Restoration of the hardwood forest:
A profitability approach
Thèse
Mariana Hassegawa
Doctorat en sciences du bois
Philosophiae doctor (Ph. D.)
Québec, Canada
Restoration of the hardwood forest:
A profitability approach
Thèse
Mariana Hassegawa
Sous la direction de :
Alexis Achim, directeur de recherche
Nancy Gélinas, codirectrice de recherche
Daniel Beaudoin, codirecteur de recherche
RÉSUMÉ
En Amérique du Nord, la préférence pour la récolte des arbres vigoureux et de gros diamètre a amené les forêts feuillues à un état d’appauvrissement. Ces forêts sont composées de grandes quantités de tiges de faible qualité et d’essences moins prisées par l’industrie. Conséquemment, le secteur forestier doit composer avec une grande quantité de matière première de faible qualité, ce qui entraîne une augmentation des coûts d’opération et cause la production des forts volumes de résidus qui doivent être valorisés. Cette situation force l’industrie à rechercher des solutions pour l’utilisation de grandes quantités de bois de faible qualité abondamment disponibles en forêt afin d’augmenter la rentabilité de ses opérations. Une option serait de miser sur les produits de haute valeur ajoutée qui pourraient être extraits à partir des résidus des scieries. Ces produits incluent les extraits de bouleau jaune (Betula alleghaniensis Britt.), lesquels ont du potentiel pour l’utilisation dans les industries nutraceutique, cosméceutique et pharmaceutique. Les produits de haute valeur ajoutée, intégrés dans l’industrie des produits traditionnels, peuvent augmenter la rentabilité de la chaîne de valeur, surtout si les résidus sont utilisés comme matière première. Afin de comprendre le potentiel et les limites de cette approche, cette étude a évalué les facteurs qui influencent la valeur monétaire de sciage, proposant ensuite une utilisation alternative pour le bois de faible qualité et l’écorce provenant des forêts feuillues de la province de Québec au Canada. Cette étude a été structurée en trois parties : l’évaluation de la relation entre la valeur monétaire des sciages et les caractéristiques du peuplement, de la station et du climat; l’évaluation de la relation entre la quantité d’extraits de bois et d’écorce du bouleau jaune et des caractéristiques des arbres et; l’évaluation de l’inclusion d’un produit à haute valeur ajoutée dans la chaîne de création valeur. Dans la première partie, la valeur monétaire des sciages du bouleau jaune et de l’érable à sucre (Acer saccharum Marsh.) a été utilisée comme indicateur de la qualité du peuplement. Les résultats ont démontré que des patrons géographiques de variation de la valeur monétaire des sciages existaient à travers la province. Cette variation pourrait être attribuée en partie aux caractéristiques du peuplement, de la station et du climat local. Même s’il persiste de l’incertitude quant à l’effet de l’historique d’aménagement forestier, on croit que la production d’arbres de haute qualité devrait être priorisée aux endroits où la valeur monétaire des bois sur pied est la plus grande. Dans la deuxième partie, les analyses réalisées pour quantifier les extraits du bois et
de l’écorce ont fourni une compréhension plus approfondie du potentiel du bouleau jaune pour les produits non traditionnels à haute valeur ajoutée. Dans la troisième partie, la rentabilité d’une coupe de jardinage a été évaluée, en plus du profit potentiel de production de la bétuline et l’inclusion de cet extrait dans la chaîne de valeur du bois. Dans certains cas, la coupe de jardinage a été très peu rentable, surtout lorsqu’appliquée dans les peuplements qui étaient composés des grandes quantités de tiges de faible qualité. Dans de tels cas, les produits à haute valeur ajoutée, comme la bétuline, pourraient augmenter les profits et, par le fait même, ajouter de la valeur à la ressource forestière. La diversification des produits est une approche qui pourrait être envisagée par l’industrie forestière pour augmenter sa résilience et promouvoir une sylviculture respectant les règles de l’art.
ABSTRACT
The historical preference for harvesting vigorous and large-diameter trees from stands in North America resulted in a forest mainly composed of poor-quality stems of less valuable species that present lower growth potential. As a result, the forestry sector has to work with large quantities of low-quality raw material, increasing operation costs and producing large amounts of residues. This situation forces the industry to find solutions to use the abundant quantity of low-quality wood available in forest stands to increase profits. One option is the high value-added products that could be extracted from sawmill residues. These products include yellow birch (Betula alleghaniensis Britt.) extracts that have the potential to be used in the nutraceutical, cosmeceutical and pharmaceutical industries. The enhanced products, when integrated with the traditional products industry, could increase profit of the wood value chain, especially if residues are used as raw material. In order to better understand the potential and limitations of this option, this study assessed the factors that influence lumber value, and proposed an alternative use for the abundant low-value wood and bark available in the province of Quebec, Canada. Structured in three parts, this work first studied the relationship between stand, site and climatic variables and stand quality, using lumber value recovery (LVR) of sugar maple (Acer saccharum Marsh.) and yellow birch as surrogate variable. In the second part, the relationship between extracts content in wood and bark of yellow birch trees with selected tree characteristics was assessed. This allowed a better understanding of the potential of yellow birch extracts as a high-value added product. In the third and final part, the potential impact of integrating a high-value added product to the processing of traditional wood products was evaluated. For this, the profitability of a selection cut was analysed, the potential financial gain of producing betulin extract was assessed, and finally, the inclusion of betulin extracts in a hardwood chain was evaluated. Results from the ensemble of this thesis show that variations in LVR could be attributed to in part to stand, site and climatic conditions. Although it remains uncertain as to what extent the variability of LVR might result from past management practices or from inherent site characteristics, we believe that efforts to produce high-quality lumber should be prioritized in sites where LVR is predicted to be the highest. In addition, simulations showed that tested selection cuts in some cases generated very little profit, especially when forest stands were composed of great quantities of low-quality
stems. In this situation, the production of high value-added products, such as betulin extracts, could be an interesting alternative to increase profits, contributing to add value to the existing forest resource. The product diversification is a pathway that could be explored by the forest industry to improve its resilience and promote a more efficient use of the resources.
