with Ground Penetrating Radar

Transcription

1 Tree Roots in Agroforestry: Evaluating Biomass and Distribution with Ground Penetrating Radar by Kira Alia Borden A thesis submitted in conformity with the requirements for the degree of Master of Science in Forestry Faculty of Forestry University of Toronto Copyright by Kira Alia Borden 2013

2 Tree Roots in Agroforestry: Evaluating Biomass and Distribution with Ground Penetrating Radar Abstract Kira Alia Borden Master of Science in Forestry Faculty of Forestry University of Toronto 2013 The root systems of five tree species (Populus deltoides nigra clone DN-177, Juglans nigra, Quercus rubra, Picea abies, and Thuja occidentalis) are described following non-intrusive imaging using ground penetrating radar (GPR). This research aimed to 1) assess the utility of GPR for in situ root studies and 2) employ GPR to estimate tree root biomass and distribution in an agroforestry system in southern Ontario, Canada. The mean coarse root biomass estimated from GPR analysis was 54.1 ± 8.7 kg tree -1 (± S.E.; n=12), within 1 % of the mean coarse root biomass measured from matched excavations. The vertical distribution of detected roots varied among species, with T. occidentalis and P. abies roots concentrated in the top 20 cm and J. nigra and Q. rubra roots distinctly deeper. I evaluate these root systems based on their C storage potential and complementary root stratification with adjacent crops. ii

3 Acknowledgments I am grateful for the guidance and support of my co-supervisors Dr. Marney Isaac and Dr. Sean Thomas. They both continually amaze and inspire me with their expertise and passion for their work. Additionally, I appreciate the contributions to my graduate research made by committee members Dr. Marie-Josée Fortin and Dr. Jing Chen. Stephanie Gagliardi was extremely integral in completing field and lab work. I thank all those associated with the Isaac and Thomas labs, notably Vasuky Thirugnanassampanthar for assistance in the lab and Jake Munroe and Janise Herridge becoming last minute field assistants. I thank Ian Kennedy, Tony Ung, and the administration staff from the Faculty of Forestry as well as Chai Chen, Tom Meulendyk, Tony Adamo, and the administration staff at UTSC. Importantly, my research benefited from the collaborative nature of the project with those at the University of Guelph Agroforestry Research Station. I thank Dr. Andrew Gordon, Dr. Naresh Thevathasan, and their staff and students for all the logistical and field work support. A special acknowledgment to Amy Wotherspoon for her exceptional organization, hard work at the field site during the excavations, and graciously providing me with root biomass data. I am grateful for funding support from the Faculty of Forestry, Agriculture and Agri- Food Canada s Agriculture Greenhouse Gases Program, and the Natural Sciences and Engineering Research Council. Additionally, the Centre for Global Change Science Graduate Award enhanced my graduate experience by allowing me to attend a short course in Montpellier, France and present my research at the International Symposium for Root Research in Dundee, Scotland. Lastly, to my family and friends who gave me encouragement during the last two years while I was out looking for tree roots with radar, I thank you. iii

4 Table of Contents Abstract... ii Acknowledgments... iii Table of Contents... iv List of Tables... vi List of Figures... vii List of Appendices... ix List of Abbreviations... x Chapter 1 : Tree Roots in Agroecosystems Introduction Current ecological issues for agriculture in Ontario Temperate tree-based intercropping The importance of tree root studies Ground penetrating radar for tree root studies Thesis approach... 7 Chapter 2 : General Methods Case study: University of Guelph Agroforestry Research Station Site conditions during radar survey Tree species Radar survey Chapter 3 : Estimating coarse root biomass with ground penetrating radar in a treebased intercropping system Abstract Introduction Ground penetrating radar for root biomass estimation Materials and Methods Radargram processing GPR index biomass relationship Coarse root biomass estimates Root carbon content Statistical analysis Results GPR images and index biomass relationship Coarse root biomass Root system C content Discussion The GPR index biomass relationship Biomass and carbon estimates of tree root systems Application and limitations of GPR in tree-based intercropping Conclusions iv

5 Chapter 4 : Evaluating vertical distributions of tree roots in a tree-based intercropping system with ground penetrating radar Abstract Introduction Root distribution in TBI systems Materials and Methods Coarse root distribution measurements with GPR Accuracy testing Root distributions Fine root distribution in crop rows Statistical analysis Results GPR detection frequency of coarse roots Coarse root distribution Root distribution into crop rows Fine root distribution Discussion Coarse root detection Tree root distribution in TBI systems Tree and crop root stratification Conclusions Chapter 5 : Conclusions The use of GPR for tree root study Tree-based intercropping Final conclusions References Appendices v

6 List of Tables Table 1 Relational equations between specific root biomass (W) representing dry weight of 10 cm long root segments (g) with root diameter (D) (cm) (n=20 per species) Table 2 Analysis of covariance for root biomass measured in exposed profiles and GPR response index. Displayed are the results for data inclusive of all species and the corrected relationship that removed species main effect on remaining pooled species data. Significant results (p<0.05) are in bold Table 3 GPR estimated coarse root biomass (BGPR) (kg tree -1 ; mean ± S.E.) with corresponding excavated biomass of five tree species (kg tree -1 ; mean ± S.E). Paired t-tests completed on the means for each species and across all study trees between BGPR and excavated biomass. Also shown are corresponding calculated allometric estimates of coarse roots dependent on DBH and the species class and group Table 4 Carbon concentration (%) and C content (kg C tree -1 ) of the coarse root system of five tree species (25 years old). Carbon concentration values are reported as total carbon of dry root weight (mean ± S.E.) (n=3, except Populus sp. and Picea abies n=2) following i) conventional oven dry sample preparation or ii) volatile inclusive methodology. Carbon content of the trees root systems were calculated using BGPR and the species-specific coarse root C concentration (volatile inclusive) Table 5 Minimum, maximun and mean (m) depth of coarse roots detected by GPR for five tree species (mean ± S.E.) (n=3 except n=2 for Populus sp. and Picea abies). Total tree and subdivided crop row and tree row data. Values followed by the same letter in a column for each parameter indicates non-significant (p<0.05) result (Tukey HSD) Table 6 Cumulative distribution coefficient (β) for five tree species at the University of Guelph Agroforestry Research Station, Canada. The distribution coefficients (β) were calculated using species means of pooled detected coarse roots in 0.10 cm depth increments and fitted to the function Y = 1- β d where Y is proportion of roots at depth (d). Results from subsetted distribution data for coarse roots located in crop or tree rows are also presented. Depth (m) to estimated cumulative 95 % of coarse roots (d95) presented vi

7 List of Figures Figure 1 Conceptual diagram of thesis objectives and structure Figure 2 Satellite image of the University of Guelph Agroforestry Research Station, Guelph, Ontario, Canada. Site was established in 1987 as a tree-based intercropping experimental site. Insert is a schematic diagram of the planting design Figure 3 Image of intercropping rows at the University of Guelph Agroforestry Research Station, Guelph, Ontario, Canada Figure 4 Image of 1000 MHz ground penetrating radar (GPR) unit used for geo-image, or radargram, data collection during this study. Unit pictured being pulled along soil surface with attached odometer measuring distance along transect near Juglans nigra Figure 5 (a) Image of grid set-up for GPR survey of Quercus rubra in tree-based intercropping system at the University of Guelph Agroforestry Research Station, Guelph, Ontario, Canada. (b) Plan-view schematic of GPR data collection grid design. The base of tree stem is represented by the circle approximately in the centre of the grid. Red lines indicate GPR data collection transects in x and y directions with 10 cm spacing Figure 6 Radargram of transect 0.70 m from Picea abies stem. Signal velocity was 0.09 m ns Figure 7 Plan view of interpolated radagrams assembled in grid orientation and visualized for signal response within depth 0.30 to m. Grid shown is surrounding Picea abies. The magnitude of reflected signals are represented on colour scale from no reflection detected (dark blue) to high reflection (red) Figure 8 GPR data processing sequence with soil profile ( m) and exposed roots (circles) of Juglans nigra (a) and the equivalent GPR geo-image (b) with applied background removal (c) surface reflections or banding is reduced. Hyperbola migration focuses root reflections to foci (c). The final data processing step is the Hilbert transformation whereby magnitude of reflection is brought into one phase (e). The extracted GPR index (area within an intensity range (175 to 255); cm 2 ) is measured (f) to develop a GPR index biomass relationship Figure 9 Transformed data used to test for species main effect and interactive effect with GPR index on measured biomass (ANCOVA). Solid symbols represent corrected pooled species data with no significant species main effect (p=0.20), or interactive effect with GPR index (p=0.68), on biomass (Table 2). Open circles are isolated data points collected below Thuja occidentalis representing the corrected data used for that species GPR index biomass estimation equation (p<0.0001) vii

