UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The detection of mountain pine beetle green attacked lodgepole pine using compact airborne spectrographic… Heath, Jamie 2001

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

Item Metadata

Download

Media
831-ubc_2001-0212.pdf [ 20.24MB ]
Metadata
JSON: 831-1.0090048.json
JSON-LD: 831-1.0090048-ld.json
RDF/XML (Pretty): 831-1.0090048-rdf.xml
RDF/JSON: 831-1.0090048-rdf.json
Turtle: 831-1.0090048-turtle.txt
N-Triples: 831-1.0090048-rdf-ntriples.txt
Original Record: 831-1.0090048-source.json
Full Text
831-1.0090048-fulltext.txt
Citation
831-1.0090048.ris

Full Text

THE DETECTION OF MOUNTAIN PINE BEETLE GREEN A T T A C K E D LODGEPOLE PINE USING COMPACT AIRBORNE SPECTROGRAPHS IMAGER (CASI) D A T A by Jamie Heath Bachelors Degree in Geography, Simon Fraser University, 1997 Certificate in Spatial Information Systems, Simon Fraser University, 1998 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE In F A C U L T Y OF G R A D U A T E STUDIES DEPARTMENT OF FOREST RESOURCE M A N A G E M E N T 1 FACULTY OF FORESTRY We accept this thesis as conforming to the required standard The University of British Columbia April 2001 ©Jamie Heath, 2001 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 The University of British Columbiir ^ Vancouver, Canada Date April Zo" /zoo\ DE-6 (2/88) Abstract The mountain pine beetle (Dendroctonus ponderosae Hopkins) is the major pest of mature lodgepole pine (Pinus contorta var. ladfolia) in western North America. The present method of surveying bark beetle attack is by sketch mapping or photographing the location of last year's damaged trees (red attack) from an aircraft. The area around the red attack is then ground probed to find currently infested trees that have not changed colour (green attack). If there was a way to locate both the "red and green attack" pine trees without the need to ground probe, it would greatly assist bark beetle management and reduce costs. Previous studies have indicated that there is a reflectance change in the near infrared and the visible wavelength regions (particularly the red rise and red edge regions). This study analyses the use of a compact airborne spectrographic imager (CASI) for its ability to detect subtle reflectance changes that are believed to occur following a mountain pine beetle attack. Imagery was acquired in 36 discrete narrow bands with a ground resolution of 60 cm. Differences between the (24) green and (25) non-attack populations could not be visually detected on the imagery, therefore the digital numbers (DN) from the central crown portion of the sample trees were examined statistically. A stepwise inclusive discriminant analysis indicated that two spectral bands, the shoulder of the green rise (539.5 +1-1.1 nm) and the toe portion of the red edge (706.4 +1-1.% nm) exhibited a difference between the populations. A jackknifed classification matrix displayed that 79 percent of the green attack and 68 percent of the non-attack trees were properly classified. A graph showing the central frequencies from the results of the canonical scores of the populations indicate that the populations are quite close together and overlap considerably. The subtle spectral differences between the populations may complicate future attempts at classifying the populations. The difficulty may be due, in part, to the complex ecological diversity of the lodgepole pine sites, insufficient needle desiccation (from cool / moist environmental conditions following initial beetle attack) and from the low sun angle at the time of the imagery acquisition. The low sun angle adds complexity to the analysis from poorly lit tree crowns, long shadows, and from having only the southern side of the tree crown illuminated. ii Table of Contents A B S T R A C T ii T A B L E OF CONTENTS i i i LIST OF T A B L E S iv LIST OF FIGURES v A C K N O W L E D G E M E N T S vii 1. INTRODUCTION 1 2. OBJECTIVES 2 3. L ITERATURE REVIEW 3.1 The Life Cycle of the Mountain Pine Beetle 3 3.2 Controls of Mountain Pine Beetle 4 3.3 Present Methods of Mountain Pine Beetle Detection 5 3.4 Green Attack Research Studies 7 3.5 Environmental Factors and Seasonal Conditions 12 4. THE C O M P A C T AIRBORNE IMAGING S Y S T E M (CASI) 4.1 An Overview of the System 14 4.2 Image Scale 16 4.3 Bi-directional Effects 17 5. METHODS 5.1 The Study Area 17 5.2 Airborne Data Collection 22 5.3 Ground Truth 23 5.4 Analysis of Spectral Data 25 5.5 Statistical Methods 29 5.6 Categorizing by Physical Attributes and Location on the Imagery 30 6. RESULTS 6.1 Reflectance Curves 30 6.2 Classifications 44 11.2 Band Ratios 46 11.3 Statistical Significance 47 12. DISCUSSION 55 13. CONCLUSIONS 63 14. RECOMMENDATIONS 64 15. REFERENCES 65 16. G L O S S A R Y 68 APPENDIX I 72 APPENDIX 2 73 APPENDIX 3 74 APPENDIX 4 79 iii List of Tables Table 1. Findings from Ahern (1988) study of lodgepole pine tree stress following bark beetle attack 2. The center wavelength (and the corresponding bandwidth) from the thirty-six (36) bands used in this study. 3. A synopsis of subzones in the Sub-Boreal Spruce zone (SBS) (Meidinger et al., 1991). 4. The band ratios used on the extracted tree crown imagery. 5. Band ratios for the green attack and non-attack trees. 6. The mean, standard deviation, and t-test results (60 cm imagery) from all 36 bands for the green attack and non-attack populations. 7. The mean, standard deviation, and t-test results (1.2 m imagery) from all 36 bands for the green attack and non-attack populations. 8. Classification matrix (cases in row categories classified into columns) 9. Jackknifed classification matrix as a result of the discriminate analysis on the 60 cm data. 10. Canonical discriminant functions ~ standardized by within variances. Please note that the discriminate analysis only indicated that bands 8 and 19 are different between the populations using the 60 cm data. 11. Canonical scores of group means, based from bands 8 and 19 using the 60 cm data. 12. Classification matrix (cases in row categories classified into columns) 13. Jackknifed classification matrix as a result of the discriminate analysis on the 1.2 m data. 14. Canonical discriminant functions ~ standardized by within variances. Please note that the discriminate analysis indicated that bands 3, 5 and 6 are different between the populations using the 1.2 m data. 15. Canonical scores of group means, based from bands 3, 5 and 6 using the 1.2 m data. I V List of Figures Figure 1. The lifecycle of the mountain pine beetle (Roswell, 1981). 2. A cross sectional example of the blue stain fungus (Roswell, 1981). 3. An example of the colour change following bark beetle attack (Safranyik, et al, 1974). 4. An example of the different foliage age classes on a conifer tree branch (Murtha et al., 1997). 5. A graph showing the results of a green attack study of ponderosa pine {Pinus ponderosa) from Heller (1968). 6. The approximate study area is indicated by the red star. 7. The species composition and their approximate percentage within the Tyee Lake study area. 8. The red boundary shows the area where the CASI imagery was flown. The smaller pink boundaries show the location of three sample sites (where ground data was collected). 9. Site #1. Sample trees were selected from both sides of the directional flight swath. The seam line in the middle of the image divides the top (forward-scatter) and the bottom (backscatter) zones. Created from bands 27, 19, 8, as R,G,B. The red boundaries represent the green attack trees and the green boundaries represent the non-attack trees. 10. Site #2. Sample trees were selected from both sides of the directional flight swath. The seam line in the middle of the image divides the top (forward-scatter) and the bottom (backscatter) zones. Created from bands 27, 19, 8, as R,G,B. The red boundaries represent the green attack trees and the green boundaries represent the non-attack trees. 11. The merged CASI imagery. The (east / west) flight lines have been processed and geo-referenced. 12. An example pitch-tubes on the trunk of a green attacked lodgepole pine. 13. Both images show closeup examples of two large trees. The image on the left is the original 60 cm imagery and the image on the right is from the resampled 1.2 m imagery. A vector line indicates the boundary of the larger tree and a cross-hair indicates the center of the crown. The numbered grid shows the order in which the digital numbers were recorded into the spreadsheet program. The #5 pixel (from the 60 cm imagery) is located at the center of the tree crown. Only the reflectance value from one (central pixel) is recorded from the resampled 1.2 m imagery. 14. The imagery from Site 1. The resampled imagery (1.2 m) is in the smaller window. 15. The imagery from Site 2. The resampled imagery (1.2 m) is in the smaller window. 16. The derived 60 cm spectral reflectance data for the green attack trees. 17. The derived 60 cm spectral reflectance data for the non-attack trees. 18. The averaged 60 cm derived spectral reflectance data for all of the sample trees. 19. The derived 1.2 m spectral reflectance data for the green attack trees. 20. The derived 1.2 m spectral reflectance data for the non attack trees. 21. The averaged 1.2 m derived spectral reflectance data for all of the sample trees. 22. The derived spectral reflectance curves (60 cm data) for all of the sample trees in Site 1. 23. The derived spectral reflectance curves (60 cm data) for all of the sample trees in Site 2. 24. The derived spectral reflectance curves (1.2 m data) for all of the sample trees in Site 1. 25. The derived spectral reflectance curves (1.2 m data) for all of the sample trees in Site 2. 26. A graph showing the red edge portion of the wavelength for green attack and the non-attack lodgepole pine trees from the 60 cm data. 27. A graph showing the red edge portion of the wavelength for green attack and the non-attack lodgepole pine trees from the 1.2 m data. 28. A Maximum Likelihood Supervised Classification on Site 1 (60 cm data). The green attack crowns are traced in red and the non-attack crowns are traced in green. The pink colour represents the green attack and the blue colour displays the non-attack classification. 29. A Minimum Likelihood Supervised Classification on Site 1 (60 cm data). The green attack crowns are traced in red and the non-attack crowns are traced in green. The pink colour represents the green attack and the blue colour displays the non-attack classification. 30. A Minimum Likelihood Supervised Classification on Site 2. The green attack crowns are traced in red and the non-attack crowns are traced in green. 31. The canonical scores results from the green attack and non-attack populations of 60 cm data. Please note that the central frequencies of the populations are close together and overlap considerably. 32. The average spectral reflectance curves and the standard deviations of the green attack and non-attack populations from 60 cm data. 33. The average spectral reflectance curves and the standard deviations of the green attack and non-attack populations from the 1.2 m data. 34. The spectral reflectance curves from the (6,0 cm imagery) colour coded based on the trees location on the imagery (fore or back scatter zones). V I Acknowledgements Several people provided assistance during this project. I would like to thank Mr. Leo Rankin of the British Columbia Forest Service as without his help and financial assistance, this project would not have been conducted. I would also like to thank Mr. Val Fletcher and Mr. Robin Brown, also from the Ministry of Forests (Victoria), for their invaluable support and financial assistance in the project. I would also like to acknowledge Dr. Doug Davidson from ITRES Research Limited (Calgary, Alberta) for his assistance in setting up the project and the assistance during the analysis. As well, I would like to thank my graduate supervisors, Dr. Peter Murtha and Dr. John McLean. Dr. Murtha provided assistance throughout my studies and his remote sensing classes were very informative. I would like to thank Dr. John McLean for his assistance with the statistics in this project. The use of the computer and the remote sensing software (PCI) at the FIRMS lab was valuable. V l l 1. Introduction The mountain pine beetle (Dendroctonus ponderosae Hopk.), poses a tremendous risk to the lodgepole pine (Pinus contorta Dougl.) forestry resource in British Columbia. The insect is widespread, destroying thousands of hectares of lodgepole pine forests every year. It is one of the roles of the forester to monitor for signs of insect infestation. This process is conducted by sketch mapping the locations and numbers of red trees, commonly called "red attack" (usually killed by insects) onto maps during the annual aerial survey. Occasionally, aerial photography is used to assist the sketch mapping survey. The surrounding area (around the red trees) is ground probed to determine the number of trees that are currently infested, but have not changed colour, commonly called "green attack." The term "green attack" is used since the foliage of the tree remains green, visually undetectable, yet the tree is slowly being killed. It would be a valuable tool for foresters if a remote sensing system could be used to detect the red and the green attack lodgepole pine trees. In order to be able to detect the changes in spectral reflectance, a proper understanding of the physiological changes of the tree must be known. The successful previsual detection of tree stress is dependent on the detection of the spectral reflectance changes of the tree. Shortly after the bark beetles bore into the soft cambium tissue, the tree becomes stressed. The tree appears "healthy" because the foliage remains green for up to a year following the bark beetle attack. Despite the "healthy" green appearance, chlorophyll production is decreased and existing chlorophyll starts to disintegrate as a result of the stress, absorbing less light in the chlorophyll absorption bands (Runesson, 1991). The decrease in light absorption causes an increase light reflectance in the corresponding reflectance regions. Field spectroscopy is the best method for examining change of the trees' reflectance. Unfortunately, in-situ studies of large conifers are very difficult and impractical. The use of an airborne sensor, such as CASI (compact airborne spectrographic imager) or a spaceborne sensor is the only practical means of surveying large stands of lodgepole pine trees. Airborne or spaceborne scanning systems also provide a "one time" survey, thus reducing complications encountered from laboratory analyses. This research study is a joint effort between ITRES Research Ltd., the Cariboo Forest Region, the Ministry of Forests (Victoria), and the Department of Forest Resource Management at the University of British Columbia. The purpose of this thesis is to evaluate the use of a CASI for the previsual identification of "green attack" mountain pine beetle infested lodgepole pine trees. Thirty-six (36) bands of hyperspectral, 60 cm pixel size, mosaicked imagery are examined. The results of the spectral reflectance data are tested to determine if the green attack stage can be separated. 1 2. Objectives The major objective of this study is to apply the CASI data to the sample trees and to determine if the green attack trees are detectable. To do this, thirty-six (36) band hyperspectral (<16nm in width) imagery from CASI are evaluated to determine if the subtle spectral changes can be detected (and interpreted), following a mountain pine beetle attack. The technology has to be demonstrated as an alternative to ground probing. The specific objectives of the study were to: 1. Visually inspect the CASI derived spectral reflectance data from the green attack and the non-attack populations to determine if any specific bands have a pronounced difference. 2. Verify previous published results that, by using supervised classifications, narrow band hyperspectral imagery can detect differences between green attack and non-attack populations. 3. Test the hypothesis that, through the use of band ratios, differences in spectral reflectance between green attack and non attack populations are more pronounced. 4. Determine statistically (examining means, t-tests, standard deviations, and discriminant analysis), i f there are any differences among the reflectance data between the green attack population and the non-attack population. 5. Conduct the same tests on the resampled imagery to determine if the resampled imagery is similar / different to the original crown reflectance data. 2 3. Literature Review 3.1 Life Cycle of the Mountain Pine Beetle Eggs are laid in the phloem of pine trees in August and September. Within a few days, the eggs hatch and the young larvae begin feeding (Furniss and Carolin, 1977). In overwintering and early summer stage, there are four larval instars and then they transform into pupae in June-July, developing into adults by July-August (Amman and Cole, 1983). Shortly thereafter, the new adults emerge from the host tree, beginning the migration and colonization phases. Figure 1. The Lifecycle of the Mountain Pine Beetle (Roswell, 1981). When the beetles reach adulthood, they emerge and search for a new host, mate and continue the species. After leaving the brood tree, beetles engage in a dispersal flight to widely scatter the new adults and to find new hosts. There is usually one generation per year (Safranyik et al., 1974). 