TABLE OF CONTENTS
RÉSUMÉ ... iii
ABSTRACT ... v
TABLE OF CONTENTS ... vii
LIST OF TABLES ... x LIST OF FIGURES ... xi ACKNOWLEDGEMENTS ... xiii PREFACE ... xv THESIS INTRODUCTION ... 1 HYPOTHESES ... 4 WORK OUTLINE ... 5
CONSIDERATIONS AND ASSUMPTIONS ... 5
Chapter 1: LARGE-SCALE VARIATIONS IN LUMBER VALUE RECOVERY OF YELLOW BIRCH AND SUGAR MAPLE IN QUEBEC, CANADA ... 8
ABSTRACT ... 8
1 INTRODUCTION ... 9
2 MATERIAL AND METHODS... 11
2.1 REASSESSMENT OF THE LUMBER VALUE ESTIMATION METHOD ... 12
2.2 LVR PREDICTIONS BASED ON STAND AND SITE CHARACTERISTICS ... 17
2.2.1 BOOSTED REGRESSION TREES ... 18
2.2.2 PROVINCIAL-SCALE ANALYSIS OF LVR ESTIMATES ... 20
3 RESULTS ... 21
3.1 BRT MODEL ... 21
3.1.1 PARTIAL DEPENDENCE PLOTS ... 22
3.1.2 BRT MODEL PREDICTIONS AND HOTSPOT ANALYSIS ... 25
4 DISCUSSION ... 27
5 CONCLUSIONS ... 31
Chapter 2: RELATIONSHIP BETWEEN ETHANOLIC EXTRACTS OF YELLOW BIRCH AND TREE CHARACTERISTICS ... 32
ABSTRACT ... 32
1 INTRODUCTION ... 33
2 MATERIAL AND METHODS... 35
2.1.1 SAMPLE PREPARATION ... 36
2.1.2 MACERATION ... 37
2.2 DETERMINATION OF TOTAL PHENOLS CONTENT ... 38
2.3 DETERMINATION OF MAJOR TRITERPENES AND PHYTOSTEROLS ... 38
2.4 DATA ANALYSIS ... 39
3 RESULTS AND DISCUSSION ... 41
3.1 EXTRACTIVES CONTENT ... 42
3.2 TOTAL PHENOLS CONTENT ... 43
3.3 MAJOR TRITERPENES AND PHYTOSTEROLS CONTENTS ... 43
3.4 STATISTICAL MODELS ... 44
3.4.1 WOOD TISSUE ... 49
3.4.2 BARK TISSUE ... 50
4 CONCLUSIONS ... 51
Chapter 3: ASSESSING THE POTENTIAL IMPACT OF BIOREFINED PRODUCTS FROM YELLOW BIRCH SAWMILL RESIDUES ON THE PROFITABILITY OF A HARDWOOD VALUE CHAIN: A CASE STUDY IN THE LAURENTIANS REGION ... 53
ABSTRACT ... 53
1 INTRODUCTION ... 54
2 MATERIAL AND METHODS... 56
2.1 STUDY AREA ... 56
2.2 PROFITABILITY OF A SELECTION CUT ... 57
2.3 POTENTIAL FINANCIAL GAIN OF PRODUCING BETULIN EXTRACT ... 61
2.3.1 BETULIN SUPPLY CHAIN ANALYSIS ... 61
2.3.2 QUANTITY OF BETULIN EXTRACTS PRODUCED AND COMMERCIALIZED PER YEAR ... 62
2.3.3 BETULIN EXTRACT PRODUCTION COSTS ... 63
2.3.4 BETULIN EXTRACT PRODUCTION YIELD ... 63
2.4 THE INCLUSION OF BETULIN EXTRACTS IN A HARDWOOD VALUE CHAIN ... 63
3 RESULTS ... 64
3.1 PROFITABILITY OF A SELECTION CUT ... 64
3.2 POTENTIAL FINANCIAL GAIN OF PRODUCING BETULIN EXTRACTS FROM YELLOW BIRCH TREES ... 69
3.2.1 QUANTITY OF EXTRACTS PRODUCED AND COMMERCIALIZED PER
YEAR ... 69
3.2.2 BETULIN EXTRACT PRODUCTION YIELD ... 70
3.2.3 BETULIN EXTRACT PRODUCTION COSTS ... 70
3.3 THE INCLUSION OF BETULIN EXTRACT IN A HARDWOOD VALUE CHAIN ... 71 4 DISCUSSION ... 72 5 CONCLUSIONS ... 74 GENERAL CONCLUSIONS ... 75 LITERATURE CITED ... 79 APPENDICES ... 94
LIST OF TABLES
Table 1.1 Relative prices by NHLA (2011) board class... 15 Table 1.2 Parameter estimates and associated standard errors (SE) for QI and Vsw predictive
models. ... 16 Table 2.1 Sampling site characteristics ... 35 Table 2.2 Definitions and abbreviations of the variables ... 41 Table 2.3 Average ethanol-soluble extraction yields, total phenols, major triterpenes and phytosterols concentrations (±SD) in the extracts of wood and bark from vigorous, non-vigorous and moribund trees. ... 42 Table 2.4 Variables used in the models for describing the extractives content, total phenols, major triterpenes and phytosterols in the wood and bark, model errors calculated from the fixed effects, and fit indices calculated from fixed and random effects of the models ... 46 Table 2.5 Parameter estimates, associated standard errors (SE) and p-values for the models
... 47 Table 3.1 Description of study area (by forest management unit) ... 56 Table 3.2 Parameter estimates and associated standard errors (SE) for Vbark predictive model ... 60 Table 3.3 Estimated raw material processing costs for sawdust and bark ... 70
LIST OF FIGURES
Figure 1.1 Comparison between LV predicted with parameters from 1976 and the observed LV ... 14 Figure 1.2 Comparison between LV predicted by the reassessed model and observed values. ... 16 Figure 1.3 Comparison between predicted and observed LVR for (a) yellow birch and (b) sugar maple ... 22 Figure 1.4a Partial dependence plots for each predictor retained in the final models for yellow birch ... 23 Figure 1.4b Partial dependence plots for each predictor retained in the final models for sugar maple ... 24 Figure 1.5a Distribution of LVR for yellow birch, across the province of Quebec. Comparison between LVR from fitting data (inset figure) and predicted LVR from BRT ... 26 Figure 1.5b Distribution of LVR for sugar maple, across the province of Quebec. Comparison between LVR from fitting data (inset figure) and predicted LVR from BRT ... 27 Figure 2.1 Structure of the triterpenes and phytosterols identified by GC–MS from the yellow birch wood and bark ... 40 Figure 3.1 Location of the study area and temporary sample plots (TSP) ... 57 Figure 3.2 Variable costs as a function of average harvested volume per stem (m3), according to silvicultural prescription and stand type ... 59 Figure 3.3 Simplified betulin extract supply chain ... 62 Figure 3.4 Profitability per plot of a hardwood value chain focused on lumber production. The grey dashed line represents the reference line, with slope of 1. ... 65 Figure 3.5 Profit sensitivity to changes in fixed costs, product value, transportation costs and stumpage fees ... 66 Figure 3.6 Profitability per plot considering a decrease in product value of 10%. The grey dashed line represents the reference line, with slope of 1. ... 67 Figure 3.7 Percentage of plots per classes of profit for sawmill products and coproducts considering the inclusion of additional profit per hectarefrom a new product .. 68
ACKNOWLEDGEMENTS
I would like to thank the Fonds de recherche du Québec – Nature et Technologie (FRQNT) for funding of this research and the Ministère des Forêts, de la Faune et des Parcs du Québec for making the inventory data available, and all the companies that participated in the market research.
I am deeply grateful to my supervisor Dr. Alexis Achim for believing in my capabilities for developing this project and guiding me during the course of my study. His passion for research and his enthusiasm helped me to keep focused and motivated to this work’s completion.
I would also like to thank my co-supervisors Dr. Nancy Gélinas and Dr. Daniel Beaudoin for always giving me valuable feedback in their areas of expertise.
To the members of the examining committee: Dr. David Pothier, Dr. Tatjana Stevanovic, who gave me a new appreciation for wood chemistry, and Dr. Isabelle Duchesne, external member of the jury. I am grateful for the time they spent evaluating my thesis, their comments and insightful questions.
I would like to extend my gratitude to Dr. David Auty and Dr. Rock Ouimet for their input in their areas of expertise and direct collaboration to this work.
I would like to thank Dr. Alain Cloutier, Guylaine Bélanger, Marthe Larouche, Yves Bédard, Martine Lapointe, David Lagueux, Daniel Borgault, and Luc Germain. We are very fortunate at the Renewable Materials Research Centre and the Department of wood and forest sciences for having such a dedicated team.
