"Forestry, Faculty of"@en . "DSpace"@en . "UBCV"@en . "Coggins, Sam"@en . "2011-01-13T15:39:38Z"@en . "2011"@en . "Doctor of Philosophy - PhD"@en . "University of British Columbia"@en . "Bark beetle infestations in western Canada have caused damage at previously unrecorded levels. Conventional forest health surveys are conducted to collect information on these infestations; however, due to the widespread nature of attack digital remote sensing technologies have the potential to offer new methods to augment forest inventories. This thesis will investigate the utility of remotely sensed data to detect and monitor insect infestations and provide innovative approaches to determine forest health information. In the first section of the thesis the accuracies of conventional forest health surveys were reviewed and assessed in a series of plots at the edge of the infestation. Mitigation levels were shown to be 43%, which was inadequate to stop a doubling expansion rate. A review of the detection rates of digital remote sensing was also conducted and used in a simple expansion model to assess the capacity of digital techniques. In the second part of the thesis a series of innovative methods were applied over a hierarchy of remotely sensed data sets. Attacked trees identified during field surveys were delineated on fine scale imagery with an accuracy of 80.2%. From these delineations, tree [stem diameter (r = 0.71, p <0.001)] and stand level [stocking density (r = 0.95, p <0.001)] information was accurately predicted and used to initiate an infestation spread model. Using this technique, an adaptive cluster sampling approach was applied in an innovative way to develop regional estimates of infestations. A relative efficiency estimator confirmed the adaptive approach was twice as efficient as conventional sampling schemes. With confidence in the approach, adaptive cluster sampling was applied to consecutive annual images determining a doubling infestation rate. Finally, an advanced remote sensing model was applied to stratify the landscape based on predictions of stocking and crown size, to predict the susceptibility of attack over the study area. Ultimately, this research successfully used a hierarchy of remotely sensed data to provide forest health and inventory information at a variety of scales from individual tree to stands and regions, which can augment existing forestry databases."@en . "https://circle.library.ubc.ca/rest/handle/2429/30610?expand=metadata"@en . " i INTEGRATION OF MULTI-SOURCE, MULTI-SCALE REMOTELY SENSED IMAGERY WITH GROUND SURVEY INFORMATION TO PROVIDE FOREST HEALTH AND INVENTORY DATA by Sam Coggins BSc Forest Resources Management, University of British Columbia, Vancouver, 2006 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Forestry) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) January 2011 \u00C2\u00A9 Sam Coggins, 2011 ii ABSTRACT Bark beetle infestations in western Canada have caused damage at previously unrecorded levels. Conventional forest health surveys are conducted to collect information on these infestations; however, due to the widespread nature of attack digital remote sensing technologies have the potential to offer new methods to augment forest inventories. This thesis will investigate the utility of remotely sensed data to detect and monitor insect infestations and provide innovative approaches to determine forest health information. In the first section of the thesis the accuracies of conventional forest health surveys were reviewed and assessed in a series of plots at the edge of the infestation. Mitigation levels were shown to be 43%, which was inadequate to stop a doubling expansion rate. A review of the detection rates of digital remote sensing was also conducted and used in a simple expansion model to assess the capacity of digital techniques. In the second part of the thesis a series of innovative methods were applied over a hierarchy of remotely sensed data sets. Attacked trees identified during field surveys were delineated on fine scale imagery with an accuracy of 80.2%. From these delineations, tree [stem diameter (r = 0.71, p <0.001)] and stand level [stocking density (r = 0.95, p <0.001)] information was accurately predicted and used to initiate an infestation spread model. Using this technique, an adaptive cluster sampling approach was applied in an innovative way to develop regional estimates of infestations. A relative efficiency estimator confirmed the adaptive approach was twice as efficient as conventional sampling schemes. With confidence in the approach, adaptive cluster sampling was applied to consecutive annual images determining a doubling infestation rate. Finally, an advanced remote sensing model was applied to stratify the landscape based on predictions of stocking and crown size, to predict the susceptibility of attack over the study area. Ultimately, this research successfully used a hierarchy of remotely sensed data to provide forest health and inventory information at a variety of scales from individual tree to stands and regions, which can augment existing forestry databases. iii TABLE OF CONTENTS ABSTRACT ..................................................................................................................................... ii TABLE OF CONTENTS ................................................................................................................. iii LIST OF TABLES........................................................................................................................... vi LIST OF FIGURES ........................................................................................................................ vii ACKNOWLEDGEMENTS ............................................................................................................... x DEDICATION................................................................................................................................. xii CO-AUTHORSHIP STATEMENT................................................................................................. xiii 1 INTRODUCTION..................................................................................................................... 1 1.1 Background..................................................................................................................... 1 1.2 Overview of forest health................................................................................................ 2 1.3 Forest inventory .............................................................................................................. 3 1.4 Remote sensing in forest inventory ................................................................................ 7 1.5 Research focus............................................................................................................. 