8 Figure 10 GPR index (area above a threshold; cm 2 ) as a predictor of dry weight coarse root biomass (g) determined from exposed soil profiles and correlated with equivalent section of GPR radargram. Corrected relationships displayed from i) pooled data (solid symbols inclusive of Populus sp., Juglans nigra, Quercus rubra, and Picea abies) identified by solid line (y = 0.214x 4.72; r = 0.55; n=51) and ii) isolated Thuja occidentalis (open circles and dotted line) (y = 0.038x 4.62; r = 0.95; n=12) Figure 11 Positive correlation (solid line) between coarse root biomass estimated by GPR (kg tree -1 ) and coarse root biomass measured from matched excavations (kg tree -1 ) (R 2 =0.75; p=0.0003; RMSE=14.4 kg; n=12). The 1:1 relationship also shown (dotted line) Figure 12 Mean detection frequency of coarse roots grouped by diameter size class (n=3 except n=2 for Populus sp. and Picea abies). No significant differences between species means of each diameter class (ANOVA) or between diameter class of same species (paired t-test) (p>0.05). Bars represent ± S.E. of the mean Figure 13 Mean detection frequency at each soil depth interval (n=13 exposed profiles) measured from subset of matched soil profiles and radargrams. Means with same letter are not significantly different (Kruskal-Wallis) (p>0.05). Bars represent ± S.E. of the mean Figure 14 Total detected coarse root frequency by depth to 0.80 m for each species (n=3 except Populus sp. and Picea abies n=2). Bars represent ± standard error of the mean Figure 15 Fine root length density (RLD; mean ± S.E. cm cm -3 ) by depth inclusive of 3 sampling distances (1.0, 1.5, and 2.0 m) from tree stem into crop rows. Five tree species (n=18 except n=12 for Populus. sp. and Picea abies). Samples collected during spring May-June Same letters represent non-significant differences among species means for each sampling depth using non-parametric Kruskal-Wallis test (p>0.05) viii

9 List of Appendices A1: Bulk density of soil within tree study plots (mean ± S.E.; n=36 except n=24 for Populus sp. and Picea abies) A2: GPR estimated (BGPR), excavated, and allometrically derived coarse root biomass (kg tree -1 ) ix

10 List of Abbreviations BGPR coarse root biomass estimated from GPR analysis (dry weight; kg tree -1 ) β d95 DBH EM GHG GPR R:S TBI cumulative proportion of root distribution coefficient depth to 95% of the cumulative root frequency tree stem diameter at 1.3 m aboveground (cm) electromagnetic greenhouse gases ground penetrating radar root:shoot tree-based intercropping x

11 1 Chapter 1 : Tree Roots in Agroecosystems 1.1 Introduction Current ecological issues for agriculture in Ontario The population of Ontario is projected to grow 28.6 %, an additional 17.4 million people, by the year 2036 (Ontario Ministry of Finance 2013). This growth will occur predominantly in Southern Ontario where urban expansion and rich agricultural soils overlap (Francis et al. 2012). Consequently, there will be additional strain on the landscape from intensifying agriculture that can negatively impact the environment, for instance increasing erosion into waterways and releasing greenhouse gases (GHG) into the atmosphere (Paustian et al. 1997). Furthermore, the resilience of landowners to withstand economic pressures from the expansion of urban development will be challenged (Francis et al. 2012). Therefore, future landscape management must aim to ensure stable and sufficient economic incentives for landowners while reducing the environmental externalities commonly associated with intensive agriculture, acknowledging that agriculture has a major role in GHG mitigation (FAO 2007; Jose 2009; Smith et al 2012). Current practices that mediate soil degradation and GHG emissions include low or no-till practices (Paustian et al. 1997), increased use of cover crops (Rosenzweig and Hillel 2000), and afforestation on marginal lands (Paustian et al. 1997; Laganière et al. 2010; Foote and Grogan 2010). Commonly, these applied management techniques protect soils and increase inputs and the stabilization of organic matter (Rosenzweig and Hillel 2000). Agroforestry is an alternative land management approach that can further incorporate organic matter into agricultural soils and capture atmospheric CO2. Agroforestry, broadly described as agricultural systems that incorporate woody perennials, are more prevalent in tropical regions in part due to access to tree species that are fast growing or N2 fixing (Jose et al. 2004). While these desirable tree qualities are limited in temperate regions, there are other

12 2 ecological benefits derived from the presence of trees in agricultural landscapes. For example, compared to conventional monocropped agriculture, increases in biodiversity, soil fertility, nutrient cycling, erosion control, water quality, and C storage have been reported in temperate agroforestry systems (Dixon et al. 1994; Schroth 1999; Jose et al. 2004; Thevathasan and Gordon 2004; Oelbermann et al. 2004; Thevathasan et al. 2008; Jose 2009). In Ontario, agroforestry is commonly found in the form of windbreaks and riparian buffers, but there is further potential in the adoption of multi-species systems that integrate trees more directly within agricultural fields such as tree-based intercropping (TBI) (Gordon and Williams 1991; Thevathasan et al. 2008) Temperate tree-based intercropping In TBI, or alley cropping, trees and crops are typically planted in alternating rows. Tree species selection and planting design (e.g. stem spacing and crop row widths) can be modified across sites and over time (e.g. through thinning practices) depending on ecological constraints and economic incentives (Gordon and Williams 1991). Ecological constraints such as tree-crop competition for soil moisture can depress crop yields (Jose et al. 2004), while economic incentives and the valuation of the crops and tree products can influence relative importance to the landowner. For example, timber or other tree products such as nuts might offset the loss in crop yield (Gordon and Williams 1991). Simultaneously, C sequestration in trees on agricultural landscapes is increasingly studied as a potential means to capture atmospheric CO2 (Winjum et al 1992; Dixon et al. 1994; Montagnini and Nair 2004; Thevathasan and Gordon 2004; Isaac et al 2005; Peichl et al.2006; Oelbermann et al. 2004; Nair et al. 2009; Evers et al. 2010) and future C credit incentives offer promise for diversified income to landowners (FAO 2007). Therefore, with the increased complexity of multi-functional systems such as TBI there are a multitude of interacting factors, both ecological and economic, to consider when attempting to optimize the system.

13 3 After decades of research, Nair (2011) voiced concern that there still remains a deficiency of research and quality of measurements pertaining to the biophysical processes occurring in agroforestry. However, there have been a number of published studies contributing to our understanding of temperate TBI systems from the University of Guelph Agroforestry Research Station, Ontario, Canada. Prior studies have been completed on interspecies competitive interactions (Reynolds et al. 2007; Clinch et al. 2009), C storage potentials (Thevathasan and Gordon 2004; Peichl et al. 2006), N inputs from leaf litter (Thevathasan and Gordon 1997), microclimate regulation (Clinch et al. 2009), earthworm populations (Price and Gordon 1999), soil fertility (Bambrick et al. 2010), and mychorrhizal communities (Bainard et al 2012). Additionally, some in situ investigations of tree root systems have been conducted. The root system of Populus deltoides nigra clone DN-177 is the most documented with biomass data from two complete excavation studies (Thevathasan and Gordon 2004; Peichl et al. 2007), one of which also excavated the root system of Picea abies (Peichl et al. 2007). However, these belowground studies were limited to very low replication (n 3) due to methodological constraints. Consequently, thorough root system studies are lacking, resulting in a gap in the literature pertaining to temperate TBI systems and for agroforestry more broadly (Oelbermann et al 2004; Nair 2011) The importance of tree root studies The distribution of roots measured on the vertical profile, referred to in this thesis as root distribution, is functionally an important area of research in TBI systems. Competition for resources can occur aboveground for light (Reynolds et al. 2007), belowground for soil resources (Jose et al. 2000), or as an interactive effect between the two (Freschet et al. 2013). As TBI is commonly cultivated with fertilizer use, thus meeting nutrient requirements of the crops, belowground competition is often attributed to reduced soil moisture that in turn inhibits nutrient

14 4 availability for the crops (Jose et al. 2000; Fletcher et al. 2012). Thus, as crop roots prevalently occupy shallower soil, trees with root systems deeper than crop roots might mediate some belowground competition (Schroth 1999; Jose et al. 2001; Mulia and Dupraz 2006). Furthermore, with an increased stratification of rooting zones between species in agroforestry systems, tree roots can act as a safety net capturing nutrients leached below the crop roots thus maximizing total resource use of a system (Van Noordwijk and Pernomosidhi 1995; Jose et al. 2004; Bergeron et al. 2011; Thevathasan et al. 2012). Finally, determining the biomass of tree roots within TBI systems is essential for accurate C budgets of these systems and by failing to include tree roots in biomass and C inventories, more than 20% of trees biomass is overlooked (Cairns et al. 1997; Mokany et al. 2006; Brunner and Godbold 2007). To study root system distribution, conventional sampling techniques include minirhizotrons, profile walls, and soil cores (Vogt et al. 1998; Polomski and Kuhn 2002). However, due to the spatial heterogeneity and the inherent fractal nature of root systems, these techniques generally have low accuracy particularly for coarse roots (diameter > 2mm) (Schroth and Kolbe 1994; Taylor et al. 2013). Conversely, partial or complete root system excavations can capture this heterogeneity, but they are time consuming, destructive, and non-repeatable. This has led to biomass study in agroforestry to adopt root:shoot (R:S) ratios or allometric equations generated from studies in forest ecosystems (e.g. Kirby and Potvin 2007; Chauhan et al. 2011). However, generalities of tree biomass allocation may produce inaccuracies when applied to agroforestry, or TBI sites specifically, due to variation of tree root growth in cultivated scenarios (Nair 2011; Kuyah et al. 2012). Owing to the challenges associated with the inaccessibility to study tree roots in situ, the belowground component of trees is known as the hidden half (Vogt et al. 1998; Bhattachan et al. 2012) and consequently there is a pressing need for techniques that non-intrusively measure tree root systems.