I wish to thank the Coopérative forestère des Hautes Laurentides for helping me select my study area, and also the people who helped me collect and prepare my data: Jean-Sébastien Perron, Louis-Vincent Gagné, Simon Delisle-Bouliane, Samuel Guy-Plourde, Laurence Martel and Aida Khemarkhen.
I would also like to thank Alexis Leroux, from Bureau de mise en marché du bois, for the assistance with the software MÉRIS.
I am also grateful to my colleagues who helped me directly or indirectly during these years: Filip Havreljuk, Emmanuel Duchateau, François St-Pierre, Mariana Royer, Etienne Leroux, Yannick Vianno, Pierre Betu Kasangana; and my friends who were part of this journey: Aline, Bruna, Cindy and Chris, Derek, Fabio, Jedi, Juliana, Luciane and Diego, Mai, Marco, Pola and Andrew, and Wei.
A huge thank you to my family Ana, Leo, Mieko, Siani, Alec and especially to my parents Celia and Toshi, who have always fueled my curiosity about everything, including forestry. Also to my grandmother Rosa, who made me discover the love for plants from an early age. To my daughter Luisa, with whom I learn so much everyday about prioritizing, multitasking and enjoying the simplest things in life. Finally, I would like to thank my husband Eduardo, whose support, patience and motivation have been essential for me to finish this work.
PREFACE
This document is a paper-based doctorate thesis, written in accordance to the criteria established by the graduate studies committee of Laval University. It begins with a general introduction, followed by the core of this thesis, which corresponds to two scientific papers presented in Chapters 1 and 2, and a third manuscript in preparation, presented in Chapter 3. These manuscripts cover specific, but complimentary themes of the research and are listed as follows:
Chapter 1: Hassegawa, M.; Havreljuk, F.; Ouimet, R.; Auty, D.; Pothier, D.; and Achim, A. 2015. Large-scale variations in lumber value recovery of yellow birch and sugar maple in Quebec, Canada. PLoS ONE 10(8). doi: 10.1371/journal.pone.0136674.
Chapter 2: Hassegawa, M.; Stevanovic, T.; and Achim, A. 2016. Relationship between ethanolic extracts of yellow birch and tree characteristics. Ind. Crops and Prod. 94:1-8. doi: 10.1016/j.indcrop.2016.08.038.
Chapter 3: Hassegawa, M.; Gélinas, N.; Beaudoin, D.; and Achim, A. 2016. Assessing the potential impact of biorefined products from yellow birch sawmill residues on the profitability of a hardwood value chain: a case study in the Laurentians region. (In preparation)
As a doctorate candidate and first author of these articles, I was responsible for reviewing the literature, planning the research, performing the analyses in the laboratory, as well as the statistical analyses, and also writing the scientific manuscripts. My supervisor Alexis Achim and my co-supervisors Nancy Gélinas and Daniel Beaudoin have advised me during my research project, also revising the scientific articles. Filip Havreljuk, David Auty, Rock Ouimet, David Pothier and Tatjana Stevanovic have also contributed with their expertise in the articles.
The results from this project have been presented in the following conferences:
Hassegawa, M.; Achim, A.; Gélinas, N.; Beaudoin, D. 2013. Estimation de la valeur des produits issus de billes de feuillus de faible qualité. Séminaire II (SBO-8001). Université Laval, Quebec, QC. April 25th, 2013.
Hassegawa, M.; Havreljuk, F.; Auty, D.; Pothier, D.; Achim, A. 2013. Applicabilité de la coupe de jardinage en forêt feuillue au Québec: évaluation de la valeur monétaire des produits transformés. 81e Congrès de l’Association Francophone pour le Savoir (ACFAS). Quebec, QC. May 8th, 2013.
Hassegawa, M.; Havreljuk, F.; Auty, D.; Achim, A. 2013. Estimating the value of sawn lumber using log characteristics. 4th International Scientific Conference on Hardwood Processing (ISCHP). Florence, Italy. October 7th -9th, 2013.
Hassegawa, M.; Havreljuk, F.; Ouimet, R.; Auty, D.; Pothier, D.; Achim, A. 2014 Predicting hardwood lumber value at the landscape level. 7th Eastern CANUSA: forest science conference (eCANUSA). Rimouski, QC. October 16th-18th, 2014.
Hassegawa, M.; Achim, A. Quantification of wood and bark extracts from yellow birch (Betula alleghaniensis Britt.). 5th International Scientific Conference in Hardwood Processing (ISCHP). Quebec, QC. September 15th-17th, 2015.
Hassegawa, M.; Havreljuk, F.; Ouimet, R.; Auty, D.; Pothier, D.; Gélinas, N.; Beaudoin, D.; Achim, A. 2016. Stand profitability in the decision-making process for hardwood forest restoration. WOOD QC 2016: Modelling wood quality, supply and value chain networks. Quebec/Baie St-Paul, QC. June 12th-17th, 2016.
Based on the results obtained during this study, a general conclusion and remarks are presented at the end of this work.
THESIS INTRODUCTION
Forestry practices in North America were conducted intensively during a long period of time, especially in the last century. They aimed the maximal productivity in volume and the forest resource was seen as something inexhaustible (Boulet 2015). In the early 1980’s, the diameter-limit cutting was the mainstream in forestry (Robitaille and Roberge 1981). This preference for harvesting vigorous and large-diameter trees resulted in residual stands that were mainly composed of poor-quality stems with lower growth potential and less valuable species (Deluca et al. 2009; Hawley et al. 2006; Nyland 1992). As a result, the forestry sector had gradually to work with larger quantities of low-quality raw material (Majcen 1994; Nyland 1992).
In an effort to improve the state of the forest, a new management system was established in Quebec in 1986 to control forestry operations on public lands focusing on the improvement of forest productivity and also on the protection of natural resources (Paille and Deffrasnes 1988). The new system would ensure that wood-based industries would have a long-term wood supply; on the other hand, they would be obliged to maintain, or even increase, forest productivity (Duchesneau 2004). According to the selection cutting system established in this province, trees that should be preferably cut would be the ones that presented good quality, but that were bound to die (Majcen 1994). The second harvest priority was the moribund, low-quality trees. This was thought to be closer to natural forest dynamics, and would improve future quality of stands by protecting the understory, leaving a larger number of vigorous trees and promoting biodiversity (Bédard and Majcen 2003; Majcen 1994).
In a wide-scale study monitoring forest growth after selection cuttings, Bédard and Brassard (2002) showed that the industrial application of this system was deficient. They attributed poor stand growth after selection cutting to inadequate estimations of tree vigour. Underestimating tree vigour meant that more good quality trees could be included in wood supplies (Bédard et al. 2004; Meunier et al. 2002). To palliate this situation the then Ministry of Natural Resources and Wildlife of Quebec developed a tree vigour estimation system, known as MSCR. This classification was conceived to establish harvesting
priorities according to tree vigour (Boulet 2007). It categorized standing trees in four harvesting priorities based on visual defects like presence of sporophores, stem checks, crown defects, etc. These characteristics were related to the probability of a tree dying before the following rotation. The classification prioritized the harvesting of moribund (class M) and non-vigorous trees (class S), and aimed to preserve growing and stock trees (classes C and R, respectively). Because moribund and non-vigorous trees are frequently low-quality trees, this tree selection system had a negative impact on the hardwood value chain (Pothier et al. 2013). Supplying the industry with high-quality wood became increasingly difficult and expensive because the forest stands with acceptable growing stock became less accessible (Boulet 2015). In addition, in hardwood forests with high quantity of low-vigour trees, the employment of selection cuts that aim at forest restoration may result in low profitability (Sabbagh et al. 2002).