13 1.5.1 Research goals ........................................................................................................ 13 1.5.2 Objectives................................................................................................................. 14 1.5.3 Methodological approaches ..................................................................................... 15 1.5.4 Study area ................................................................................................................ 16 1.5.5 Digital aerial imagery................................................................................................ 22 1.5.6 Document structure .................................................................................................. 23 1.6 References ................................................................................................................... 27 2 COMPARING THE IMPACTS OF MITIGATION AND NON-MITIGATION ON MOUNTAIN PINE BEETLE POPULATIONS................................................................................ 32 2.1 Introduction ................................................................................................................... 32 2.1.1 The impact of mountain pine beetle infestations...................................................... 32 2.1.2 Objectives................................................................................................................. 35 2.2 Summary of mitigation techniques ............................................................................... 36 2.3 Methods ........................................................................................................................ 39 2.3.1 Study plots................................................................................................................ 39 2.3.2 Modelling .................................................................................................................. 41 2.4 Results .......................................................................................................................... 43 2.5 Discussion .................................................................................................................... 50 2.6 References ................................................................................................................... 56 3 LINKING SURVEY DETECTION ACCURACY WITH ABILITY TO MITIGATE POPULATIONS OF MOUNTAIN PINE BEETLE.......................................................................... 59 3.1 Introduction ................................................................................................................... 59 3.2 Objectives ..................................................................................................................... 66 3.3 Review of current methods for red attack detection ..................................................... 67 3.3.1 Regional scale.......................................................................................................... 67 3.3.2 Landscape scale ...................................................................................................... 69 3.3.3 Local scale ............................................................................................................... 70 3.4 Effects of mitigation on mountain pine beetle infestations ........................................... 72 3.5 Use of survey data to monitor mountain pine beetle infestations................................. 80 3.6 References ................................................................................................................... 83 iv 4 INITIALISATION OF AN INSECT INFESTATION SPREAD MODEL USING TREE STRUCTURE AND SPATIAL CHARACTERISTICS DERIVED FROM HIGH-SPATIAL RESOLUTION DIGITAL AERIAL IMAGERY................................................................................ 87 4.1 Introduction ................................................................................................................... 87 4.2 Materials and methods ................................................................................................. 91 4.2.1 Survey data .............................................................................................................. 91 4.2.2 Individual tree crown delineation.............................................................................. 95 4.2.3 Assessment of the crown delineation..................................................................... 100 4.2.4 Mountain pine beetle infestation simulations ......................................................... 103 4.3 Results ........................................................................................................................ 106 4.4 Discussion .................................................................................................................. 114 4.4.1 Information derived from ground surveys and digital aerial imagery ..................... 114 4.4.2 Issues affecting the tree crown delineation ............................................................ 115 4.4.3 Mountain pine beetle infestation simulations ......................................................... 117 4.5 References ................................................................................................................. 119 5 IMPROVEMENT OF LOW LEVEL BARK BEETLE DAMAGE ESTIMATES WITH ADAPTIVE CLUSTER SAMPLING............................................................................................. 124 5.1 Introduction ................................................................................................................. 124 5.1.1 Mountain pine beetle .............................................................................................. 124 5.1.2 Forest health monitoring......................................................................................... 125 5.1.3 Role for sampling ................................................................................................... 127 5.1.4 Objectives............................................................................................................... 128 5.2 Materials and methods ............................................................................................... 129 5.2.1 Phase 1: Individual tree crown delineation on 10 cm imagery............................... 129 5.2.2 Phase 2: Adaptive cluster sampling ....................................................................... 132 5.2.3 Phase 3: Non-adaptive approach........................................................................... 136 5.2.4 Phase 4: Relative efficiency ................................................................................... 