UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The application of spectral unmixing and supervised classification remote sensing techniques to Landsat… Norquay, Alan 2000

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2000-0508.pdf [ 3.08MB ]
Metadata
JSON: 831-1.0089591.json
JSON-LD: 831-1.0089591-ld.json
RDF/XML (Pretty): 831-1.0089591-rdf.xml
RDF/JSON: 831-1.0089591-rdf.json
Turtle: 831-1.0089591-turtle.txt
N-Triples: 831-1.0089591-rdf-ntriples.txt
Original Record: 831-1.0089591-source.json
Full Text
831-1.0089591-fulltext.txt
Citation
831-1.0089591.ris

Full Text

THE APPLICATION OF S P E C T R A L UNMIXING AND S U P E R V I S E D CLASSIFICATION REMOTE SENSING TECHNIQUES TO LANDSAT 7 DATA FOR DETECTING A R B O R E A L LICHEN A B U N D A N C E by Alan Norquay B.Sc. The University College of the Cariboo, 1998 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE D E G R E E OF M A S T E R OF S C I E N C E In FACULTY OF G R A D U A T E STUDIES DEPARTMENT OF F O R E S T R E S O U R C E M A N A G E M E N T We accept this thesis as conforming to the required standard The University of British Columbia August 2000 ©Alan J . Norquay, 2000 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of fOt&T /jeSOQ^S/^^m^f: The University of British Columbia Vancouver, Canada Date OCT 151'oo-DE-6 (2/88) Abstract In British Columbia, mountain caribou (Rangifer tarandus caribou) rely on arboreal lichen as a main source of food during the late winter months. The animals spend a majority of the year at these higher elevations where old growth forests provide a suitable microclimate for the lichen genera Bryoria and, to a lesser extent, Alectoria. Caribou migrate there from lower areas when the snowpack solidifies sufficiently to allow efficient travel, possible by their unique hooves that spread to distribute weight. Forestry operations also have an interest in these old growth stands, and harvesting operations have begun to threaten the continued existence of mountain caribou in British Columbia. Efforts to determine caribou habitat, and the extent of that habitat, have been underway for many years. However, there has been an increase in the amount and intensity of research during the past six years, due largely to higher level planning processes that have identified the animals as a priority for inventory research programs. This research was focused on identifying stands of trees that contain arboreal lichen by using remotely - sensed data. While it is known that older stands are required for lichen growth, not all older stands have produced useful quantities of lichen. Since stands containing heavy loadings of arboreal lichen appear different to the naked eye than those with lesser loadings, it is reasonable to assume these differences can be measured with remote sensing techniques. The purpose of this work is to identify these differences using Landsat 7 data. I used a 30-metre pixel resolution Landsat scene collected on August 22 1999. Spectral unmixing, supervised classification and correlation analysis techniques where conducted, but no method was able to distinguish stands by the amount of lichen they contained. The reason this attempt did not succeed is probably due to the pixel size of the data (30 metres) versus the effect small clumps of lichen have on the reflectance recorded for those pixels. If data from a sensor with higher spatial resolution (e.g. IKONOS 4 meter data) where used in future work, the results may be more successful. TABLE OF CONTENTS ABSTRACT ii LIST OF FIGURES iv LIST OF TABLES v ACKNOWLEDGEMENTS vi INTRODUCTION 1 Caribou 1 Arboreal Lichen 2 Lichen Habitat 3 BACKGROUND 4 Previous Research 4 Succession Characteristics and Remote Sensing 5 NULL HYPOTHESIS 6 STUDY AREA 6 Image Type Selection 9 Image Specifications 10 METHODS 11 Field Data 11 IMAGE ANALYSIS METHODS 12 Supervised Classification 13 Masks 14 Training Classes 14 Number of Pixels Required 15 Classifiers 15 Spectral Unmixing Process 16 Normalized Difference Vegetation Index (NDVI) 18 Tasseled Cap Transformation 18 iii Results Assessment 19 Supervised Classification 19 Spectral Unmixing 19 Tasseled Cap and NDVI 20 RESULTS AND DISCUSSION 20 Supervised Classification 20 Spectral Unmixing 22 Correlation Analysis 23 Analysis of Variance 24 Future Direction 25 SUMMARY 26 REFERENCES 27 APPENDIX A 30 TABLE CODES 32 APPENDIX B 33 List of Figures Figure 1 Defoliated zone (after Goward 1998) 3 Figure 2 Map of study area 7 Figure 3 Landsat 7 image of study area 8 Figure 4. Good separation between classes: 14 Figure 5 Ellipse overlap 21 iv LIST OF TABLES Table 1. List of Landsat 7 bands 10 Table 2. Supervised classification 21 Table 3. Unmix Results: Low lichen 22 Table 4. Unmix Results: Medium lichen 23 Table 5. Unmix Results: High lichen 23 Table 6 Pearson correlation 24 Table 7 Analysis of Variance of Bands 24 Table 8 Raw Field Data (Part A) 33 Table 9 Raw Field Data (Part B) 34 Table 10 Raw Field Data (Part C) 35 v Acknowledgements I would like to thank Dr. Peter Murtha, Michael Burwash of the British Columbia Ministry of the Environment Lands and Parks, Dr. David Shackleton, and Dr. Valery Lemay. Special thanks goes to Rajeev Sharma for his unending patience during my numerous software queries. vi Introduction The caribou in North America are of the subspecies Rangifer tarandus caribou, or woodland caribou, and inhabit the boreal forests throughout Canada. The woodland species are further divided into three ecotypes: northern, boreal and mountain, with the latter being of interest in this study (Shackleton, 1999; BC Ministry of Environment, Lands and Parks 1997). Remote sensing is a valuable tool in wildlife management in that it efficiently identifies numerous characteristics important to the wildlife manager, including: land cover types such as forests, wetlands, grasslands; coniferous versus deciduous, and, perhaps of equal importance, offers the ability to detect and record change due to human activity. Soon, more satellites will be offering higher resolution data that will (hopefully) cause a reduction in data costs, making innovative research less costly. My study was directed toward the application of remote sensing to locating caribou winter habitat. Caribou The relevance of my study is linked to the needs of caribou and how they conflict with those of lumber companies. Mountain caribou often make four elevational migrations each year. In the fall and early winter, caribou leave the higher Engelmann Spruce-Subalpine Fir (ESSF) and Alpine Tundra (AT) elevations (~1700 meters) (BC MOF, 1991) for the lower Interior Cedar Hemlock (ICH) zone where there is less snow and more forage (Edwards and Ritcey, 1959). As the season progresses, the snow becomes firmer at higher elevations, making travel and access 1 to arboreal lichen easier and they return to the E S S F (Antifeau, 1987). Since