Aside from impacts caused by past management practices, environmental conditions may also have an effect on tree quality. Yellow birch (Betula alleghaniensis Britt.), for instance, tends to thrive in sites with higher annual precipitation, producing high-quality trees (Gagné et al. 2013). Another example is the inverse relationship between the proportion of discoloured heartwood in sugar maple (Acer saccharum Marsh.) and the lowest daily minimum temperature (Havreljuk et al. 2013). This species has also shown to be sensitive to soil acidification, which may possibly contribute to a decline in tree growth and vigour (Duchesne et al. 2002). The spatial distribution of these environmental conditions may lead to regional variations in the quality of the resource, possibly affecting the wood supply.
In many regions, the large quantities of poor-quality trees are still a problem for the forest industry because the weak market for pulpwood adds pressure to harvest trees that can potentially yield more lumber (Boulet 2015). Due to the lack of buyers, the pulpwood is frequently left in the forest (Boivin 2003), representing losses in capital for the industry. The large volumes of unused and underutilized wood, along with residues from sawmills (Alderman et al. 1999; Briedis et al. 2011; Levin et al. 2007; Wood and Layzell 2003), represent an important volume of potential raw material for products with high added value (Kuznetsov et al. 2015; Mabee and Saddler 2010). Enhanced wood products have been seen
as one of the solutions to overcome the problem generated by the great amount of low-quality wood present in the stands (Briggs 2010).
Since the last few decades, a large diversity of enhanced products from wood and bark has been developed to take advantage of the available raw material, representing a new avenue for this underutilized material (Lavoie and Stevanovic 2006; Royer et al. 2012) without putting at stake the production of veneer, lumber and pulp. These high value-added products include bio-fuels (Mabee and Saddler 2010), bio-chemicals (D'Souza et al. 2014), extracts (Laavola et al. 2016; Royer et al. 2012), among others. Tree species, such as black spruce (Picea mariana (Mill.) Britton, Sterns & Poggenb.), jack pine (Pinus banksania Lamb.), yellow birch, sugar maple and red maple (Acer rubrum L.) to name a few, have the potential to be used in the nutraceutical, cosmeceutical and pharmaceutical industries (Krasutsky 2006; Royer et al. 2011). This potential is mainly due to the extracts composition and abundance found in those species (Royer et al. 2012). Because of their properties, extracts from birch species have been the object of several studies (Garcia-Perez et al. 2008; Habiyaremye et al. 2002; Seshadri and Vedantham 1971).
Cole et al. (1991), Habiyaremye et al. (2002), Lavoie and Stevanovic (2007) and Seshadri and Vedantham (1971) characterized the main triterpenes found in the wood and bark of yellow birch. Diouf et al. (2009) and St-Pierre et al. (2013) quantified also the total phenols, flavonoids and proanthocyanidins. Some of the main compounds found in the yellow birch wood are betulonic acid and acetyl methyl betulinate (Diouf et al. 2009; Lavoie and Stevanovic 2007). The bark of the same species is also rich in pentacyclic triterpenes, more specifically lupenone, lupeol and betulin (Cole et al. 1991; Habiyaremye et al. 2002; Seshadri and Vedantham 1971).
Many of those chemical compounds have properties that could be explored by the pharmaceutical and nutraceutical industry. Their efficacy against cancer (Alakurtti et al. 2006; Dehelean et al. 2012; Serafim et al. 2014), diabetes (Anhê et al. 2013) and certain types of viruses (Baltina et al. 2003; Pavlova et al. 2003; Tolstikov et al. 2005) has been demonstrated in several studies. In some cases, plant extracts such as phytosterols, have potential to reduce cholesterol levels (Best et al. 1954; Rideout et al. 2012) or, like the total
phenols, to act as antioxidant (Garcia-Perez et al. 2008; Royer et al. 2011). Some authors have suggested that these enhanced products should be integrated with the traditional products industry to increase profit of the value chain and strengthen it through the diversification of products (Chambost et al. 2008; FPAC 2011; Simard et al. 2012).
The integration of high value-added products in the value chain would be in line with the proposed strategies for sustainable management of Quebec’s forests, which aims the use of forest resources in conjunction with the conservation of natural forests main attributes and restoration of ecosystems (MFFP 2015b). One of the challenges of sustainable forest management is to maximize wood value while respecting the production capacity of the different ecosystems. With that in mind, this thesis proposes a possible approach for restoration of the northern hardwood forests that could concomitantly diversify the range of hardwood-based products and increase profitability without changing hardwood harvesting levels. Taking into consideration the evolving forest industry, this work is based on two major points: understanding the factors that influence lumber value, and identifying an alternative use for the low-value wood and bark common in the study region.
HYPOTHESES
This work was set up to test two hypotheses. The first one is that the distribution of lumber value in a large landbase is explained by a combination of stand- and site-level characteristics. That means that at least part of the expected variation is associated with factors other than tree characteristics, such as soil properties and climate conditions.
The second hypothesis is that the production of high value-added extracts from low-value wood and bark would increase profit generation in at least 20% when compared to the status quo, where only traditional wood products are obtained from low-quality forest stands.
WORK OUTLINE
This thesis is composed of three chapters. The first chapter presents how lumber value varies across the province of Quebec. Its main objective was to improve the understanding of the factors associated with the large-scale variation in wood quality in northern hardwoods. The lumber value recovery was used as an indicator of stand quality, evidencing the variations within the hardwood forests of eastern Canada. Stand, site and climatic variables were used to describe how certain areas have a higher potential for producing high-quality hardwoods.
When analysing the value of the wood resource, one usually takes into consideration only the traditional wood products, i.e. lumber, veneer and pulp. However, the forest industry has the potential and capacity to explore other resources and generate other products, such as energy and enhanced wood-based products. The second chapter evaluated the potential for producing high value-added extracts from sawmill coproducts and residues. This was achieved by assessing the quantity of major ethanolic extracts (total phenols, major triterpenes and phytosterols) in yellow birch wood and bark samples, and also by analysing the relationship between the ethanolic extracts content and selected tree characteristics. The idea was to demonstrate that other avenues could be explored by the forest industry.
The third chapter estimated to which extent the inclusion of betulin extract to the forest products portfolio could extend the conditions in which a selection cut is profitable. It demonstrates, through a simulation, how some stands generate very low profit when only the traditional products are aimed as source of revenue. This chapter also pointed to the possibility of increasing profit by adding the production of a selected type of extract to the current value chain.
CONSIDERATIONS AND ASSUMPTIONS
As in any work, it is nearly impossible to capture all the angles of a subject. Therefore, some considerations are made to better set the scope of this thesis. In the first
chapter, the effect of past management strategies was not evaluated, despite its high influence on the current stand characteristics (Nyland 2002). The reason for not including past management decisions in the lumber value recovery analysis was because of the lack of consistent data. Including incomplete or unreliable data might lead to erroneous inferences (Gelman and Hill 2007).
In the second chapter, where quantification of total phenols, triterpenes and phytosterols was performed, ethanol was chosen as solvent exactly for being able to extract all of those components (Conde et al. 2013; Rizhikovs et al. 2015). There are other options for solvents and methods of extraction (Hua et al. 1991; Krasutsky 2006; Lavoie and Stevanovic 2006). However, since the objective of this work was to quantify total phenols, triterpenes and phytosterols, a tested and proven method for extracting all of those components was chosen (Diouf et al. 2009; St-Pierre et al. 2013).