138 5.3 Results ........................................................................................................................ 138 5.4 Discussion .................................................................................................................. 140 5.5 References ................................................................................................................. 145 6 ESTIMATES OF BARK BEETLE INFESTATION EXPANSION FACTORS WITH ADAPTIVE CLUSTER SAMPLING............................................................................................. 148 6.1 Introduction ................................................................................................................. 148 6.1.1 Global forest resource ............................................................................................ 148 6.1.2 Forest insects ......................................................................................................... 148 6.1.3 Sampling to detect and quantify insect infestations ............................................... 150 6.1.4 Objectives............................................................................................................... 152 6.2 Methods ...................................................................................................................... 153 6.2.1 Study area .............................................................................................................. 153 6.2.2 Tree crown delineation ........................................................................................... 154 6.2.3 Adaptive cluster sampling ...................................................................................... 156 6.2.4 Sensitivity analysis ................................................................................................. 161 6.3 Results ........................................................................................................................ 161 6.4 Discussion .................................................................................................................. 164 6.5 References ................................................................................................................. 171 7 AUGMENTING FOREST HEALTH AND INVENTORY DATA WITH ATTRIBUTES DERIVED FROM LANDSAT THEMATIC MAPPER IMAGERY WITH ADVANCED REMOTE SENSING MODELS..................................................................................................................... 174 7.1 Introduction ................................................................................................................. 174 7.1.1 Forest inventory...................................................................................................... 174 7.1.2 Remote sensing in forest inventory........................................................................ 176 7.1.3 Geometric optical modelling................................................................................... 177 v 7.1.4 Aims........................................................................................................................ 179 7.2 Material and methods ................................................................................................. 180 7.2.1 Remotely sensed data............................................................................................ 180 7.2.2 Susceptibility modelling .......................................................................................... 181 7.2.3 Application of the geometric optical model............................................................. 182 7.2.4 Model validation ..................................................................................................... 184 7.2.5 Susceptibility model development.......................................................................... 184 7.3 Results ........................................................................................................................ 186 7.4 Discussion .................................................................................................................. 195 7.5 References ................................................................................................................. 201 8 CONCLUSION .................................................................................................................... 205 8.1 Summary .................................................................................................................... 205 8.2 To investigate the implications of varying survey detection accuracies on mitigating and bringing under control mountain pine beetle populations................................................. 206 8.3 To assess whether a simple insect infestation spread model can be parameterised using tree structure and spatial characteristics derived from high spatial resolution digital aerial imagery. ................................................................................................................................... 210 8.4 To demonstrate how current forest inventory data can be augmented with estimates supplied by innovative sampling techniques and advanced remote sensing models ............. 211 8.5 Future research .......................................................................................................... 215 8.6 Recommendations...................................................................................................... 217 8.7 References ................................................................................................................. 220 vi LIST OF TABLES Table 2.1. Stand characteristics and measurements within the study sites. ...................................................40 Table 2.2. Estimated expansion factors in 2007 and 2008 for each non-mitigated (top) and mitigated plot (bottom), where GA = green attack, RA = red attack, and GR = grey attack. ..............................45 Table 2.3: Mitigation efficacy determined from field observations in the mitigated plots, where GA = green attack and RA = red attack. ........................................................................................................47 Table 2.4: The number of years (t) required to suppress infested stands of mountain pine beetle using a range of initial infestation (N0) and a range of detection accuracies where the number of red attack trees was derived from those remaining after mitigation and the expected infestation after 10 years in the non-mitigated plots. .....................................................................................................49 Table 3.1: Characteristics associated with the population states of mountain pine beetle and the likely rate of population expansion associated with each population