15 Ground penetrating radar for tree root studies Over the last 15 years, ground penetrating radar (GPR) has been tested and applied as a geo-imaging tool to detect trees coarse roots under controlled conditions (Barton and Montagu 2004; Dannoura et al. 2008; Hirano et al. 2009; Bassuk et al. 2011; Cui et al. 2013; Tanikawa et al. 2013; Guo et al. 2013b; Guo et al. 2013c) or in field experiments (Hruska et al. 1999; Butnor et al. 2001; Butnor et al. 2003; Stover et al. 2007; Samuelson et al. 2008; Zenone et al. 2008; Samuelson et al. 2010; Hirano et al. 2012; Isaac and Anglaaere 2013; Raz-Yaseef et al. 2013; Day et al. 2013). GPR emits electromagnetic (EM) signals into the ground and records the reflected signals amplitude, polarity, and travel time, which can be used to interpret belowground features. EM signals are reflected where there is contrast of dielectric permittivity between two media such that: R = k 1 k 2 k 1 + k 2 (1) where R is the reflection coefficient and k denotes the dielectric constant, of the first medium (k1) and second medium (k2) (Davis and Annan 1989). Importantly, water has a high dielectric constant compared to soil (k water = 80 vs. ɛ dry sand = 5) (Davis and Annan 1989). Therefore, as coarse roots can contain higher water content than the surrounding soil matrix, coarse roots provide the necessary interface for radar signal reflections (Hirano et al. 2009; Guo et al. 2013b; Guo et al. 2013c). Depth of GPR signal penetration is limited by the frequency of the GPR unit and the electrical conductivity of the subsurface (Davis and Annan 1989). Additionally, the water and clay content of the soil will influence the rate of radar signal attenuation due to their dielectric properties (Butnor et al. 2001). Thus, ideal conditions for GPR study of coarse roots are well drained soils with low clay content and with coarse roots of sufficient water content (Guo et al. 2013a).

16 6 As a GPR unit operates along a transect, sequential signal responses can be compiled to create an interpreted subsurface image of the soil profile, or radargram. A hyperbolic reflection pattern is generated in the radargram as the GPR moves above a coarse root perpendicular to the direction of GPR travel. These root reflection patterns can be correlated to biomass, as addressed in Chapter 3, or identified across multiple radargrams to chart the coarse root distribution, as addressed in Chapter 4. The benefit of this approach is that it provides researchers the opportunity for repeated measurements of coarse roots and the capacity to examine the uniqueness of tree root system response along gradients of biotic and abiotic constraints (Isaac and Anglaaere 2013). Recently, the first application of GPR to measure distribution of coarse roots in tropical agroforestry systems was used for comparative analysis of tree rooting patterns across different edaphic and management conditions (Isaac and Anglaaere 2013). This study was the first to use GPR for multi-species tree root investigation, specifically at a temperate agroforestry site, and conducted both an analysis of biomass and root distribution.

17 7 1.2 Thesis approach Throughout my thesis, there are two main objectives: 1) to investigate the utility of GPR to quantify tree roots in situ, and 2) to contribute to the knowledge of tree root architecture in temperate TBI systems. I organized my thesis chapters not by these objectives, but in the two forms of measurements made. Chapter 3 focuses on amount, specifically root biomass and C content whereas Chapter 4 focuses on location, specifically vertical root distribution. My approach to data analysis was different between these chapters with distinct methodological and ecological applications (Figure 1). As such, I derived my research questions for Chapter 3: 1) Can ground penetrating radar estimate coarse root biomass of trees in the multispecies scenario of a TBI system and how accurate are such measures? 2) Does root biomass vary between tree species in TBI systems thereby making some trees preferential for C storage purposes? And for Chapter 4: 1) Can ground penetrating radar accurately describe the vertical distribution of coarse roots in the studied TBI system? 2) How do the vertical root distributions vary between tree species in this system, and notably within crop rows, thereby making some trees preferential for optimal belowground resource use? As the study site, study trees, and GPR data collection protocol are common to both Chapters 3 and 4, these components are described first in Chapter 2: General Methods.

18 Figure 1 Conceptual diagram of thesis objectives and structure. 8

19 9 Chapter 2 : General Methods 2.1 Case study: University of Guelph Agroforestry Research Station GPR data and root samples were collected below 13 individual trees at the University of Guelph Agroforestry Research Station, Guelph, Ontario, Canada (44 32'28 N, 80 12'32 W; elevation 325 m) (Figure 2). The 30 ha site was established in 1987 as an experimental agroforestry systems system (specifically tree-based intercropping). A variety of tree species are planted in tree rows that are spaced 12.5 or 15 m apart, in between which a conventional crop rotation is practiced under no-till cultivation and annual intercropping with Zea mays (maize), Glycine max (soybean), Triticum aestivum (winter wheat), or Hordeum vulgare (barley) (Thevathasan and Gordon 2004) (Figure 3). The soil is classified as Grey-Brown Luvisol with a sandy loam texture (65% sand, 25% silt, and 10% clay) (Oelbermann and Voroney 2007). The Ap horizon continues to a depth of 28 to 53 cm (Price and Gordon 1999) and a moraine till is located approximately 1 m below the soil surface Site conditions during radar survey Water content of coarse roots and soil within study plots was determined by weighing samples before and after drying at 65 C for 7 days. Coarse root gravimetric and volumetric (assuming cylindrical root shapes) water contents was 62 ± 1% (mean ± S.E.; n=65) and 76 ± 11% (n=93) respectively. Soil gravimetric and volumetric (using known soil sampler volume of 100 ml) water content of the soil at the site was 9 ± 1% and 12 ± 1% respectively (mean ± S.E.; n=20) and did not vary significantly by depth. Thus, the necessary soil to root water content gradient was satisfied for GPR detection and interpretation (Hirano et al. 2009; Guo et al. 2013c) with coarse roots containing more than six times more water than the soil on a volume basis. Soil bulk density ranged between 0.98 ± 0.02 g cm -3 (mean ± S.E.; n=36) and 1.20 ± 0.04 g cm -3

20 Figure 2 Satellite image (Google Earth 2013) of the University of Guelph Agroforestry Research Station, Guelph, Ontario, Canada. Site was established in 1987 as a tree-based intercropping experimental site. Insert is a schematic diagram of the planting design. 10

21 Figure 3 Image of intercropping rows at the University of Guelph Agroforestry Research Station, Guelph, Ontario, Canada. 11

22 12 (n=24) in the top 0.20 m of soil and for soil at 0.40 and 0.60 m depths bulk density ranged between 1.15 ± 0.03 (n=36) and 1.40 ± 0.03 (n=36) (Appendix 1). 2.2 Tree species The 13 study trees included five species (Populus deltoides nigra clone DN-177, Juglans nigra, Quercus rubra, Picea abies, and Thuja occidentalis) that are commonly selected for agroforestry systems in the region (Gordon and Williams 1991). Populus deltoides nigra clone DN-177 (hybrid poplar) has low-value wood, but poplar hybrids are the more commonly studied species for calculating C storage potential in these systems as they are extremely fast growing (Evers et al. 2010). While root architecture among poplar hybrids can vary, Populus deltoides is known to have roots nearer to the surface, but with some deep vertical roots in sandy and moist soil conditions (Burns and Honkala 1990). Juglans nigra L. (black walnut) is a commonly intercropped tree in temperate regions valued for both its wood and nut production (Jose and Gillespie 1998; Gordon and Williams 1991). The root system of J. nigra is known to produce a both a distinct taproot and also have a strong lateral root system (Burns and Honkala 1990) and of note is the alleopathic exudates, juglone, from roots (Jose and Gillespie 1998). Quercus rubra L. (red oak), is also valued for its wood, has a large root system that includes a taproot and many oblique roots (Jose et al. 2001). Picea abies L. Karst. (Norway spruce) is characterized by a strong horizontal spreading root system with vertically descending roots that emerge from the lateral roots (Drexhage and Gruber 1998). For this study, these four aforementioned tree species had stem diameters (DBH) 18 cm (see Appendix 2) and aboveground heights > 7 m. The fifth study species was the multi-stem Thuja occidentalis L. (eastern white cedar), which can have a root system that can grow in extremely shallow substrate and is a common choice for hedgerows (Burns and Honkala 1990; Kelley et al. 1992). The T. occidentalis examined this study had aboveground heights < 6 m. The trees were planted with 6

