety of forest stands (e.g. Kool

et al.,2014;Schlesinger and Jasechko,2014;Warren et al.,2018;Liu et al.,2020). Transpiration is traditionally considered to be the dominant component of evapotranspiration in closed canopies. In the boreal forest, past studies have shown that, on average, ET/Eamounts to 65±

18%(Schlesinger and Jasechko,2014). This result shows that the contributions of EC and EG

to the total evapotranspiration are far from being negligible. Despite its importance, in several hydrological models, evaporation from intercepted rainfall is often disregarded or lumped together with soil evaporation (Savenije,2004). Studies byGrelle et al.(1997) andKozii et al.

(2020) in two boreal forests of Sweden have shown that evaporation from intercepted rainfall represented 20% − 34% of the total evapotranspiration. Again, all these results highlight the importance of each component of evapotranspiration and the necessity to explicitly include them within the models.

The ratio of ET/E and EC/E will vary throughout the growing season. For instance, ET/E

is often dominant during dry months when the available energy is used for photosynthesis

(Savenije, 2004). On the other hand, EC/E tends to be large during wet months, as the

available energy is used to evaporate the wet canopy surface before photosynthesis (transpi-ration) can take place (Savenije, 2004). The dynamics of ET/E and EC/E are obviously

very important in humid boreal forests, where the growing season is short and precipitation is abundant. In a recent study looking at the impact of high precipitation on evapotranspiration in the humid boreal forest, Isabelle et al. (2020) reported precipitation of ≈ 950 mm during the growing season and evapotranspiration of ≈ 360 mm for the period, two of the highest values across the 13 boreal sites investigated. The authors suggested that with such high precipitation, evaporation of the intercepted water should account for a signicant portion of total evapotranspiration. Moreover, the canopy wetness conditions can evolve over short time scales. Therefore, an evapotranspiration partitioning study at such sites requires a ne temporal scale of measurement of the dynamics of ET/E and EC/E.

While the sap ow method can monitor transpiration at ne temporal scales, observations of short-lived variations in canopy wetness conditions can be challenging. By design, the approaches of Rutter and Gash are constrained to operate at daily time steps, or on an event basis. One approach to monitor canopy conditions at high temporal resolution is to deploy leaf wetness sensors (e.g.Aparecido et al.,2016;Ringgaard et al.,2014). However, because of the fact that the size and shape of such sensor dier from the actual leaves, this measurement may not provide an accurate quantitative description of canopy wetness status. While gross precipitation, throughfall, and stemow can be observed directly (e.g. Plamondon et al.,1984;

Fleischbein et al., 2005; Pypker et al., 2005), measuring directly the amount of water on

the canopy is a challenge. A novel non-destructive approach to do so is the so-called stem compression approach (Friesen et al.,2015). This method monitors the small compression on the tree trunk imposed by water stored on the canopy (Friesen et al., 2008). This approach has been successfully used to determine intercepted snow and used to estimate the maximum

snow storage of a western hemlock tree (Martin et al.,2013). However, it has never been used to continuously measure rainfall interception during a whole season.

0.2.2 Challenges in estimating stand transpiration

One of the advantages of the eddy covariance technique is that it provides a measurement that is representative of a large "footprint" area (Wilson et al., 2001). On the other hand, stand transpiration is a result of extrapolation from sap ow measurements on sampled trees

(Hogg et al.,1997;Granier et al.,2000a;Wilson et al.,2001). Therefore, spatial variability can

contribute to the errors in the stand transpiration calculation (Köstner et al.,1998). Another source of error from sap ow measurement is the calibration coecient in the calculation of sap ux density. A study by Peters et al. (2018) has recently demonstrated that species-specic calibration coecients were required to minimize uncertainties.

While the contribution of transpiration to total evapotranspiration in boreal forests has been extensively studied (e.g.Hogg et al.,1997;Grelle et al.,1997;Warren et al.,2018;Kozii et al.,

2020), the information regarding the variation of ET under dierent canopy wetness levels is

minimal. Studies unveiling the transpiration process around rainfall events are very limited (e.g.Granier et al.,2000a;Aparecido et al.,2016), despite its importance to investigate factors controlling the transpiration rate during such conditions.