wolves do not experience this increased mobility this tactic also serves as a predator avoidance strategy (Bergerud, 1978). However, at higher elevations, deep snow prevents feeding on plants below the surface. Instead, caribou diet consists almost exclusively of Alectoria sarmentosa, Bryoria fremontii and B. pseudofuscescens arboreal lichen (Goward 2000, Child et al. 1991, Stevenson and Hatler 1985), consuming an estimated 5 kgs per day (BC Ministry of Environment, Lands and Parks 1997). As spring approaches, caribou migrate down to lower elevations and feed on emerging vegetation. As the summer progresses they again return to the ESSF/AT, again as a predator avoidance strategy. Arboreal Lichen Arboreal lichens grow slowly and have generally been associated with late successional and oldgrowth stands (BC MoELP 1997, Rose 1976, 1992). However, recent studies have shown lichens require a combination of structure and microclimate conditions for growth, rather than old growth stands per se (Goward, 1998). As a result, ensuring that an adequate high elevation (for predator avoidance) late winter food source is available for caribou after logging requires more effort than simply leaving tree stands of a certain age, since they may not always contain the required lichen. The purpose of this study was to investigate the use of remotely sensed data to identify forest stands with high quantities of arboreal lichen versus stands with lesser loadings. 2 Lichen Habitat Lichen is a combination of algae and fungus. The relationship is symbiotic as neither the fungus nor the algae could survive independently in the environment where they are found together. Lichen growth occurs only when wet, but the algae itself can only grow during the last stages of the wetting cycle, i.e. nearly dry. A prolonged wet cycle can effectively kill off the algae and thus the lichen (Coxson 1999). The bark of coniferous trees is the most common substrate for thallus to establish, and Stevenson (1985) found that age-related factors (such as the bark's ability to hold moisture or bark chemistry) have no impact on lichen abundance. Requirements for lichen establishment and growth include: sufficient radiation, regular wetting and drying cycles, high evaporation rates, a source of thallus fragments, a suitable substrate for thallus attachment, and time (Goward 1998, Stevenson and Enns 1992). Goward (1998) observed a defoliated zone in the crown of older trees where the greatest lichen loads are found (Figure 1). Defoliation occurs when needles nearer the bole die and fall off as the foliage ages. Figure 1 Defol iated zone (after Goward 1998) 3 Since lichens are occasionally observed on foliated as well as defoliated portions of the same branch, another mechanism besides the lack of needles must be responsible for lichen growth. Goward (1998) suggested that a process whereby branches shed water outward, away from the defoliated zone, is responsible. This process provides the required amount of water in the defoliated zone, but so much on the outer, foliated portions of the same branch that it prevents lichen growth. Furthermore, the trend to lower stem density as a stand ages (Stevenson, S. and K. Enns 1992) alters airflow and thus contributes to the drying cycle. In summary, lichen requires an adequate combination of conditions that result in few stress events: • Correct duration of wetting episodes • Suitable substrate • Correct frequency of wetting and drying cycles Furthermore, it is generally recognized that the forest attributes found in oldgrowth ecosystems favoring lichen colonization are: • Rougher bark that hold fragments more securely (Stevenson 1992) • Available wood and bark in a variety of decay stages (Stevenson 1992) • A relatively stable environment in which to grow (Goward 1998) Background Previous Research In my literature review, I found few references where remote sensing had been applied to detect lichen. Steiner and Gutermann reported in 1966 that "reflectance data for mosses and lichens are scanty", and little has changed since then. In fact, the only reference I found for arboreal lichen reflectance was Stevenson (1978), who 4 found weak but promising results in the use of infrared photographs for detecting arboreal lichen. Stevenson found that it was possible to correlate lichen predictions with biomass, but in only 39% of the cases was the correlation significant at the 0.05 confidence level. Succession Characteristics and Remote Sensing Changes in the structure of a forest stand as it ages include: • Tree height increases until site limits are reached (Deuling 1999, Spies et al. 1990) • Bole diameter increases (Deuling 1999) • The number of living trees/hectare declines (Stevenson and Enns 1992) • Basal area/hectare increases (Stevenson and Enns 1992) • Gaps in canopy tend to become larger (Spies etal. 1990) • Gaps are less abundant than in younger stands (Spies et al. 1990) • Heterogeneity of tree sizes increases (Spies et al. 1990) Remote sensing techniques can use a combination of these structural changes to separate younger from older stands. Franklin (1986) found that as basal area increases, spectral reflectance in the visible bands (bands 1, 2, 3,) decreases due to the increased absorption in these bands, caused by increasing leaf area as the trees age. Reflectance is further reduced by shadows that form from this increased vegetation. Fiorella and Ripple (1993) and Cohen and Spies (1992) found the Wetness component of the Tasseled Cap Transformation (Crist and Cicone 1984) and the ratio of bands 4:5, to be most useful for distinguishing between mature and oldgrowth stands. However, the presence of these characteristics does not necessarily indicate the presence of arboreal lichen (Goward 1998). Goward (1998) 5 found that while older trees generally do contain more lichen than younger trees, those growing in sheltered areas often have less lichen than trees growing in exposed sites. He concluded that sensitivity to prolonged wetting was most important; with other environmental factors such stem density and wind playing a lesser role in lichen growth and distribution. Null Hypothesis Goward (1998) stated the distinct dark green coloration of old trees in the E S S F (as compared to younger trees) is from Bryoria loadings within the trees themselves. These differences may be detectable using remote sensors, and formed the basis of my research. The null hypothesis I investigated is: In Landsat 7 imagery, there is no significant difference between spectral signatures of forest stands containing high lichen loads and those with low lichen loads. Study Area The study area is located north of Kamloops, British Columbia in the general vicinity of Clearwater, BC (Figure 2 and Figure 3). The approximate center is located at 119.587° W 51.802° N. Elevation range is 1310 - 1810m a.s.l. 6 Caribou Habitat Study Area Figure 2 Map of study area. Area is north of Clearwater, British Columbia, adjacent to Wells Gray Park, and within the Caribou Habitat Study boundary. Figure 3 Landsat 7 image of study area. Sites numbers in red. Recent clearcuts have a slightly reddish hue. Shades of green progress from lighter to darker as the stand ages; lighter green areas are young regeneration. Landsat 7 data copyright NOAA. Received by CCRS. Processed and distributed by RADARSAT International. Reproduced with permission by Space Imaging. 