3 The beetle attacks also serve to introduce two blue stain fungi (Ceratocystis montiaRumb. And Europhium spp.) into the trees (Reid et al.,1967). The fungi aid in killing the tree by interrupting the flow of translocates, causing moisture stress in the tree. Not only does the fungi help the beetle kill the tree, but it also serves as a food source to the beetle during part of its life cycle. Once the bluestain becomes successfully established, it will also retain moisture in the sapwood and prevent excessive dehydration of the phloem, which is essential for brood survival (Unger, 1993). Figure 2. A cross sectional example of the blue stain fungus (Roswell, 1981). Pines trees have an ability to resist the attacks from the beetles. Vigorous, healthy trees exude pitch in response to the injury inflicted from the insect trying to bore into it. If the pitching is rapid enough, the beetle can be forced out or be entombed within the tree (Furniss and Carolin, 1977). Not only does the pitch remove the beetle, but the resin interferes with communication between beetles, by preventing aggregation pheromones, like trans-verbenol (produced by the beetles), from getting out of the tree (Raffa, 1988). The monoterpenes contained in the pitch act as fungal inhibitors, disrupting the growth of the symbiotic blue stain fungus that the beetles carry (Raffa, 1988). A l l of these reactions are a normal response to any injury. However, there is usually an additional, induced response in the form of a necrotic lesion forming in advance of the injury. Accompanying the lesion is an increase in fungicidal secretions in the area of the wound and infection (Raffa, 1988). This is often referred to as a hypersensitive response. 3.2 Controls of MPB For the beetle population to go from endemic to epidemic levels, the beetles must be able to overcome the trees' threshold of resistance. The transition from endemic to epidemic is also dependent on whether there is a high number of hosts available to the beetle population. If there is a low availability of good hosts, the beetle population can collapse. The fall from epidemic to endemic 4 population occurs as the larger diameter trees are killed by previous generations of beetles, leaving fewer suitable hosts. This results in lower survival rates. After that, there are not enough beetles to overcome the strongest trees' threshold of resistance, resulting in a-further loss of host trees, resulting in lower beetle survival, and so on (Raffa, 1988). Planning is the best control strategy to prevent mountain pine beetle problems. Due to the beetles' preference for large diameter, mature trees (80+ years), it appears reasonable to harvest stands before they reach over-maturity (Runesson, 1991). Shore and Safranyik (1993) identified fours variables as indicators of a stand's susceptibility to mountain pine beetle: relative abundance of larger diameter pine, age of dominant and codominant pine, stand density and location. Lodgepole pine forested areas can be evaluated using this criteria, those that are within the zone of "high risk" from a mountain pine beetle infestation should be harvested prior to becoming infested by beetles. The only direct control method for mountain pine beetle is by salvaging the infected wood or burning the infested trees. The present method of salvaging is effective when the area is easily accessible. If there are already roads to the infested area, then salvage is the best sanitation option. When the infested area is not accessible, then it is much more difficult to salvage the wood, and other sanitizing methods must be performed. The most common method is by falling the infested trees and burning them. This "fall and burn" technique is quite costly and is somewhat wasteful since the wood is lost. Throughout the Cariboo Forest Region (and the around much of the province of British Columbia) the overwinter mortality (in the 1998/1999 and the 1999/2000 winters) of mountain pine beetle was very low (pers. comm. L. Rankin). This is in part due to the mild temperatures experienced during these winter seasons. The low winter beetle mortality may contribute to a high population of beetles. These epidemic beetle populations will have devastating results on the susceptible, overmature pine forests of British Columbia. 3.3 Present Methods ofMPB Detection Every year, the survey of forest damage is conducted from a fixed wing plane or helicopter and the location and intensities are sketched onto maps. Damage is seen as discoloured "red attack" trees. The aerial survey is primarily concerned with detecting and monitoring the more serious and economically important forest pests that produce symptoms that can be classified by aerial observations (Moody, 1981). 5 Annual aerial surveys can be used to detect and map mountain pine beetle infestations once the foliage of beetle killed trees turns colour. Figure 3 shows the stages of colour change: 1) green, 2) chlorotic, 3) red. The highly visible orange - red foliage resulting from M P B attack is most noticeable in late July and August, and thus overview flights should be scheduled within this period (Hall and Maher, 1985). Special surveys of damaged trees are done in more detail on the ground, but the extensive forests of the region require aerial surveys (Moody, 1981). Figure 3. An example of the colour change following bark beetle attack (Safranyik, et al, 1974). Although annual sketchmapping surveys supply resource managers with current and useful information about the status of pests, more reliable remote sensing techniques could provide improved data on the location and extent of the pest damage (Myhre and Silvey, 1992). Conventional nine-inch aerial photographs are used daily in a number of forestry applications. Colour and colour infrared nine-inch aerial photographs have proven especially valuable for locating and appraising mountain pine beetle damage. Aerial photographs are more expensive than overview flights but they have the advantage of a higher accuracy for detection and mapping. Harris and Dawson (1979) found that depending on the scale used, aerial photography is as, or more accurate than aerial sketch mapping. But they also found that by using a combination of the two methods (sketch mapping and aerial photography) the advantages of each are well utilized. 6 Multispectral imaging systems have proven useful for overview surveys but they are not cost effective for single tree detection. Satellite multispectral imaging systems, such as SPOT and Landsat are able to show differences on a regional level, and to varying degrees, the forest stand area, but are unable to show single trees or small groups of red attack. Higher resolution airborne multispectral systems are able to detect red attack trees, but are often not cost effective when compared to conventional aerial photography. Airborne multispectral digital scanning systems (such as MEIS-2 and CASI) are able to detect individual red attack trees. Kneppeck and Ahern (1989) have shown success with push-broom scanning for mortality "red attack" mapping. Their work was based on the use of MEIS-2 airborne scanner flown with a (coarse) spatial resolution of 1.4 m. Airborne multispectral imaging is not a cost effective means for red attack detection due to the poor spatial resolution of the sensors. More flightlines are required to sufficiently cover an area (than aerial photography), thus the increased operating costs of airborne multispectral imaging. Multispectral satellite data has also been used to detect "red attack," but the accuracy is insufficient for operational use. There have been studies that examined SPOT multispectral and panchromatic data to determine their ability to detect recent mountain pine beetle mortality ("red attack"). Sirois and Ahern (1988) indicated that the minimum level of detection is an area of approximately 1-2 ha extent with 80 to 100 per cent of the trees having red crowns. This degree of mortality is too great for control programs, but SPOT data would possibly be useful for an inventory update following a mountain pine beetle outbreak that caused extensive mortality (Sirois and Ahern, 1988). 3.4 Green Attack Research Studies Traditional studies of "previsual" tree stress indicate that normal plants have 6 to 10 times larger chlorophyll concentrations than the chlorotic plants, because of the chlorotic plant's deformed cellular structure (Gausman, 1974). Chlorophyll is responsible for absorbing light and processing it into high energy compounds. Decreases in chlorophyll absorption cause an increase in reflectance. Therefore, the detection of stressed trees is dependent on knowing the spectral response characteristics, which in turn, is based on the plants chlorophyll concentrations. When a lodgepole pine tree is subjected to stress from a mountain pine beetle infestation, the tree undergoes physiological changes. This is brought on by the interruption of translocates and deterioration of chloroplasts as the beetle bores into the trees soft cambium tissue. Chlorophyll production is 7-decreased and existing chlorophyll starts to disintegrate as a result of stress, absorbing less light in the chlorophyll absorption bands (Runesson, 1991). The decrease in light absorption causes an increase light reflectance. The change in spectral reflectance following tree stress occurs relatively slowly. The Canadian summer season is too short to sufficiently dry out the trees during the same summer that they are attacked. The attacked trees do not turn visibly orange - red until the following summer. Previous studies have concentrated on the use of colour infrared film to monitor "green attack" pine. There has been varying levels of success, by measuring the desitometric changes on large scale (> 1:2000) colour infrared film. Hobbs and Murtha (1984) successfully detected "green attack" pine using this method, but Runesson (1991) could not replicate their success. The difficulty is partially due to the broad band nature of film and partially due to biological variability of the trees and insects. Even if the variabilities were somehow calibrated, large scale (ex: 1:2000) colour infrared photographs are not operationally feasible, due to the expense of obtaining and interpreting large scale colour infrared photographs over wide areas. Film is a broad band device, and as such, are unlikely to detect subtle changes that occur in narrow band portions of the wavelength. Aerial films have excellent spatial resolutions but low radiometric resolution. Bandwidths less than 50 nm (which are simply achieved through the use of interference filters) are not normally possible with aerial films. Therefore, the small spectral changes (as a result of stress) in vegetation, when measured (as radiances) by various broad-band sensors, are often masked by the high degree of variation in radiance caused by factors such as varying view geometry, illumination, and canopy density (Runesson, 1991). Most plant stress studies are conducted in laboratories using agricultural crops. The majority of the studies indicate that there are changes in the near infrared and red reflectance wavelengths following plant stress. Sensors with spectral bands in the red (RED) and near-infrared (NIR) lend themselves well to vegetation monitoring since the difference between the red and near-infrared bands has been shown to be a strong indicator of the amount of photosynthetically active green biomass (Tucker, 1979). Some models have been developed to quantify these changes. Vegetation indices such as the Normalized Difference Vegetation Index (NDVI - a ratio between the near infrared and the red wavelengths) and the Ratio Vegetation Index (RVI), are widely used to show such photosynthetic activity, nitrogen deficiency, other forms of vegetation stress (Tucker, 1979). Through ratioing, N D V I will show healthy vegetation has having a higher number than stressed vegetation even in the presence of non-uniform illumination. 8 Unfortunately, it is difficult to apply these indices to the detection of tree stress in forests. Deering (1989) found that indices are still sensitive to changing illumination geometry. This is partially because the shadow structure of a pine forest is much more complex than a uniform agricultural crop. It is very difficult to model the changing illumination geometry. The influence of shadows is more severe in the near infrared than in the red; hence, indices that utilize both red and near infrared reflectance will be affected (Runesson, 1991). Even though the near-infrared region has been identified as sensitive to stress at the foliar level in many studies, attempts to use such results in practical remote sensing applications have often been frustrated by the great variability caused by extraneous effects, such as, tree vigor in the micro-site and effects from understory vegetation. Detection of previsual conifer (individual tree level) stress has had little success with the use of NDVI . Instead, previous previsual stress detection studies that have shown that the greatest change in stressed conifers occurs in the visible wavelengths. Hobbs (1983), stated that results of the total film response analysis of damage classes, where significant, tend to indicate that there are changes in the total green and total red response, suggesting that reflectance changes occur in the green-red portion of the spectrum. Other studies have supported these results. Murtha and Wairt (1989) found that significant differences existed between non-attack and current-attack digital numbers (DN) in the green and red data, and that there was no significant difference between the NIR data. Murtha (1985) used visual analysis to detect green attack lodgepole pine. His analysis examined the colour and the hue of the tree crown to determine whether the tree had been recently attacked. He identified different foliage age classes on a normal (healthy tree) tree have different tones and hues (as shown in Figure 4). The detection of recently killed (green attack) lodgepole pine trees can be seen when the crown loses its variegated pattern. The variegated patterns (from different foliage age classes) in non attacked trees are evident in remotely sensed data when the spatial resolution is fine enough to resolve the foliage differences in a tree crown. We suspect that a finer spatial resolution than is currently used in airborne scanners (i.e., 0.7 milliradian resolution for MEIS-2) is required to detect and separate non attacked from current attacked pine crowns (Murtha, 1985). 9 Branch From Normal Tree w i l h Dif ferent Fo l iage Age C l a s s e s C//^ N E W N E W = Current Year Growth 1 B 1 Year O ld Fol iage 2 = 2 Year Old Fo l iage 3 - 3 Year Old Fol iage (senescent fol iage) 4 » Bare Branch U g h t Gr«en Dark Green Darkest Green Dead R e d - B r o w n Bark Beetle Successful Green Attack (Tree has been killed) NEW Foliage Growth in Year of Attack N e w , 1 , 2 , and 3 year old foliage have the same tone and hue. Figure 4. An example of the different foliage age classes on a conifer tree branch (Murtha et al., 1997) Through the use of narrow band imaging, subtle spectral changes can be more easily detected. Narrow band spectral changes have proved to be useful in the detection of previsual tree stress. Carter et al. (1996) found that the reflectance of blighted needles of slash pine was higher in the 680 nm portion of 10 the wavelength. In another study, Carter (1998) stated that a vegetation index from a narrow band filter centered at about 700 nm and a near infrared band should provide increased accuracy in estimating photosynthetic capacity of lobolly pine. Carter et al. (1996) found that the 694 nm narrow band wavelength yielded the best results for early and previsual stress detection of lobolly pine. The majority of narrow band plant stress detection studies focus on the red portion of the wavelength particularly the red edge and the red rise. The red rise (627 nm to 635 nm) is the small reflectance region that is reported to fall following stress among conifers. Reflectance change in the area of the red rise can be observed in the spectra of ponderosa pine under attack by mountain pine beetle (Heller, 1968). In this study, Heller (1968) reported that there is a rise in this region following an attack from the mountain pine beetle. These results differ from Ahern's (1988) study, stating that the reflectance at the red rise (633 nm) is lower for the current foliage of attacked trees. Figure 5. A graph showing the results of a green attack study of Ponderosa Pine (Pinus Ponderosa) from Heller (1968). The red edge as it is often referred to, is located between the chlorophyll well at 680 nm and the near infrared plateau at 740 nm. It is the transition zone between the red chlorophyll absorption region and NIR reflection region, where internal leaf scattering causes the high NIR reflectance (Murtha et al., 1997). When a plant or tree is stressed the greatest changes occur in narrow band reflectance regions, l l especially in the area of the red edge (680-740 nm). The red edge, which responds to a decrease in total chlorophyll (as well as to additions of tannins) by shifting towards shorter wavelengths, has proven a successful indicator of stress in forest vegetation (Curtiss and Ustin, 1989; Lichtenthaler, 1989). Ustin and Curtiss (1990) found that not only lodgepole pine, but Douglas-fir and ponderosa pine all showed the greatest changes in the 600 - 725 nm range when they were stressed. Ahern (1988) published a detailed report where he outlined the specific reflectance regions