0.2.3 Estimation of wet canopy evaporation

Estimation of wet canopy evaporation at a ne temporal scale is important to understand the hydrological processes taking place around rainfall events. In theory, it is possible to estimate timeseries of 30 min canopy evaporation by estimating the potential evaporation -from the residual of evapotranspiration and sum of two other evapotranspiration components (transpiration and understory evapotranspiration) - and the available water for evaporation. Furthermore, this approach requires a proper estimation of evapotranspiration, which is typi-cally not possible as rain droplets tends to obstruct open-path eddy covariance optics, resulting in gaps in the timeseries data around rainfall events.

0.2.4 Simulation of evapotranspiration partitioning

To the best of our knowledge, with the exception of Kozii et al. (2020), no previous study has ever evaluated the ability of a LSM to partition E with eld measurements including eddy covariance, sap ow, and rainfall interception in a boreal forest. Although the ability of CLASS to simulate energy and water balance has been evaluated (e.g Bartlett et al.,2000;

Verseghy,2000;Alves et al.,2019), its performance to partition evapotranspiration has never

been tested. Since it is a commonly used LSM for climate projection studies (Kothavala

et al., 2005; Alves et al., 2020) and the recent development of CLASS towards Canadian

seems critical and appropriate to focus on its ability to reproduce water pathways through the canopy and evapotranspiration partitioning observed in the eld.

0.3 Objectives

The goal of this thesis is to quantify each component of evapotranspiration, especially overstory transpiration and wet canopy evaporation, in a humid boreal forest of eastern Canada, the Montmorency Forest. This main objective is declined in three specic objectives:

1. To assess the impact of high precipitation on the dynamics of transpiration to evapotran-spiration ratio from half-hourly to seasonal time scales. To achieve this goal, we measure the transpiration of balsam r (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. Then, we monitor the state of the canopy around rainfall events and link this to the dynamics of evapotranspiration and transpiration. 2. To observe and estimate the canopy storage capacity and evaporation at the sub-daily

and seasonal time scale. To achieve this, we analyze the canopy interception components of two young balsam r stands with contrasting characteristics. Then, we compare time series and maximum values of canopy storage estimated using the canopy water balance and directly observed using the stem compression approach. Lastly, we estimate half-hourly and seasonal canopy evaporation using the canopy water balance and detailed measurements of evapotranspiration other components.

3. To evaluate the performance of the Canadian Land Surface Scheme (a physically-based land surface model) in simulating evapotranspiration partitioning over two growing sea-sons. This is based on a highly detailed canopy water balance dataset. To achieve this, we compare observed versus modeled transpiration and wet canopy evaporation during transitions from wet to dry canopy conditions. Then, we evaluate the modeled evapotranspiration partitioning at sub-daily, daily and seasonal time scales.

Methods

0.4 Study site

This study is conducted in the "Bassin Expérimental du Ruisseau des Eaux-Volées" (BEREV), located in Montmorency Forest (47◦₁₇0_{18”N; 71}◦₁₀0_{05.4”W), an experimental watershed}

man-aged by Université Laval since 1965, some 80 km north of Quebec City (Tremblay et al.,2008,

2009). This region is under the inuence of a continental subarctic climate (Köppen clas-sication Dfc) characterized with a mean annual temperature of 0.5o_{C and a short growing}

season that stretches from June to October (Guillemette et al., 2005; Senez-Gagnon et al.,