8 In this study area, caribou use three biogeoclimatic zones: 1. Engelmann Spruce-Subalpine Fir (ESSF) with sub-zones and variants: • Wet Cold (wc2) • Very Wet Cold (vc) • Very Wet Cold Parkland (wcp) 2. Interior Cedar Hemlock (ICH) zone with sub-zones and variants: • Moist warm (mc3) • Wet Cool (wk1) • Very Wet Cool (vk1) 3. Alpine Tundra (AT) with one sub-zone: • Parkland (p) During the years between 1996 and 1998, 88% of caribou G P S locations (n=1391) in a BC Ministry of the Environment study of caribou habitat were in the E S S F biogeoclimatic zone. Therefore, since my intention was to study the detection of arboreal lichen as it relates to late winter habitat, I placed all plots in that zone. Image Type Selection Ideally, a Compact Airborne Spectrographic Imager (CASI) sensor would be used for this project. CASI collects hyperspectral data in 288 channels at a bandwidth of 3 nm, and can produce a variety of ground resolutions depending on the aircraft flying height. Because arboreal lichen is smaller than the trees they are on, CASI would likely do a better job of separating them than any other sensor. However, the estimated cost for a test flight was $50,000 and beyond the budget of this project. The choices therefore were among currently available space platforms, such as 9 S P O T or LANDSAT satellites. While S P O T data offered better spatial resolution than LANDSAT, (20 meters versus 30) it does not have the radiometric resolution of other platforms. IKONOS offers high spatial resolution data, but was not available when the project began, and the costs were again prohibitive. Therefore the final choice was Landsat 7 with its sensor called the Enhanced Thematic Mapper or " E T M + " . Image Specifications The image I used is described as Path 49 Row 24, recorded on August 22 1999. The Landsat 7 E T M + sensor consists of 8 spectral bands: They are: Table 1. List of Landsat 7 bands Band Frequency (um) Spatial resolution (m) 1 (Blue) 0 . 4 5 - 0 . 5 2 30 2 (Green) 0 . 5 2 - 0 . 6 0 30 3 (Red) 0 . 6 3 - 0 . 6 9 30 4 (Near infra-red) 0 . 7 7 - 0 . 9 0 30 5 (Mid infra-red) 1 .55 -1 .75 30 6 (Thermal) 1 0 . 4 - 1 2 . 5 60 7 (Mid infra-red) 2 . 0 8 - 2 . 3 5 30 8 (Panchromatic) 520 - 920 15 Note that Landsat 7 differs from previous Landsat missions by including the panchromatic band (at 15 m resolution) and reducing the resolution of band 6 from 120m to 60 m. I used only the bands 1-5 and 7 for analysis as they are of comparable resolution. 10 Methods Field Data Coincident with the Landsat pass, fieldwork was completed between August 18 t h and September 2 2 n d 1999. First, I identified 30 homogenous stands from a helicopter and recorded their location using a Trimble GeoExplorer II Global Positioning System (GPS) receiver. These were then differentially corrected with Pathfinder Office (version 2.11) and corresponding files from the Williams Lake base station. Base files from that locale are appropriate because it is only ~ 200 km from the study area, well below the recommended maximum of 500 km (Trimble 1994). After creating coverages (using Arclnfo GIS software) I determined the geographic coordinates of the center of these stands. An experienced field crew then traveled to each stand, and, using forest cover and TRIM maps, located (as accurately as possible) the center of each stand. Three 3.99 m radius plots were established in each stand; the first at the center and two others 50 m away on a random compass bearing for a total of 90 plots. The following data were recorded: • Quantity of lichen per plot • Stand age • Diameter at breast height • Crown closure • Number of stems per hectare • Number of snags per hectare • Height of leading species, determined by the most frequent species of the stand age The averages of each of these variables for the three plots in each stand were recorded as the value for that stand. Lichen quantity measurement followed the 11 Estimating Abundance of Arboreal Lichens handbook (Armeleder et al. 1992), using the method that estimates the proportion and abundance of the two main types of forage lichen (Alectoria and Bryoria). I chose this method because: 1) the handbook displays very clear photographs of trees and branches with various quantities of lichen to use for comparison, and having these comparisons at hand when conducting the survey helped reduce bias among surveyors and sites; and 2) the technique has been used in various caribou field studies. The objective when using this method is to place each tree into one of six abundance classes (0 - 5); these classes are based on the actual weight of the lichen on the trees. Note that only lichen below 4.5 meters from the ground is measured, as this is the estimated limit of access by caribou when the snowpack is present (Armeleder et al. 1992). While arbitrary, it does provide a reasonable standard that can be applied between sites. I then grouped these 6 classes into 3 groups: Low (0 - 2.4), Medium (2.5 - 3.5), and High (3.6 - 5) to compare image classification results. Image Analysis Methods I investigated the following methods to identify differences in reflectance that correspond to the quantity of arboreal lichen present: • Supervised Classification • Sub-Pixel Unmixing • Pearson Correlation on the raw data and with the transformations: o Normalized Difference Vegetation Index (NDVI) o Tasseled Cap Transformation 12 Supervised Classification The objective of classifying an image is to have the software automatically categorize all the pixels in an image into "themes" or "classes". This is accomplished by using the spectral pattern, derived from the numerical value of the pixels in each class, as a basis for categorization (Lillesand and Kiefer 1994). By identifying unique ground features (e.g. grassland versus wetland), training classes are created that software programs can use to classify unknown pixels. This technique works for feature types that have spectral reflectance properties that produce unique digital numbers (DN). The DN is produced through the process of converting the analog electrical signal to positive digital integers. Figure 4 is an example of how two distinct classes are separated in the imagery used for this study. In this example, the combination of channels 4 (770 - 990 nm) and 5 (1550 - 1750 nm) can be used to separate exposed soil from emerging vegetation. 