and narrow band wavelengths where reflectance change is most significant for mountain pine beetle attacked lodgepole pine trees. His findings are similar to those reported in other previsual tree stress articles. He identified three spectral bands (the green peak, red edge and near-infrared (NIR) shoulder regions) as being the most promising for detecting early effects of bark beetle attack. Ahern (1988) conducted this study by measuring the reflectance from needles of different age classes in both non-attacked and green attacked lodgepole pine trees. He found that the current foliage for healthy trees was brighter in the regions of 525 - 565 nm and at the 633 nm (red rise). Table 1 shows several narrow band wavelengths where the largest difference between non-attacked and attacked lodgepole pine trees (according to Ahern, 1988). The results also showed that there were age class differences between current and older foliage in successfully attacked lodgepole pines. These findings support the variegated visual description used by Murtha (1985). In the visible part of the spectrum (including the red edge), only the current foliage shows reliable differences between attacked and non-attacked trees. Ahern (1988) concluded by stating that more work remains to be done at the whole tree level to determine whether "green attack" can be detected in the face of natural variability and confusion classes which we have not examined. Table 1. Results based on Ahern (1988) Study of Lodgepole Pine Tree Stress Following Bark Beetle Attack. Findings from Ahern (1988) Study of Lodgepole Pine Tree Stress Following Bark Beetle Attack Green peak 525 - 565 nm Lower for current foliage in attacked trees Red rise 633 nm rise Brighter for current foliage in non-attacked trees Red edge 690 - 730 nm Red shifted for current foliage of attacked trees NIR shoulder 730 - 760 nm Lower for attacked trees Runesson (1991) had similar findings to Ahern's (1988) study. Both studies consisted of a detailed ground survey examining the reflectance from individual needles from both non-attacked and current attacked lodgepole pine trees. The main difference between the studies, was that Runesson found that 12 the red edge for "green attack" lodgepole pine trees shifted to shorter wavelengths (a blue shift) instead of a shift to longer wavelengths (a red shift) as indicated in Ahern's study. Runesson (1991) also found that the peak of the red edge of healthy trees had higher values (average of 715.6 nm) than those of the successfully attacked trees (average of 710.5 nm). The changes were so subtle that he stated it would be premature, based on this study, to suggest a suitable classification strategy. 3.5 Environmental Factors and Seasonal Conditions It was once believed that all objects had a spectral signature, that is, a distinct spectral curve that could be used to describe a particular object's reflectance pattern. Once a spectral signature is known, any object matching that pattern can be classified. Unfortunately, distinct spectral signatures rarely occur in the natural world. For these reasons, spectral signature curves must not be assumed to be homogeneous throughout the world even if the same species is involved. Since the mountain pine beetle attacks the new host trees in mid summer, any spectral or physiological changes begin to occur from the attack date. Certain reflectance changes may occur following a bark beetle attack on lodgepole pine trees, but the reflectance effects are not uniform. For instance, reflectance changes stemming from a bark beetle attack occur more quickly in southerly areas of North America. In most of the southern United States, the pine trees will turn yellow or red during the fall following a mid-summer bark beetle attack. This was recently illustrated by Everitt, 1998. The trees that initially showed slight yellow - tinged foliage in August had a significant increase in discolouration by September with some leaves turning yellow-tan (Everitt, 1998). There are also numerous biological and environmental factors that can influence a tree's reflectance patterns. Silvennoinen et al. (1995) found that longitudinal distinguishing characteristics of Scots pine spectra were greater than latitudinal ones. There are numerous other factors that cause reflectance values to differ. Factors such as age of the tree, available nutrient status, amount of water available, sun angle, time of day and time of year are other factors that can affect reflectance values. Time of the year is an important factor affecting reflectance. As leaves (or needles) mature over the growing season, the spectral reflectance decreases in both the visual and the near infrared (Steiner and Gutermann, 1966). Physical factors such as the amount of foliage in the crown, density of the foliage in the crown, nutrient status, time since last rainfall, elevation, and other less important factors are believed to influence the reflectance. In any ecosystem, very complex intrasite interactions take place, and 13 comparisons of spectral reflectance variations from the same species across sites are tenuous (Murtha et al., 1997). Environmental conditions have an impact on the rate of the trees physiological change after it has been attacked. Very dry weather following a successful mountain pine beetle attack should accentuate the physiological changes in the tree, allowing for stress detection via remote sensing sooner. Attacked lodgepole pine were detected within approximately six weeks of infestation by mountain pine beetle, with a 91% accuracy at a scale of 1:3000 (Hobbs, 1983). Hobbs success of previsual "green attack" detection may, in part, be due to the noted dry fall season during her study. The optimum time to conduct an aerial survey for green attack lodgepole pine trees is late in the summer, after the beetles have attacked the trees (and sufficient time has elapsed for physiological changes to take place), and before the late fall when the low sun angles begin to impede surveys. Typically, for these reasons, August and early September are the best times to conduct aerial green attack surveys. Appendices 1 and 2 show the sun angles at 50°N and 55°N. With low sun angles, all forms of aerial imaging suffer from the long shadows obscuring the target. Also, in British Columbia, the weather during the months of October, November, and December is more unpredictable than earlier in the fall, further complicating late season surveys. 4. The Compact Airborne Imaging System (CASI) 4.1 An Overview of the System Used in this Study The commercial CASI (Compact Airborne Spectrographic Imager) instrument developed by ITRES Research Limited is a remote sensing instrument designed to measure and record spectral differences from imaged surfaces across the visible and near-infrared wavelength region, (400 to 950 nm) (Davidson and Latialle, 2000). Due to the systems high radiometric resolution, it is rarely limited by available light. It can acquire all 48 narrow bands if there is sufficient light over the target area. In situations of insufficient lighting, less handsets are used, or the width of a band is increased (ex: from +/-8 nm per band to +/- 15 nm per band). The system is capable of acquiring imagery even in winter months in Canada. Sensors with high radiometric resolution (such as CASI) have the ability to detect subtle changes in narrow bandwidths. The CASI acquires 16-bit imagery for increased radiometric resolution. The digital numbers of the imagery range from 0 to 16384, opposed to 8-bit data that ranges from 0 to 255. Narrow band imaging is achieved by restricting the sensor, through the use of narrow band pass filters, to only "see" in a certain portion of the wavelength. This type of sensor is called a hyperspectral sensor. 14 Narrow band pass filters are selected for the regions of the wavelength where the change (of the target) is most distinguishable. Only 36 bands were used during the data acquisition portion of this study due to low levels of illumination. The CASI is typically installed in rotary wing aircraft, although occasionally, fixed wing aircraft are used. The imagery can be accurately geo-referenced by using geographic positioning system(s) (GPS) and an inertial measurement unit (IMU) to record the aircraft's position and attitude. The CASI image data are geo-referenced using the aircraft motion data, a reference ground-based GPS station with the application of a digital elevation model (DEM) and appropriate geoid (earth shaped model) model (Davidson and Lataille, 2000). Table 2 shows the bands (and their corresponding bandwidths) that were acquired for this study. The width of the narrow bands ranges from 15.2 nm to 15.8 nm. Table 2. The center wavelength (and the corresponding bandwidth) from the thirty-six (36) bands used in this study. CASI Band Frequencies for Tyee Lake Project Channel Center Frequency Range CASI Detector Rows 1 434.7 nm +/- 7.6 nm (rows 281-288) 2 449.6 nm +/- 7.6 nm (rows 273-280) 3 464.5 nm +/- 7.6 nm (rows 265-272) 4 479.5 nm +/- 7.6 nm (rows 257-264) 5 494.4 nm +/- 7.7 nm (rows 249-256) 6 509.4 nm +/- 7.7 nm (rows 241-248) 7 524.4 nm +/- 7.7 nm (rows 233-240) 8 539.5 nm +/- 7.7 nm (rows 225-232) 9 554.5 nm +/- 7.7 nm (rows 217-224) 10 569.6 nm +/- 7.7 nm (rows 209-216) 11 584.7 nm +/- 7.7 nm (rows 201-208) 12 599.9 nm +/- 7.7 nm (rows 193-200) 13 615.0 nm +/- 7.7 nm (rows 185-192) 14 630.2 nm +/- 7.7 nm (rows 177-184) 15 645.4 nm +/- 7.8 nm (rows 169-176) 16 660.6 nm +/- 7.8 nm (rows 161-168) 17 675.9 nm +/- 7.8 nm (rows 153-160) 18 691.2 nm +/- 7.8 nm (rows 145-152) 19 706.4 nm +/- 7.8 nm (rows 137-144) 20 721.8 nm . +/-7.8nm (rows 129-136) 21 737.1 nm +/- 7.8 nm (rows 121-128) 22 752.4 nm +/- 7.8 nm (rows 113-120) 23 767.8 nm +/- 7.8 nm (rows 105-112) 24 783.2 nm +/- 7.8 nm (rows 97-104) 25 798.6 nm +/- 7.8 nm (rows 89-96) 26 814.0 nm +/- 7.8 nm (rows 81-88) 27 829.4 nm +/- 7.9 nm (rows 73-80) 28 844.0 nm +/- 7.9 nm (rows 65-72) 29 860.3 nm +/- 7.9 nm (rows 57-64) 15 30 875.8 nm +/- 7.9 nm (rows 49-56) 31 891.3 nm +/- 7.9 nm (rows 41-48) 32 906.8 nm +/- 7.9 nm (rows 33-40) 33 922.3 nm +/- 7.9 nm (rows 25-32) 34 937.9 nm +/- 7.9 nm (rows 17-24) 35 953.4 nm +/- 7.9 nm (rows 9-16) 36 969.0 nm +/- 7.9 nm (rows 1-8) The individual flight lines are "stitched" together based on sensor / target distance information provided by the aircraft's GPS and INU. It generates a digital multi-band image in the chosen spectral wavelengths along a swath representing an individual flight line (Davidson and Lataille, 2000). The sensor is slightly off-nadir facing. Bi-directional reflectances are more pronounced on off-nadir sensors such as the CASI used in this study since the sensor angle is oriented at a slight low oblique angle. Providing that there is sufficient lighting, almost any band frequencies and bandwidths can be chosen. The selection of bands is user definable, permitting identical bands (and bandwidths) to be used in future studies (if needed). For proper analysis, calibration would be required. This process may be difficult due to the numerous complexities associated with changing illumination, seasonal differences, and changing biological phenomena. 4.2 Image Scale The scale of the aerial photo is taken by a simple equation to determine the representative fraction (RF). Scale of the photographs defines the relationship between the size of the object being represented on the photo and the actual size of the object on the ground. Determining scale is slightly different when we make the move to digital softcopy. A digital image file does not have a scale based on a representative fraction, since it can be viewed and printed at many different sizes. A new method of determining an image's interpretability, called Ground Sample Distance (GSD) is used. It is comprised of collection, product and display GSD's (Comer et al, 1998). • Collection GSD = (array element size)(height above the ground level) / (focal length) • Product GSD = is the real world size of a pixel in a digital image product after all rectification and resampling procedures have occurred • Display GSD = (display pixel size) / (collection GSD) In general, GSD is simply the linear dimension of a single pixel's footprint on the ground (Comer et al, 1998). It does not imply that a 1 meter object can be resolved from a 1 meter GSD image. For visual identification, 4 to 9 pixels within the object are needed to resolve the object of interest (Comer etal, 1998). 16 This study examines airborne hyperspectral (60 cm) imagery for its ability to detect the subtle changes that occur in the currently infested "green attack" pine tree. Although the spatial resolution is not as high as aerial photographs, it is still much higher than the current satellites, making it the ideal sensor for surveys of this nature. The scale depends on the height of the aircraft above the target area. 4.3 Bi-directional Effects Bi-directional reflectance is the term used when, for a given wavelength, the brightness of plant leaves and canopies changes as the view angle changes relative to the angle of incident radiation (Suits, 1972; Colwell, 1974; Kalensky and Wilson, 1975, and Murtha et al., 1997). The bi-directional reflectance distribution function (BRDF) can significantly affect the radiometric characteristics of remotely sensed data, particularly when the data are collected off-nadir (Schill, 2000). Canopy reflectance increases with increasing solar zenith angles (Losee, 1951). Bi-directional reflectance effects are basically categorized into a backscatter zone (front-lit) and a forward-scatter zone (backlit). 5. Methods 5.1 Study Area The study area is located at approximately 52° °21'00" north, 122 ° 04"00" west in central British Columbia (Figure 5). It is situated just south of Tyee Lake, approximately 15 kilometers north of the town of Williams Lake, B.C. 17 Figure 6. The location of the study area is indicated by the red star. The study site is located in the Sub-Boreal Spruce (SBS) biogeoclimatic zone. Although the SBS biogeoclimatic zone name does not suggest a large pine component, lodgepole pine is common throughout the range. Lodgepole pine is a serai species in this zone especially widespread in the drier parts of the zone. The SBS biogeoclimatic zone is characterized in the following paragraph: "The SBS (Sub-Boreal Spruce) zone is the montane zone dominating the landscape of the central interior of British Columbia. It occupies the gently rolling terrain of the Nechako and Fraser Plateaus and the Fraser Basin and fingers into more mountainous areas along its western, northern, and eastern boundaries. The SBS is found over a wide latitudinal range, from 51° 30' to 59° N latitude. The zone generally occurs from the valley bottoms to 1100 - 1300 m elevation. The forests of the SBS are between the true montane forests of Douglas-fir to the south; the drier, colder pine - spruce forests to the southwest; boreal forests to the north; and subalpine forests at higher elevations." (Meidinger et al., 1991) 18 Table 3. A synopsis of subzones in the SBS (Sub-Boreal Spruce) zone (Meidinger et al., 1991). Subzone Code Dry Hot SBS SBSdh Dry Warm SBS SBSdw Dry Cool SBS SBSdk Moist Hot SBS SBSmh Moist Warm SBS SBSmw Moist Mild SBS SBSmm Moist Cool SBS SBSmk Moist Cold SBS SBSmc Wet Cool SBS SBSwk Very Wet Cool SBS SBSvk There are a total of 10 subzones in the SBS (as shown in Table 3). The study site is located in the SBS zone, dw subzone. The SBSdw subzone, is one of the three drier subzones within the SBS. Douglas-fir (Pseudotsuga menziesii var. glauca), lodgepole pine (Pinus contortd), hybrid spruce (Picea engelmannii x glauca), and subalpine fir (Abies lasiocarpa) are the most common conifer tree species. Douglas-fir is usually a long-lived serai species in the SBS, occurring abundantly on dry, warm, rich sites and as a consistent, although small, component of many mesic forests, especially in the southeastern part of the zone (Meidinger et al., 1991). • Douglas-f ir • Lodgepole pine • Hybrid spruce • Subalp ine fir g Hardwoods (mostly trembling aspen) Figure 7. The species composition and their approximate percentage within the Tyee Lake study site. 19 The largest trees in the study area are the Douglas-fir (in both crown width and height). Some of the lodgepole pine crowns are slightly obscured or are shaded by these large Douglas-fir trees. Forest stands that have more than a single canopy level (such as the sites used in this study) further complicate the study. The study area was chosen because there was an active mountain pine beetle population and the site was indicative of much of the Cariboo Forest Region. The site was also chosen due to its proximity to the Williams Lake airport (to minimize helicopter costs), the relatively flat terrain (to minimize orographic shadows), and the road access (for ground truthing). ITRES Research Ltd. acquired approximately 12 km 2 of CASI imagery on September 30 t h, 1999 (Figure 11). Within the imaged area, two sample sites were selected based on the initial ground truth. The location and boundary of these sample sites: site #1, and site #2 (Figure 8). Figure 8. The red boundary shows the area where the CASI imagery was flown. The smaller pink boundaries show the location of three sample sites (where ground data were collected). The map was produced using digital forest cover information (courtesy of the Ministry of Forests). Sections of the CASI imagery were extracted from the mosaic to perform a detailed analysis. Figures 9 and 10 show the extracted sample sites [images composed from bands 5 (494.4 nm), 10 (569.6 nm), and 16 (660.6 nm), as blue, green, and red]. These bands were chosen since the combination of these bands best separates the lodgepole pine from the Douglas-fir trees. These sample sites were chosen 20 for several reasons: 1) they are easily accessible, 2) they have active beetle damage, 3) they have a large pine component, 4) and the sample site is on the edge of the flight line, so that both directions of image acquisition are shown. Within these sites, the trees were numbered and the health status of each pine tree was recorded. The pine trees were recorded as non-attacked, green attacked, or red attacked. Also, the physical qualities were recorded, such as, tree height, diameter at breast height (DBH), and age. Trees were not included if there was any presence of disease or mistletoe. Figure 9. Site #1. Sample trees were selected from both sides of the directional flight swath. The seam line in the middle of the image divides the top (forward-scatter) and the bottom (backscatter) zones. Created from bands 27, 19, 8, as R,G,B. The red boundaries represent the green attack trees and the green boundaries represent the non-attack trees. 