23 13 m in-row tree stem spacing, except T. occidentalis with 1 m spacing. Each species was replicated with 2 or 3 randomly selected trees. All trees included in the study were approximately 25 years old and were located in flat areas of the site. 2.3 Radar survey Tree root detection typically occurs with a GPR antenna emitting a centre frequency of 500 MHz (e.g. Barton and Montagu 2004) up to 2000 MHz (e.g. Cui et al. 2013). For this study, we used a 1000 MHz GPR unit (NogginPlus; Sensors and Software Inc., Mississauga, ON, Canada) (Figure 4) for all data collection. Previous studies using this frequency have reported detection coarse roots of diameter 0.5 cm and greater (Guo et al. 2013a). The area around the base of each target tree was cleared of leaf litter and other organic material. A m grid frame was installed surrounding the base of each target tree so the tree stem was situated in the approximate centre (Figure 5a). The grid frame was constructed of plastic pipe, which served to anchor guide-rope at 0.10 m increments in both the x and y directions (Figure 5a). To ensure straight and square transects, the GPR unit was pulled by an attached handle so that the antennae remained alongside the grid guide-rope. Data collection of 13 trees occurred between April and June, Transect increments of 0.10 m were selected for GPR data collection to reduce dependency on interpolation. Each set of tree GPR data were collected on the same day and under consistent conditions with the GPR programmed to EM signal emission intervals of 0.1 ns stacked with 16 traces every 5 mm along each transect. To measure the average EM signal velocity in the subsurface, metal rods were inserted horizontally at depths of 0.40 m into soil profiles adjacent to target trees. The travel time and distance of the GPR signal to the rods were measured and thus average velocity of the radar signal was calculated. Velocities were measured the day of data collection for each tree and ranged between 0.08 and 0.10 m ns -1. Radar signal attenuation became severe at depths of

24 14 approximately 0.80 cm. Each radargram contained the radar signal response data adjusted to the appropriate velocity, interpreting depth (Figure 6). The radargrams were associated with x and y data from the orientation of the grid design (Figure 7) and were the starting point for the analyses described in the subsequent two chapters.

25 15 Figure 4 Image of 1000 MHz ground penetrating radar (GPR) unit used for geo-image, or radargram, data collection during this study. Unit pictured being pulled along soil surface with attached odometer measuring distance along transect near Juglans nigra. Figure 5 (a) Image of grid set-up for GPR survey of Quercus rubra in tree-based intercropping system at the University of Guelph Agroforestry Research Station, Guelph, Ontario, Canada. (b) Plan-view schematic of GPR data collection grid design. The base of tree stem is represented by the circle approximately in the centre of the grid. Red lines indicate GPR data collection transects in x and y directions with 10 cm spacing.

26 16 Figure 6 Radargram of transect 0.70 m from Picea abies stem. Signal velocity was 0.09 m ns -1. Figure 7 Plan view of interpolated radargrams assembled in grid orientation and visualized for signal response within depth 0.30 to m. Grid shown is surrounding Picea abies. The magnitude of reflected signals are represented on colour scale from no reflection detected (dark blue) to high reflection (red).

27 17 Chapter 3 : Estimating coarse root biomass with ground penetrating radar in a tree-based intercropping system 3.1 Abstract Conventional measurements of tree root biomass in tree-based intercropping (TBI) systems may not accurately represent the heterogeneity of rooting patterns or are highly destructive and nonrepeatable. In this study, I estimated total coarse root biomass using ground penetrating radar (GPR) of 25-year-old trees inclusive of five species (Populus deltoides nigra, Juglans nigra, Quercus rubra, Picea abies, and Thuja occidentalis) at a TBI site in Southern Ontario, Canada. Subsurface images generated by GPR were collected in grids ( m) centred on tree stems. The predictive relationship developed between GPR signal response and root biomass was corrected for species effect prior to tree scale estimates of belowground biomass. Accuracy and precision of the tree scale estimates was assessed comparing coarse root biomass measured from complete excavations of the corresponding tree. The mean coarse root biomass estimated from GPR analysis was 54.1 ± 8.7 kg tree -1 (mean ± S.E.; n=12), within 1 % of the mean coarse root biomass measured from excavation, and for all trees there was a root mean square error of 14.4 kg between measured and estimated biomass with no detectable bias despite variable conditions within the in-field and multi-species study. Root system C storage is reported by species, calculated with species-specific root carbon concentrations, to range between 5.4 ± 0.7 to 34.8 ± 6.9 kg C tree -1 at this site. GPR is an effective tool for non-destructively predicting coarse root biomass in multi-species environments such as temperate TBI systems.

28 Introduction With more than 20 % of total tree biomass allocated belowground, tree roots comprise a substantial but understudied component of biomass in many ecosystems (Cairns et al. 1997; Mokany et al. 2006; Brunner and Godbold 2007). Ecological benefits derived from the presence of tree roots, particularly in modified agricultural systems, include soil amelioration via root exudation, turnover, and sloughing, improved water infiltration and aeration from root channels, and prevention of erosion (Schroth 1999; Jose et al. 2004; Thevathasan and Gordon 2004; Jose 2009; Nair et al. 2009). Furthermore, the incorporation of trees into agricultural landscapes can increase the C storage potential of a landscape, considerably within belowground biomass, and is touted as a viable and potentially significant land-use approach to sequester atmospheric CO2 (Dixon et al. 1994; Isaac et al. 2005; Peichl et al. 2006; Bambrick et al. 2010; Kuyah et al. 2012). While potential aboveground C sequestration in temperate agroforestry systems is estimated at Mg C year 1 (Oelbermann et al. 2004), data are limited on the belowground contribution. Improved approaches to chart and predict the belowground biomass are required to fully capture the role of roots in C budgets for temperate TBI. Biomass inventories within agroforestry systems are primarily focused on measuring the aboveground component with a limited number of studies also calculating root mass (e.g. Oelbermann et al. 2005; Peichl et al. 2006; Kirby and Potvin 2007; Moser et al. 2010; Kessler et al. 2012). Conventional sampling techniques of tree root systems include minirhizotrons, profile walls, and soil cores (Vogt et al. 1998; Polomski and Kuhn 2002). However, due to the spatial heterogeneity and the inherent fractal nature of root systems, these techniques generally have low accuracy particularly for coarse roots (diameter > 2 mm) (Schroth and Kolbe 1994; Taylor et al. 2013). Conversely, partial or complete root system excavations are time consuming, destructive, and non-repeatable. This has led to biomass study to adopt root:shoot ratios (e.g. Jackson et al.

29 ; Cairns et al. 1997; Mokany et al. 2006) or allometric equations (e.g. Kurz et al. 1996; Jenkins et al. 2003) generated from biomass studies in forest ecosystems. However, generalities of tree biomass allocation may produce inaccuracies when applied to agroforestry, or TBI sites specifically, due to variation of tree root growth in cultivated scenarios (Nair 2011; Kuyah et al. 2012). Thus, there is a need for new methodologies to study and measure root biomass in TBI systems Ground penetrating radar for root biomass estimation Previous research in the use of ground penetrating radar (GPR) for biomass estimation have been conducted in controlled experiments (Barton and Montagu 2004; Dannoura et al. 2008; Hirano et al. 2009; Cui et al. 2013; Guo et al. 2013c; Tanikawa et al. 2013) and in situ within tree monocultures or two-species mixtures (Butnor et al. 2003; Stover et al. 2007; Samuelson et al. 2008; Samuelson et al. 2010; Hirano et al. 2012; Raz-Yaseef et al. 2013; Day et al. 2013). Generally, physical samples of root biomass equating to corresponding areas of GPR radargrams have been compared to develop predictive relationships between a GPR signal response index and root biomass (Butnor et al. 2003; Stover et al. 2007; Guo et al. 2013a). A growing body of research has used this approach to explore root responses to variation in atmospheric CO2 (Stover et al. 2007; Day et al. 2013), forest stand management techniques (Butnor et al. 2003; Samuelson et al. 2008; Samuelson et al. 2010), and precipitation patterns (Raz-Yaseef et al. 2013). However, the utility of GPR biomass estimation has yet to be tested in a multi-species scenario or in an agroforestry system specifically. Therefore, the objectives of this study were to test the utility of GPR for coarse root biomass estimation across a variety of tree species and in field conditions and to calculate the C content of the belowground biomass in a temperate TBI system.