13 200 190 180 170 160 150 r 140 a 120 ] n 110 n 100 e 90 I 80 70 60 50 40 30 20 10 0 Exposed soil Emerging vegetation 0 10 20 30 40 50 60 70 80 80 100110 120130 140150160 170 Channel 5 Figure 4. Example of good separation between classes: Exposed soil and emerging vegetation. Channel digital numbers on axis. Cross indicates class mean. Masks To limit the processing to relevant sites, I employed a mask routine to exclude lakes, exposed soil (i.e. new cutblocks, landings, gravel roads), ice or snow, urban areas and fire scars by excluding Band 6 DN values <18 and > 49. This reduced the variance in the data and produces separate classes in materials that otherwise would be grouped together. Training Classes I created classes for six types: High lichen, Medium lichen, Low lichen, Recent Regeneration, Young regeneration, and Old regeneration. 14 Number of Pixels Required The theoretical minimum number of pixels required in a training set is n + 1, where n is the number of spectral bands. However, it would be problematic to attempt to evaluate the variance and covariance (covariance is the amount two datasets vary with respect to each other) from only 7 pixels (6 bands + 1) because of the variance within each class. Practical estimates place the number of pixels needed at between 10/7 and 100/7 (Lillesand and Kiefer 1994). Therefore, in the supervised classification stage, I selected 5 x 5 pixel grids (25 pixels each) around sites. As there were 11 Low sites, 275 pixels were selected; the 13 Medium sites resulted in 325 pixels; and the 6 High sites had 150 pixels used for training classification. Classifiers Three of the most common classifying routines are the Maximum Likelihood, Parallelepiped and Minimum Distance to Means. I selected the Maximum Likelihood function because it assesses the probability of a pixel belonging to a particular category, is sensitive to covariance, and is particularly suited to ellipse shaped data (as in Figure 4). The Maximum Likelihood classification uses both the variance and covariance of the category's spectral response patterns to classify an unknown pixel. In essence, this process considers the shape of the ellipse formed from the data making up the classes, and calculates the probability of the unknown pixel belonging to each class. 15 Should the values of an unknown pixel fall too far from any of the classes, it is labeled "unclassified". Spectral Unmixing Process To detect features smaller than the nominal pixel size of an image, it is possible to use a method called spectral unmixing (Huguenin 1994). This process is based on the assumption that each pixel consists of a linear combination of ground features, called Materials of Interest (MOI), plus residual error. Spectral unmixing has been successfully applied to estimate the fraction of loblolly pine (Pinus taeda) and longleaf pine (Pinus palustris) (Huguenin 1994) as well as amounts of bald cypress (Taxodium distichum) and Tupelo (Nyssa aquatica) in forest stands in Georgia and South Carolina (Huguenin 1997). In my study, I investigated the usefulness of this process by taking it to the next step: instead of separating species, my intention ultimately was to identify a particular MOI present on trees of the same species, namely lichen abundance. Essentially, the analysis is accomplished by separating the MOI from its background. To do this, pixels that contain a high quantity of the MOI are identified and selected (at this stage they are called "endmembers"). Their spectral signatures are stored in a spectral "library", and used for reference. The procedure follows this general formula: DNb = t FiDNib + e b 16 Where DNb = Digital Number (DN value) of the pixel under analysis in band b; F, = the fraction of the endmember /'; DN,>Di is the DN value of endmember / in b, and eD is the error in b. This allows for the estimation of the fraction of each endmember in each pixel. The process may be easier understood if only one band is considered. Assume in Band 4, the pure endmember for fir trees has a DN of 40. However, the B4 in our sample has a pixel DN of 60. All other endmembers (called the background) are removed (while minimizing the error, eD). In this example, the remaining DN value is 10. Therefore, the estimated fraction of the fir endmember (F) is .25 (10/40). However, when multiple bands are used, the process seeks to minimize the error eD of the estimates and the observed values of all the bands. The final estimate in a multi-band case is the sum of all the bands for that endmember. Unique solutions of the endmember fractions are only possible if the number of bands is equal to the number of endmembers, plus one (used to account for effects from shadow). Since I am using Landsat 7, there are 6 bands (1-5, 7) for consideration. Bands 6 and 8 (Table 1) are of different spatial resolution and thus cannot be used in this process. I employed endmembers for Low, Medium and High lichen pixels; Young vegetation, Old vegetation; exposed soil and shadow. 17 Normalized Difference Vegetation Index (NDVI) NDVI uses the red and near infrared bands to produce an index that is sensitive to the presence and condition of green vegetation. It also has been correlated with leaf area index, biomass estimates, estimates of percentage ground cover, and photosynthetically active radiation estimates (Goward 1991). It is calculated with the formula: NDVI = M R - r e d NIR + red I investigated whether NDVI fluctuates in response to lichen abundance. Tasseled Cap Transformation The Tasseled Cap Transformation was developed by Kauth and Thomas (1976) and further extended by Crist and Cicone (1984). This method transforms the three visible and one near infra red (NIR) bands into three components that are directly related to physical characteristics. The components are: Brightness, Wetness and Greenness. Brightness measures variation in soil reflectance and is not relevant to my study. Coefficients for Landsat-7 transformations were provided by Sharma (personal communication 2000). Greenness Greenness is the contrast between the near infrared and visible bands. The scattering of near infrared energy by the cells of green vegetation, and the absorption of visible energy by plant pigments, combine to produce high Greenness values in areas of high vegetation density (Crist and Cicone 1984). Since Goward 18 (1998) has postulated that the reason some older stands are darker than others is due to the amount of lichen they contain, it is logical to investigate whether Greeness can be correlated to lichen quantity. Wetness Wetness contrasts the sum of the near infrared and visible bands with the longer infrared bands and highlights moisture related characteristics in the scene (Crist and Cicone 1984). Since Goward (1998) noted that a correct amount of moisture is critical to lichen growth I investigated whether this component could be related to lichen quantity. Results Assessment Supervised Classification The supervised classification process is assessed by producing an error matrix of known pixels and classified results. The matrix takes the form of a table containing the classification results in rows and the known ground features in columns. Spectral Unmixing Results are analyzed by comparing the processing results to the known lichen quantity value. If differences in reflectance between sites are detected and are due to the quantity of lichen present, sites with (ground measured) lichen ratings of Low will have a high fraction value in the unmixing results Low Lichen endmember category. Conversely, the same site should have a low fraction value in the High Lichen endmember category. 