21 Figure 10. Site #2. Sample trees were selected from both sides of the directional flight swath. The seam line in the middle of the image divides the top (forward-scatter) and the bottom (backscatter) zones. Created from bands 27, 19, 8, as R,G,B. The red boundaries represent the green attack trees and the green boundaries represent the non-attack trees. 5.2 Airborne Data Collection A Bell Longranger helicopter was used to acquire approximately 12 km 2 of CASI imagery. Several flight lines (travelling east / west) were needed to sufficiently cover the area using a vertically mounted CASI. Thirty-six (36) discrete spectral bands were collected. The band frequencies and the bandwidths are shown in Table 2. The wavelengths range from 427 nm to 977 nm. The imagery was flown at 400 m above the ground. At this altitude, the imagery has a ground sample distance (GSD) (also referred to as pixel resolution) of 60 cm. The images were taken between 9:30am and 10:00am (local time) on September 29, 1999. The weather was cloud free, but windy. During this time of year (end of September), the sun angle is quite low (see Appendix 1 and 2). The high radiometric resolution of the CASI instrument allows imagery to be acquired during low light levels (sun angle approximately 26°) associated with the late fall at 52° latitude. Preflight lighting tests of the system indicated that 36 bands could be acquired. The ability to record 36 bands at this time of the year is really a testament to show that CASI has excellent radiometric resolution. 22 The black "slivers" in the imagery are caused by very turbulence conditions during the imagery acquisition. According to ITRES, these sort of missing regions (slivers) are experienced during windy conditions. The (east / west) flight lines have been processed and geo-referenced (see Figure 11). No ground references or "rubber sheeting" was involved in generating the mosaic (Davidson and Lataille, 2000). **3€*0<> ^Sa** •*§§*> **J>0W *»25<» s sgo 0 0 e*35«> e*4 01** *«45<J« M J 0 O I > Figure 11. The merged CASI imagery. The (east / west) flight lines have been processed and geo-referenced. The red box indicates the road junction near to the selected study sites. 5.3 Ground Truth The ground truthing began on November 3rd, 1999. The two individual study sites (Site #1 and Site #2) (Figures 9 and 10) were surveyed. In order to correctly identify each stem / crown in the study sites, large scale positive prints of the CASI images were used as field "maps." Each crown was numbered and circled using a thin, waterproof pen. Each corresponding tree stem was also marked on the ground, by spray painting the tree number onto the trunk of the tree. Marking both the crown on the imagery and the trunk of the tree would be important i f future visits or further analysis had be conducted. In site #1 the locations of 61 trees were recorded; site #2 had 49 trees. The health of every marked tree was recorded. The red attack trees were obvious due to their red needles. To distinguish between green attack pine and healthy pine trees a more detailed examination was required. The trunk of green attack trees was inspected for the presence of pitch tubes (Figure 12). 23 Pitch tubes are a small amount of creamy looking resin about the size of a marble that have been exuded from the tree where a bark beetle has bored through the bark. The base of the tree was also inspected to see if frass (sawdust) was present. Frass at the base of the tree is caused by the beetles boring through the bark. If both pitch tubes and sawdust were present, then the bark was partially removed to determine if the beetle attack was successful. To ensure a more higher conformation that the tree is a successful green attack, a small portion of bark was removed to look for live beetles and beetle feeding galleries. Positive conformation that the trees were indeed killed by the beetles would not be known until the foliage turned red the following summer. Figure 12. An example pitch-tubes on the trunk of a green attacked lodgepole pine. In April, 2000 (the following spring after the image acquisition) physical attributes of the sample trees were acquired. The following physical attributes were acquired: tree height, age, and diameter at breast height. Unfortunately, the green attacked trees had been removed by winter salvage loggers Therefore, it was only possible to record the age and diameter from the stumps for the green attack population. The "true" breast height diameter of these trees was extrapolated from the buttress diameter. The tops of the stumps were measured, then the results were slightly reduced (by 1/6*, not including butt portions) to more accurately estimate the diameter at breast height (DBH). The age, diameter, and height data were acquired in order categorize sample trees into classes for analysis. 24 5.4 Analysis of Spectral Data In order to properly identify the ground truthed trees, several vector files were created and geo-referenced to the sample study sites. The first vector file contained a line around each of the tree crowns that were referenced on the ground. The second vector file was a cross-hair placed on the center of each tree crown. The boundary around the tree crown and the cross-hair placed on the central pixel (top of the crown) helped to divide the pixels in the crown so that they could be identified and analyzed on an individual pixel by pixel basis throughout all the bands. Separate vector files were created for the boundaries of the selected green attack and non attack trees for each site. Several image manipulation methods were attempted to differentiate between the green attack and the non attack on the CASI imagery. The analysis of the (16 bit) CASI data was done using the PCI remote sensing software. Initially, three different spectral reflectance curves from each tree were created. The first method of obtaining the derived spectral reflectance curve was by plotting a single digital number (DN) value from each band. The second method of creating a derived spectral reflectance curve was more robust. Nine digital numbers (DN) were recorded in the same order from every crown in all 36 bands. The nine DN's were averaged to better represent the tree crown as a whole. The samples were taken around the top of the crown and were indicated by the cross-hair vector point. The nine D N samples were taken in a certain order as shown in Figure 13 (center pixel is the #5 pixel). In cases where the tree crown was smaller than 9 pixels in area, nine pixels were still recorded but were not included when the crown average was tabulated. The recorded D N values that are outside the tree crown or that appear to be skewed by reflectance from understory vegetation or soil were not included in the averaging. The third method of obtaining a derived spectral reflectance curve was by recording a single D N value at the center of the tree crown (brightest portion) after the imagery had been resampled to 1.2 m imagery. 25 Figure 13. Both images show closeup examples of two large trees. The image on the left is the original 60 cm imagery and the image on the right is from the resampled 1.2 m imagery. A vector line indicates the boundary of the larger tree and a cross-hair indicates the center of the crown. The numbered grid shows the order in which the digital numbers were recorded into the spreadsheet program. The #5 pixel (from the 60 cm imagery) is located at the center of the tree crown. Only the reflectance value from one (central pixel) is recorded from the resampled 1.2 m imagery. Each tree's reflectance curve was created and used to display reflectance similarities (and differences) between the populations. Close-ups of certain sections of the graphs can be further analyzed. Graphs of the important red edge area can be analyzed since each tree is colour coded to compare / contrast the results for each population. The imagery was resampled to average the spatial / spectral information for a separate analysis. The images were resampled to a 1.2 m pixel size. This meant that four (60 cm) pixels were merged into a single (1.2 m) pixel. This image analysis technique is a procedure used when complicating spatial / spectral patterns are present. Averaged values for the tree crowns do not have to be derived. Since the spatial extent of a single (1.2 m) pixel covers a large portion of the tree crown only the single "brightness" value of the tree crown is recorded. The DN values of the 16-bit imagery were analyzed using PCI software. Vector files (of the tree crown boundaries) were also created on the resampled imagery. The original and the resampled imagery are shown in Figures 14 and 15. The green attack tree crowns are outlined in red and the non-attack boundaries are outlined in green. 26 Fib Edit View Tools Classify 1' If si £1 1 Q 12 3 4 / iUi,r +7 PCI lmageWoiksVG.3.0: (VDO)... File Edit View Tools Classify Help 1' If ra ra 12 3 4 / ili,li,r + J 122P141L 532:1 266:2 116:3 [xi 1305P277L 12023:1 655:2 160:3 |R:1 G:2B:3 Figure 14. The imagery from Site 1. The resampled imagery (1.2 m) is in the smaller window. 27 File Edit View Tools Classify Help PCI ImageWorks VG.3.0: (VDO)... tf l ' 11 Pol 1 2 34 / ilJl,r *J File Edit View Jools Classify Help tf I f si ei |q| n 1 2 34 / 7" iliJi,r +7 Figure 15. The imagery from Site 2. The resampled imagery (1.2 m) is in the smaller window. Fol lowing an analysis o f the spectral reflectance curves, suitable supervised classifications and spectral enhancement transformations were performed. The bands that indicated the strongest difference between the populations (bands 8, 19, and 27 as indicated by the discriminant analysis and the t-test results) were chosen as the input bands for the classification. Portions o f the crown that most closely represented the populations average were selected as the training areas. 28 Band ratioing is powerful analysis technique that determines the health status of vegetation. It is especially useful for minimizing varying illumination effects. Several band ratios have been shown to be an effective way to measure vegetation stress. Presumably, more information is provided by two or more bands than by any one spectral band (Murtha et al., 1997). Traditional band ratios, such as N D V I , are not really applicable since they were created for broad band imagery, not hyperspectral imagery. The following band ratios were conducted to the hyperspectral imagery (Table 4). Table 4. The band ratios used on the extracted tree crown imagery. 1) toe of red edge compared to the shoulder portion of green rise (706.4-539.5) / (706.4+539.5) 2) slope of red rise (645./ 660.6) 3) lower red edge slope (706.4-721.8) / (706.4+721.8) 4) red edge slope (706.4-737.1) / (706.4+737.1) 5) green rise compared to red well (554.5-675.9) / (554.5+675.9) 6) toe of red edge compared to red rise (691.2-645.4) / (691.2+645.4) The band ratios were performed on the 60 cm data as well as the resampled 1.2 m data. The results of the band ratios were displayed in table form for analysis. Due to the number of bands (36 discrete hyperspectral bands), many more ratios are possible. The basis for choosing the band ratios used in this study were from the results of the discriminant analysis and from results indicated in previous literature. 5.5 Statistical Methods The following statistical methods were applied to determine if the green attack could be separated from the non-attacked: • the data from each band were recorded and the pixel arrangements were averaged (derived the mean) • the standard deviations of each band were recorded • prior to conducting a t-test, the data were examined to see i f it is normally distributed (Shiparo-Wilks test) • the individual band data that were normally distributed, were examined using a t-test • the individual band data that were not normally distributed, were examined using a Mann-Whitney test for non-parametric data • a discriminant analysis was performed to determine what (if any) bands are significant in distinguishing the green attack from the non-attack. 29 5.6 Categorizing by Physical Attributes and Location on the Imagery The initial analysis of the data raised questions about the physical attributes of the trees and the trees location on the imagery (bi-directional reflectance effects). The trees were categorized and graphed based on the trees location (forward-scatter or backscatter zone on the imagery). These classifications were not part of the study design, but have been included to analyze other influencing factors. 6. Results 6.1 Reflectance charts The derived reflectance charts are a very descriptive way to show the differences (or similarities) of the trees. In a study of the change in spectral reflectance of lodgepole pine trees following a mountain pine beetle, it is important to show each tree's derived spectral reflectance curve rather than just a single averaged derived reflectance curve of green attack and non-attack. The derived spectral reflectance curve for the averaged 60 cm data of each green attack tree is graphed in Figure 16. The derived spectral reflectance curve for the averaged 60 cm data of each non-attack trees in Figure 17. The averages of all the 60 cm derived reflectance curve samples are shown in Figure 18. The derived spectral reflectance curves were also created for the resampled 1.2 m imagery: green attack imagery (Figure 19), and non-attack imagery (Figure 20). The averages of all the 1.2 m derived reflectance curve samples are shown in Figure 21. 30 co co CO CO CO c o CO co CO CO CO co co co CM CM CM CM CM CM CM I CO r o CO CO 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 CM h~ CO 05 00 CO CO T cn CO r - CD o cn OO O IT) CO CM O) CD r~- OO O) CM CO CO t o i n cn cn cn CD T— CM •<r T •<r i n i n CU CD cu 0) <u cu cu cu CU cu cu cu cu 0 0 0 0 0 0 0 0 0 0 0 0 cu 0 cu cu cu 0 0 cu cu cu cu cu 0 0 0 0 0 a> CU cu CD CD CD Cl) CD H (- 1- r - H (- 1— H 1- H t - H h- H 1- h- h- h- t- H h- H t— H r— 5 O O 3 O O 3 m o CO CD (NQ) eouejoaiiay |ejjoads The derived spectral reflectance curves were also created for the green attack and the non attack data for each site, site 1 imagery (Figure 22), and site 2 imagery (Figure 23). The derived spectral reflectance curves were also created for the resampled 1.2 m imagery, site 1 imagery (Figure 24), and site 2 imagery (Figure 25). The derived spectral reflectance curves were created for each site to determine if differences could be seen between the sites. Significant differences between the sites were not expected since the sites are close together (< 1 km). The graphs indicate that there is a high variability within the samples of each site rather than between the sites. 37 CO £ o (0 •o £ o o C D 0. -sc O ro +•» < c o T 3 C ro c o 0) 1-O £ w *-< c 0 > ro 0 0 Jg cn o w o C3 -o o CD & 3 o o o ca a o CD u oo ro o CD a o o o o o o o o o o o o o o o o o o o o o o o o LO o u o o UO o u o a u o o UO o CD CD u o u o r o r o C N CM o o UO (NQ) eouBjoaijey |ejjoeds — CD > CD -a CD eC H «s -3 WD The results from the derived spectral reflectance graphs and the forward-scatter / backscatter derived spectral reflectance graphs indicate that there is significant variability within each population. There is spectral reflectance variability within the green and non-attack 60 cm reflectance curves, the green and non-attack 1.2 m reflectance curves, and from each site. Generally, the trend of the derived spectral reflectance curves is similar. Although, the spectral variability complicates interpretation. It is believed that the spectral variability is mostly caused by the low sun angles at the time of the data acquisition (9:30am, September 30) and the orientation of the sensor. These confounding problems are analyzed in the discussion section. The red edge area has been indicated as the region of initial reflectance change in the Ahern (1988), Runesson (1991), and (Davidson and Latialle, 2000). Figure 26 shows the red edge area (bands 17-23) for the green attack and the non-attack lodgepole pine trees from the 60 cm data. Figure 27 shows the red edge area (bands 17-23) for the green attack and the non-attack lodgepole pine trees from the 1.2 m resampled imagery. 