The wood value chain analysed in the third chapter is a simplified version of actual value chains (D'Amours et al. 2008). Costs and losses associated with many of the intermediate sections of supply and value chains, such as log terminals and warehouses, were not taken into consideration. Since the objective of this chapter was to evaluate the profitability of the hardwood industry after the inclusion of betulin in the product portfolio, many important aspects of the wood value chain, such as the transportation costs, were simplified. Another important observation regarding the simplification of the value chain is that betulin was the only non-traditional product added to the portfolio. However, there are several chemical compounds that can be obtained in the same extraction and that could be added to the value chain. Moreover, the residues from this process could also be used for producing energy or they could be sold to other industries.
In this work, value chain is defined as “a strategic collaboration of organizations for the purpose of meeting market objectives over the long term and for the mutual benefit of all ‘links’ of the chain”, based on Kozak and Maness (2005). They also state that a value chain approach should be based on a collaborative effort, rather than a competitive approach. This was taken into consideration when analysing the hardwood value chain in the third chapter.
Finally, the assumptions made in the third chapter regarding the profitability of selection cuts are based on results from simulations performed using MÉRIS, a software program used in the province of Quebec (MFFP 2015a). As with any model, the results from these simulations are not real, and should not be considered as such. Some of the models that are part of MÉRIS have a stochastic nature; therefore, certain variations in the results are expected (Grigoriu 2012).
Chapter 1: LARGE-SCALE VARIATIONS IN LUMBER VALUE RECOVERY OF
YELLOW BIRCH AND SUGAR MAPLE IN QUEBEC, CANADA
ABSTRACT
Silvicultural restoration measures have been implemented in the northern hardwoods forests of southern Quebec, Canada, but their financial applicability is often hampered by the depleted state of the resource. To help identify sites most suited for the production of high-quality timber, where the potential return on silvicultural investments should be the highest, this study assessed the impact of stand and site characteristics on timber quality in sugar maple (Acer saccharum Marsh.) and yellow birch (Betula alleghaniensis Britt.). For this purpose, lumber value recovery (LVR), an estimate of the summed value of boards contained in a unit volume of round wood, was used as an indicator of timber quality. Predictions of LVR were made for yellow birch and sugar maple trees contained in a network of more than 22000 temporary sample plots across the Province. Next, stand-level variables were selected and models to predict LVR were built using the boosted regression trees method. Finally, the occurrence of spatial clusters was verified by a hotspot analysis. Results showed that in both species LVR was positively correlated with the stand age and structural diversity index, and negatively correlated with the number of merchantable stems. Yellow birch had higher LVR in areas with shallower soils, whereas sugar maple had higher LVR in regions with deeper soils. The hotspot analysis indicated that clusters of high and low LVR exist across the province for both species. Although it remains uncertain to what extent the variability of LVR may result from variations in past management practices or in inherent site quality, we argue that efforts to produce high-quality timber should be prioritized in sites where LVR is predicted to be the highest.
1 INTRODUCTION
Forestry practices in the northern hardwood forests have for many decades favoured the selective harvesting of the most valuable trees available, which has resulted in the general depletion of the resource (Deluca et al. 2009; Nyland 1992; Robitaille and Roberge 1981). To reverse this trend and promote forest restoration, new stem marking rules were introduced in the public forests of Quebec, Canada, to ensure harvesting of low-vigour trees in selection cuts (Boulet 2007; Delisle-Boulianne et al. 2014). However, the current state of these forests can affect the financial applicability of this silvicultural system (Pothier et al. 2013). The often low-quality wood obtained from low-vigour trees and the reduced demand for pulpwood limit the capacity to apply such forest restoration measures in northern hardwood forests.
The search for solutions to this problem has mainly focused on improving the stem selection process during harvesting operations. Pothier et al. (2013) argued that among non-vigorous stems expected to die before the next scheduled cut, those that have maintained high quality should be selected for harvest. This could be achieved by establishing a marking priority for non-vigorous trees exempt from cracks and external signs of fungal infections, and with a diameter at breast height approaching 40 cm (Havreljuk et al. 2014). In addition to applying such rules within a given cutblock, the strategy should also consider the variability among sites, so that restoration measures can be applied where the potential return is the highest (Cockwell and Caspersen 2014). However, evaluating the propensity of a site for the production of high-quality timber is complicated by 1) the fact that the characteristics of the current resource are influenced both by the intrinsic characteristics of the site and by the effects of past disturbances, particularly high-grading practices and 2) the multiple potential definitions for ‘timber quality’.
Whereas the long-term effects of various silvicultural scenarios have been documented to some extent in the literature (Bédard and Brassard 2002; Nyland 2005), there is much less information about how site characteristics might affect the quality of timber from northern hardwoods. In one of the few studies available on the subject, Havreljuk et al. (2013) described the regional pattern of variation in the proportion of discoloured heartwood in yellow birch (Betula alleghaniensis Britt.) and sugar maple (Acer
saccharum Marsh.). Although tree size and growth rate descriptors accounted for most of the regional variability, results also showed a negative correlation between the red heartwood proportion in sugar maple stems and extreme minimum temperature at a given site. For yellow birch, Gagné et al. (2013) found a correlation between stem quality and the mean annual precipitation. Such correlations between wood properties and site characteristics have also been observed in other species and forest types (Jiang et al. 2007; Lenz et al. 2014; Moore et al. 2009; Pokharel et al. 2014). Despite the fact that true causal links between site characteristics and wood properties have yet to be elucidated, these studies tend to confirm that some sites have a higher potential than others for producing wood of high quality in a given species. However, among the several factors known to induce variations in wood quality within and between stems, the effects of site characteristics arguably remain the least documented (Macdonald and Hubert 2002).
The concept of timber quality implies the association with a specific end-use (Briggs 2010; Briggs and Smith 1986). In northern hardwoods, the production of sawn boards for the subsequent manufacture of ‘appearance’ wood products, such as flooring and furniture, is usually the primary processing option that generates the most monetary value (Havreljuk et al. 2014). Consequently, lumber value recovery (LVR) may be used as an indicator of timber quality. First described by McCauley and Mendel (1969) in the context of sawmilling studies, LVR is essentially an estimation of the summed monetary value of boards contained in a unit volume of round wood. Because stumpage prices are also determined per unit volume of round wood, LVR can be considered as a tangible estimate of ‘value’, as perceived by a sawmill purchasing logs and selling boards on the National Hardwood Lumber Association market (NHLA 2011). An advantage of this metric is that it is less dependent on tree size than the summed value of boards contained in an entire tree (Auty et al. 2014; Barrette et al. 2012), making it more useful for making comparisons between different sites. In the context of forest management, over time LVR could also reflect the effectiveness of forest restoration initiatives in northern hardwood forests.
To improve our understanding of the factors associated with the large-scale variation in wood quality in northern hardwoods, our aim in this study was to explore site-to-site variations in LVR for yellow birch and sugar maple in the forests of Quebec,
Canada. These species were chosen both for their abundance in the North American temperate deciduous forests and for their economic importance. To pursue our objective, we first reassessed an existing lumber value model in order to make reliable estimates of LVR for trees contained in a large provincial forest inventory database. Next, using ‘ensemble’ methods, we developed a statistical model to link estimates of this indicator to site- and plot-level descriptors. Specifically, we used a combination of boosting and regression trees for data analysis and model prediction. These methods allowed us to include complex nonlinear interactions between predictors in the models. Finally, we used the predictions of LVR to produce ‘hotspot’ maps to visualize the variation in timber quality potential across the distribution range of sugar maple and yellow birch in the Province.