30 Materials and Methods See Chapter 2: General Methods for description of study site, study trees, and radar survey Radargram processing Prior to image analysis of the GPR data, non-root anomalies (e.g. plane reflectors and signal noise ) were reduced with a sequence of noise reduction steps (DC shift, dewow, and background removal) (Figure 8b and c). Subsequently, dipping features, such as hyperbolic reflections of roots, were repositioned to their foci with a migration algorithm (2d FK migration with Stolt equation using known signal velocity and the angle of incidence) (Figure 8d). Finally, an envelope algorithm known as the Hilbert transformation (amplitudes of the reflected EM waves are used to interpret the data into one phase) was applied so that reflectors are more discernible (Figure 8e). In order to enhance subsurface root reflections representatively with depth, a spreading and exponential compensation gain (SEC2) was applied to all processed GPR radargrams based on the rate of energy decay, similar to methodology in Cui et al. (2013) and Guo et al. (2013c). A colour palette of bipolar grey was selected for visualization whereby low to high amplitude response was displayed as grey to white. All GPR data processing steps were completed in EKKO_View Deluxe (Sensors and Software Inc.). GPR radargrams were imported into ImageJ (US National Institutes of Health, Bethesda, MD, USA) as 8-bit bmp files. Radargrams were standardized to ensure consistent measures of distance within an image using a ratio of 400 pixels:1 m. The final data processing step measured the number of pixels within a threshold range (ImageJ) and subsequently converted to a cross sectional area (cm 2 ) (Figure 8f), defined as the GPR index GPR index biomass relationship Following GPR data collection, soil profiles (0.25 m across and 1 m deep) were exposed at distances ranging from 0.5 to 2.0 m from the tree stem (n=64) (Figure 8a).

31 Figure 8 GPR data processing sequence with soil profile ( m) and exposed roots (circles) of Juglans nigra (a) and the equivalent GPR geo-image (b) with applied background removal (c) surface reflections or banding is reduced. Hyperbola migration focuses root reflections to foci (c). The final data processing step is the Hilbert transformation whereby magnitude of reflection is brought into one phase (e). The extracted GPR index (area within an intensity range (175 to 255); cm 2 ) is measured (f) to develop a GPR index biomass relationship. 21

32 22 These soil profiles were orientated along GPR data transects. To calculate total coarse root dry weight biomass in each excavated soil profile, the diameters of all coarse roots crossing through the exposed profiles were measured and applied to species-specific relationships of root dry weight biomass to root diameter as described in Hirano et al. (2012) and Guo et al. (2013b) (Table 1). Relationships were developed from root samples of various diameters collected at each tree and assuming a length of 10 cm, equivalent to the GPR transect spacing. To determine the GPR index, three pixel intensity thresholds (165 to 255, 175 to 255, and 200 to 255) were selected based on visually delineating root features and minimizing the incorporation of non-root anomalies. The area derived from each threshold level were compared for their correlations to coarse root biomass. The pixel intensity threshold of 175 to 255 (Figure 8f) was ultimately selected to generate the area index of GPR signal response as it produced the optimal correlation with measured biomass Coarse root biomass estimates The radargram processing sequence as described in Radargram processing was applied to all GPR data within each tree grid. The resulting GPR index determined for each transect was used to calculate the predicted root biomass using the GPR index biomass relationship. All transect biomass estimates were summed for total tree coarse root biomass (BGPR; kg tree -1 ). Since tree roots are best detected when crossing between 45 and 135 to the plane of the radargram (Butnor et al. 2001; Tanikawa et al. 2013), the biomass estimates from grid transects in both x and y directions were included to maximize root detection by capturing roots irrespective of direction of growth. An additional estimate of coarse root biomass for the study trees was completed using conventional allometric equations from Jenkins et al. (2003) that employed measured DBH and

33 23 Table 1 Relational equations between specific root biomass (W) representing dry weight of 10 cm long root segments (g) with root diameter (D) (cm) (n=20 per species). Tree species Biomass relationship to diameter a r Populus sp. W = D J. nigra W = D Q. rubra W = D P. abies W = D T. occidentalis W = D a It was assumed that root diameter was constant for 10 cm to be comparable to GPR transect spacing of 10 cm.

34 24 parameters associated with species class (hardwood or softwood) and species group (hardwoods: Populus sp. = aspen/alder/cottonwood/willow, J. nigra = mixed hardwood, and Q. rubra = hard maple/oak/hickory/beech ; softwoods: P. abies = spruce, and T. occidentalis = cedar/larch ) (Jenkins et al. 2003). The coarse root biomass was estimated such that: BGBratio = exp(β0 + β1/dbh) (2) where β0 and β1 are parameters fitted from species class data (Jenkins et al. 2003), DBH is the diameter at breast height (cm), and BGBratio is the ratio of coarse root biomass to aboveground biomass (AGB) whereby: AGB = exp(β0 + β1 lndbh) (3) where AGB is total aboveground biomass (kg), DBH is the diameter at breast height (cm), and β0 and β1 are parameters fitted from species group data (Jenkins et al. 2003) Root carbon content Coarse roots were randomly selected during complete excavation. Sample preparation and analysis follow C volatile-inclusive methodology as suggested by Thomas and Martin (2012) as volatile compounds lost during oven drying are being realized as a non-negligible amount of C necessary for more accurate C content estimates (Lamlom and Savidge 2003; Thomas and Malczewski 2007; Martin and Thomas 2011). Roots were placed in air-tight bags and transported in a cooler to the laboratory at which point roots were washed to remove soil and stored in airtight bags at -5 C. Coarse roots were cut into cylindrical segments ~1 cm in length to retain representative proportions of root tissue. Root samples were dried in an 8 L freeze dryer (Labconco Co., Kansas City, MO, USA) for seven days. The largest diameter roots were weighed for constant mass on the final day to ensure complete drying. For each tree species (n=3, except P. abies and Populus sp. n=2), coarse root samples were prepared as composite samples inclusive of four coarse root diameter classes (0.2 to 0.5,

35 to 1.0, 1.0 to 2.0, and > 2.0 cm). Samples were ground in a ball grinder (Restch MM400 Mixer Mill) and stored in snap-cap 1 ml containers at -5 C. Total C for each sample was determined using a CHN analyzer (Thermo Flash 2000). Samples were weighed on a microbalance for total sample mass. Elemental analysis calibrations were completed prior to each sample run using aspartic acid. Known standards (SRM , Thermo Scientific) were tested during analysis to confirm instrument accuracy. As biomass estimates are reported on a dry weight basis, the C concentrations (%) measured from freeze-dried samples were converted onto a dry weight basis whereby: C = [MC / (MF (VMF MF))] 100 (4) where MF was the mass of the freeze-dried sample used for elemental analysis and MC was the mass of C in MF. The species mean volatile mass fractions (VMF) applied to equation 4 was the species-specific fraction of biomass lost during heating methods and was calculated whereby: VMF = (MF MH) / MF (5) where an additional subset of freeze-dried samples were weighed (MF) and oven dried at 105 C for 48 hours and weighed again after drying (MH). Carbon concentration analysis was repeated with oven dried roots to evaluate methodology. Finally, the total root C content for each tree species was calculated by applying the resultant C concentration values of coarse roots to GPRderived estimates of coarse root biomass Statistical analysis The GPR index biomass relationship was developed by correlating the GPR response (area of processed radargrams above an intensity threshold; cm 2 ) and the coarse root biomass (g) from spatially matched subsurface soil profiles to a depth of 1 m to develop the predictive equation. Cook s distance identified a datum with significant influence on the regression. The datum represented an excavated soil profile section containing an extremely large J. nigra root (>

36 26 10 cm) of non-cylindrical shape and deemed justifiable to be discarded due to the uncertainty of applying a diameter biomass relational equation (Table 1). An analysis of covariance (ANCOVA) tested for species main effect on biomass as well as the interactive effect of species and GPR signal response (GPR index species) on biomass in order to identify required corrected predictive equations among species. An assessment of precision of BGPR was completed by comparing BGPR to the coarse root biomass measured from matched excavated study plots (Wotherspoon et al. unpublished data) using paired t-tests on means and a linear regression for all study trees. Differences among species for BGPR and C concentration were tested using one-way analysis of variance (ANOVA). Prior to parametric tests, data were confirmed for equality of variance using Bartlett test and for normal distribution of residuals using Shapiro-Wilk test. Statistical analyses were completed in R v (R Foundation for Statistical Computing, Vienna, Austria) and the level of significance was set at p< Results GPR images and index biomass relationship Signal noise and planar reflections, specifically from surface reflections, were reduced following the image processing sequence (Figure 8b and c). Hyperbola migration and the Hilbert transformation were successful in emphasizing root reflections in the radargrams (Figure 8d and e). The selected image intensity threshold (pixel intensity between 175 and 255) delineated these areas of high GPR signal response (Figure 8f). This allowed for detectable roots to be converted quantitatively to the cross sectional area (cm 2 ) on the radargram, which became the GPR index. The area bounded in detected root signals within the subset of radargram profiles ranged from 1.56 cm 2 to cm 2.