42 The Red Edge Slope from the (60cm) Imagery 6000 , 675.9 691.2 706.4 721.8 737.1 752.4 767.8 Bands 17-23 (center bandwidths 675.9nm to 767.8nm) i i Figure 26. A graph showing the red edge portion of the wavelength for green attack and the non-attack lodgepole pine trees from the 60 cm data. 43 The Red Edge Slope from the (1.2m) Imagery 7000 6000 5000 z Q g 4000 c ro +J o 0) cu CU 2 3000 o cu CL C7) 2000 1000 .Green Attack Non Attack 675.9 691.2 706.4 721.8 737.1 752.4 767.8 Bands 17-23 (center bandwidth 675.9nm to 767.8nm) Figure 27. A graph showing the red edge portion of the wavelength for green attack and the non-attack lodgepole pine trees from the 1.2 m data. 6.2 Classifications Several supervised classifications were performed on the imagery. The results are shown in Figures 28 and 29. The blue colour represents the green attack and the pink colour represents the non attack. The green attack crowns are outlined in red and the non-attack crowns are outlined in green. 44 Figure 28. A Maximum Likelihood Supervised Classification on Site 1 (60 cm data). The green attack crowns are traced in red and the non-attack crowns are traced in green. The pink colour represents the green attack and the blue colour displays the non-attack classification. i. 6.o + • Figure 29. A Minimum Likelihood Supervised Classification on Site 1 (60 cm data). The green attack crowns are traced in red and the non-attack crowns are traced in green. The pink colour represents the green attack and the blue colour displays the non-attack classification. 45 6.3 Band Ratios Table 6 displays the results from band ratios between the green attack and the non-attack populations (band ratios are displayed in order of significant finding). The most significant band ratio is the toe of red edge compared to the shoulder of green rise (706.4 - 539.5) / (706.4 + 539.5). These bands displayed the most significant difference between the green and non attack populations, originally indicated by the discriminant analysis. Although the difference of 0.265 (green attack) and 0.251 (non-attack) is slight, the results could be amplified to increase the difference. The second strongest difference shown by the band ratios was between the toe of the red edge compared to the red rise. Although these bands were not indicated by the discriminant analysis or by the t-tests there is a slight difference between the populations. The other band ratios indicated slight differences between the populations that are probably insignificant. The lower red edge slope ratio (721.8 nm and 706.4 nm) (as indicated by Davidson and Latialle, 2000) is the only band ratio that does not display any difference. Table 5. Band ratios for the green attack and non-attack trees. Band Ratios (averaged for each population) Bands Used (in nm)* Green Attack Non-Attack Toe of red edge compared to shoulder of green rise (706.4 - 539.5) / (706.4 + 539.5) 0.265 0.251 Toe of red edge compared to red rise (691.2-645.4)/ (691.2 + 645.4) 0.038 0.026 Slope of the red rise 645.4 / 660.6 1.124 1.134 Red edge slope (706.4-737.1)/ (706.4 + 737.1) -0.381 -0.367 Green rise compared to red well (554.5 - 675.9) / (554.5 + 675.9) 0.290 0.288 Lower red edge slope (706.4-721.8)/ (706.4 + 721.8) 0.39 0.39 * Center bandwidth frequency, range +/- 7 nm. 46 6.4 Statistical Significance Prior to conducting a t-test, the bands were tested to determine if the data were normally distributed. The bands that failed the normality test (Shapiro-Wilks) were compared using the Mann Whitney (nonparametric) t-test. The bands that passed the normality test (Shapiro-Wilks) were compared using a 2-sample t-test. The results of the t-tests are shown in Table 6 and 7. Table 6. The mean, standard deviation, and t-test (60 cm imagery) results from all 36 bands for the green attack and non-attack populations. Non Attack Green Attack Band# Wavelength Mean Standard Deviation Mean Standard Deviation T-Test Mann -Whitney Test Band # 1 434.7nm +/-7.6 nm 327.38 89.40 294.36 69.27 * 0.1556 Band # 2 449.6nm +/-7.6 nm 426.82 116.20 377.04 83.99 0.092 ** Band # 3 464.5nm +/-7.6 nm 476.40 124.40 423.45 97.11 0.103 ** Band # 4 479.5nm +/-7.6 nm 503.51 135.88 450.68 108.46 0.139 ** Band # 5 494.4nm +/-7.6 nm 498.17 140.56 444.17 110.86 0.141 ** Band # 6 509.4nm +/-7.7 nm 556.24 162.22 481.98 128.47 0.082 ** Band # 7 524.4nm +/-7.7 nm 755.24 226.38 644.23 172.08 0.059 ** Band # 8 539.5nm +/-7.7 nm 951.41 300.43 806.21 216.02 0.058 ** Band # 9 554.5nm +/-7.7 nm 1034.86 342.12 882.78 261.12 * 0.0949 Band#10 569.6nm +/-7.7 nm 933.31 323.01 799.84 245.43 * 0.1164 Band #11 584.7nm +/-7.7 nm 828.28 295.07 715.64 227.06 * 0.1707 Band#12 599.9nm +1-7.7 nm 799.08 290.93 689.78 223.68 * 0.1902 Band#13 615.0nm +/-7.7 nm 738.64 272.19 642.35 213.45 * 0.2187 Band#14 630.2nm +/-7.7 nm 700.34 262.88 610.35 207.18 * 0.2585 Band#15 645.4nm +/-7.8 nm 677.04 256.81 587.53 204.22 * 0.2757 Band #16 660.6nm +/-7.8 nm 597.50 230.14 523.92 186.19 * 0.4065 Band* 17 675.9nm +/-7.8 nm 561.37 217.36 493.41 179.01 * 0.5029 Band* 18 691.2nm +/-7.8 nm 720.64 273.60 634.66 227.11 * 0.3843 Band* 19 706.4nm +/-7.8 nm 1615.24 554.39 1429.13 474.54 * 0.2846 Band # 20 721.8nm +/-7.8 nm 2322.39 702.16 2060.80 621.38 * 0.1527 Band # 21 737.1nm +/-7.8 nm 3544.73 975.84 3145.49 889.31 * 0.0949 Band # 22 752.4nm +/-7.8 nm 3821.02 1024.19 3362.56 939.37 * 0.0643 Band # 23 767.8nm +/-7.8 nm 2950.90 794.42 2618.34 732.81 * 0.0949 Band # 24 783.2nm +/-7.8 nm 3894.28 1070.86 3427.33 973.39 0.0588 Band # 25 798.6nm +/-7.8 nm 3653.63 1022.73 3214.33 923.95 * 0.0703 Band # 26 814.0nm +/-7.8 nm 3164.19 911.78 2775.46 814.75 * 0.0615 Band # 27 829.4nm +/-7.9 nm 3058.89 909.14 2678.87 802.11 * 0.0466 Band # 28 844.0nm +/-7.9 nm 3256.76 995.27 2840.01 866.03 * 0.0561 Band # 29 860.3nm +1-7.9 nm 3065.79 972.04 2664.07 828.59 0.126 ** Band # 30 875.8nm +/-7.9 nm 2986.68 984.52 2588.32 826.80 0.131 ** Band #31 891.3nm +/-7.9 nm 2618.99 889.43 2256.59 731.71 0.126 ** Band # 32 906.8nm +1-7.9 nm 2029.89 711.22 1758.80 581.04 0.15 ** Band # 33 922.3nm +/-7.9 nm 1914.87 692.35 1646.01 545.71 0.137 ** Band # 34 937.9nm +/-7.9 nm 877.69 320.36 752.58 248.81 0.133 ** Band # 35 953.4nm +/-7.9 nm 1036.03 380.96 889.39 292.68 0.137 ** Band # 36 969.0nm +/-7.9 nm 1495.58 576.39 1280.68 415.07 0.14 ** * Where d ata are non normal, Mann Whitney test conducted ** Where data are normal, t-test conducted .. . Areas highlighted indicate difference between the green and non-attack population. 47 Table 7. The mean, standard deviation, and t-test (1.2 m imagery) results from all 36 bands for the green attack and non-attack populations. Non Attack Green Attack Band# Wavelength Mean Standard Deviation Mean Standard Deviation T-Test Mann -Whitney Test Band # 1 434.7nm +/-7.6 nm 338.88 108.37 295.46 80.15 0.117 ** Band # 2 449.6nm +/-7.6 nm 442.96 130.04 388.13 106.87 0.113 ** Band # 3 464.5nm +/-7.6 nm 500.64 145.22 437.58 120.19 0.104 ** Band # 4 479.5nm +/-7.6 nm 514.40 141.85 480.71 119.84 0.373 ** Band # 5 494.4nm +/-7.6 nm 504.68 148.33 485.83 118.16 0.624 ** Band # 6 509.4nm +/-7.7 nm 566.16 172.40 532.54 135.46 0.451 ** Band # 7 524.4nm +/-7.7 nm 756.64 263.65 702.63 200.44 0.423 ** Band # 8 539.5nm +/-7.7 nm 954.00 350.60 891.50 268.43 0.486 ** Band # 9 554.5nm +/-7.7 nm 1037.16 398.98 973.33 305.23 0.532 ** Band#10 569.6nm +/-7.7 nm 936.40 376.01 883.38 286.89 0.581 ** Band #11 584.7nm +/-7.7 nm 826.44 342.52 789.08 261.25 0.669 ** Band#12 599.9nm +/-7.7 nm 798.72 335.94 761.29 253.89 0.661 ** Band#13 615.Onm +/-7.7 nm 736.64 308.82 706.00 249.54 0.7 ** Band#14 630.2nm +/-7.7 nm 697.48 300.96 670.17 231.05 0.723 ** Band#15 645.4nm +/-7.8 nm 672.52 293.20 643.38 225.18 0.698 ** Band#16 660.6nm +/-7.8 nm 592.88 261.13 571.33 202.18 * 0.803 Band#17 675.9nm +/-7.8 nm 570.00 265.24 538.54 195.34 0.638 ** Band#18 691.2nm +/-7.8 nm 727.68 324.58 695.79 250.47 * 0.881 Band#19 706.4nm +/-7.8 nm 1608.28 650.24 1593.79 559.49 0.934 ** Band # 20 721.8nm +/-7.8 nm 2355.60 808.97 2301.33 768.30 0.811 ** Band #21 737.1nm +/-7.8 nm 3604.72 1133.45 3506.88 1127.44 0.763 ** Band # 22 752.4nm +/-7.8 nm 3898.72 1193.04 3728.13 1196.57 0.62 ** Band # 23 767.8nm +/-7.8 nm 3019.48 931.00 2888.13 926.03 0.623 ** Band # 24 783.2nm +/-7.8 nm 3997.12 1249.39 3757.71 1221.86 0.501 ** Band # 25 798.6nm +/-7.8 nm 3756.84 1192.32 3504.13 1147.62 0.454 ** Band # 26 814.0nm +/-7.8 nm 3259.84 1052.48 3007.96 1002.37 0.395 ** Band # 27 829.4nm +/-7.9 nm 3152.80 1041.59 2888.88 973.87 0.364 ** Band # 28 844.0nm +/-7.9 nm 3364.56 1141.34 3045.33 1042.98 0.312 ** Band # 29 860.3nm +/-7.9 nm 3164.44 1104.66 2844.54 988.55 0.291 ** Band # 30 875.8nm +/-7.9 nm 3076.20 1112.77 2746.83 980.54 0.277 ** Band # 31 891.3nm +/-7.9 nm 2692.40 998.08 2391.67 865.54 0.265 ** Band # 32 906.8nm +/-7.9 nm 2089.24 796.16 1859.63 681.95 0.283 ** Band # 33 922.3nm +/-7.9 nm 1635.75 615.57 1488.04 506.54 0.137 ** Band # 34 937.9nm +/-7.9 nm 879.75 327.46 756.96 249.81 0.133 ** Band # 35 953.4nm +/-7.9 nm 1030.45 381.71 894.47 297.00 0.137 ** Band # 36 969.0nm +/-7.9 nm 1487.40 574.76 1283.92 416.49 0.14 ** * Where d ata are non-normal, Mann Whitney test conducted ** Where data are normal, t-test conducted Areas highlighted indicate difference between the green and non-attack population. A discriminant analysis was performed on the 60 cm and the resampled 1.2 m imagery. The discriminant analysis was conducted in a forward stepwise method with an alpha value for inclusion / exclusion, of 0.150. O n the 60 cm imagery, the discriminant analysis initially indicated that band 8 (539.5 +/-7.7 nm) was the strongest indicator for showing a difference between the green attack and the 48 non attack populations. After band 8 (539.5 +1-1.1 nm) was identified, it was automatically removed as a predictor and then the remaining bands were analyzed. Band 19 (706.4 +1-1.% nm) was identified next. It is the second strongest indicator for showing the difference between the green attack and the non-attack populations. Bands 8 and 19 were the only bands that were indicated in the discriminant analysis. The complete discriminant analysis for the 60 cm imagery is shown in Appendix 3. The classification matrix (Table 8) indicated (by using bands 8 and 19) 83 percent of the green attack trees and 68 percent of the non-attack trees in both sites 1 and 2 are correctly classified. This classification process is biased since the population of each sample is known. A more accurate classification method is the Jackknifed classification matrix since it is unbiased. It classifies each sample into a category without knowing which population each sample belongs to, then checks the results. The Jackknife classification matrix results (Table 8) indicate that 79 percent of the green attack trees and 68 percent of the non-attack tress in both sites 1 and 2 are correctly classified. The total of correctly classified sample trees was 73 percent. The canonical discriminant functions — standardized by within variances are shown in (Table 10) and the canonical scores of group means are shown in (Table 11). A graph showing the central frequencies from the results of the canonical scores of the populations (Figure 30) indicate that the populations are quite close together and overlap considerably. Table 8. C lass i f i ca t ion matrix ( c a s e s in row ca tegor ies c lass i f ied into co lumns) G r e e n N o n %cor rec t G r e e n 20 4 83 N o n 8 17 68 Tota l 28 21 76 Table 9. Jackknifed classification matrix as a result of the discriminant analysis on the 60 cm data. Green Non %correct Green 19 5 79 Non 8 17 68 Total 27 22 73 49 Table 10. Canonical discriminant functions ~ standardized by within variances. Please note that the discriminate analysis only indicated that bands 8 and 19 are different between the populations using the 60 cm data. 1 Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 Band 7 Band 8 5.596 Band 9 Band 10 Band 11 Band 12 Band 13 Band 14 Band 15 Band 16 Band 17 Band 18 Band 19 -5.217 Band 20 Band 21 Band 22 Band 23 Band 24 Band 25 Band 26 Band 27 Band 28 Band 29 Band 30 Band 31 Band 32 Band 33 Band 34 Band 35 Band 36 Table 11. Canonical scores of group means, based from bands 8 and 19 using the 60 cm data. 1 Green -0.620 Non 0.596 50 o O ATTACKSTATUS • green • non Figure 30. The canonical scores results from the green attack and non-attack populations of 60 cm data. Please note that the central frequencies of the populations are close together and overlap considerably. A discriminant analysis was also performed on the 1.2 m imagery. The discriminant analysis initially indicated that band 3 (464.5 +1-1.6 nm) was the strongest indicator for showing a difference between the green attack and the non-attack populations. After band 3 (464.5 +1-1.6 nm) was identified, it was automatically removed as a predictor and then the remaining bands were analyzed. Band 5 (494.4 +1-1.1 nm) was identified next. It is the second strongest indicator for showing the difference between the green attack and the non-attack populations. The last band to be identified was band 6 (509.4 +1-1.1 nm). Bands 3, 5, and 6 were the only bands that were indicated in the discriminant analysis of the resampled 1.2 m imagery. The complete discriminant analysis for the 1.2 m imagery is shown in Appendix 4. The classification matrix (Table 12) indicated (by using bands 3, 5 and 6) 67 percent of the green attack trees and 64 percent of the non-attack trees in both sites 1 and 2 are correctly classified. The Jackknife classification matrix results (Table 13) indicate that 63 percent of the green attack trees and 64 percent of the non-attack tress in both sites 1 and 2 are correctly classified. The total of correctly classified sample trees was 63 percent. The canonical discriminant functions ~ standardized by within variances are shown in (Table 14) and the canonical scores of group means are shown in (Table 15). Discriminant analysis can be used not only to test multivariate differences among groups, but also to explore: 1) which variables are most useful for discriminating among groups, 2) if one subset of variables performs equally well as another, 3) which groups are most alike and most different. Table 12. Classification matrix (cases in row categories classified into columns) Green Non %correct Green 16 8 67 Non 9 16 64 Total 25 24 65 51 Table 13. Jackknifed classification matrix as a result of the discriminant analysis on the 1.2 m data. Green Non %correct Green 15 9 63 Non 9 16 64 Total 24 25 63 Table 14. Canonical discriminant functions ~ standardized by within variances. Please note that the discriminate analysis indicated that bands 3, 5 and 6 are different between the populations using the 1.2 m data. 1 Band 1 Band 2 Band 3 1.878 Band 4 Band 5 -5.614 Band 6 4.103 Band 7 Band 8 Band 9 Band 10 Band 11 Band 12 Band 13 Band 14 Band 15 Band 16 Band 17 Band 18 Band 19 Band 20 Band 21 Band 22 Band 23 Band 24 Band 25 Band 26 Band 27 Band 28 Band 29 Band 30 Band 31 Band 32 Table 15. Canonical scores of group means, based from bands 3, 5 and 6 using the 1.2 m data. 1 Green -0.504 Non 0.483 52 Figure 31 (60 cm data) shows the average spectral reflectance and the (+/-) standard deviation of each average for the green attack and non-attack trees. Figure 32 (1.2 m data) shows the average spectral reflectance and the (+/-) standard deviation of each average for the green attack and non-attack trees. The average / standard deviation graphs indicate that there is significant overlap between the populations. Average Green Attack (60cm) and Average Non Attack (60cm) Spectral Reflectance Curves with the Standard Deviat ions r O ( D C 3 3 C M i n o O T - ^ . | ^ 0 ( 0 < D O > C « J < D O > C M U l • » ' « a - ' * r m u > w c o < D C D h ~ i ^ r ~ r ~ c o c o c o o > a > Wavelengths (nm) Figure 31. The average spectral reflectance curves and the standard deviations of the green attack and non-attack populations from 60 cm data. 53 Average Green Attack (1.2m) and Average Non Attack (1.2m) Spectral Reflectance Curves with Standard Deviations 6000 , z Q £ 3000 to a 2000 -Average Non Attack I -Average Green Attack -Non Attack (+/-) Standard Deviation 1000 Green Attack (+/-) || Standard Deviation Wavelengths (nm) Figure 32. The average spectral reflectance curves and the standard deviations of the green attack and non-attack populations from the 1.2 m data. A graph displaying the position (forward-scatter or backscatter zone) of the each sample tree was made (Figure 33). Although, the four brightest trees are located in the backscatter zone, the graph does not show a clear separation between the populations on the basis of their location on the imagery. 54 Fore and Back Scatter Zones Compared from both Green Attack and Non Attack Trees 7000 5000 u 4000 a •5 3000 i3 Forward-scatter Zone Backscatter Zone 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 Wavelengths (nm) Figure 33. The spectral reflectance curves from the (60 cm imagery) colour coded based on the trees location on the imagery (fore or back scatter zones). 