2 MATERIAL AND METHODS
This study was conducted within the mixed and deciduous forests of Quebec, Canada. As no field work was involved in our study, no life form was put at risk during our work and no field permit was necessary. Temporary sample plot (TSP) data covering the period from 1991 to 2012 were provided by Quebec’s Ministry of Forests, Wildlife and Parks. To be included in the analysis, a plot needed to contain at least one yellow birch or one sugar maple tree with a diameter at breast height (DBH, 1.3 m above the ground) larger than 23 cm. This corresponds to the lowest merchantable diameter limit for sawlogs, in accordance with the hardwood tree grading system used in the province (Monger 1989). A total of 22579 400-m2 TSPs were selected, with the two species occurring concomitantly in 7331 of these. The assessment of LVR is organised in two main parts in the following sections. The first part illustrates how an existing method was reassessed and recalibrated to obtain lumber value estimates for individual trees in each TSP. The second part describes how stand and plot-level characteristics were used to predict LVR values at the landscape level.
2.1 REASSESSMENT OF THE LUMBER VALUE ESTIMATION METHOD
The lumber value of individual stems was estimated using a combination of predictive models and equations. As the first step of this process, the volume by log grade was estimated using the models developed by Fortin et al. (2009) for yellow birch and sugar maple. These models can be applied to predict the occurrence and volume of different log grades in standing trees, as described by Petro and Calvert (1976). According to this classification, logs can be arranged into three categories, namely F1, F2 and F3, in descending order of quality (Vaughan et al. 1966). A fourth grade (F4) was also included to account for the possibility of obtaining sawn wood from short logs (Clement et al. 2005; Giguère 1998), i.e. those with a small-end diameter ranging from 16 to 20 cm and length from 1.2 to 2.4 m. The log grade and volume predictions are based on variables measured in Quebec’s forest inventory, namely tree species, DBH, and tree quality class. The latter is based on Monger’s (1989) classification system, which is analogous to that developed by the US Forest Service for northern hardwoods (Hanks 1976). Using a combination of stem DBH thresholds and the distribution and size of defects along the stem, trees are categorized into four classes, indicating their processing potential (Havreljuk et al. 2014).
Once the volume by log grade was estimated for the TSP dataset, the second step consisted in using the method from Petro and Calvert (1976) to estimate the lumber value contained in a log (LV) of a given quality class (Eq. 1.1):
(1.1)
where LV is lumber value ($US), QI is the quality index (see Eq. 1.2), P is the market price for a class 1C (1 Common (NHLA 2011)) sawn board ($US m-3), and Vsw is the sawn wood
yield of each log (m3), estimated from volume tables by the authors.
This method was created as a way to estimate the value of different log grades while accounting for fluctuations in current sawn wood prices. Petro and Calvert (1976) chose the National Hardwood Lumber Association’s (NHLA) board class 1C as a market price reference, since it was the most important category for both supply and demand. Market prices for the class 1C sawn boards were presented from 1953 to 1972 in their study, when
average prices for yellow birch varied from 65 to 83 $US m-3, and from 58 to 73 $US m-3 for sugar maple.
The QI indicates the relative value of a log in terms of yield of 25-mm-thick (1 inch) lumber pieces. It represents the sum of the sawn wood yield by board class multiplied by the relative price by board class (Eq. 1.2):
(1.2)
where Yc is the sawn wood yield by board class (%) obtained by the authors in a sawmilling
study, and RPc is the ratio (dimensionless) between the market price for sawn wood of a
given NHLA board class and the price for the reference grade (i.e. 1C).
Petro and Calvert (1976) showed that the ratios between the market prices of each NHLA board grade were fairly constant through time. To assess the current applicability of the method, we used data from a sawing study conducted by Havreljuk et al. (2014). The database comprised 32 yellow birch and 64 sugar maple trees sampled from multiple stands in two regions of the province of Quebec. Trees were measured and categorized before being converted into boards. Independently of their dimensions, logs were sawn to maximize the production of high grade lumber (i.e. knot-free and sapwood) (Steele 1984). Boards were graded according to the NHLA standards (NHLA 2011) after being kiln dried. The board classes were later regrouped to match those described by Petro and Calvert (1976), i.e. FAS, 1C, 2C and 3C. The board market prices for calculating RPc were
obtained from the Hardwood Market Report (HMR 2012), averaged for a five-year period, from 2008 to 2012. Equation 1 was then used to calculate the LV.
Even when the prices were updated to current values, the method proposed by Petro and Calvert (1976) tended to underestimate the LV of individual trees (Figure 1.1). The main reason for this bias is the smaller variation in market price among the board quality classes observed by those authors compared to the current price variation (Table 1.1). The range of this variation directly influences the relative price (RPc), one of the key variables
in the calculation of QI. Another factor that influenced the underestimation of LV estimates using Petro and Calvert (1976)’s equations was the volume of sawn boards (Vsw). The
original method underestimated this volume for all log classes, possibly as a result of changes in sawing techniques that occurred over the past four decades. Furthermore, Petro and Calvert (1976) did not consider the use of short logs for producing lumber. We hence considered that a reassessment accounting for current sawmilling practices and market conditions was necessary.
FIGURE 1.1 COMPARISON BETWEEN LV PREDICTED WITH PARAMETERS FROM 1976 AND THE OBSERVED LV
The black dashed line is the reference line, with slope of 1, and the shaded area represents the 95% confidence interval.
TABLE 1.1 RELATIVE PRICES BY NHLA (2011) BOARD CLASS.
Board class 1976 2012
a
Yellow birch Sugar maple
FAS 1.44 1.81 1.65
1C 1.00 1.00 1.00
2C 0.71 0.70 0.69
3C 0.43 0.25 0.37
a Averaged values from 2008 to 2012
After confirming that patterns of dispersion and variation in QI and Vsw were similar
for both species, we then recalibrated the models for the two species combined. Because our intention was to propose a practical way of estimating LV, we tried to eliminate the use of proxy variables. Therefore, instead of simply reassessing Yc and updating RPc using current market values, we decided to simplify the whole concept by modeling QI as a function of gross log volume (Vg) and the volume for the various log classes (VF1, VF2, VF3
and VF4), variables that can be easily obtained (Eq. 1.3):
(1.3)
where VF12, VF22, VF32, and VF42 are the squared lumber volume in each log class and 0,
1, 2, 3, and 4 are the model parameters to be estimated.
Predictions of Vsw were a function of Vg (Eq. 1.4).
(1.4)
where 0 and 1 are the model parameters to be estimated.
The adjusted R-squared values for the QI and Vsw predictive models were 0.50 and
TABLE 1.2 PARAMETER ESTIMATES AND ASSOCIATED STANDARD ERRORS (SE) FOR QI AND VSW PREDICTIVE MODELS.
Model Parameter Estimates SE
QIa 0 0.64 0.090 1 1.09 0.187 2 0.60 0.064 3 0.49 0.039 4 0.48 0.029 Vsw b 0 -0.01 0.001 1 0.58 0.007
a QI, quality index; b V
sw, sawn wood yield of a log.
Once we had obtained predictions for QI and Vsw, we then estimated LV using
Equation 1. After the recalibration, the model provided unbiased estimates of LV (Figure 1.2).
FIGURE 1.2 COMPARISON BETWEEN LV PREDICTED BY THE REASSESSED MODEL AND OBSERVED VALUES.