37 27 Measured coarse root biomass in the exposed soil profiles were positively correlated with the GPR index extracted from the matched radargrams (r=0.47; n=63). However, there was a species main effect on biomass (p = ) (Table 2) that necessitated the development of corrected relationships based on species. When the data collected below T. occidentalis were separated from the other four species ( pooled ), the GPR index remained significantly correlated to biomass (p<0.0001). There was no species main effect or GPR index and species interactive effect on biomass for the remaining pooled species data (Table 2; Figure 9). The resulting GPR index biomass predictive equation for the corrected pooled species was y = 0.214x 4.7 (r=0.55; n=51) and the GPR index biomass predictive equation of the corrected T. occidentalis was y = 0.039x 4.6 (r=0.95; n=12). These two relational equations were used for biomass estimation at the tree scale (Figure 10) Coarse root biomass BGPR was 54.1 ± 8.7 kg tree -1 (mean ± S.E.) (n=12), regardless of species, and the mean coarse root biomass measured from excavation was 54.8 ± 8.3 kg tree -1 (n=12), and not significantly different (p=0.876; Table 3) (see Appendix 2). One tree replication of J. nigra was omitted for concern of it being a source of error due to an unusually small measurement of its excavated biomass of 8.3 kg (close to 10% to coarse roots measured from either of the other two J. nigra trees). The remaining two replications of J. nigra resulted in equivalent means of BGPR and excavated biomass. BGPR of Q. rubra was a slight overestimate of 4 % and the BGPR of P. abies was the largest overestimate of 24 %. In contrast, the BGPR of Populus sp. was 54.6 ± 6.0 kg (n=2) opposed to an excavated mean of 71.9 ± 10.8 kg (n=2), an underestimation of 32 %. Thuja occidentalis had a BGPR of 11.8 ± 1.5 kg tree -1 (n=2), an underestimation of the excavated amount by 16 %. The resulting correlation of BGPR and excavated coarse root biomass resulted in a linear relationship with no evident bias of prediction capabilities of the GPR methodology

38 28 Table 2 Analysis of covariance for root biomass measured in exposed profiles and GPR response index. Displayed are the results for data inclusive of all species and the corrected relationship that removed species main effect on remaining pooled species data. Significant results (p<0.05) are in bold. df SS MS F P All data: GPR index < species GPR index species residuals Corrected pooled data without T. occidentalis: GPR index < species GPR index species residuals

39 29 Table 3 GPR estimated coarse root biomass (BGPR) (kg tree -1 ; mean ± S.E.) with corresponding excavated biomass of five tree species (kg tree -1 ; mean ± S.E). Paired t-tests completed on the means for each species and across all study trees between BGPR and excavated biomass. Also shown are corresponding calculated allometric estimates a of coarse roots dependent on DBH and the species class and group. Tree species BGPR (kg tree -1 ) Excavated (kg tree -1 ) n t-test p-value Allometric (kg tree-1) a Populus sp ± ± ± 3.5 Juglans nigra 75.0 ± ± ± 5.3 Quercus rubra 77.0 ± ± ± 2.3 Picea abies 62.0 ± ± ± 12.8 Thuja occidentalis b 11.8 ± ± ± 4.7 all study trees 54.1 ± ± ± 7.9 a Allometric equations from Jenkins et al. (2003). b Due to 1 m stem spacing for T. occidentalis, 1 or 2 additional trees were located within the 3 replicates of the root study area. The trees located near the study plot boundary were assumed to contribute 50% of their root biomass and adjusted accordingly during analyses.

40 Figure 9 Transformed data used to test for species main effect and interactive effect with GPR index on measured biomass (ANCOVA). Solid symbols represent corrected pooled species data with no significant species main effect (p=0.20), or interactive effect with GPR index (p=0.68), on biomass (Table 2). Open circles are isolated data points collected below Thuja occidentalis representing the corrected data used for that species GPR index biomass estimation equation (p<0.0001). 30

41 Figure 10 GPR index (area above a threshold; cm 2 ) as a predictor of dry weight coarse root biomass (g) determined from exposed soil profiles and correlated with equivalent section of GPR radargram. Corrected relationships displayed from i) pooled data (solid symbols inclusive of Populus sp., Juglans nigra, Quercus rubra, and Picea abies) identified by solid line (y = 0.214x 4.72; r = 0.55; n=51) and ii) isolated Thuja occidentalis (open circles and dotted line) (y = 0.038x 4.62; r = 0.95; n=12). 31

42 32 (r 2 =0.75; p=0.0003) and a root mean square error (RMSE) of 14.4 kg (Figure 11). Overall, the estimates derived from the applied allometric equations were less accurate or precise than BGPR, underestimating the mean excavated biomass by 19 % (Table 3), and resulted in a weaker correlation (R 2 =0.60; RMSE=16.7 kg) Root system C content Among the five species, the concentration of C in the coarse roots was 45.9 ± 0.6 % (mean ± S.E.; n=5 species) (Table 4). Carbon concentrations varied from 44.7% to 48.1 % (in J. nigra and P. abies, respectively), though no significant variation was found among species (p=0.361). Of note, the volatile inclusive methodology captured and additional 2.1 ± 0.8 % (mean ± S.E.; n=5 species) of C lost during oven drying methods. The C content of tree root systems using coarse root estimates from GPR data ranged between T. occidentalis with 5.4 ± 0.7 kg C tree -1 (mean ± S.E.; n=3) to Q. rubra 34.8 ± 6.9 kg C tree -1 (n=3) (Table 4), although no significant variations were detected among species (p=0.361). Overall, the mean C content of tree root systems at this site was estimated at 25.7 ± 5.4 kg C tree -1 (n=5 species), which scales to the landscape level as 2.9 Mg C ha Discussion The GPR index biomass relationship The use of a linear GPR index biomass relationship was suitable in this study due to low variability of coarse root water content (Guo et al. 2013c) and a large difference between root and soil water contents (Hirano et al. 2009). A correlation of r=0.89 between biomass from soil cores and GPR index was found in a study completed by Samueslon et al. (2008). They used a 1.5 GHz unit in a Pinus taeda plantation on sandy loam soils and correlated subsurface data exclusively to a depth of 30 cm. The correlation from Day et al. (2013), also using a 1.5 GHz GPR unit, was r=0.69 between biomass from soil cores and a GPR index inclusive of subsurface

43 Figure 11 Positive correlation (solid line) between coarse root biomass estimated by GPR (kg tree -1 ) and coarse root biomass measured from matched excavations (kg tree -1 ) (R 2 =0.75; p=0.0003; RMSE=14.4 kg; n=12). The 1:1 relationship also shown (dotted line). 33

44 34 Table 4 Carbon concentration (%) and C content (kg C tree -1 ) of the coarse root system of five tree species (25 years old). Carbon concentration values are reported as total carbon of dry root weight (mean ± S.E.) (n=3, except Populus sp. and Picea abies n=2) following i) conventional oven dry sample preparation or ii) volatile inclusive methodology. Carbon content of the trees root systems were calculated using BGPR and the species-specific coarse root C concentration (volatile inclusive). Tree species C concentration of coarse roots (%) i) conventional dry C concentration of coarse roots (%) ii) volatile inclusive Coarse root C content from BGPR (kg C tree -1 ) Populus sp ± ± ± Juglans nigra 44.5 ± ± ± Quercus rubra 42.2 ± ± ± Picea abies 47.8 ± ± ± Thuja occidentalis Average (n=5 species) 41.4 ± ± ± ± ± ± a Study site tree stem density is assumed to be 111 trees ha -1. Coarse root C content at site level (Mg C ha -1 ) a

45 35 data to a depth of 60 cm in a scrub-oak ecosystem co-dominated by Quercus myritfolia and Quercus geminate on sandy soils in Florida, USA. Although it should be expected that GPR index biomass correlations will be reduced when there is an increase of depth of radar analysis and an increase of the variability in subsurface conditions, here I include roots to a depth of 1 m while still maintaining a reasonably strong correlation (r=0.55) inclusive of four different tree species. The corrected relationship for T. occidentalis showed a very strong correlation (r=0.95) in part due to the shallower root system of this relatively smaller tree species. The utility of GPR biomass estimation across a landscape would be greater given the applicability of one GPR index biomass relationship to apply to all radargram data. From the results of this study, two corrected predictive equations were appropriate in order to remove any significant species effect on predicted biomass. Similarly, Butnor et al. (2003) developed corrected relationships of GPR index biomass for two contrasting scenarios of fertilizer use or no fertilizer use in a Pinus taeda stand, which altered the soil conditions for radar signals. Given reasonably consistent subsurface conditions (e.g. clay content) and soil-root moisture gradients, corrected GPR index biomass relationships may be required for scenarios of distinct biomass gradients, a reality for temperate TBI systems Biomass and carbon estimates of tree root systems Inclusion of the fraction of volatile C lost during high-heat (105 C) drying was confirmed for improved accuracy of C content estimates. The highest root C concentration was for coniferous P. abies, consistent with previous reported trends where coniferous trees have a higher concentration of C than deciduous trees in temperate regions (IPCC 2006; Thomas and Martin 2012). Peichl et al. (2006) reported coarse root C concentrations, using conventional oven-drying methods, of 13-year-old P. abies at 51 %, which is ~3 % greater than those found in our study. Conversely, carbon concentration for Populus sp. roots found by Peichl et al. (2006)