state (adapted from Wulder et al. 2006b). .................................................................................................................................................61 Table 3.2: Data sources for remote sensing of mountain pine beetle red attack mapping including sources and detection accuracies..........................................................................................................68 Table 3.3: Number of years (t) required to suppress infested stands of mountain pine beetle using a range of initial infestation (N0) and a range of detection accuracies representative of the survey methods discussed................................................................................................................................75 Table 4.1: Summary of ground survey data collected from twenty-three variable radius plots during the summers of 2006 and 2007. ...........................................................................................................93 Table 4.2: Assessment criteria of mountain pine beetle infestation status on individual lodgepole pine trees. .....................................................................................................................................................94 Table 4.3: Summary of data derived from digital aerial imagery, which corresponds to ground survey data.....................................................................................................................................................108 Table 4.4: Number of trees infested by mountain pine beetles over a 10 year period in each simulation...........................................................................................................................................114 Table 5.1: A summary of input variables and estimates provided by adaptive cluster sampling and the non-adaptive approach. ......................................................................................................................139 Table 6.1: Inputs and estimated variables provided by adaptive cluster sampling in 2007 and 2008 for Sites A and B. ....................................................................................................................................162 Table 7.1: Attributes required in forest inventories and susceptibility models with similar attributes provided by geometric optical models. ..............................................................................................179 Table 7.2: Field data input for the geometric optical model, vertical crown height, stand density, crown radius, and tree height. ............................................................................................................183 Table 7.3: Variation for the model output layers, measured by mean (\u00C2\u00B1 standard deviation), minimum and maximum values. ........................................................................................................................187 Table 7.4: Comparison of susceptibility estimated by forest inventory data and by geometric optical models given as the percent of the forest land base. ..........................................................................194 vii LIST OF FIGURES Figure 1.1: The location of the study area in western Canada, with surrounding townships, cities, and road network defined. ..........................................................................................................................17 Figure 1.2: The sample plots within the focus area, plots 3, 7. 8, and 9 were utilized in the study as access to these areas was easiest. .........................................................................................................19 Figure 1.3: The linkages between ground data and a hierarchy of multi-source, multi-scale remotely sensed imagery to determine the severity and extent of insect infestations. The thesis is separated into two sections, the first discusses current practices (Chapter 2) and demonstrates how remotely sensed data provide additional information to guide these surveys (Chapter 3). The second section first locates and then determines the presence of attacked trees. It then creates relationships between ground data and high spatial resolution imagery (Chapter 4). Estimates of beetle attack are the extrapolated to the per hectare level (Chapters 5 and 6), and finally these are used for stratification at the landscape level (Chapter 7). ..........................................24 Figure 2.1: The number of infested trees expected to be present without mitigation using the number of infested trees in the non-mitigated plots to initiate the model and using expansion factors derived from field-based observations in 2007 and 2008, with an expansion factor of 2 used in 2009 onwards. The average infestation per year is shown by the thick black line, with the range of infestation that could be expected in each year in dark grey, and the minimum and maximum amount of infestation shown in light grey (after Carroll et al. 2006)...................................................46 Figure 2.2: The minimum proportion of infested trees requiring treatment during mitigation can be calculated for the average, minimum, and maximum expansion factors in the non-mitigated plots, and for the average mitigation efficacy in the mitigated plots. The minimum proportion of infested trees requiring treatment to control an infestation expansion of 5.1 (the average expansion factor calculated) is approximately 80%. The maximum expansion rate was 18 which required a minimum of 93% of the infested trees to be treated to provide control. Infestation in the mitigated plots requires a minimum of 10% of infested trees to be treated to control current levels of infestation. The average mitigation efficacy of 43% is ineffective at controlling a doubling population (after Carroll et al. 2006). ...................................................................................48 Figure 2.3: Number of years required to suppress infestations of mountain pine beetle using the number of red attack trees was derived from those remaining after mitigation and the expected infestation after 10 years in the non-mitigated plots. The average number of years required to complete mitigation is shown by the thick black line, with the range of infestation that could be expected in each year in dark grey, and the minimum and maximum amount of infestation shown in light grey (after Carroll et al. 2006)......................................................................................49 Figure 3.1: Prediction of the number of trees killed using a range of detection accuracies. .........................76 Figure 3.2: Generalized prediction of the effect of detection accuracies given by two detection accuracies provided by recent