19 Tasseled Cap and NDVI Both of these conversions were compared with image bands and tested for correlation with lichen loads using the Pearson correlation function in Systat 8.0. Results and Discussion Supervised Classification The classified supervision process indicates that, using Landsat 7 imagery bands 1 -5 and 7, it is not possible to distinguish between stands based on the amount of arboreal lichen. The spectral signatures are not distinct among stands containing differing quantities of lichen. A comparison of channels (bands) 3 versus 5 (Figure 5), using all sites as training areas, is representative of the overlap found between all combinations. Because the classes are not distinct the supervised classification process resulted in over 78% of the sites being unclassified (Table 2). Again, this is because the spectral reflectance of unknown pixels is too different than the established classes to be assigned to one of those classes. 20 150 -I 140 • 130 • 120 • 110 • c 100 • h a 90 -n 80 -n e 70 • 1 GO -5 50 • 40 -30 -20 -10 -0 Red: Hiqh lichen —i— 10 Blue: Low l ichen Yellow: Medium lichen ~i 1 1 — 40 50 GO Channel 3 70 —i— 80 —i— 90 100 Figure 5 Ellipse overlap between Low (blue), Medium (Yellow) and High (Red) lichen sites for bands 3 and 5. Cross indicates class mean. Table 2. Supervised classification results of known sites. Sites marked with a * were used as training sites. Site Actual Result Site Actual Result 1 LOW UNCLASS 18 MEDIUM UNCLASS 2* MEDIUM N/A 19 HIGH UNCLASS 3 HIGH MEDIUM 20 LOW UNCLASS 4 MEDIUM UNCLASS 21* LOW N/A 5 MEDIUM UNCLASS 22 MEDIUM UNCLASS 6 LOW UNCLASS 23 MEDIUM UNCLASS 8 MEDIUM UNCLASS 24 LOW UNCLASS 9 MEDIUM UNCLASS 25 HIGH UNCLASS 10 MEDIUM UNCLASS 26 LOW UNCLASS 11 HIGH UNCLASS 27 LOW UNCLASS 12* HIGH N/A 28 LOW UNCLASS 13 MEDIUM UNCLASS 29 LOW UNCLASS 14 LOW UNCLASS 30 MEDIUM UNCLASS 15 MEDIUM UNCLASS 32 MEDIUM UNCLASS 16 LOW UNCLASS 17 HIGH MEDIUM 21 Spectral Unmixing The spectral unmixing process failed to distinguish between stands at any level. The process produced fractions of low, medium and high lichen quantity that could not be correlated to the amounts measured on the ground (Tables 3, 4 and 5). Materials of Interest (MOI) are in three categories: U N M I X L o w , for sites that had a low lichen rating when measured on the ground; UNMIX_Med for sites with a medium rating; and UNMIX_High for sites that rated high. If the process could detect differences based on the amount of lichen present, sites with a high quantity of MOI would have relatively high fractions (e.g. 0.75 - 1.0). Table 3. Results from unmixing processing on sites known to have a low quantity of lichen. Site 21 (*) was used to collect the endmember spectrum. One would expect the UNMIX_Low fraction to be relatively high for these sites. Site UNMIX_Low UNMIX_Med UNMIX_High Actual 1 0.00 0.13 0.00 LOW 6 0.00 0.00 0.00 LOW 14 0.00 0.39 0.41 LOW 16 0.00 0.64 0.24 LOW 20 0.00 0.37 0.49 LOW 21* 0.00 (N/A) 0.00 0.00 LOW 24 0.00 0.71 0.00 LOW 26 0.00 0.00 0.00 LOW 27 0.00 0.00 0.00 LOW 28 0.07 0.54 0.00 LOW 29 0.00 0.00 0.00 LOW 22 Table 4. Results from unmixing process on sites known to have a medium quantity of lichen. Site 9 (*) was used to collect the endmember spectrum. These sites were expected to have relatively high UNMIXMed fraction values Site UNMIX_Low UNMIX_Med UNMIXJHigh Actual 2 0.00 0.00 0.00 MEDIUM 4 0.00 0.00 0.00 MEDIUM 5 0.00 0.00 0.00 MEDIUM 8 0.00 0.00 0.00 MEDIUM 9* 0.00 0.72(N/A) 0.00 MEDIUM 10 0.49 0.36 0.00 MEDIUM 13 0.00 0.00 0.00 MEDIUM 15 0.00 0.00 0.00 MEDIUM 18 0.00 0.00 0.00 MEDIUM 22 0.00 0.00 0.78 MEDIUM 23 0.00 0.00 0.74 MEDIUM 30 0.52 0.00 0.00 MEDIUM 32 0.50 0.00 0.00 MEDIUM Table 5. Results from unmixing process on sites known to contain high quantities of lichen. Site 12 (*) was used to collect the endmember spectrum. I expected these sites to have high UNMIX_High fractions. Site UNMIX_Low UNMIX_Med UNMIXJHigh Actual 3 0.00 0.00 0.22 HIGH 11 0.00 0.35 0.55 HIGH 12* 0.00 0.29 0.61 (NA) HIGH 17 0.00 0.00 0.00 HIGH 19 0.26 0.32 0.30 HIGH 25 0.00 0.00 0.00 HIGH Correlation Analysis Pearson's correlation analysis showed virtually no correlation between lichen quantity and any of the six bands under investigation, nor with the tasseled cap transformation or NDVI (Table 6). Scatterplots are found in Appendix A. 23 Table 6 Pearson correlation results comparing lichen quantity with ETM+ bands 1 - 5 and 7, the Tasseled Cap and NDVI Transformations. Band Pearson Correlation r P B1 0.259 0.167 B2 -.0109 0.565 B3 -.062 0.744 B4 -0.246 0.190 B5 0.042 0.825 B7 0.242 0.197 Greenness -0.244 0.194 Wetness 0.028 0.885 NDVI -.320 0.840 Analysis of Variance An Analysis of Variance was conducted on the pixel DN numbers of each band within each of the lichen classes. Again, the results show no significant difference between classes in any of the ands (Table 7). Table 7 Analysis of Variance of Bands Band F P 1 2.292341 0.120359 2 0.627181 0.541697 3 0.155747 0.856539 4 0.498825 0.612731 5 0.349891 0.707913 7 0.810038 0.45536 All Bands 0.126888 0.880912 This effort indicates the application of spectral unmixing or supervised classification methods to LANDSAT 7 imagery to detect stands with differing quantities of lichen is unsuitable. Therefore, I failed to reject the null hypothesis. However, the concept of 24 removing the background (trees) from the MOI (lichen) would be valuable for determining lichen quantity and thus assessing caribou habitat, and should be considered for future research using other methods and/or data. I believe the reason Spectral Unmixing failed to separate the stands based on lichen quantity was because I did not have the spectral curve for lichen. For the process to be successful, it will require either this curve be known, or an area on the ground (of sufficient size to fill an entire pixel) must contain nothing but the substance (lichen) that can be used to extract the curve. A patch of pure lichen that is larger than 30 m x 30 m (i.e. one pixel) can not be obtained. However, hyperspectral data from aircraft - mounted sensors can obtain very small pixels (< 60 cm), and it may be possible to isolate larger lichen collections using such data. In fact, even 4 metre IKONOS data would produce pixels of greater purity than does Landsat. Future Direction To discern features in an image, their reflectance values must be separable. In my study, the lichen was not separable from the trees, nor was there any (or as of yet undiscovered) characteristic in stands themselves that could be used to distinguish stands by lichen quantity using Landsat 7 imagery. However, this could be overcome by: • Determining the spectral reflectance of arboreal lichen, then investigating the application of hyperspectral data to separate trees and lichen. • Using data with finer spatial resolution, such as CASI. By adjusting flying height, 60 cm resolution data is possible. 