11. Discussion The derived reflectance graphs are useful to see the spectral variability within each population, green attack and non-attack. The derived spectral reflectance graphs (Figures 16 and 17, for the 60 cm data) were created to visually assess the data to determine if any spectral bands (especially the red rise and red edge) may show a difference between the green and non-attack trees. The averaged population derived spectral reflectance curves seem to be very similar, yet the derived spectral reflectance curves within the green attack population and the non-attack population vary considerably. The spectral variability within each population seems to be greater than any other factors between the populations. The reflectance graphs from the resampled (1.2 m) imagery greatly resemble the averaged (60 cm) reflectance graphs. The resampled derived spectral reflectance graphs (Figures 19 and 20) also failed to visually indicate any bands that display a difference between the 55 populations. The derived spectral reflectance curves from each individual tree within the populations vary considerably, although the derived spectral reflectance curves of each population as a whole seem to be very similar. Previous studies have indicated that the most pronounced area of reflectance change occurs at the red edge portion of the wavelength. Therefore, a detailed graph of the red edge area was created (60 cm imagery is seen on Figure 26, and the 1.2 m imagery is seen on Figure 27). The graphs (Figures 26 and 27) display varying derived reflectance patterns of both the green attack and the non-attack populations. The green attack and non-attack populations are not separated along the red edge as predicted from previous studies. The red edge slopes in each population vary considerably at both the red edge shoulder point and the point of maximum red edge slope. Interesting enough, the highest and the lowest reflectance values were from the green attack population. Although the tree species classifications were not included in this report, the 60 cm spatial resolution imagery was sufficient to distinguish different tree species. The supervised and even the unsupervised classifications enabled different tree species to be easily distinguished on the imagery (ex: lodgepole pine, Douglas-fir, hybrid spruce, etc.). The resampled imagery (reprocessed to 1.2 meter data) was not as accurate at separating tree species, but was able to detect the red attack trees. This supports similar tests, which indicated that red attack was detected by using a multispectral imaging system (MEIS-2) with a 1.2 meter pixel size. The CASI instrument provides precise location (lat/long) of the red attack trees and accurately displays the numbers of trees affected. If CASI were to be used for mortality "red attack" surveys, >36 bands would not be needed. Three (3) bands would be sufficient to record the red attack, thus reducing the amount of data, and subsequent data processing. The supervised classifications (Figures 28 and 29) were unable to distinguish between green attack and the non-attack lodgepole pine trees on the imagery (both 60 cm and the resampled 1.2 m). Sections of the tree crowns that were selected as training areas most closely represented the mean values of the bands used in the classification. In spite of the fact that the training areas selected represented the population mean, the classifications were still unsuccessful. This is mostly due to the spectral variability within each population. A classification method may be used, but the development of the classification algorithm is beyond the scope of this study. The classification algorithm should include: 56 • mask all areas of the image that are not in the lodgepole pine central crown portion • create a third band that is a ratio of bands 8 (539.5 +1-1.1 nm) and 19 (706.4 +/-7.8 nm) • perform a discriminant analysis on bands 8 (539.5 +/-7.7 nm), 19 (706.4 +/-7.8 nm) and the band containing the ratio • classify all of the unmasked areas as either non or green attack • create a Jackknifed classification matrix to display the results The imagery from the green attack and the non-attack populations were analyzed using band ratioing methods. Selected bands from the imagery were ratioed to assist in determining tree stress (due to bark beetle attack) and to minimize different illumination effects. The ratios are displayed in (Table 5). The following band ratios were performed: 1) shoulder of the green rise compared to the toe portion of the red edge such as the red edge slope, 2) the slope of the red rise, 3) the red edge slope, 4) the green rise compared to the red well, and 5) the lower red edge slope. The most pronounced difference between the populations was the toe of the red edge compared to the shoulder of the green rise (green attack = 0.265, non attack = 0.251) and by the toe of the red edge compared to the red rise (green attack — 0.038, non attack — 0.026). It is possible that the results of the toe of the red edge compared to the red rise are due to differences between the populations, but since the red rise band (645.4 +/- 7.8 nm) was not indicated by the discriminant analysis or the t-tests, it is unlikely. Runesson (1991) indicated that the spectral area of 470 nm to 510 nm and the 620 nm to 660 nm (red rise) would be two major reflectance areas where differences will occur between green attack and non-attack. The spectral differences in these wavelength areas (between the populations) were not detected by the ITRES report. Separate detailed plots were made of these regions and they showed no significant differences between healthy and green stage populations of pixels (Davidson and Lataille, 2000). As previously stated, the red rise indicated a difference between the populations through band ratioing, but was not indicated by the discriminant analysis or the t-tests. The 470 nm to 510 nm reflectance area, indicated by Runesson (1991) as being a major reflectance area where there is a difference between green and non-attack trees, was not indicated as significant. The ITRES report stated that a ratio of the lower red edge slope (as indicated by bands 706.4 nm and 721.8 nm) delivered the strongest results to distinguish the green attack from the 57 non-attack trees. The same ratio (lower red edge slope ratio) was performed on the imagery from sites 1 and 2. The results (see Table 5) of the lower red edge ratio are identical for the green attack and non-attack populations. The mean and standard deviation were tabulated from the average reflectance values of each tree (in all 36 bands). Figure 31 (60 cm data) shows the average spectral reflectance and the (+/-) standard deviation of each average for the green attack and non-attack trees. Figure 32 (1.2m data) shows the average spectral reflectances and the (+/-) standard deviations of each average for the green attack and non-attack trees. The standard deviations from each population are high, indicating a large variance within the population. It is interesting to note that the trend of the spectral reflectance curves of both green attack and non-attack are similar, yet a slightly higher (+ standard deviation) was from the non-attacked trees. When the population averaged red edge slopes are examined, the green attack red edge slope seems to have shifted towards the infrared portion of the wavelength. The "infrared shift" is only observed in the averaged data, since the spectral variability of each individual trees red edge and the region of maximum red edge slope is very high. The "infrared shift" as seen when the populations are averaged is opposite to the ITRES findings presented by Davidson and Latialle (2000). Davidson and Lataille (2000) found that the shift in the wavelength of the maximum first derivative can be seen to become "bluer" for the green stage MPB pines. Their findings support the Runesson (1991) results that indicate that the red edge of green attack shifts towards the blue region. The results (as shown in Figures 31 and 32) of this study support the Ahern (1988) results which indicate that the red edge of the green attack shifts to longer wavelengths for the current foliage of attacked trees. To further analyze the data, t-tests were conducted. Many of the averaged (60 cm data and 1.2 m) spectral bands failed the Shapiro-Wilks normality test, therefore the non-normally distributed data (indicated on Tables 7 and 8) were tested using the Mann Whitney nonparametric test. T-tests were run on the normally distributed bands. Only band 27 (829.4 +/-7.9 nm) from the 60 cm data rejected the hypothesis when the confidence level was set at 95 percent. Band 27 (829.4 +/-7.9 nm) has been previously shown to be an area of initial reflectance in the Heller (1968) study. More recent green attack reflectance studies have not indicated the infrared portion of the wavelength as an area where change occurs following change following a successful bark beetle attack. The infrared wavelength area has significant variability in both derived spectral reflectance curves from each population. The result of the t-test indicates that there may be a difference between the populations in band 27 (829.4 +/-7.9 nm), although the difference is weak 58 (0.0466 < 0.05). Band 27 was shown as non-significant in the discriminant analysis, therefore the t-test results may just be due to noise within the population data. The statistical method of discriminant analysis is the only method that can classify the green and the non-attack trees. The analysis of the 60 cm imagery examined all of the 36 hyperspectral bands and identified two bands where there is a difference between the green and non-attack populations: band 8 (539.5 +/-7.7 nm) and band 19 (706.4 +/-7.8 nm). When the results were displayed in a Jackknifed classification matrix (unbiased classification method where each sample is classified regardless of previous categorical belonging) 79 percent of the green attack trees and 68 percent of the non-attack trees in both sites 1 and 2 are correctly classified. The overall success rate is 73 percent. It is important to note that there was only slight separation between the populations (displayed by a graph showing the central frequencies from the results of the canonical scores of the populations in Figure 30). The graph indicates that the populations are quite close together and overlap considerably. The discriminant analysis of the resampled 1.2 m imagery indicated that three bands display the difference between the populations. Bands 3 (464.5 +/-7.6 nm), 5 (494.4 +1-1.1 nm), and 6 (509.4 +1-1.1 nm) were indicated. The results of the resampled imagery are weaker than the results exhibited from the 60 cm imagery analysis. When the results were displayed in a Jackknifed classification matrix only 63 percent of the green attack trees and 64 percent of the non-attack trees in both sites 1 and 2 are correctly classified. The overall success rate is 63 percent. The success rate is not very high considering the classification of the two populations initially has a 50 percent chance of correct classification. The discriminant analysis results (from the 60 cm imagery) support the findings from other similar studies. Hobbs (1983), stated that results of the total film response analysis of damage classes, where significant, tend to indicate that there are changes in the total green and total red response, suggesting that reflectance changes occur in the green-red portion of the spectrum. Murtha and Wairt (1989) found that significant differences existed between non-attack and current-attack digital numbers (DN) in the green and red data, and that there was no significant difference between the M R data. Ahern (1988) found that the reflectance was lower for current foliage of attacked trees in the regions of 525 nm - 565 nm, lower reflectance at the red rise (620 to 650 nm), and that the red edge shifted towards longer wavelengths (displayed in Table 1). It is surprising that the subtle spectral effects occurring in the green and red 59 wavelengths have been detected using broad-band film based studies. These studies may have coincided with dry summers increasing the needle desiccation. The shoulder of the green rise (band 8, 539.4 +1-1.1 nm) indicated by the discriminant analysis supports the Ahern (1988) results that the reflectance (in wavelengths 525 - 565 nm) is lower for the attacked trees. The second wavelength area indicated by the discriminant analysis (band 19, 706.5 +/-7.8 nm) does not support Ahern's results. The discriminant analysis of the resampled 1.2 m imagery displayed a difference in three bands: band 3 (464.5 +1-1.6 nm), band 5 (494.4 +/-7.7 nm), and band 6 (509.4 +/-7.7 nm). The three bands are in the blue region of the wavelength. Unfortunately, light scattering in the blue portion of the wavelength is very pronounced. Light scattering adds complexity to the spectral survey and accurate measurements in this region are often unreliable. Although the blue region is associated with light scattering, the bands which were indicated by the discriminant analysis are mentioned. Figure 33 shows the derived spectral reflectance curves from the (60 cm imagery) colour coded based on the trees location on the imagery (fore or back scatter zones). Although there is not a strong separation between the trees from forward-scattering and backscattering zones, there does seem to be categorical significance. The four "brightest" trees are from the backscatter zone and (with the exception of one tree) six of the "least brightest" trees are from the forward-scatter zone. A larger population (>25) would have to be sampled to determine average physical attribute categories. I suspect that there will be significant differences in spectral values based on tree physical attributes (age, height, and diameter). These subtle spectral differences are evident even in monoculture lodgepole pine stands (ie. much of the Chilcotin region of the British Columbia) where tree crown height is not uniform. The taller trees (even slightly taller) receive direct sunlight at the top of their crown and the portion of the crown above the surrounding trees, while the trees in the lower part of the canopy receive much less direct sunlight. In some cases, the lower trees are shaded completely by the surrounding trees, leaving their crowns in shadow or illuminated by diffuse light. The heavily shaded tree samples were not used in this study, but during an operational survey of this nature, shading effects would be present. The complexity introduced by low sun angles would be reduced if the aerial survey were to be completed earlier 60 in the season. Unfortunately, the impact of the mountain pine beetle damage would be even less earlier in the season. The height of the trees' crowns in comparison with surrounding tree crowns seems to influence overall spectral reflectance. Tree crowns that are lower in the canopy compared to the surrounding tree crowns generally have a lower spectral reflectance. Tree # 6 2 and tree # 11_2 are non-attacked trees with a low spectral reflectance. Tree 6_2 is 18.07 m and tree 1 1 2 is 18.19 m; while the average is 21.04 m. These trees are almost 3 meters lower than the average tree height. Therefore at the time of the aerial survey, these tree crowns received very little direct sunlight. The opposite is also true. Tree 56 2 has a similar age and diameter breast height (DBH) to tree 11_2 (both are also apparently healthy), yet tree 56_2 has a much brighter spectral reflectance trend. This is due to tree 56_2 being taller than the trees surrounding it, thus receiving direct sunlight on the top and sides of the crown (especially on the south facing branches). The problem of improper lighting is amplified with low sun angles when surveys are conducted during spring (February to April) or fall (September to November), in Canada. Warm, dry summers accentuate the colour change by increasing the desiccation process of the lodgepole pine tree. As an example, in warmer areas of the United States, the phenomena of green attack is very brief. Higher mean average temperatures limit the green attack stage to less than a couple of months, and often the trees change from green to red during the same year that they are attacked. Obviously, warmer / drier environmental effects enhance the tree drying process. A warm summer may impact needle desiccation in two ways. During warm summers, the beetles may fly earlier in the season. The longer duration of beetle larval feeding increases the amount of damage to the tree prior to the fall survey. The other way that the warm weather can impact the needle desiccation is by extenuating the summer drought conditions. With warm weather, the effects of the bark beetle attack may be more pronounced. Successful detection of the green attack lodgepole pine may even be possible on an operational level using modern remote sensing (airborne or spaceborne) survey techniques. In Canada, environmental factors are crucial to the success of green attack detection. Due to our shorter summer season, the tree drying process is seriously limited. During cooler than average summer / fall seasons there is very little drying and subsequent reflectance change. 