research; 92% and 71%........................................................................77 Figure 3.3: Number of years required to suppress infestations of mountain pine beetle using detection accuracies (\u00C2\u00B1 known error) generated from remotely sensed and conventional data sources (after Carroll et al. 2006). ..............................................................................................................................80 Figure 4.1: Processing steps taken to complete the object-based classification to delineate individual tree crowns on digital aerial imagery. Processing involved: 1) Differentiation of forest and non- tree vegetation, bare ground, and roads; 2) Segmentation of forest and non-forest; 3) Classification of forest within the image; 4) Segmentation of forest to form individual tree crowns; and 5) Classification of forest to give polygons of individual tree crowns; and 6) Export of resultant polygons to a GIS..............................................................................................................97 Figure 4.2: Demonstration of the effects of iteratively changing the filter kernel size to identify tree crown edges prior to crown delineation on 10 cm spatial resolution digital aerial imagery. Filter settings used ranged from no filter (a), 5 x 5 filter (b), and 7 x 7 filter (c), and 11 x 11 filter (d). ......98 viii Figure 4.3: The relationship between crown area (m2) and stem diameter of lodgepole pine trees measured during ground surveys which was used to develop stem diameter measurements from crowns delineated on digital aerial imagery.......................................................................................102 Figure 4.4: Spatial distribution of random tree positions, Simulation 1 (a) and positions derived from digital aerial imagery, Simulations 2 and 3 (b). .................................................................................105 Figure 4.5: Stem diameter distributions of simulations 1 and 2 (a), using random number values for diameter, and the distribution of diameters derived from digital aerial imagery used in simulation 3 (b). .................................................................................................................................106 Figure 4.6: Comparison of crown area (m2) measurements derived from the field and digital aerial imagery using simple linear regression (r2 = 0.55, se = 1.19, p <0.001, n = 57, digital aerial image crown = 0.5208x + 1.4065). ....................................................................................................107 Figure 4.7: Comparison of stem diameters (cm) measured during ground surveys and stem diameters (cm) estimated from digital aerial imagery using simple linear regression (r2 = 0.51, se = 2.63, p <0.001 n = 23). The 1: 1 relationship between parameters is indicated by the dashed line. ..............109 Figure 4.8: Comparison of stocking density (stems per hectare) measured during ground surveys and stocking density (stems per hectare) estimated from digital aerial imagery using simple linear regression (r2 = 0.90, se = 506.65, p<0.001, n = 23). The 1: 1 relationship between parameters is indicated by the dashed line. ..............................................................................................................110 Figure 4.9: The scale of mountain pine beetle infestations over a 10 year period depicted by the number of trees infested. The sensitivity of the model to the quality of data entered is depicted as the difference in the number of trees infested in simulation 1 and 2 (solid line) and 3 (dashed line). ...................................................................................................................................................111 Figure 4.10: Output from simulation 1 illustrating the number of trees infested in each stage of attack. The simulation used randomized stem diameters and tree locations to predict the number of trees infested by mountain pine beetles over a 10 year period...........................................................112 Figure 4.11: Output from simulation 3 illustrating the number of trees infested in each stage of attack. The simulation used stem diameters and tree locations derived from the digital aerial imagery to predict the number of trees infested by mountain pine beetles over a 10 year period........................113 Figure 5.1: Flow chart of the crown delineation and individual tree object-based classification algorithm. ...........................................................................................................................................131 Figure 5.2: An example of an initial sample unit located within a grid square in an adaptive cluster sampling design (a). Additional sample units positioned at the cardinal directions of the initial sample unit (b). The final sample network (c) and the edge units which contain no instances of the object of interest (d). The presence of red attack in cells is indicated by RA. .............................133 Figure 5.3: The initial sample grid of 60 x 60 m overlaid on the digital remotely sensed imagery with the transect lines shown in solid colour (a). The randomly placed transect lines positioned within the sample area with the sample networks (b) and the resulting red attack crown delineation within the sample networks and transect lines (c). ..........................................................137 Figure 6.1: Progression of network in adaptive cluster sampling, starting with the initial sample plot overlaid on 2007 imagery (a), sample units are added at the cardinal directions (b), the final sample network (c) and the edge units which contain no mountain pine beetle attack (d). Finally, the 2008 imagery is examined and the network is extended to account for infestation spread (e). ...158 Figure 6.2: The 60 x 60 m sample grid (top) on which the adaptive cluster sampling was implemented in Site A. Five randomly placed transect lines are shown by black lines. The object-based classification defined each network placed along the transect lines in 2007 (black boxes). The transect lines were re-examined in 2008, new networks were established and old networks were extended to capture instances of mountain pine beetle attack. The 2008 networks are shown by the grey shading. ................................................................................................................................159 Figure 6.3: The mean number of infested trees/ha with confidence interval for each site in 2007 and 2008. ..................................................................................................................................................163 Figure 7.1: Endmembers from the SMACC algorithm that represent sunlit canopy (top left), sunlit