25 Summary My study supports the findings of Stevenson (1978) that the detection of arboreal lichen by remote sensing is a process not feasible using the techniques tested. For this study, I hired a field crew to establish 90 plots in 30 stands; the average of 3 plots in each stand was used as the value for that stand. I used a Landsat 7 image (collected on August 22 1999) for analysis. The methods I used in my attempt to discern stands containing different quantities of lichen were sub-pixel unmixing, supervised classification, and Pearson's correlations on the tasseled cap transformation and the normalized difference vegetation index. No method was effective in discerning stands by lichen quantity. I believe the reason these results occurred is because the amount of reflectance change due to lichen is relatively small compared to the size of the trees and the pixel in the image. Future research should employ higher resolution data, such as IKONOS (4m pixels) or the CASI (60cm pixels) because differences in their DN values will more likely be due to differences within the tree itself, rather than other influences (such as the ground between trees). 26 References Ant i feau, T. D. 1987. T h e s ign i f icance of s n o w and a rborea l l i chens in W e l l s G r a y Prov inc ia l P a r k with spec ia l re ference to their impor tance to mounta in car ibou {Rangifer tarandus caribou) in the North T h o m p s o n wa te rshed of Brit ish Co lumb ia . M S c T h e s i s , Univers i ty of Brit ish C o l u m b i a , 142 p a g e s . A p p s , C . and T. Kinley. 1997. Mult ivariate S t a n d - L e v e l Habitat M o d e l s for Mounta in C a r i b o u in the Sou thern Purce l l Mounta ins , Brit ish C o l u m b i a . Unpub l i shed . A rme leder , H. and S . S t e v e n s o n , S . Wa lke r . 1992. Estimating the Abundance of Arboreal Forage Lichens. Land Management Handbook. F ie ld G u i d e Insert 7. Brit ish C o l u m b i a R e s e a r c h P r o g r a m . 22 p a g e s . Bergerud , A . T. 1978. The Status and Management of Caribou in British Columbia. Ministry of Recrea t ion and Conse rva t i on , Vic tor ia, B C . Brit ish C o l u m b i a Ministry of the Env i ronment , L a n d s and P a r k s . 1997. Toward a Mountain Caribou Management Strategy for British Columbia, Background Report. M A I A Publ ish ing Ltd. V a n c o u v e r , B C . 72 p a g e s . Brit ish C o l u m b i a Ministry of Fores ts . 1991. Ecosystems of British Columbia, rev. ed . Me id inge r D. and J . Pojar. R e s e a r c h B ranch , M o F , Vic tor ia , B C . Ch i l d , K. N. , 1976. P a g e s 1 7 - 2 0 . Ca r i bou in Brit ish C o l u m b i a . Wildlife Review; V l l l : 2 Winter . Ch i l d , K . N . , S . K . S t e v e n s o n , and G . S . Wat ts . 1991 . Mountain Caribou in Managed Forests: Cooperative Ventures for New Solutions. P r o g r e s s Repor t . Ministry of the Env i ronment , L a n d s and P a r k s , B C . C o h e n W . B. and T. S p i e s . 1992. Est imat ing Structural Attr ibutes of D o u g l a s F i r /Western H e m l o c k Fores t S t a n d s f rom Landsa t and S P O T Imagery. Remote Sensing Environment. 41 :1 -17 C o x s o n , D. 1999. Presenta t ion at Manag ing Fo res t s for L i chen seminar . C o l u m b i a Moun ta ins Institute of Techno logy . Reve ls toke . S e p t e m b e r 29 - 30 1999. Cr ist , E. , and R ichard C . C i c o n e . 1984. A P h y s i c a l l y - B a s e d Transformat ion of Themat i c M a p p e r Da ta - T h e T a s s e l e d C a p . IEEE Transactions on Geoscience and Remote Sensing. V o l G E - 2 2 , No . 3. Deul ing, M . 1999. P r o g r e s s Repor t #2. Field Collection Methodology and Statistical Analysis of Forest Structural Parameters. Depar tment of G e o g r a p h y , Univers i ty of Ca lgary .31 pages . DeWul f , R., Ro land E. G o o s s e n s . 1990. Extract ion of Fo res t S t a n d P a r a m e t e r s from panchromat i c and Mul t ispectra l S P O T - 1 Data . International Journal of Remote Sensing, V o l 11, No . 9, p. 1571 -1588. E d w a r d s , R., and W . Ri tcey. 1959. Migrat ions of C a r i b o u on a Moun ta inous A r e a in W e l l s G r a y Park , Brit ish C o l u m b i a . Canadian Field Naturalist. 73 :21 -25 . Fiore l la , M. and W . Ripp le . 1993. Determining S u c c e s s i o n a l S t a g e of T e m p e r a t e Con i f e rous Fo res t s with Landsa t Satel l i te Data . Photogrammetric Engineering and Remote Sensing. V o l 59 , No . 2 F e b 1993. P p 239 -246 . Frank l in , J . 1986. Themat i c M a p p e r A n a l y s i s of Con i f e rous Fores t Structure and C o m p o s i t i o n . International Journal of Remote Sensing. V o l 7 No . 10. P p 1287 - 1301 . 27 Goward S. 1991. Normalized Difference Vegetation Index Measurements from the Advanced Very High Resolution Radiometer. Remote Sensing of the Environment. Vol 35 Pp. 257 - 277. Goward T. 2000. Personal Communications. Goward, T. 1998. Observations on the Ecology of the Lichen Genus Bryoria in High Elevation Conifer Forests. The Canadian Field Naturalist. Vol 112 pages 496-501. Huguenin, R. 1994. Subpixel Analysis Process Improves Accuracy of Multispectral Classifications. Earth Observation Magazine July 1994. Pp 37 - 40. Huguenin, R. 1997. Subpixel Classification of Bald Cypress and Tupelo Gum Trees in Thematic Mapper Imagery. Photogrammetric Engineering and Remote Sensing. Vol. 63, No. 6. June 1997. Pp. 7 1 7 - 7 2 5 Kauth, R and G. Thomas. 1976. The Tasseled Cap - A Graphic Description of the Spectral -Temporal Development of Agricultural Crops as Seen by Landsat. In Proceedings from the Symposium on Machine Processing of Remotely Sensed Data. Purdue University, West Lafayette, Indiana. Pp 7 0 5 - 7 2 1 . Lillesand, T, and R. Kiefer. 