61 The subtle reflectance changes that are supposed to occur are masked within the natural reflectance variability o f the tree species as a whole. A warmer than average summer / fall season would increase the likeliness o f green attack detection. Other environmental factors may also complicate aerial spectral surveys. L o w sun angle is a major problem during fall (after September 15 t h) surveys (north of the 49 t h latitude). The effects o f low sun angle are more pronounced in northerly latitudes, surveying may have to be restricted to summer only (May to August). The low sun angle only illuminates the side o f the tree often obscuring other trees around it. Other environmental factors such as the season, time of day, slope, and weather are difficult to overcome, yet sometimes these problems can be calibrated. Although specific regions have been shown to display the earliest signs o f bark beetle induced tree stress, these can be masked by the natural variability within the forest stands. The variability is not from the other tree species, but variability within the same tree species due to complex factors that would be very difficult to quantify. Factors such as tree age, tree height, crown height in comparison to surrounding tree crowns, and numerous micro-site factors strongly influence the trees spectral reflectance. The variability o f the reflectance within tree species is well documented in recent literature. Murtha et al. (1997) outlined numerous examples o f reflectance variability within identical tree species. Factors such as age o f the tree, nutrient condition, moisture status, geographic location, slope, seasonal change, and even reflectance change on a daily basis, w i l l create significant obstacles when implementing a system that can detect subtle reflectance changes such as "green attack." Therefore, unless sufficient needle desiccation has occurred (as a result o f a dry, warm weather period following the beetle flight), the subtle reflectance changes (in early fall) are not enough to separate the green attack from the non-attack lodgepole pine trees. Many local M P B experts c laim to have (during dry years) seen a different shade o f green in the green attack trees1. These visual changes are associated with hotter than average summers. During the period of late August to late September in 1998 (the warmest summer for the last 6 years), green attack trees were visually detectable 2. The detection o f green attack is dependent on the amount o f physiological changes (from the tree drying out) that occur. A basic guide (or 1 Murtha, P. A. , 2001. Pers.Comm. Remote Sensing Professor, University of British Columbia, Vancouver, B.C. 2 Cortese, J., 1999. Pers. Comm. Aerial Survey Contractor, Alta Vista Contracting, Tatla Lake, B.C. 62 chart) should be created that determines the amount of "tree drying" based on environmental factors, to determine the likelihood of successful green attack detection. 12. Conclusions The derived spectral reflectance curves (for the green attack and the non-attack populations) display greater variability within each population than between the populations. It is not possible to visually detect a difference between the populations. Subtle spectral differences between the populations became evident only through statistical analysis. The result of an individual band t-test, indicated a difference between the populations in band 27 (829.4 +1-1.9 nm), although the difference is weak (0.0466 < 0.05). Unfortunately, band 27 was shown as non-significant in the discriminant analysis, therefore the t-test results may just be due to noise within the population data. Only two bands indicated a difference in the discriminant analysis: band 8 (539.5 +/-7.7 nm) and band 19 (706.4 +/-7.8 nm). When the results were displayed in a Jackknifed classification matrix (unbiased classification method where each sample is classified regardless of previous categorical belonging) 79 percent of the green attack trees and 68 percent of the non-attack trees in both sites 1 and 2 are correctly classified. The overall success rate was 73 percent. The imagery was resampled to a 1.2 meter pixel size for further analysis. The 1.2 m imagery indicated that three (different) bands display the difference: bands 3 (464.5 +/-7.6 nm), 5 (494.4 +/-7.7 nm), and 6 (509.4 +/-7.7 nm) were indicated. The results of the resampled imagery are weaker than the results exhibited from the 60 cm imagery analysis. When displayed in a Jackknifed classification matrix only 63 percent of the green attack trees and 64 percent of the non-attack trees in both sites 1 and 2 are correctly classified. The overall success rate is 63 percent. The success rate is not very high considering the classification of the two populations initially has a 50 percent chance of correct classification. Several band ratio techniques and supervised classifications were applied to the bands indicated by the statistical analysis. The very subtle spectral differences between the non-attack and green attack populations was quantitatively accentuated using the band ratios. Although spectral differences were accentuated, the band ratio techniques were still unable to be a reliable method for discrimination. Due to the spectral variability within each population, supervised classification methods were unable to distinguish a difference between the populations. 63 13. Recommendations The objectives for future methods of aerial beetle damage appraisal should be based on cost and accuracy. Although conventional sketchmapping is substantially cheaper than all the other forms of surveying, the accuracy is questionable. Many forest companies find the sketch map results unacceptable for operational purposes. Due to the costs associated with beetle probing and cutblock layout, having the location of red and green attacked trees would present major time and money savings. Prior to the operational use of green attack detection, further studying should be conducted. Future studies should focus on the following: • Create a "likelihood of green attack detection chart," based on the flight date of the beetles and the degree days following attack. That way, foresters could gauge the environmental drying that occurs to the green attack pine trees. • A larger, multi-year study of green attack detection (using hyperspectral imagery) is needed to verify the results of this paper. Two bands were indicated as having promise for separating green attack and non-attack pine trees: the shoulder of the green rise (539.5 +1-1.1 nm) and the toe of the red edge (706.4 +1-1.% nm). Should results should be verified. • Two other regions were also identified as having some promise for separating green attack from non-attack detection: the infrared portion (829.4 +/- 7.9 nm) and the red rise (645.4 +/-7.8 nm). These regions should also be studied further. • Attempts should be made to quantify the changes on the imagery. The following phenomena may need quantifying: varying tree heights, seasonal changes, differences in sun angle, year to year differences, biogeoclimatic zone differences, and other site factors. Many of these environmental discrepancies may be quantified with the development of a likelihood of green attack detection chart. 64 15. References 1. Airborne Operations Section, 1977. Information Bulletin Airborne Operation, Third Operation, Canada Centre for Remote Sensing, Department of Energy, Mines, and Resources, Canada, pages E-4 - E - 6 . 2. Amman G.D. and W.E. Cole, 1983. Mountain Pine Beetle Dynamics in Lodgepole Pine Forests, USDA Forest Service, GTR-INT-145, Ogden, Utah. 3. Ahern, F., 1988. The Effects of Bark Beetle Stress on the Foliar Spectral Reflectance of Lodgepole Pine, International Journal of Remote Sensing, Vol . 9, No 9, pages 1451-1468. 4. Carter, G.A., 1998. Reflectance Wavebands and Indices for Remote Sensing of Photosynthesis and Stomatal Conductance in Pine Canopies, Remote Sensing of the Environment, Vol . 63, pages 61-72. 5. Carter, G.A., Cibula, W.G., and R.L. Miller, 1996. Narrow-band Reflectance Imagery Compared with Thermal Imagery for Early Detection of Plant Stress, Journal of Plant Physiology, 148: pages 515-522. 6. Carter, G.A., Dell, T.R., and W.G. Cibula, 1996. Spectral Reflectance Characteristics and Digital Imagery of Pine Needle Blight in Southeastern United States, Canadian Journal of Forestry Research, 26 (3), pages 402-407. 7. Comer, R.P., Kinn, G., Light, D., and C. Mondello, 1998. Talking Digital, Photogrammetric Engineering & Remote Sensing, pages 1139-1142. 8. Colwell, J.E., 1974. Vegetation Canopy Reflectance, Remote Sensing of Environment, 3:175-183. 9. Curtiss B. and S.L. Ustin, 1989. Parameters Affecting Reflectance of Coniferous Forest in the Region of Chlorophyll Pigment Absorption. Proceedings of IGARRS '89 /12th Can. Symp. On Remote Sensing, Vancouver, B.C. pp.2633-2636. 10. Davidson, D., and S. Lataille, 2000. Assessment of Possible Spectral Indicators for Green Stage Mountain Pine Beetle Attack Using Airborne CASI Imagery at Tyee Lake. Prepared for the B.C. Ministry of Forests. Prepared by ITRES Research Ltd, Calgary, Alberta. 11. Deering, D.W., 1989. Field Measurements of Bidirectional Reflectance: In Theory and Applications of Optical Remote Sensing , Wiley and Sons, New York, pages 14-61. 12. Everitt, J.H., Escobar, D.E., and M.R. Davis, 1998. Application of Remote Sensing and Spatial Information Technologies for Detecting and Mapping Insect Infestations, Natural Resource Management Using Remote Sensing and G.I.S.; Proceedings of the Seventh Forest Service Remote Sensing Applications Conference, Nassau Bay, Texas, pages 240-256. 13. Furniss, R.L., and V . M . Carolin, 1977. Western Forest Insects, Misc. Publication, 1339, Washington D.C., USDA Forest Service, page 654. 14. Gausman, H.W., 1974. Leaf Reflectance of Near Infrared, Photogrammetric Engineering, pages 183-191. 15. Hall, P .M. and T.F. Maher, 1985. Proceedings of the Mountain Pine Beetle Symposium, Smithers, 1985. Ministry of Forests, Pest report #7, Queens Printer for British Columbia, Victoria, Canada. 65 16. Harris J.W.E. and A.F. Dawson, 1979. Evaluation of Aerial Forest Pest Damage Survey Techniques in British Columbia, Pacific Forest Research Centre, Victoria, B.C. 17. Heller, R.C., 1968. Previsual Detection of Ponderosa Pine Trees Dying from Bark Beetle Attack, Proceeding in 5 t h Symposium on Remote Sensing of Environment. Willow Run Lab, University of Michigan, pages 387-434. 18. Hobbs, A.J. , 1983. Effects of Air Photoscale on Early Detection of Mountain Pine Beetle Infestation. Masters thesis, University of British Columbia, Faculty of Forestry. 19. Hobbs, A.J . , and P. A. Murtha, 1984. Visual Interpretation of Four Scales of Aerial Photography for Early Detection of Mountain Pine Beetle Infestation, Renewable Resources Management; Applications of Remote Sensing. American Society of Photogrammetry, Falls Church, V A . , pages 433-444. 20. Kalensky, Z. and D. Wilson, 1975. Spectral Signatures of Forest Trees, Proceedings: Canadian Symposium on Remote Sensing, 3 r Ottawa: Canadian Remote Sensing Society, pages 155-171. 21. Kneppeck, I.D., and F. Ahern, 1989. A Comparison of Images from a Pushbroom Scanner with Normal Colour Aerial Photographs for Detecting Scattered Recent Conifer Mortality; Photogrammetric Engineering & Remote Sensing, Vo l . 55, No. 3, pages 333-337. 22. Lichtenthaler, H.K., 1989. Possibilities for Remote Sensing of Terrestrial Vegetation by a combination of Reflectance and Laser-induced Chlorophyll Fluorescence. Proceedings of IGARRS '89 /12th Can. Symp. On Remote Sensing, Vancouver, B.C. pp. 1349-1354. 23. Losee, S.B.T., 1951. Photographic Tone in Forest Interpretation, Photogrammetric Engineering, 17:785-799. 24. Meidinger, D., Pojar, J., and W.L. Harper, 1991. Ecosystems of British Columbia: Chapter 14: Sub-Boreal Spruce Zone, Research Branch, Ministry of Forests, Victoria, B.C. 25. Moody, B.H. , 1981. The Role, Informational Requirements, and Major Pest Problems of the Forest Insect and Disease Survey, Northern Forest Research Centre, Canadian Forestry Service, Environment Canada, Edmonton, Alberta, Proceedings of Seminar, Uses of Remote Sensing in Forest Pest Damage Appraisal, compiled by R.J. Hall. 26. Murtha, P. A. , 1985. Interpretation of Large-scale Colour Infrared Photographs for Bark Beetle Incipient Attack Detection. PECORA 10 Symposium Proceedings, American Society for Photogrammetry and Remote Sensing, Falls Church, Va, USA, pages 209-219. 27. Murtha, P.A., Deering, D.W., Olson, C E . Jr., and G.A. Bracher, 1997. Chapter 5: Vegetation, In Philipson, W. (ed). Manual of Photographic Interpretation, Second Edition, American Society of Photogrammetry and Remote Sensing, Bethesda, M D . , pages 225-255. 28. Murtha, P.A. and R. J. Wiart., 1989. PC-Based Digital Analysis of Mountain Pine Beetle Current-Attacked and Non-Attacked Lodgepole Pine, Canadian Journal of Remote Sensing. Pages 70-76. 29. Myhre, R. J. and B. Silvey, 1992. An Airbourne Video System Developed Within Forest Pest Management - Status and Activities. In Proceedings of Fourth Forest Service Remote Sensing Applications Conference, Orlando, Florida, April 1992, pp. 291-300. 66 30. Raffa, K.F. , 1988. The Mountain Pine Beetle in Western North America, Plenum Public Corp., pages 505-530. 31. Reid, R.W., Whitney, H.S., and J.A. Watson, 1967. Reactions of Lodgepole Pine to Attack by Dehdroctonus ponderosae Hopkins and blue stain fungi, Canaidan Journal of Botany, Vol . 45, pages 1115-1126. 32. Roswell, R.P., 1981. Bark beetle Detection Manual Prince Rupert Forest Region. Ministry of Forest, Queens Printer for British Columbia, Victoria, Canada. 33. Runesson, U.T., 1991. Considerations for Early Remote Detection of Mountain Pine Beetle in Green-Foliaged Lodgepole Pine. Doctorate thesis, University of British Columbia, Department of Forestry. 34. Safranyik, L. , D . M . Shrimpton and H.S. Whitney, 1974. Management of Lodgepole Pine to Reduce Losses from the Mountain Pine Beetle. Can For. Serv. Tech. Report 1, Env. Canada: 24pp. 35. Schill, S., 2000. Bidirectional Reflectance Distribution Function Modelling, web site: http://steve.schill.com/research.html, Site accessed on January 15 th, 2001. 36. Shore, T.L. and L . Safranyik, 1992. Susceptibility and Risk Rating Systems for the Mountain Pine Beetle in Lodgepole Pine Stands, Forestry Canada, Information Report BC-X-336. 37. Silvennoinen, R., Jaaskelainen, T., Nygren, K. , Hiltunen, J., and J. Parkkinen, 1995. Temporal, Spatial, and Environmental Classification of Pine Reflectance Spectra., Environmental Science & Technology, Volume 29, pages 1456-1459. 38. Sirois, J., and F. Ahern, 1988. An Investigation of SPOT H R V Data for Detecting Recent Mountain Pine Beetle Mortality, Canadian Journal of Remote Sensing, Vol . 14, No 2, pp 104-108. 39. Steiner, D., and T. Gutermann, 1966. Russian Data on Spectral Reflectance of Vegetation, Soil and Rock Type, Zurich: University of Zurich, Dept.of Geography, 232p. 40. Suits, G.H., 1972. The Calculation of Directional Reflectance of a Vegetation Canopy, Remote Sensing of Environment, 2:117-125. 41. Tucker, C.J., 1979. Red and Photographic Infrared Linear Combinations for Monitoring Vegetation, Remote Sensing of the Environment, Vo l . 8, pages 127-150. 42. Unger, L., 1993. Mountain Pine Beetle, Forestry Canada, Forest Insect and Disease Survey, Forest Pest Leaflet No.76, page 7. 43. Ustin, S.L., and B. Curstiss, 1990. Spectral Characteristics of Ozone Treated Conifer Species, Environmental and Experimental Botany, 30: pages 293-308. 44. Glossary taken from http://www.ccrs.nrcan.gc.ca/ccrs/eduref/ref/glosndxe.html The Canadian Centre for Remote Sensing, Date accessed: March 30', 2001. 67 16. Glossary (From Canadian Centre for Remote Sensing Webpage) 1. Atmospheric Effects - Physical phenomena occurring in the atmosphere and affecting electromagnetic radiation. 2. Backscatter Zone - The backscatter zone is the portion of the image that has the highest reflectance, caused when the illumination source is behind the sensor and the canopy is front-lit. 3. Bidirectional Reflectance Factor (BRF) - The surface reflects according to the cosine law, where the reflected beam, is measured as radiance, by a sensor with an aperture of specified angular field of view; is the angle off-nadir. BRF is a function of the sun's position and spectral properties, the spectral reflectance properties of the surface, and the position (viewing angle) and spectral response of the radiometer/spectrometer. 4. Brightness - Brightness may be a result of variations in tone, texture, or in the case of radar imagery, radar artifacts. The topography and surface roughness of the terrain will affect the image brightness. Where the local incidence angle is large, the image will be dark. Conversely, the image will be brighter where the local incidence angle is small. 5. Electromagnetic Radiation (EMR) - Energy propagated through space or through material media in the form of an advancing interaction between electric and magnetic fields. 6. Forescatter Zone - The forescatter zone is the portion of the image that has a lower reflectance, caused when the illumination source is forward of the sensor and the canopy is back-lit. 7. Greenness - The difference between normalized near infrared (0.7-1.1 microns) and visible (0.5-0.7 microns) radiances of vegetation representing the state of growth of a crop. 8. Green Attack - Where the foliage of a bark beetle killed tree is still green. The foliage remains green for up to a year after the tree has been attacked and killed by bark beetles. 9. Green Rise - The peak of the reflectance region (centered at approximately 550 nm) where the greatest green spectral reflectance occurs. 68 10. Infrared Wavelengths - This area of the electromagnetic spectrum can be divided in three sections: 1) the near-IR (700 - 1300nm), 2) the mid-IR (1300 - 3000nm), and 3) thermal (far) IR (3000nm <). Within the IR portion of the spectrum, it should be noted that only thermal IR energy is directly related to the sensation of heat; near-IR and mid-IR energy are not. Wweim nptft tm) frequency (H 9$ Lortittw Lower Infrared 10* -I 8 craw*') 0 —1 % s*r> -+1 ry -I 1 a'^  10 SfrcxtiM- I 5v« •10' t- 1 rt>.r Higher Wavelength (metres} 10*H 1 0 -4 io*H 10"7H «3 41 .E > s ,«2 .a cs 41 > SJ U 01 0 = 41 [ 41 r o 11. Latitudinal Effect - A quantitative variation of a measurement due to its position with respect to the equator. 