The black dashed line is the reference line, with slope of 1, and the shaded area represents the 95% confidence interval.
2.2 LVR PREDICTIONS BASED ON STAND AND SITE CHARACTERISTICS
The TSP database contained stand and site descriptors such as stand height class, age class, basal area, slope, altitude, stand density, ecological type, surficial deposit, and drainage class (Berger et al. 2004; Berger and Leboeuf 2008). All these variables were considered as potential predictors in a statistical model that aimed to predict LVR values. In all cases, the predictor variables excluded the characteristics of the subject trees, so that only the influence of the site and stand characteristics on LVR would be considered. The LVR for each tree was obtained by dividing the reassessed LV by the total roundwood volume of that tree.
The species and structural diversity indices were added to the database as potential predictor variables to describe the stand structure and composition. Both these indices were based on the exponential form of Shannon’s index (H’exp) (Shannon 1948), as follows:
(1.5)
In the species diversity index, pi was the proportion of occurrence of one species
and S the total number of species. In the structural diversity index, pi was the proportion of
one DBH class and S the number of DBH classes. A value of zero for these indices indicated that there was no stand-level variability for the index in question.
To test if climatic conditions were associated with variations of LVR, data obtained using BioSIM (Régnière et al. 2014), averaged for the period of 1981 to 2010, were tested along with the previously described variables. These included, for each plot location, the annual mean of daily minimum, mean and maximum temperatures (C), annual total precipitation (mm year-1), annual total snowfall (mm of water), annual mean of daily relative humidity (%), mean wind speed (km h-1), growing season (days), annual potential evapotranspiration (mm), aridity (accumulation of monthly water deficit, mm), annual total radiation (MJ m-2), and number of days with precipitation.
The relationship between LVR and the limits of atmospheric acid deposition in the soil and its exceedance were also tested. The elements used to determine these limits are the maximum critical load of sulfur (CLmax) deposition and its exceedance, both in molar
equivalent for potential acidity (eq ha-1 yr-1). The maximum critical load was obtained based on the methodology described by Ouimet et al. (2001) (Eq. 1.6).
(1.6)
where BCdep is the sum of K, Ca, Mg and Na deposition rates, Cldep is the Cl deposition
rate, BCw is the soil weathering rate of K+Ca+Mg+Na, Bcu is the net K+Ca+Mg uptake rate,
and Alkle(crit) is the critical alkalinity leaching rate, all variables in eq ha-1 yr-1. The
exceedance is the difference between the averages of total annual depositions of sulfur and the maximum critical load for years 1999-2002. The details on the CLmax and exceedance
mapping for Quebec can be found in Ouimet and Duchesne (2009).
2.2.1 BOOSTED REGRESSION TREES
The statistical analyses were performed with the full set of stand, climatic and soil acidity variables as the predictors, and LVR as the response variable. This base LVR was obtained using the reassessed equations of Petro and Calvert (1976). Candidate models were developed and selected using Boosted Regression Trees (BRT), a machine-learning technique that produced a predicted LVR model based on an ensemble of decision trees. The BRT method combines two statistical techniques, namely boosting and regression trees. The latter is a technique that uses decision trees formed by binary splits to build a predictive model, taking into consideration the interactions between variables (Hastie et al. 2001; Sutton 2005). Boosting uses a forward stage-wise procedure, where the regression trees are fitted iteratively to a subset of the training data (Sutton 2005). These subsets are randomly selected without replacement, and the proportion of the training data (bag fraction) can be specified. This procedure, known as stochastic gradient boosting, introduces some randomness into the boosted model, improving accuracy and reducing overfitting (Friedman 2002). Although BRT is normally used for predicting presence/absence data, the works of Moisen et al. (2006), Carslaw and Taylor (2009), and Carty (2011), among others, have shown that it can be efficiently used for modelling continuous variables.
The analyses were made using the R statistical programming environment, version 3.1.2 (2014), using the 'gbm' (Ridgeway 2013) and 'dismo' (Hijmans et al. 2013) packages. Separate models were developed for each species. Before fitting the models, the database for yellow birch and sugar maple was randomly split into training (70%) and validating datasets (30%). Several models were created to verify the combination of BRT parameters, namely the tree complexity (tc), learning rate (lr) and regression trees, that would result in the minimum predictive error. Those models were fitted to the training dataset and, once the best models were selected, the validating dataset was used for evaluating the model predictions.
The BRT method is capable of modelling complex variable interactions (Carslaw and Taylor 2009), by increasing the tc. This parameter was set either as 1 (indicating that no interactions among variables would be allowed), 2 (two-way interaction), 5, 7 or 10. The lr, which determines the contribution of each tree to the model, varied from 0.1 to 0.0001. Slower learning rates are normally preferred because they shrink the contribution of each decision tree, giving more reliable estimates of the response, but at the cost of increased computation time (Elith et al. 2008). Faster learning rates may increase the predictive deviance too rapidly after reaching the minimum, indicating that the method is overfitting the final model. We chose a combination of parameters that would give us reliable predictions with the fastest computing time. For all models, we used a bag fraction of 0.5, meaning that, at each iteration, 50% of the data would be drawn at random, without replacement. The models were run and the results compared to determine the best combination of parameters. The predictive performance of the models was based on the proportion of the deviance explained (D-squared) (Guisan and Zimmermann 2000), and on the root mean squared error (RMSE).
To evaluate the contribution of each term in reducing the overall model deviance and help eliminating non-influential variables (Abeare 2009), an index of relative importance was used. The relative importance of predictor variables was based on the number of times a variable was selected for splitting in the tree weighted by the squared improvement to the model as a result of each split (Breiman et al. 1984). The sum of the
relative contributions was added to 100, and the contribution of each variable was scaled accordingly.
The nature of the relationship between the predictor variables and the response (Friedman and Meulman 2003) was verified with the aid of partial dependence plots. Once the relevant predictors were selected, based on the change of predictive deviance, the model was then simplified to include only the variables that would contribute to the predictions.
2.2.2 PROVINCIAL-SCALE ANALYSIS OF LVR ESTIMATES
The LVR estimates from the BRT method were plotted to maps and a hotspot analysis was performed using the Getis-Ord Gi* method (Getis and Ord 1992) from ArcMap 10.1 (ESRI 2012). This analysis was used to verify whether the predicted values were clustered, meaning that there would be areas with significantly higher and significantly lower LVR within the Province. The interaction between a value and its neighbours was set as a fixed distance band, where the scale of the analysis was constant across the study area (Alqadi et al. 2014). The distance band for each species was determined by using the Spatial Autocorrelation (Global Moran’s I) method from ArcMap 10.1 (ESRI 2012), assuring a 99% likelihood of real clusters. For yellow birch, the distance was 40 km, and for sugar maple, 72 km. This procedure ensured that all features had at least one neighbour during the analysis and that all occurrences outside the fixed distance did not influence the TSP in question.
The clusters were originated by the combination of the z-score (standard deviation) and the p-value (probability of having a spatial pattern created by some random process). The resulting clusters were then categorized as either (i) very significantly higher/lower than mean (=0.01), (ii) significantly higher/lower than mean (=0.1), and (iii) mean.
3 RESULTS 3.1 BRT MODEL
A total of 35 base LVR predictive models per species were tested to verify the combination of parameters that would result in the minimum predictive error, with a stipulation that at least 1000 decision trees should be used for fitting the models. The learning rate and tree complexity values that gave the minimum predictive error for yellow birch were lr=0.003 and tc=10, while for sugar maple the values were 0.01 and 10, respectively. For yellow birch, the D-squared for the full model was 0.13 and the RMSE was 16.13, while for sugar maple the values were 0.17 and 17.79, respectively.