46 36 (43 %) were ~3 % lower than our reported values inclusive of volatiles, but proximal to the results from oven dry methods. Bert and Danjon (2006) detected variation between the interior root wood and the exterior root bark (bark + phloem) of Pinus pinaster, with the root bark was ~3 % greater in concentration. They also reported diameter-dependent variation notably for roots < 4 cm (Bert and Danjon 2006). I did not test for variation within root tissues, but acknowledge that there may be within-root variation of C concentrations dependent on the ratio of root components, which would be inherently affected by the diameter of root samples used during elemental analysis. Additional sources of variability in C concentration of tree roots may arise from variation in sampling and C analysis protocol as well as physiological variation (Lamlom and Savidge 2003), such as tree root carbohydrate storage (Bert and Danjon 2006). In order to show the C storage potential of trees in temperate tree-based intercropping systems, system level root C quantification was calculated using the current hardwood tree density of 111 trees ha -1. However, it should be noted that if only coniferous trees are integrated, the tree density will be much higher due to lower spacing used for coniferous trees. The estimated root C content at this site indicates an increase of belowground C storage over the last 12 years when compared to the root C content of the average reported values for 13-year-old Populus sp. and P. abies at the same site (1.8 Mg C ha -1 ) (Peichl et al. 2006). In temperate TBI systems, stem density, species composition, and the age of trees are highly variable. Thus reporting species root biomass and root C content at the tree scale is valuable for operational purposes that are specific to these variables (Thevathasan and Gordon 2004) Application and limitations of GPR in tree-based intercropping There are limitations to the amount of biomass GPR can detect. For example, coarse roots smaller than 1.0 cm in diameter are less likely to cause radar signal response than larger roots ( 1 cm) (Hirano et al. 2012), coarse roots located deeper than GPR signal penetration can be

47 37 undetected or misinterpreted by the GPR signal response (Hirano et al. 2009), and coarse roots located outside of the field of the radar signals, such as directly below the stem, will be undetected (Samuelson et al. 2008). As a result of these detection limitations there is an anticipated bias towards GPR underestimation of coarse root biomass, assuming proper calibration and appropriate conditions. With the exception of Populus sp., this bias was not seen in our results suggesting that the subsurface conditions and the morphological characteristics of the root systems at this study site were conducive for radar study. However, some detected biomass might be attributed to false positive GPR signal response incurred from in-field conditions. Although GPR does require some destructive calibration sampling, such as soil cores, the amount of physical sampling required to estimate the coarse root biomass is drastically reduced compared to conventional studies in TBI systems. During the current study, 243 m 2 of surface area were scanned with GPR, an area equating to over 13,500 soil cores (of 15 cm diameter). GPR techniques can provide more thorough understanding of the heterogeneity of the root systems without total excavation in agroforestry systems (Isaac and Anglaaere 2013) and unlike destructive sampling, this method of root data collection can be repeated, critical for temporalscale studies on root system dynamics (Norby and Jackson 2000). I tested the use of preestablished species-based allometric and root:shoot equations, a more traditional approach to quantify root biomass, and found less accurate estimates of the excavated root biomass as compared to the GPR estimates. Generalized equations derived from forest ecosystem data might be unsuitable for trees in agricultural landscapes where variations in management (e.g. planting density and fertilizer application) can induce differences in biomass allocation. Results from this study suggest that the overall precision of the allometric estimate was outperformed by GPR estimations in comparison to the excavated biomass and supports the need for more site and

48 38 species-specific tree root data. Recent advancements in modelling radar signal response given variable root and soil conditions have been reported following controlled experiments (Guo et al. 2013b; Tanikawa et al. 2013). With these advancements, there is potential of enhancing accuracy of root estimation for in-field conditions and lessening the need for destructive sampling for calibrations. 3.6 Conclusions Coarse root biomass of 12 trees, inclusive of five species, was accurately estimated with the use of GPR at a TBI site in Southern Ontario, Canada. Therefore, the use of GPR technique to quantify belowground biomass at the system level with diverse tress species may be of importance. Subsequently, C content of tree root systems was quantified using species-specific coarse root C concentrations. This was the first in-field study to test the robustness of GPR as a coarse root biomass estimation tool across multiple species. Corrected predictive relationships between GPR signal response and root biomass were required to remove species effect, namely isolating data from a species with a distinctly smaller and shallower root system. I argue that this technology can be suitable for use in temperate TBI systems under well drained, sandy loam soils. Ultimately, these results contribute to furthering methodological techniques of GPR root study for direct quantification of belowground biomass and C storage in agroforestry systems and other tree-based ecosystems.

49 39 Chapter 4 : Evaluating vertical distributions of tree roots in a treebased intercropping system with ground penetrating radar 4.1 Abstract Within tree-based intercropping (TBI) systems, tree root architecture contributes to important ecological interactions, such as belowground resource competition that can influence the success of adjacent crops. Yet the belowground component of trees in TBI systems remains largely understudied due to methodological constraints. I used ground penetrating radar (GPR) to detect coarse root locations and extracted fine roots from soil cores to determine the root distributions below thirteen study trees of five species (Quercus rubra, Juglans nigra, Populus sp., Picea abies, and Thuja occidentalis) at a TBI site in Guelph, Ontario, Canada. Coarse root locations were identified across the radargrams (visualized as radar signal reflections) from m survey grids and provided root distribution data in the soil profile. Coarse roots detected by GPR accounted for 80.6 ± 0.8% (mean ± S.E.; n=13) of large coarse roots ( 1 cm) and 43.4 ± 0.5% (n=13) of small coarse roots (< 1 cm) that were later exposed in a subset of matched soil profiles. There was significant variation in mean depth of detected coarse roots among species at the treescale (p=0.001) and also for coarse roots detected solely in crop rows (p=0.031). Fine root densities at each sampling depth varied among species (p<0.0001). Results suggest that a conventionally applied asymptotic curve that describes the cumulative fraction of coarse roots with depth might not be appropriate in describing the rooting profile of trees in TBI systems. I evaluate tree species root distribution in the context of selection and management of trees that promotes complementary root stratification between trees and crops.

50 Introduction Important belowground ecological processes in tree-based intercropping (TBI) systems, such as nutrient and water acquisition, are highly influenced by tree root architecture (Ong et al. 1991; Schroth 1999; Jose et al. 2001; Jose et al. 2004; Oelbermann et al. 2004). Furthermore, roots play a significant role in increasing soil aggregate stability (Udawatta et al. 2008) and C sequestration potential as biomass storage and as contributors to soil organic matter (Peichl et al. 2006; Oelbermann and Voroney 2007; and see Chapter 3). However, the benefits of carbon storage and soil quality must balance with the economic requirements for TBI, specifically the yield demands for adjacent crops (Schroth 1999; Livesley et al 2000; Thevathasan and Gordon 2004). Reductions in yields were identified at the study site in Southern Ontario, particularly for corn (Thevathasan and Gordon 2004; Reynolds et al. 2007). While these reductions were attributed to aboveground competition for light (for corn in particular as it is a C4 plant) (Thevathasan and Gordon 2004; Reynolds et al. 2007), belowground competition for soil resources was reported following tree root exclusion experiments in other temperate TBI systems (Gillespie et al. 2000; Miller and Pallardy 2001). Furthermore, other studies found belowground tree-herbaceous interactions to be more influential than competition for aboveground resources (Bloor et al. 2007; Fletcher et al. 2012). There remains conflicting evidence of the degree to which above and belowground competition influence tree-crop interactions. However, it is generally understood that minimizing belowground competition is critical in successful TBI systems whereby a stratification of resource pools, spatially or temporally, is desired (Schroth 1999; Van Noordwijk and Purnomosidhi 1995; Livesley et al. 2000). Additionally, deep-rooted trees can encourage water uptake and hydraulic redistribution from areas of moist soil to dry soil or from deeper soils to shallower soils (Oliveira et al. 2005; Fernández et al. 2008; David et al. 2013). Tree roots may

51 41 capture nutrients that have leached beyond the crop rooting zones (Das and Chaturvedi 2008; Bergeron et al. 2011). For example, Dougherty et al. (2009) found lower amounts of nitrate runoff from the Guelph TBI system compared to a control monocrop field, suggesting that tree root systems might be acting as a safety net (Jose et al. 2004) Root distribution in TBI systems The distribution of a trees roots are dependent upon internal factors (i.e. genotypic) and external factors (Puhe 2003). External factors can include wind (Tamasi et al. 1995), interspecies competition (Upson and Burgess 2013), and edaphic conditions (Sudmeyer et al. 2004; Isaac and Anglaaere 2013). Thus, site-specific understanding of root distribution is required to explain belowground processes generally in TBI systems and arguably for the Guelph site specifically. Current knowledge of root distribution in TBI systems is based on limited research due the difficulty of studying roots in situ (Livesley et al. 1999; Taylor et al. 2013). Knowledge of tree root distribution can be used to better infer zones of competition or complementarity and ultimately contribute to decision making of species selection, planting design, and identifying if management prescriptions are required to mediate belowground treecrop interactions. However, studying tree roots in situ is an acknowledged methodological challenge. Fine roots provide the primary surface area for water and nutrient absorption by plants (Jackson et al. 1997). Some advancements of non-intrusive study of tree fine roots include the use of electrical conductance (Aubrecht et al. 2006; Čermák et al. 2006) or modeling. However, as tree fine root distribution is more uniform than tree coarse roots (Danjon and Reubens 2008; Taylor et al. 2013), for this study fine roots were assumed to be accurately assessable through soil sampling techniques, specifically soil cores, which result in a measured value for fine roots in a known soil volume (root length density; cm cm -3 ). Coarse roots are methodologically more challenging to study than fine roots. Generally, destructive sampling is involved in TBI research