12. Normalized Difference Vegetation Index (NDVI) - An index calculated from reflectances measured in the visible and near infrared channels. It is related to the fraction of photosynthetically active radiation. 13. Plant Reflectance - Ratio of the intensity of reflected radiation to that of the incident radiation from plant surfaces. 14. Previsual - Incipient damage is commonly called previsual damage because the leaf still appears green. 15. Push-broom scanning - Describes the technique of using the forward motion of the platform to sweep the linear array of detectors across the ground scene. 69 16. Radiance - A measure of the energy radiated by the object together with the frequency distribution of that radiation. 17. Red Attack - Where the foliage of a bark beetle killed tree has turned red. This usually happens the following summer after the tree was attacked. 18. Red Edge - Spectral region at the limit of the red and near-infrared wavelengths characterized by a sharp rise in the plant reflectance. 19. Red Rise - The region of reflectance (620 - 650nm) where the current foliage of unattacked trees is brighter. 20. Scattering - The process in which a wave or beam of particles is diffused or deflected by collisions with particles of the medium which it traverses. 21. Spectral Signature - The frequency distribution patterns of radiation reflected and/or emitted by an object. 22. Vegetation Index - The reduction of multispectral scanning measurements to a single value for predicting and assessing vegetative characteristics. Examples of such characteristics include plant leaf area, total biomass, fresh and dry above-ground phytomass, chlorophyll content, plant height, percent ground cover by vegetation, grain or forage yield and general plant stress and vigour. 23. Visible Wavelengths - The portion of the electromagnetic spectrum that is sensitive to the human eye. The region is between 400 - 700nm. 70 Wavetangth (m) F r e q u e n c y \Hz) tDnoer Lower u<ru 10 ItiiH-t 1 m-1,0-1 1cm-+10 1min-*1#-n r - J 1 nm-f I F Shatter 1 0 " ^ 1 0 - J ) I at I r9 1 0 s <-1 CK* 1 0 1 0 | _ 1 0 1 2 . . r m , 10*^1 EHI 10 10 Higher Visible Waw ©length (met res | 0.7 x 10* 0.6 x 10 l l 0.5 x 10 4 0.4 x 10 24. Wavelength - Minimum distance between two events of a recurring feature in a periodic sequence, such as the crests in a wave. It is represented by the Greek letter, X. 71 Appendix 1 The sun angle for 50°N throughout the year (Airborne Operations Section, 1977). Date Sun Angle in De< jrees 0 10 15 20 25 30 40 50 60 1-Jan 7:56 4:55 2:35 11-Jan 8:11 5:18 3:14 21-Jan 8:34 5:50 4:03 31-Jan 9:02 6:28 4:54 2:44 10-Feb 9:34 7:09 5:45 4:03 0:57 20-Feb 10:09 7:52 6:35 5:08 3:16 2-Mar 10:46 8:35 7:24 6:06 4:36 2:33 12-Mar 11:24 9:16 8:09 6:58 5:40 4:08 22-Mar 12:03 9:58 8:54 7:47 6:36 5:17 0:59 1-Apr 12:42 10:37 9:34 8:30 7:23 6:12 3:15 11-Apr 13:20 11:14 10:12 9:09 8:05 6:58 4:25 21-Apr 13:56 11:49 10:46 9:44 8:41 7:36 5:17 1:59 1-May 14:31 12:20 11:17 10:15 9:12 8:09 5:56 3:15 11-May 15:03 12:48 11:44 10:41 9:39 8:36 6:27 4:01 21-May 15:30 13:11 12:06 11:02 10:00 8:57 6:51 4:32 0:47 31-May 15:51 13:28 12:22 11:18 10:15 9:13 7:07 4:52 1:51 10-Jun 16:04 13:39 12:32 11:28 10:24 9:22 7:17 5:04 2:16 20-Jun 16:09 13:43 12:36 11:31 10:28 9:25 7:20 5:08 2:24 30-Jun 16:05 13:39 12:33 11:28 10:25 9:22 7:17 5:05 2:17 10-Jul 15:52 13:29 12:23 11:19 10:16 9:13 7:08 4:53 1:53 20-Jul 15:31 13:12 12:07 11:03 10:01 8:58 6:52 4:33 0:55 30-Jul 15:04 12:49 11:45 10:42 9:40 8:38 6:29 4:03 9-Aug 14:33 12:22 11:19 10:16 9:14 8:11 5:59 3:18 19-Aug 13:59 11:51 10:48 9:46 8:43 7:39 5:19 2:06 29-Aug 13:22 11:16 10:19 9:11 8:07 7:00 4:29 8-Sep 12:44 10:39 9:37 8:32 7:26 6:15 3:20 18-Sep 12:06 10:01 8:56 7:50 6:39 5:21 1:15 28-Sep 11:27 9:20 8:13 7:03 5:45 4:14 8-Oct 10:49 8:38 7:28 6:11 4:42 2:43 18-Oct 10:12 7:56 6:40 5:14 3:24 28-Oct 9:37 7:13 5:50 4:09 1:19 7-Nov 9:04 6:32 4:58 2:51 17-Nov 8:36 5:53 4:07 0:44 27-Nov 8:13 5:20 3:18 7-Dec 7:57 4:57 2:38 17-Dec 7:50 4:45 2:15 27-Dec 7:52 4:48 2:21 72 Appendix 2 The sun angle for 55°N throughout the year (Airborne Operations Section). Date Sun Angle in Degrees 0 10 15 20 25 30 40 50 60 1-Jan 7:01 2:46 11-Jan 7:20 3:28 21-Jan 7:49 4:20 31-Jan 8:24 5:16 2:56 10-Feb 9:04 6:12 4:22 1:02 20-Feb 9:46 7:07 5:33 3:32 2-Mar 10:31 8:01 6:37 5:00 2:46 12-Mar 11:17 8:53 7:35 6:10 4:30 1:28 2 2-Mar 12:04 9:43 8:30 7:12 5:47 4:03 1-Apr 12:50 10:30 9:19 8:06 6:48 5:21 11-Apr 13:36 11:15 10:05 8:54 7:40 6:21 2:59 21-Apr 14:20 11:56 10:46 9:36 8:25 7:11 4:20 1-May 15:03 12:34 11:23 10:14 9:03 7:52 5:15 0:44 11-May 15:42 13:07 11:56 10:45 9:35 8:25 5:56 2:41 21-May 16:16 13:35 12:22 11:11 10:01 8:51 6:26 3:32 31-May 16:43 13:57 12:42 11:30 10:20 9:10 6:47 4:03 10-Jun 17:00 14:10 12:54 11:42 10:31 9:22 7:00 4:20 20-Jun 17:07 14:15 12:59 11:46 10:35 9:26 7:04 4:26 30-Jun 17:01 14:11 12:55 11:42 10:32 9:22 7:00 4:21 10-Jul 16:44 13:58 12:43 11:31 10:21 9:11 6:48 4:05 20-Jul 16:18 13:37 12:23 11:12 10:02 8:53 6:28 3:35 30-Jul 15:44 13:09 11:57 10:47 9:37 8:27 5:58 2:45 9-Aug 15:05 12:36 11:26 10:16 9:06 7:54 5:18 1:00 19-Aug 14:23 11:59 10:49 9:39 8:28 7:14 4:24 29-Aug 13:39 11:17 10:08 8:57 7:43 6:25 3:06 8-Sep 12:53 10:33 9:22 8:09 6:51 5:25 18-Sep 12:07 9:46 8:33 7:16 5:51 4:09 28-Sep 11:21 8:57 7:40 6:16 4:37 2:14 8-Oct 10:35 8:06 6:42 5:06 2:57 18-Oct 9:50 7:12 5:39 3:41 28-Oct 9:07 6:17 4:28 1:25 7-Nov 8:27 5:21 3:04 17-Nov 7:52 4:25 0:47 27-Nov 7:23 3:32 7-Dec 7:02 2:49 17-Dec 6:53 2:25 27-Dec 6:55 2:31 73 Appendix 3 60 cm data inclusive discriminant analysis Forward stepwise with Alpha-to-Enter=0.150 and Alpha-to-Remove=0.150 Group frequencies Green Non 24 25 Group means Green Non Band 1 294.363 327.382 Band 2 377.042 426.824 Band 3 423.446 476.404 Band 4 450.682 503.507 Band 5 444.165 498.172 Band 6 481.978 556.237 Band 7 644.229 755.241 Band 8 806.214 951.409 Band 9 882.780 1034.856 Band 10 799.838 933.308 Band 11 715.637 828.284 Band 12 689.776 799.083 Band 13 642.354 738.641 Band 14 610.350 700.340 Band 15 587.525 677.035 Band 16 523.924 597.500 Band 17 493.409 561.369 Band 18 634.658 720.640 Band 19 1429.126 1615.235 Band 20 2060.798 2322.385 Band 21 3145.492 3544.729 Band 22 3362.565 3821.020 Band 23 2618.341 2950.897 Band 24 3427.330 3894.283 Band 25 3214.332 3653.626 Band 26 2775.457 3164.187 Band 27 2678.866 3058.888 Band 28 2840.007 3256.757 Band 29 2664.065 3065.786 Band 30 2588.316 2986.678 Band 31 2256.593 2618.992 Band 32 1758.804 2029.885 Band 33 1646.015 1914.874 Band 34 752.581 877.694 Band 35 889.388 1036.028 Band 36 1280.676 1495.576 Classification functions Green Non C O N S T A N T -0.693 -0.693 V a r i a b l e F-to-remove Tolerance V a r i a b l e 6 Band 1 7 Band 8 Band 9 Band 10 Band 11 Band F-to-enter Tolerance 08 93 74 25 22 14 000000 000000 000000 000000 000000 000000 74 12 Band 7 3 71 1 000000 13 Band 8 3 75 1 000000 14 Band 9 3 04 1 000000 15 Band 10 2 64 1 000000 16 Band 11 2 23 1 000000 17 Band 12 2 16 1 000000 18 Band 13 1 89 1 000000 19 Band 14 1 76 1 000000 20 Band 15 1 81 1 000000 21 Band 16 1 51 1 000000 22 Band 17 1 42 1 000000 23 Band 18 1 43 1 000000 24 Band 19 1 59 1 000000 25 Band 20 1 90 1 000000 26 Band 21 2 23 1 000000 27 Band 22 2 66 1 000000 28 Band 23 2 31 1 000000 29 Band 24 2 54 1 000000 30 Band 25 2 48 1 000000 31 Band 26 2 47 1 000000 32 Band 27 2 40 1 000000 33 Band 28 2 44 1 000000 34 Band 29 2 41 1 000000 35 Band 30 2 34 1 000000 36 Band 31 2 41 1 000000 37 Band 32 2 12 1 000000 38 Band 33 2 27 1 000000 39 Band 34 2 32 1 000000 40 Band 35 2 27 1 000000 41 Band 36 2 23 1 000000 Between groups F-matrix -- d f : Green Non Green 0.000 Non 3.745 0.000 1 47 Wilks' lambda Lambda = 0.9262 df= 1 1 47 Approx. F= 3.7452 df = 1 47 prob = 0.0590 V a r i a b l e 13 Band i F-to-remove 3.75 Tolerance 1. 000000 V a r i a b l e F-to-enter Tolerance 6 Band 1 0 50 0 178873 7 Band 2 0 09 0 114206 8 Band 3 0 38 0 087431 9 Band 4 1 85 0 066740 10 Band 5 3 71 0 040342 11 Band 6 0 84 0 021942 12 Band 7 0 00 0 004588 14 Band 9 5 56 0 005641 15 Band 10 9 17 0 009070 16 Band 11 11 42 0 014537 17 Band 12 9 25 0 019519 18 Band 13 9 10 0 028363 19 Band 14 8 15 0 036436 20 Band 15 5 87 0 045664 21 Band 16 5 34 0 069490 22 Band 17 4 23 0 091254 23 Band 18 7 13 0 059294 24 Band 19 13 02 0 029138 25 Band 20 5 79 0 041624 26 Band 21 1 45 0 081269 27 Band 22 0 32 0 108474 28 Band 23 0 84 0 103952 29 Band 24 0 46 0 107321 75 30 Band 25 0 56 0 105650 31 Band 26 0 53 0 110445 32 Band 27 0 57 0 115669 33 Band 28 0 45 0 125112 34 Band 29 0 41 0 135069 35 Band 30 0 40 0 148360 36 Band 31 0 28 0 158773 37 Band 32 0 47 0 173891 38 Band 33 0 30 0 182737 39 Band 34 0 25 0 185584 40 Band 35 0 25 0 194173 41 Band 36 0 25 0 201929 Between groups F-matrix -- df = 2 46 Green Non Green 0.000 Non 8.860 0.000 Wilks' lambda Lambda = 0.7219 df= 2 1 47 Approx. F= 8.8599 df = 2 46 prob = 0.0006 V a r i a b l e F-to-remove Tolerance 13 Band 8 15.64 0.029138 24 Band 19 13.02 0.029138 V a r i a b l e F-to-enter Tolerance 6 Band 1 0 37 0 178872 7 Band 2 0 10 0 114182 8 Band 3 0 23 0 087403 9 Band 4 0 99 0 066484 10 Band 5 1 51 0 039451 11 Band 6 0 15 0 021566 12 Band 7 1 16 0 004062 14 Band 9 0 00 0 003128 15 Band 10 0 45 0 004816 16 Band 11 1 30 0 008055 17 Band 12 0 55 0 010814 18 Band 13 0 47 0 015393 19 Band 14 0 25 0 019810 20 Band 15 0 03 0 027708 21 Band 16 0 02 0 043746 22 Band 17 0 02 0 058058 23 Band 18 0 02 0 024562 25 Band 20 0 07 0 019867 26 Band 21 0 12 0 066140 27 Band 22 ' 0 42 ' 0 095157 28 Band 23 0 06 0 092613 29 Band 24 0 23 0 095563 30 Band 25 0 13 0 095041 31 Band 26 0 11 0 100512 32 Band 27 0 05 0 106674 33 Band 28 0 05 0 117050 34 Band 29 0 02 0 128411 35 Band 30 0 01 0 142092 36 Band 31 0 01 0 153401 37 Band 32 0 00 0 168408 38 Band 33 0 00 0 178811 39 Band 34 0 00 0 181929 40 Band 35 0 00 0 190913 41 Band 36 0 01 0 199670 Classification matrix (cases in row categories classified into columns) Green Non %correct Green 20 4 83 Non 8 17 68 Total 28 21 76 Jackknifed classification matrix 76 Green Non %correct Green 19 5 79 Non 8 17 68 Total 27 22 73 Eigenvalues 10.385 | Canonical correlations 10.527 | Cumulative proportion of total dispersion 11.000 | Wilks' lambda= 0.722 Approx.F= 8.860 df= 2, 46 p-tail= 0.0006 Pillai's trace= 0.278 Approx.F= 8.860 df= 2, 46 p-tail= 0.0006 Lawley-Hotelling trace= 0.385 Approx.F= 8.860 df= 2, 46 p-tail= 0.0006 Canonical discriminant functions 1 Constant -3.378 Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 Band 7 Band 8 0.021 Band 9 Band 10 Band 11 Band 12 Band 13 Band 14 Band 15 Band 16 Band 17 Band 18 Band 19 -0.010 Band 20 Band 21 Band 22 Band 23 Band 24 Band 25 Band 26 Band 27 Band 28 Band 29 Band 30 Band 31 Band 32 Band 33 Band 34 Band 35 77 Band 36 Canonical discriminant functions - standardized by within variances 1 Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 Band 7 Band 8 5.596 Band 9 Band 10 Band 11 Band 12 Band 13 Band 14 Band 15 Band 16 Band 17 Band 18 Band 19 -5.217 Band 20 Band 21 Band 22 Band 23 Band 24 Band 25 Band 26 Band 27 Band 28 Band 29 Band 30 Band 31 Band 32 Band 33 Band 34 Band 35 Band 36 Canonical scores of group means 1 Green -0.620 Non 0.596 78 1.2 m data inclusive discriminant analysis Forward stepwise with Alpha-to-Enter=0.150 Group frequencies Green Non 24 25 Group means Green Non Band 1 295.458 338.880 Band 2 388.125 442.960 Band 3 437.583 500.640 Band 4 480.708 514.400 Band 5 485.833 504.680 Band 6 532.542 566.160 Band 7 702.625 756.640 Band 8 891.500 954.000 Band 9 973.333 1037.160 Band 10 883.375 936.400 Band 11 789.083 826.440 Band 12 761.292 798.720 Band 13 706.000 736.640 Band 14 670.167 697.480 Band 15 643.375 672.520 Band 16 571.333 592.880 Band 17 538.542 570.000 Band 18 695.792 727.680 Band 19 1593.792 1608.280 Band 20 2301.333 2355.600 Band 21 3506.875 3604.720 Band 22 3728.125 3898.720 Band 23 2888.125 3019.480 Band 24 3757.708 3997.120 Band 25 3504.125 3756.840 Band 26 3007.958 3259.840 Band 27 2888.875 3152.800 Band 28 3045.333 3364.560 Band 29 2844.542 3164.440 Band 30 2746.833 3076.200 Band 31 2391.667 2692.400 Band 32 1859.625 2089.240 Classification functions Green Non CONSTANT -0.693 -0.693 V a r i a b l e F-to-remove Tolerance ix4 and Alpha-to-Remove=0.150 V a r i a b l e F-to-enter Tolerance 6 Band 1 2.53 1.000000 7 Band 2 2.59 1.000000 8 Band 3 2.73 1.000000 9 Band 4 0.80 1.000000 10 Band 5 0.24 1.000000 11 Band 6 0.57 1.000000 12 Band 7 0.65 1.000000 13 Band 8 0.49 1.000000 79 14 Band 9 0 .39 1 .000000 15 Band 10 0 .31 1 .000000 16 Band 11 0 .18 1 .000000 17 Band 12 0 .19 1 .000000 18 Band 13 0 .15 1 .000000 19 Band 14 0 .13 1 .000000 20 Band 15 0 .15 1 .000000 21 Band 16 0 .10 1 .000000 22 Band 17 0 .22 1 .000000 23 Band 18 0 .15 1 .000000 24 Band 19 0 .01 1 .000000 25 Band 20 0 .06 1 .000000 26 Band 21 0 .09 1 .000000 27 Band 22 0 .25 1 .000000 28 Band 23 • 0 .25 1 .000000 29 Band 24 0. 46 1. 000000 30 Band 25 0 .57 1 .000000 31 Band 26 0 .73 1 .000000 32 Band 27 0 . 84 1 .000000 33 Band 28 1 .04 1 .000000 34 Band 29 1 . 14 1 .000000 35 Band 30 1 .20 1 .000000 36 Band 31 1 .27 1 .000000 37 Band 32 1 . 17 1 .000000 Between groups F-matrix - df= 1 47 Green Non Green 0.000 Non 2.729 0.000 Wilks' lambda Lambda = 0.9451 df= 1 1 47 Approx. F= 2.7293 df= 1 47 prob = V a r i a b l e F-to-remove Tolerance | 8 Band 3 2.73 1.000000 | 0.1052 V a r i a b l e F-to-enter Tolerance 6 Band 1 0 05 0 187408 7 Band 2 0 03 0 122936 9 Band 4 1 52 0 203219 10 Band 5 4 92 0 187444 11 Band 6 2 33 0 203859 12 Band 7 0 77 0 343096 13 Band 8 0 99 0 360882 14 Band 9 1 20 0 365938 15 Band 10 1 52 0 359766 16 Band 11 2 09 0 356474 17 Band 12 2 07 0 353246 18 Band 13 2 25 0 359430 19 Band 14 2 45 0 355664 20 Band 15 2 22 0 361499 21 Band 16 2 46 0 368854 22 Band 17 1 69 0 378445 23 Band 18 2 04 0 380225 24 Band 19 3 02 0 421354 25 Band 20 1 89 0 461637 26 Band 21 1 36 0 504283 27 Band 22 0 73 0 521652 28 Band 23 0 84 0 499508 29 Band 24 0 46 0 494761 30 Band 25 0 35 0 485755 31 Band 26 0 22 0 475115 32 Band 27 0 17 0 464491 33 Band 28 0 07 0 460685 34 Band 29 0 05 0 455114 35 Band 30 0 03 0 452870 36 Band 31 0 02 0 451728 80 I 37 Band 32 0.05 0.445110 Between groups F-matrix - df = 2 46 Green Non Green 0.000 Non 3.939 0.000 Wilks' lambda Lambda = 0.8538 df= 2 1 47 Approx. F= 3.9387 df = 2 46 prob = 0.0264 V a r i a b l e F-to-remove Tolerance 8 Band 3 7.60 0.187444 10 Band 5 4.92 0.187444 V a r i a b l e F-to-enter Tolerance 6 Band 1 0 11 0 186697 7 Band 2 0 19 0 120826 9 Band 4 2 49 0 040596 11 Band 6 3 17 0 019311 12 Band 7 1 37 0 144685 13 Band 8 0 89 0 158451 14 Band 9 0 60 0 164295 15 Band 10 0 35 0 159934 16 Band 11 0 11 0 152666 17 Band 12 0 12 0 149896 18 Band 13 0 06 0 154874 19 Band 14 0 02 0 153695 20 Band 15 0 06 0 157514 21 Band 16 0 01 0 166014 22 Band 17 0 21 0 173378 23 Band 18 0 11 0 165894 24 Band 19 0 02 0 192630 25 Band 20 0 02 0 253376 26 Band 21 0 05 0 314613 27 Band 22 0 28 0 339499 28 Band 23 0 22 0 322777 29 Band 24 0 55 0 320770 30 Band 25 0 73 0 313413 31 Band 26 0 99 0 305947 32 Band 27 1 11 0 300920 33 Band 28 1 46 0 299986 34 Band 29 1 65 0 294784 35 Band 30 1 72 0 296054 36 Band 31 1 86 0 293976 37 Band 32 1 63 0 290076 Between groups F-matrix -- df = 3 45 Green Non Green 0.000 Non 3.806 0.000 Wilks' lambda Lambda = 0.7976 df= 3 1 47 Approx. F= 3.8060 df= 3 45 prob =0.0163 V a r i a b l e F-to-remove Tolerance 8 Band 3 6.95 0.187435 10 Band 5 5.75 0.017756 11 Band -6 3.17 0.019311 V a r i a b l e F-to-enter Tolerance 6 Band 1 0 05 0 167979 7 Band 2 0 05 0 119085 9 Band 4 1 40 0 038867 12 Band 7 0 10 0 108366 13 Band 8 0 00 0 112454 14 Band 9 0 05 0 115617 15 Band 10 0 16 0 114338 16 Band 11 0 39 0 114031 17 Band 12 0 28 0 116656 18 Band 13 0 30 0 126118 19 Band 14 0 33 0 129716 20 Band 15 0 14 0 138194 81 21 B a n d 16 0 16 0 152564 22 B a n d 17 0 02 0 167742 23 B a n d 18 0 06 0 148159 24 B a n d 19 0 94 0 156627 25 B a n d 20 0 30 0 217220 26 B a n d 21 0 11 0 283142 27 B a n d 22 0 00 0 307309 28 B a n d 23 0 01 0 291389 29 B a n d 24 0 02 0 284567 30 B a n d 25 0 05 0 274034 31 B a n d 26 0 11 0 263011 32 B a n d 27 0 13 0 252894 33 B a n d 28 0 24 0 247441 34 B a n d 29 0 28 0 237063 35 Band 30 0 29 0 234563 36 B a n d 31 0 32 0 228058 37 Band 32 0 20 0 221411 Classification matrix (cases in row categories classified into columns) Green Non %correct Green 16 8 67 Non 9 16 64 Total 25 24 65 Jackknifed classification matrix Green Non %correct Green 15 9 63 Non 9 16 64 Total 24 25 63 Classification matrix (cases in row categories classified into columns) Green Non %correct Green 16 8 67 Non 9 16 64 Total 25 24 65 Jackknifed classification matrix Green Non %correct Green 15 9 63 Non 9 16 64 Total 24 25 63 Eigenvalues 10.254 I Canonical correlations 10.450 | Cumulative proportion of total dispersion 11.000 | Wilks' lambda= 0.798 Approx.F= 3.806 df= 3, 45 p-tail= 0.0163 Pillai's trace= 0.202 Approx.F= 3.806 df= 3, 45 p-tail= 0.0163 Lawley-Hotelling trace= 0.254 Approx.F= 3.806 df= 3, 45 p-tail= 0.0163 Canonical discriminant functions 1 Constant -0.423 82 Band 1 Band 2 Band 3 0.014 Band 4 Band 5 -0.042 Band 6 0.026 Band 7 Band 8 Band 9 Band 10 Band 11 Band 12 Band 13 Band 14 Band 15 Band 16 Band 17 Band 18 Band 19 Band 20 Band 21 Band 22 Band 23 Band 24 Band 25 Band 26 Band 27 Band 28 Band 29 Band 30 Band 31 Band 32 Canonical discriminant functions — standardized by within variances 1 Band 1 Band 2 Band 3 1.878 Band 4 Band 5 -5.614 Band 6 4.103 Band 7 Band 8 Band 9 Band 10 Band 11 Band 12 Band 13 Band 14 Band 15 Band 16 Band 17 Band 18 Band 19 Band 20 Band 21 Band 22 Band 23 Band 24 Band 25 Band 26 83 Band 27 Band 28 Band 29 Band 30 Band 31 Band 32 Canonical scores of group means 1 Green -0.504 Non 0.483 

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.831.1-0090048/manifest

Comment

Related Items