1994. Remote Sensing and Image Interpretation. Third Edition. John Wiley and Sons Inc. Chapter 7. Ritcey, R.W., 1991. Provincial Caribou Statement for British Columbia 1991 - 1996. Draft prepared for the Wildlife Branch, Ministry of Environment. Rose F. 1976. Lichenological Indicators of Age and Environmental Continuity in Woodlands. Lichenology: Progress and Problems, rev. ed. D. H. Brown, D. L. Hawksworth and R. H. Bailey. Academic Press, London. Pp 279-307. Rose F. 1992. Temperate Forest Management: Its effects on bryophyte and lichen floras and habitats. Bryophytes and Lichens in a Changing Environment, rev. ed. J.W. Bates and A. M. Farmer, Oxford University Press. Pp 211-233. Shackleton, D. 1999. Hoofed mammals of British Columbia. University of British Columbia Press and the Royal British Columbia Museum. Vancouver, BC. Pp 167 - 177. Sharma, R., 2000. Detection of Mountain Pine Beetle infestations using Landsat TM Tasseled Cap Transformations. Dept. of Forest Resources Mangement, University of British Columbia, Vancouver. M.Sc. Thesis (draft). P 82. Simpson, K. 1996. Towards a Provincial Caribou Management Strategy: Results of a BC Environment Workshop. Keystone Wildlife Research, White Rock, BC and BC Environment, Victoria, BC. 38 pp. Spies, T., J . Franklin and M. Klopsch. 1990. Canopy Gaps in Douglas Fir Forests in the Cascade Mountains. Canadian Journal of Forestry Research. 20:649-658. Stevenson, S. K. 1978. Distribution and Abundance of Arboreal Lichens and Their Use as Forage by Blacktailed Deer. Masters Thesis, University of British Columbia. Stevenson, S.K. and D.F. Hatler. 1985 Woodland Caribou and Their Habitat in Southern and Central British Columbia. B.C. Ministry of Forests Report 23, Victoria, B.C. Stevenson, S.K. 1985. Enhancing the Establishment and Growth of Arboreal forage Lichens in Intensively Managed Forests: Problem Analysis. B.C. Ministry of Environment and B.C. Ministry of Forests. IWIFR-26. Victoria, B.C. 40p. 28 Stevenson, S. K. and K. Enns. 1992. Integrating Lichen Enhancement With Programs For Winter Range Creation. Part 1. Stand/Lichen model. B.C. Ministry of Forests. IWIFR-41. Victoria, BC. 34 p. Steiner, D. T. Gutermann. 1966. Russian Data On Spectral Reflectance Of Vegetation, Soil And Rock Types. Technical Report. European Research Office Of The United States Army. Trimble Navigation Limited. 1994. Mapping Systems General Reference. P 4-28. 2 9 Appendix A Scatterplots of Bands versus Lichen Loading Class 2 3 4 Lichen Class 100,— 9 0 -8 0 -7 0 -6 0 -5 0 -4 0 -~ i s—i 1 r _i i i i _ 1 2 3 4 Lichen Class 401 r 35 ~i r B1 30 25 20_L _l I I L_ 1 2 3 4 Lichen Class 1 2 3 4 5 Lichen Class 50, 40 m 30 20 ~i 1 1 r j i i i_ 0 1 2 3 4 5 Lichen Class 30 25U 15L 10 0 1 2 3 4 Lichen Class 0 1 2 3 - 4 4 Table codes Site: Site identification number Band: Contains the digital number (DN) value for each site in specified band Bright: Pixel value after Tasseled Cap Brightness transformation Green: Pixel value after Tasseled Cap Greeness transformation Wet: Pixel value after Tasseled Cap Wetness transformation NDVI: Pixel value after Normalized Differential Vegetation Index transformation UNMIX_Low: Low lichen fraction value after Spectral Unmixing Processing U N M I X M e d : Medium lichen fraction value after Spectral Unmixing Processing UNMIX_High: High lichen fraction value after Spectral Unmixing Processing Supervised_Class: Pixel Classification after Supervised Classification Class: Lichen load rating Label: Low, Medium or High grouping of Class values L J J n e : Line of lichen growth on trees S P _ 1 : Leading Species H T 1 : Height of Leading Species DBH: Diameter of Leading Species Age_1: Age of Leading Species Stems: Stems/Hectare SP_2 : Secondary Species H T 2 : Height of Secondary Species DBH_2: Height of Secondary Species B E L S N A G S : Number of snags counted by field crew B S N A G S _ H a : Number of snags/hectare DTOP_1: Trees with dead top within plot D T O P H a : Trees with dead tops/hectare A B V S N G S : Dead trees counted from helicopter DTOP_2: Trees with dead tops counted from helicopter Size_Ha: Size of site TopClass: Estimated lichen abundance form helicopter 32 co co CD >< c <D Q. Q. < ro O I •o co « ± Z 3 X z s > Q UJ 5 LU UJ cr: o x CO cr: CD C D Q Z < 0 3 m D Z < m •fl-Ca Z < m CO Q Z < m CN Q z < < CO co CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO < 3 3 < < < < < < < < < < < < 3 < < < < < < < < < < < < < _ i _ l 1 _ l _ l _ l _ l _ J _i _ l _ l _ l _ l o L J UJ L J LU o O o o o O o o o o O o I 1 111 o O O CJ i O o o o o o o o O O z _i z z z z z z z Z z z z z z z z z z z z z z z z z z 3 r~ 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 o o C M o o o o o o L O o O o O O C D o CO o o o o o o o o o o C M o o o o o o L O C D o o C M o o C O o o o o o o o o o o d d d d d d d d d d d d d d d d d d d d d d d d d d d d d ro o o o o o o C M C D L O o o o o C M o o o o o o o o o o o o o o o C O C O C M o C O o C D o o C O C O o o o o o o in o o o d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d o o o o o o o o o o o o o o o o C D o o o o O o o o o C M o o o o o o o o o o o o o o o o o C M o o o o O o o o o o in in d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d o C D C D 00 o T— C D CO oo o o C D C O co C O C D C D o C M 00 in C O C O co C O C O C M C O C O C O C O C O C O in C O L O C O C O in C O C O m C O •<a-d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d C O C M C M 00 x — C M r » o T — C O C D C O r-- C D C O CO T_ C M o T_ C M CO l-~ t~~ C D C D C M 00 CM o L O 00 CM C D o C O C D co o> m m m in CD CO C M CXI oi O) oS L O L O cb oi C M C M C M C M C M C M c\i C O CO C M c\i CO C O C O in C M d CO C M d C M C D L O L O L O L O L O •sr C D C D C D C D C D C D C D C D C D L O C D C D m C D co C D C D C D in CO m C D C D O C M o C D L O o C D o •t C D C M C D C M C D C D C D ,_ C D C D CO in CO cn C O C D C D CO CO C D cx> O O ) 00 o L O C D o L O a> in •sr C O C M C O C M in C M C D C D C O 00 cd d 00 i-^  CO C M C O ib r-: d T— in co C M cb in cb x — cb cb cb d C O C O T~ C M T~ T _ T— T— T~ C M C M T— C O T _ T _ C M C M C M C M C O T - C O T — C D L O o L O co T— co O m x— C D 00 O ) m C O • t cn O CO 00 T— T— CM CO C D ro o CO co T— C O \J> L O C M C M cb L O C O cb o I—) cb O J cb cb d oo 00 t—\ in C M d d C M C M C O (—\ d r-: h - C D 00 C D C D CO h- C D h- r- co oo 00 CO C D C D C D r-- V > C D r^-_^ C O C O \ T C D •t co C D C O C O o T— o o> o co m co o o> ai C D co C M C M C M C M C M C M C M C M C M C M C M CM C M C M C M C M T — C M X— C M \— co C O r-- L O L O o L O C M in l O •sr C M CO C M C M C D CJ) r-- C D o C O C O C O ¥ C O C O C M C O C O C O C O C O C O C O C O C M C M C M C M C M C O CO C O C O C D cn L O CO C D in L O in L O C D o m •^ 1- o> co C M C O C D a> in C O m C M L O CD C D C D L O C D m in m l O L O m C D C D co m in 00 in m L O o> m C D co C D C M o L O •^ r O O CO o o o cn o C M 00 o cx> CO o o o C D L D C O L O L O L O L O CO -a- L O L O l O m L O m in in in •<3- in in in m C O L O CO h- C O C O C O C O C O C O C D L O C O C O C M o C M C M C D C O C O C M C O C O C O C O 