2018). Isabelle et al. (2020) classied Montmorency Forest as a humid boreal forest, owing to

the large annual precipitation: 1583 mm (40% snow; 60% rain) over the period of 19812010. The BEREV lies in the Laurentian Mountains with a mean altitude of 750 m above mean sea level (AMSL), and peaks at 1000 m AMSL (Isabelle et al., 2020). The overstory vegetation consists mostly of balsam r (Abies balsamea (L.) Mill) along white birch (Betula papyrifera Marsh) and white spruce (Picea glauca (Moench) Voss) (Tremblay et al.,2008,2009). Two ux towers were installed in the BEREV in 2015 as a part of the EVAP project (Hy-drological modelling with an energy balance approach) as presented in Figure 0.3. The rst tower (47◦₁₇0_{17”N; 71}◦₁₀0_{4”W) was installed in Basin 7A, which is located at the head of}

the BEREV and has an area of 1.2 km2_{. Vegetation surrounding this tower is the result of}

natural regeneration after the harvesting of 85% of the trees in 1993. This site was then named "Juvenile" to represent the maturity stage of the trees. As of 2017, the density of the trees at the Juvenile site was 6083 ± 946 trees per ha, with canopy height, diameter at the breast height, and leaf area index being 10.4±3.6 m, 10.5±3.2 cm, and 3.6, respectively. The second tower (47◦₁₇0_{5”N; 71}◦₉0_{5”W) was installed within Basin 7, some 1.3 km east of the Juvenile}

ux tower. Vegetation surrounding this tower is younger than that at the Juvenile site and is the product of natural regeneration after the progressive forest harvesting stretching from 2000 to 2010. Hence, this site was named "Sapling" to represent the maturity stage of the trees. Vegetation surrounding this tower had a higher tree density (8417 ± 1443 trees per ha), shorter canopy height (5.9 ± 1.4 m), smaller diameter at breast height (5.7 ± 1.7 cm), and lower leaf area index (2.9) than that at the Juvenile site. This study focused on three circular forest plots of 400 m2 _{at each of the two sites, where measurements of canopy water balance,}

Figure 0.3: Location of the Juvenile in Basin 7A and Sapling sites in Basin 7 at BEREV.

vegetation metrics, and evapotranspiration partitioning were conducted. Each of the plot was located in a footprint of the ux tower. The period of interest herein is during the growing period in 2017 and 2018, particularly between 5 July and 18 October.

0.5 Field observations

The water vapor transport from the vegetated surface to the atmosphere due to vertical turbulent transport was measured using an eddy covariance system. The latter consists of a three-dimensional sonic anemometer coupled with a CO2/H2O open-path gas analyzer to

measure high frequency uctuations of wind velocities and water vapor concentrations. These instruments were installed at a height of ≈ 1.5h, where h is the mean forest canopy height, to be located in the so-called "constant-ux layer". The ux measurements represent the conditions of an eective area of inuence upwind of the sensor called the ux footprint area (Foken and Napo, 2008), whose size and position depend on wind direction, measurement height, and atmospheric conditions. Hence, the eddy covariance systems installed at the top of the ux towers in the Juvenile and Sapling sites are facing northwest, as this is the direction of the prevailing winds at the two sites. Since the ux tower at the Juvenile site consists of a bulk scaolding structure, which could interfere with the airow, a second eddy covariance system was installed facing the opposite direction (southeast). At each time step, the measurements

of one or the other eddy covariance systems are therefore carefully selected, depending on the wind direction. The three-dimensional sonic anemometer also measures the sonic temperature which is used to calculate the sensible heat ux. The ux towers also feature meteorological instruments to allow the measurement of energy balance components (see Eq. 1). Details on the eddy covariance setup, how each term is measured, and data processing are presented in Section3.4.2.

0.6 Overstory transpiration

Forest stand overstory transpiration was estimated using two steps. First, sap ux was mea-sured using the thermal dissipation method (Granier, 1985, 1987). This method measures water transport in tree xylem from root to the canopy using two sensor probes, each equipped with a thermocouple to determine the temperature dierence between the two probe locations