The relative importance of the predictors, along with the change of predictive deviance produced by BRT was used in the selection of the independent variables to simplify the final models. In these final models, D-squared for yellow birch remained 0.13 and the RMSE was 16.13. For sugar maple, the corresponding values were 0.18 and 17.16, almost the same as the values for the full model. A comparison between the predicted and observed LVR values performed to the validating dataset can be seen in Figures 1.3a and 1.3b.
FIGURE 1.3 COMPARISON BETWEEN PREDICTED AND OBSERVED LVR FOR (A) YELLOW BIRCH AND (B) SUGAR MAPLE
3.1.1 PARTIAL DEPENDENCE PLOTS
The partial dependence plots were obtained from term-wise plots of fitted functions versus observed values for the yellow birch and sugar maple final models (Figures 1.4a and 1.4b, respectively). Despite the partial dependence plots not being an exact representation of the relationship between the predictors and the explanatory variables, they can be useful for understanding the nature of these relationships.
FIGURE 1.4A PARTIAL DEPENDENCE PLOTS FOR EACH PREDICTOR RETAINED IN THE FINAL MODELS FOR YELLOW BIRCH
The respective relative importance for each independent variable is in parentheses.
Note: Stand age class: JEQ – young even-aged stand, JET – young stratified stand, JIN – young uneven-aged stand, VEQ – old even-aged stand, VET – old stratified stand, VIN – old uneven-aged stand; Ecological types: FC1x – red oak, FE1x – sugar maple – bitternut hickory, FE2x – sugar maple – basswood, FE3x – sugar maple – yellow birch, FE4x – sugar maple – beech – yellow birch, FE5x – sugar maple – hop hornbeam, FE6x – sugar maple – red oak, MJ1x – yellow birch – sugar maple, MS1x – balsam fir – yellow birch, RT1x – hemlock; Surficial deposit: 1A – glacial deposits without specific morphology, 1B – glacial deposits characterized by their morphology, 2 – fluvioglacial deposits, 3 – fluvial deposits, 4 – lacustrine deposits, 5 – marine deposits, 6 – littoral marine deposits, 7 – organic deposits, 8 – slope and weathering deposits, R –rock substrate. A complete description of the ecological types and surficial deposit can be found at Berger and Leboeuf (2008).
FIGURE 1.4B PARTIAL DEPENDENCE PLOTS FOR EACH PREDICTOR RETAINED IN THE FINAL MODELS FOR SUGAR MAPLE
The respective relative importance for each independent variable is in parentheses.
Note: Stand age class: JEQ – young even-aged stand, JET – young stratified stand, JIN – young uneven-aged stand, VEQ – old even-aged stand, VET – old stratified stand, VIN – old uneven-aged stand; Ecological types: FC1x – red oak, FE1x – sugar maple – bitternut hickory, FE2x – sugar maple – basswood, FE3x – sugar maple – yellow birch, FE4x – sugar maple – beech – yellow birch, FE5x – sugar maple – hop hornbeam, FE6x – sugar maple – red oak, MJ1x – yellow birch – sugar maple, MS1x – balsam fir – yellow birch, RT1x – hemlock; Surficial deposit: 1A – glacial deposits without specific morphology, 1B – glacial deposits characterized by their morphology, 2 – fluvioglacial deposits, 3 – fluvial deposits, 4 – lacustrine deposits, 5 – marine deposits, 6 – littoral marine deposits, 7 – organic deposits, 8 – slope and weathering deposits, R –rock substrate. A complete description of the ecological types and surficial deposit can be found at Berger and Leboeuf (2008).
For both species, LVR from fitting data was positively correlated with stand age class, stand height class, structural diversity index and merchantable basal area, but showed a negative correlation with the number of merchantable stems and the species diversity index. The ecological types associated with high LVR were FC10, FE13, FE50, FE51, FE55, FE61, MJ11P, MJ12P for yellow birch, and FE33P, FE40, FE43, FE45, FE53P,
FE54, MJ18, MS12P, MS14, RT13, RT18 for sugar maple. Those higher LVR values for yellow birch occurred in areas with soil depths varying from medium to shallow with abundant outcrops. For sugar maple, sites with soil depth ranging from medium to deep deposits gave higher LVR values.
A positive correlation between potential evapotranspiration and LVR, and a negative correlation between the average minimum temperature and LVR were observed for yellow birch. There was a complex pattern of variability of LVR as a function of wind speed; LVR values tended to decrease before rising steeply for sites with mean wind speed ranging from 15 to 20 km h-1. From the entire range of snow water equivalent where sugar maple is found in the study region, only the middle of that range gave higher LVR values. A negative correlation between soil critical load exceedance of sulfur and LVR was observed only for sugar maple.
3.1.2 BRT MODEL PREDICTIONS AND HOTSPOT ANALYSIS
‘Hotspot' maps of the BRT model predictions on the validation dataset were plotted for yellow birch (Figure 1.5a) and sugar maple (Figure 1.5b), along with ‘hotspot’ maps performed to the fitting data for comparison. The presence of clusters on the maps indicates the existence of underlying spatial patterns. Even though the predicted values dispersion (=109.9; =4.5; range= 95.5 – 135.7 for yellow birch, and (=116.9; =6.3; range= 94.4 – 187.7 for sugar maple) differed from the fitting data dispersion (=109.8; =16.9; range= 86.1 – 281.7 for yellow birch, and =117.0; =18.5; range= 87.4 – 302.3 for sugar maple),the similarity between the pairs of plots suggests that most of the regional-scale variation in LVR was captured by the BRT model.
Clusters of statistically significant high values for yellow birch were observed in the west side of the province, as well as a small area south of Rimouski. Clusters significantly lower than the mean for the same species were found in the centre of the province, as well as east of Rimouski. For sugar maple, the high and low values occurred in several locations across the province; the regions surrounding Mont-Laurier, Montreal and south of
Rimouski produced higher values, while east and far south of Quebec City, as well as southwest of Mont-Laurier presented lower values.
FIGURE 1.5A DISTRIBUTION OF LVR FOR YELLOW BIRCH, ACROSS THE PROVINCE OF QUEBEC. COMPARISON BETWEEN LVR FROM FITTING DATA (INSET FIGURE) AND PREDICTED LVR FROM BRT
FIGURE 1.5B DISTRIBUTION OF LVR FOR SUGAR MAPLE, ACROSS THE PROVINCE OF QUEBEC. COMPARISON BETWEEN LVR FROM FITTING DATA (INSET FIGURE) AND PREDICTED LVR FROM BRT
4 DISCUSSION
Due to the observed price gap between the most valuable and the least valuable board grades from 1976 and 2012, the original model from Petro and Calvert (1976) was no longer providing a reliable, accurate estimate of LV. When RPc was eliminated from the
reassessed model and only the volume of each log class and the gross log volume were retained as predictors, the results were improved. In the original model, Vsw was obtained
from volume tables. The use of a simple linear model with only one independent variable (Vg) yielded more precise estimates for Vsw, bringing the predictions closer to the observed
values. The discrepancy in predicted Vsw between the models developed in this study and
those of Petro and Calvert (1976) can be largely explained by the inclusion of short logs, which implies that the recalibrated model predicts the maximum LV that could be extracted from a tree. The use of Fortin et al. (2009)'s model to predict volume by log grade likely