5- C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O co C M C M o C D C O C M in C D C O co o m C O C M co C M C M C O C D m C M C M C M co C O co C O C M C M C M C M C M C M C M C M C M co C M C M C M C M C M C M C M C M C M C M C M C M C M < re & re re Q 5 re Cd eo _CD (0 55 ^ C M r J ~ " ) m r a o 5 T - T - ^ T - T - T - T - T - T - T - C M ^ C M C M < N C M C M C M C M C N c r ) « CO o o C O C O o o C O C O CD C O C O r-C D C O C O C O C O o o C O C O t ^ C D c p h -C D C O C O C D C O C O N -C D C D r--c p o o c o C O C O o o o o C D C O C O o o o o C D C O C O 00 C O in C O C O C O C D C N C O C M T— C N 00 C N C O C N l b C O C O C O C D C N co C D C N 00 C D C M C O C O C D C M C D C O C O L O c o C D C M 00 C N d i b C N cb C O C O C N d C O d c b C N C O L O CM CO X CD Q C O o C O O C O o o o C O o C O C O C O o C O C O o C O C O o C D C O o C D C O O C D C O o o o C O c o C O o C O C O C D C O o C O C O C O C O C D C D o C O C O o o oo C O C N C O o C D C O C O ^ J - C D C M C N C O L O r-~ C O oo T— C O L O L O 00 00 T C M C N C M *;— C M C M C M C O o O o o o o o o o o O o o o o o o r-- o o o o o o o o c o co C O o O o o o o o o o o O o o o o o o c o C D o o o o o o o o c b c b 00 ^r' L O L O i b i b i b i b i b i b i b i b i b i b i b i b C O C O c b i b i b i b i b i b i b i b o o r-C D o o co o o o o C D o o o o o o o o o o C O r^ -C D O O o o o o o L O o o C O C D o o C O C O o o r~-co o o o o o o o o ib cb d cb ib ib ib ib ib d C O •<t ib C O cb cb cb C M cb 1^ ib ib ib C M 0. co 5 X co CD co DQ DQ CD CO CO CO CO CD CD CD CD CD CD CD CO CO O CD CO CO CO CO CO CO CO CO CD !_1 STEMS C O C O C M C O C O C M o o r--co 00 C O C O C O C O CJ) o o L O C O C O L O C O C O L O C O C O 00 C O C O L O C O C O L O C O C O o o o C M C D t~-co o C O C O f-~ o o C D co o r--co o C O C O 00 C O C O C D 00 o o C D o o o> o o 00 C O C O o r^ -co Oi o o o o C D !_1 STEMS T— T — T ~ T _ T — T — !_1 STEMS o 00 o C O C O Oi co L O T— L O Oi o C M o C O co o C O C O 00 o C N co o C O o cu L O •<*• C O C O L O C D 00 •t UJ o T — C M T — T — T — < co C O o o C D o o C D C O C O C D o o h-C D h-C D o o o o C O C O o o C O C O C D o o C O C O o o C O C O C O C O C O C O o o C O C O o o co C O r~-C D h-C D o o co DBH. cb d C D C O ib C M cb T C O L O cb C O ib •<d-C O C N d C M ib C N cb C O ib C N 00 C N cb C O ib 00 C O ib C O 00 C O cb C O cb C O ib cb C O ib C O cb C M C M C O cb C O ib cb r~-C D o o O O O o o o o o o o o o o o o o o o o o C O C O o o C O C O C O C O o o C O C O o o o o o o o o C O cp o o o o o o o o o o co H ' X cb C M ib C M lb C N ib d C N ib C M d C M d C O d C O d C N d C M ib cb C M d C M cb C M C O C M ib C M cb C N ib C N ib C M ib C M ib C M cb C N C M ib C M d C M ib C N ib C N ib C N cb C M 00 re 0. m Q 5 re CC Oi « re O . W O CC CL X CO 3 X X CO X CO 5 X X CO _ X X — CO co CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO CO 00 o CO co CO CO CO CO co CO CO CO o o o o O o o CO o o o o CO o CO o o o o o CO CO o o LO CM CM 00 r— 00 CD T— CN CD LO co CO co 1^ - CD CD Oi 00 Oi •<3- LO Oi CD LO < CO < < < o CO o < BC CO o o CO o o CO CO CO CO < < CO CO CO CO CO CO CO CO o CO CO < CO CO o CO < CO CO o < CO < CO o CO CO o o o o CO o CO o o CO o CO o o o o CO o o o co o o o o co oo oo o LO o LO co 00 o CO o CD o co o oq o LO o LO 00 o o o 00 o LO LO CM •r- 1— CM 1— CM CM CM CM CM CM 1— CN T" 1 CM I - 1 CM CM CN T— CM LU CD 5 to CO o £ to co co C M 2 ZD ZD ZD Z) ZD ZD ZD ZD ZD ZD ZD EDI IGH EDI EDI o EDI EDI IEDI IGH IGH IEDI o IEDI o IGH IEDI IGH o o IEDI IEDI o IGH o o o o IEDI IEDI X _ i X X _ i _ i X X _ i _ i _ i X _ i _ i _ l _ i h ~ o CO o r^- CO o CO r*- CO r - o o o CO CO CO r-- CO CO CD CO CD o CO o CD c o c o CD CO CD o CO CO CO CD o o o c o CD CO co c p c p c p c o CO CM CO CM CO CM CO CM CM CO c b 1— CO CM CO CO CO CO CO T— CM c b c b Q L U w C O ^ T L O C D O O C D O T - C O ^ lO (O s 00 CD O T - 1 - C N ( N n ^ i n t D s c o o o i N C N C N C N C M C M C M C M C M C O C O CN co o CL O C O CM CO C M CM T ~ T - CM L D L O L O - " C M C M CM •<- CM CM CM CO m s tf M - CM O °? cb cri O ) V T -£ 1 0 & £ o i ° > v— L O LO LO CD LO LO CO CO CD LO CM CO 1^- LO O) CM CD CO CO CO co LO O) CO CM CO LO O CM CM CD 00 CD CD CD x — •«- CM CO CM CM LO CM CM LO d CO cb 10 o i ra X I CM CL o o o cb 0 0 0 0 0 0 0 CO O O 0 0 0 0 0 0 0 0 CD q CD 00 0 0 CO CD ( _ ) CD O O 0 0 0 CD O 0 CO cb o i cb cb CO CO cb 10 CO cb o i cb 10 10 CO CM CM x - CM x - x - x -0 0 0 O r - 0 O O L O 0 L O c o 0 L O C O c b CM c b c b CM CD c b a O H Q —^ LO CO 00 CM 10 CD CO CO I-- CD O CO CO CO 0 1^- CD CD CM CO ^3- 10 CD CO CM LO 10 CD CO Oi LO CO • « — O CM CO 00 CD CO 0 CD CD co CO d o i cb o i 10 10 i b cb i b cb CM CO cb cb CO d i b x - x - CM X - x - x - CM CM X - CO CM CM X - CO CO CM CO X - ^ 4 CM C3 z CO > m < C M 0- ^J" CO CO CD O CM CO h- 00 Q CO o z CO > m < 10 C M c o c o c o o o a > o o a > c o - ^ - T - C M C M T - C M T - T -1 - 1 - C M 10 1 - r-- 10 C M C M C M r~- C M T - T -co o co c o L O C O C O CM CD C O T— LO x — CM c o o CD CD O CO 00 CO L O CM L O O 1 - CM CM CD L O CD CD L O C O C O C D C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O C O 0 0 C O C O C O C O 1^  C D cb C O cb co C O C O C O C O cb C O cb C O C O C O C O C O C O C O cb C O d cb C O cb C O cb C D 00 a> 0 00 CO 0 CO CD O x— T— CM x— O 0 0 0 C O C O 0 0 C O C O 0 1— C D 0 0 C O C O 0 0 0 O 0 0 0 0 0 0 0 0 0 0 0 d d d d cb C O d d cb C O d cb C D d d cb C O d d d o t: co \ t O) O O) M O (O C O C D C M C O C D C M L O (0 re a s re o 0) £t re co O < z CO m CO CO 0 CO CO CO CD 0 CO cb cb cb d cb CO CO co 0 CO CM CO CM T— C O C O C D C D C O C O cb cb cb C O C D C D C O C O C D C D O CO CO 0 CO CO CD CD 0 cb cb cb cb d CO CO CD CD 0 CM T — 10 CO CO • co ID to " CO CO 1^  0 CO CO r C D CD 0 cb cb cb CD cb d CO CO CD 0 CM CM x — T— CO CO CD CO CO cb cb cb CO CO CO 0 0 0 0 0 0 o o d o CD cb CD C O C O C O C O C O C O C O C O C O 0 C O C O 0 0 0 0 C O 0 0 C O 0 r-- 0 0 C O 0 0 0 C O C D C O C O C O C O C O C O C O C O 0 C O C O C D 0 0 0 0 C O 0 0 C O 0 C D 0 0 C O 0 0 0 d d d T— d d d d d d d d d d d d d d d d d d d d d d d d d d CO CO 1^- 0 CO CO CO r--cp CO CO 0 CO CO CO CO 0 CO CO 0 0 0 0 CO CO CD 0 CO CD CD CO CO CO CO CD CD 0 CD CO CO CD CD q CD CO CO 0 0 0 q CD CM cb CM •<- d d d d d CM •<- •«- •«* i b d O) CM CM d d d d d d d d LU CO LOW MEDIUM HIGH MEDIUM MEDIUM LOW MEDIUM MEDIUM MEDIUM HIGH HIGH MEDIUM LOW MEDIUM LOW HIGH MEDIUM HIGH LOW LOW MEDIUM MEDIUM LOW HIGH LOW LOW LOW LOW MEDIUM MEDIUM CO CO CD h-CD CD 0 0 CO CO 0 0 CD CD CD CD CO CO CO 0 0 CO CO CO CO CO co 0 q 0 q 0 0 CO CO r-co CO CO CO CO h-cq CD r- CO CO CO CO CM CM cb CM cb CM cb CM CM cb cb T— cb CM cb cb cb T— T— cb cb T ~ CM cb cb * CM CO 1^- 10 CD 00 CD 0 * CM CO LD CD CO O) O CM * CM CM CM CO CM CM 10 CM CD CM CM 00 CM O) CM 0 CO CM CO 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0089591/manifest

Comment

Related Items