(Lu et al.,2004). These probes were inserted radially into the tree trunk and were spaced 4

cm apart. The upper probe consisted of a heating element connected to a current regulator constantly supplying 0.2 W (Lu et al.,2004). The bottom probe was left unheated to measure the ambient temperature of the wood tissue and act as reference (Lu et al., 2004). Water transport in the conducting sapwood when the tree is actively transpiring reduces the tem-perature dierence between two probes (∆T ), as the sap ow cools down the heated sensor (hence the name of the technique, thermal dissipation method). Inversely, in the absence of transpiration (at night for instance) the temperature dierence between two probes increases. The maximum temperature dierence (∆Tmax) between the two probes, which occurs at night

becomes the reference. Because each sensor has its specic electrical resistance, ∆Tmax must

be determined individually. ∆Tmax can vary from one night to another, and as such, it

is challenging to identify the periods when there is no sap ux. Granier (1987) proposed a systematic approach to estimate ∆Tmaxusing a 7 to 10 days window to minimize the inuence

of occasional nighttime ow if it does occur. On the other hand, Oishi et al.(2008) proposed a method to estimate ∆Tmax based on environmental conditions. The dierence of ∆T in

the presence and absence of transpiration was used to calculate the sap ux (see Eq. 1.2). Since sap ux index is dimensionless, converting from the unit of sensor output in mV to temperature (i.e., Seebeck coecient: 40 µV/o_{C for copperconstantan thermocouples) is not}

necessary (Lu et al.,2004).

Water transport inside the stem only takes place inside the conducting sapwood. Therefore, the calculation of sap ux index had to be corrected for the sensors which length exceeded the sapwood depth, using a method described in Clearwater et al.(1999). Sap ow per unit sapwood area (sap ux density) was then computed using the corrected sap ux index and the empirical relationship described in Granier (1985). This relationship includes empirical coecients that can be obtained by calibration in the laboratory. Most studies relying on the

thermal dissipation technique for sap ow measurements use calibration coecients provided

by Granier (1985), that were obtained from stem samples of Pseudotsuga menziesii (Mirb.)

Franco, Pinus nigra (Arnold), and Quercus robur (Ehrh.). However, several recent studies re-ported variability in the empirical coecients, when species other than those tested by Granier were calibrated in the laboratory (e.g., Steppe et al., 2010;Bosch et al.,2014; Peters et al.,

2018). Therefore, in this study, the sap ow sensors were calibrated using stem segments of balsam r harvested in the vicinity of our measurement plots, carefully following the procedure

of Steppe et al.(2010). Details of the calibration procedures are presented in Section 1.4.4.

Finally, the last step to obtain the stand transpiration is to extrapolate the sap ux density to the whole sapwood area of the stand. This upscaling process requires information on the sapwood area of trees in the measurement plots. To do so, 15 balsam r trees with dierent diameters were harvested to obtain a relationship between sapwood area and diameter at the breast height. The stand transpiration was calculated by multiplying the average sap ux density of the measured trees with sapwood area per ground area (Granier et al.,2000a). Details on thermal dissipation sensors installation and data processing are presented in Section

1.4.3.

0.7 Rainfall interception by the canopy

Canopy interception in Eq. 3was estimated from the dierence between gross precipitation and net precipitation using rain gauges to obtain high resolution data. The stemow was collected using spiral collar mounted on the stem of the trees. For the throughfall, the collection area was expanded using gutters. This system increases the collection area by 164% compared to that of rain gauges used in this study. Further, the throughfall systems were randomly relocated within the measurement plots to increase the eective number of measurement points within the plot (Lloyd and Marques,1988). Details of the installation are presented in Section2.4.2. The intercepted rainfall was also directly observed using the stem compression approach. The basic principle of this approach is to monitor the extra load from the accumulated precipita-tion on the canopy, which causes a compression of the tree trunk. This compression can be calculated according to Hooke's law by assuming that the tree trunk behaves as linear elastic material (Friesen et al.,2008). The compression (and subsequent decompression) of the trunk was measured using a set of linear motion potentiometers. However, the travel length of the linear motion potentiometer is very short (0.00998 m), and as such, the vertical displacement caused by the load from the intercepted water on the canopy would be too small to measure. Therefore, the length of the observed trunk section was extended to 1 m by deploying 1-m quartz tubes along the tree trunk (Friesen et al., 2008; Martin et al., 2013). Quartz tubes were selected because of their thermal stability (coecient of linear thermal expansion = 0.59 µm m−1 ◦C−1) (seeFriesen et al.,2008).