52 42 (e.g. measurements from exposed roots in a trench wall (Upson and Burgess 2013)), which only capture a portion of the root system s distribution, or partial or complete harvesting (Das and Chaturvedi 2008), which is labourious and non-repeatable. Thus, GPR offers an alternative to reduce destructive sampling, while potentially capturing more of the coarse root system. GPR has been used in tropical agroforestry to detect distribution of roots in mixed-species systems at certain distances from trees (Isaac and Anglaaere 2013). Raz-Yaseef et al. (2012) estimated biomass with depth at the plot-level, using a similar form of analysis as described in Chapter 3. However, this study is the first to chart vertical root distribution at the tree-scale. Accordingly, I hypothesized that GPR can accurately locate the distribution of the study trees coarse roots due to the conducive subsurface conditions for radar survey at the study site. Furthermore, the vertical root distribution of trees can be charted. In doing so, tree root distribution can be compared among species and importantly in crop rows where tree-crop interactions might be greatest. 4.3 Materials and Methods See Chapter 2: General Methods for description of study site, study trees, and radar survey Coarse root distribution measurements with GPR GPR geo-images were compiled into the grid orientation in which they were collected in GFP_Edit (Sensors & Software, Mississauga, ON, Canada). Prior to image analysis, non-root anomalies (e.g. plane reflectors and signal noise ) were minimized by applying a sequence of processing steps (DC shift, dewow, background removal). A spreading and exponential gain (SEC2) was applied to enhance delineation of reflection patterns with depth (EKKO_View Deluxe; Sensors & Software). Subsequently, all geo-images were examined for root reflections. Root reflections were visually identified and recorded using EKKO_Interp (Sensors & Software), which provided the x,y,z coordinates for each detected root. This coordinate data was

53 43 used to quantify coarse root distribution in the vertical profile for each tree as well as for subsetted coarse roots detected below equivalent areas of crop rows or tree rows. Detected coarse roots located in the T. occidentalis tree rows were not analyzed as these trees were planted at 1 m spacing Accuracy testing In order to assess the level of accuracy of the technique, randomly selected soil profiles equating to 1 m 2 were exposed along subsets of matched GPR transects and root locations crossing the profiles were compared to identified root reflections in radargrams for each study tree (n=13). A positive detection was confirmed when a coarse root was found within 0.10 m and root diameter was measured using digital callipers. When one detected coarse root matched with multiple roots in same exposed area, the largest diameter root was assumed as the positive detection and the smaller diameter root(s) as missed detection(s). GPR coarse root detection frequency was the proportion of positive coarse root detections compared to the total number of detected and missed coarse roots. When no root was found where an identified reflection was observed in the radargram, a false positive was counted Root distributions For each tree, detected coarse roots were pooled in 0.10 m depth intervals from 0 to 1 m. Root counts were converted into a proportion of roots detected within each depth increment so that all increments summed to 1. Vertical root distribution was fitted to the function described by Gale and Grigal (1987) and Jackson et al. (1996): Y = 1 - β d (6) where Y was the cumulative proportion of roots at depth (d) and β was the distribution coefficient. A greater proportion of roots closer to the surface is common with lower β while a

54 44 higher β is indicative of a greater proportion of detected roots distributed deeper in the profile. This curve (Equation 6) assumes tree roots have an asymptotic increase of cumulative root distribution with depth and thus projects a fraction of roots to depths beyond that of the maximum detected root Fine root distribution in crop rows Soil cores were collected using a metal cylindrical sampler (100 cm 3 ) inserted horizontally into exposed soil profiles at three distances from the tree stem (0.5, 1.0, 1.5, and 2.0 m) and at four depths (10, 20, 40, and 60 cm). Sampling was completed along two transects into the crop rows originating from the tree stem. Soil cores were stored in air-tight bags at 5 C until processing. Samples were wet sieved through mesh sized 2.0, 1.0, 0.5, and 0.25 mm (as suggested by Livesley et al. (1999) for fine root length measurements) and fine roots were manually collected using forceps. Dead roots were omitted, characterized with a loss of plasticity and often very dark in colour (Das and Chaturvedi 2008, Gwenzi et al. 2011). Soil samples were collected prior to the germination of soybean crops, controlling for tree fine roots in the crop rows. The exception was for T. occidentalis sampled last, but soybean roots were differentiated as very light in colour and with distinctive morphology (e.g. with nodules) from the fine roots of T. occidentalis. Fine root samples were scanned with flatbed scanner at 600 dpi and length was measured using WinRHIZO (Regents Instruments, Montreal, Canada). Root length density (RLD) was calculated such that: RLD = L V (7) where L is the measured length (cm) within a given volume V (100 cm 3 ).

55 Statistical analysis GPR detection frequencies were grouped by diameter class (< 1 cm or 1 cm), species, and depth intervals. One-way analysis of variance (ANOVA) was used to test for differences by species and depth. When detection frequencies did not meet parametric assumptions, Kruskal- Wallis test was used and subsequent post hoc test for species pairwise variation using the agricolae package in R. Paired t-tests compared means of detected large and small coarse roots by species. Un-pooled GPR detected root depth data were used to identify the minimum, maximum, mean, and skewness of detected coarse root depths for each tree as well as for the subsets of detected coarse roots located below crop or tree rows. Skewness (a measure of departure from symmetry) of detected roots were calculated using the moments package in R. Larger positive skewness suggests a right-skew distribution with a longer tail to the right, or with depth, while a skewness value closer to 0 indicates symmetrical distribution of root depths. ANOVA was completed for each statistic across the five species and, when significant, were followed by Tukey HSD. Paired t-tests were used to test for variation between roots in the cropped rows and the tree rows. Species means of fine roots at each sampling depth followed non-normal distribution and were tested for significant variation using Kruskal-Wallis. Assumptions of normality (Shapiro-Wilk) and equal variance were met prior to parametric tests. Statistical analyses were completed in R v (R Foundation for Statistical Computing, Vienna, Austria) and the level of significance was set at p< Results GPR detection frequency of coarse roots The frequency of coarse root detection was 51.3 ± 0.3 % (mean ± S.E.; n=13) with a distinctly higher detection frequency (80.6 ± 0.8 %; n=13) for coarse roots 1 cm and a lower

56 46 detection frequency (43.4 ± 0.5 %; n=13) for smaller coarse roots < 1 cm. Smaller coarse roots were consistently under-detected compared to larger coarse roots for each tree species although not significantly (Figure 12). There were no significant differences between detection frequencies across species for either diameter class. Detection frequencies varied significantly by depth (p=0.015) (Figure 13) as it became increasingly difficult to accurately identify radar reflections of coarse roots further down the soil profile. Overall, false positives accounted for 26.8 ± 0.3 % (n=13) of the identified coarse root reflections. Of these false positives, most were located above 0.50 m (84 %) Coarse root distribution There was significant variation across species for both the mean depths of detected coarse roots (p=0.009) and the maximum depths of detected coarse roots (p=0.001) (Table 5). Furthermore, all species-specific distribution coefficients (β) (Table 6) indicated a similar species pattern in regards to rooting depth distributions. The J. nigra and Q. rubra had the deepest root systems (Figure 14) with mean detected root depths of 0.29 ± 0.01 and 0.29 ± 0.02 m (n=3) respectively and maximum detectable rooting depths of 0.67 ± 0.05 and 0.70 ± 0.01 m respectively (mean ± S.E.; n=3) (Table 5). These maximum depths were further extrapolated asymptotically to 0.95 m for the depth at which 95 % of roots are predicted (d95) (Table 6). While the Populus sp. had a shallower coarse root distribution with a mean coarse root depth of 0.21 ± 0.02 m (n=2) and a lower β resulting in d95 of 0.70 m, proximal to its maximum detected root depth (0.66 ± 0.01 m). Therefore, the deep rooted Populus sp. had a less evenly distributed coarse root system that was concentrated closer to the surface, which was further supported by the highest skew (0.68 ± 0.19) associated with more roots at shallow depths and a longer tail of detected roots deeper in the profile (Table 5). The coarse roots of the P. abies followed a similar vertical distribution pattern to the Populus sp.,

57 47 Figure 12 Mean detection frequency of coarse roots grouped by diameter size class (n=3 except n=2 for Populus sp. and Picea abies). No significant differences between species means of each diameter class (ANOVA) or between diameter class of same species (paired t-test) (p>0.05). Bars represent ± S.E. of the mean. Figure 13 Mean detection frequency at each soil depth interval (n=13 exposed profiles) measured from subset of matched soil profiles and radargrams. Means with same letter are not significantly different (Kruskal-Wallis) (p>0.05). Bars represent ± S.E. of the mean.