background (top right), and shadowed (bottom) pixel fractions in the study area. ............................188 ix Figure 7.2: Output from the geometric optical model that represent biophysical attributes of the forest stands within Landsat pixels. .............................................................................................................189 Figure 7.3: The relationship between crown closure provided by forest inventory data and crown radius (r) estimated by the geometric optical model with standards error around the mean..............190 Figure 7.4: The relationship between stocking density (stems per hectare) provided by forest inventory data and stand density (treeness; m) estimated by the geometric optical model with standard errors around the mean. .......................................................................................................191 Figure 7.5: The relationship between quadratic mean diameter at breast height (cm) by forest inventory data and stand density (treeness; m) estimated by the geometric optical model with standard errors around the mean. .......................................................................................................192 Figure 7.6: Susceptibility of the forest to mountain pine beetle attack in the study area estimated from model output. .....................................................................................................................................193 x ACKNOWLEDGEMENTS The last four and a half years has been a rollercoaster ride where I started graduate studies, got married, got divorced, became a Canadian citizen, and wrote a PhD thesis. As such there are several family and friends who kept me on the straight and narrow and without whom this thesis would not have reached this stage. First and foremost I would like to acknowledge my family for being encouraging and always on the end of the phone, Skype, and Facebook when needed: Dad and Julie, Mum and Big-Joe, Josie, and Joe (and Lizzie). Secondly, I have a great network of friends in Vancouver who provide a welcome break from academia and help relieve stress with: trips to Elwoods, climbing, hiking, running, cycling, and snowboarding. Thanks to: Chris, Dave, Rachel, Caragh, Vanita, Athena, Dave and Heather, Babita, Amy, Aisling, all my other friends in the Forestry building, and Janelle for editing skills. Third, I want to thank everyone who has been in the lab over the course of my degree, you have made graduate studies more worthwhile and a very enjoyable experience. Fourth, I would like to thank the staff in the Faculty for their help, Gayle Kosh, Erika Helmerson, Tracey Teasdale, and Cindy Prescott. I would also like to thank my committee for help, guidance, and encouragement: Dr. Nicholas Coops, Dr. Mike Wulder, Dr. John Nelson, and Dr. John McLean. Next, I would like to thank Suzi Q for her care, attention, and tolerance in the last year, and especially during the final weeks before the thesis was handed in. Lastly, I would like to thank red wine and chocolate for helping to keep me awake during the early mornings and long nights. This research would not have been possible without funding, the following agencies have contributed: 1) the Government of Canada, through the Mountain Pine Beetle Program, a xi 6-year, $40 million program administered by Natural Resources Canada \u00E2\u0080\u0093 Canadian Forest Service; 2) the Canadian Forest Service, Pacific Forestry Centre Graduate Student Award to Sam Coggins, administered by Natural Resources Canada \u00E2\u0080\u0093 Canadian Forest Service; 3) a University Graduate Fellowship (UGF) award to Sam Coggins; and 4) several University of British Columbia Faculty of Forestry awards. Lastly, I thank the editors of the journals and from many anonymous reviewers who strengthened the manuscripts with helpful comments. xii DEDICATION To family and friends\u00E2\u0080\u00A6 thank you xiii CO-AUTHORSHIP STATEMENT This thesis consists of six scientific papers (with five published in peer-reviewed journals) of which I am the lead author. The initial project overview was proposed by my supervisor, Dr. Nicholas Coops. For the scientific journal submissions, I performed all the research, data analyses, and interpretation of the results, and prepared the final manuscripts. Co-authors provided advice on methodology and made editorial comments as required. 1 1 INTRODUCTION 1.1 Background The intent of this thesis is to investigate the utility of remotely sensed imagery to support insect monitoring and mitigation through the development of innovative methods to locate infestations and generate attributes analogous to those in a forest inventory to support models of susceptibility and spread. To demonstrate this utility remotely sensed imagery is used in a hierarchy, starting at the field plot level to predict common forest inventory attributes, then a novel sampling approach is used to generate per hectare level estimates of the number of attacked trees, and finally a technique is applied to stratify landscapes to predict the likelihood of infestations. This first chapter will describe the need for up-to-date forest inventory information in on-going monitoring activities, and will present roles and opportunities for remote sensing in operational inventory and forest health monitoring. The objectives and methodologies used, followed by an overview of the thesis structure, are also presented in this chapter. The need for forest inventory was first introduced to me when I was a forester in the United Kingdom where I was often required to undertake forest mensuration activities. Inventory on small forest estates is usually completed as short surveys to generate information on tree volume and forest stand value. I became acutely interested in forest health issues after arriving in British Columbia in 2002 when I completed field based research on bark beetles and I quickly became interested in the impact of forest insects on the forest resource as well as developing methods to control the spread of infestations. I have learnt that when considering issues around forest health it is important to first 2 understand how pathogens, insects and diseases can impact the forest resource and secondly the role of inventory to describe the affected areas in terms of estimates of the area killed, the impact on aesthetic values and timber volume. Combining these topics has been a rewarding learning experience that will hopefully provide additional measurement and monitoring opportunities to address my interests in forest heath and sustainability. 1.2 Overview of forest health Forest insects caused damage to 37 million hectares of forested land between 1998 and 2002, which represents approximately 1.4% of the worldwide forest cover (United Nations Food and Agriculture Organisation 2009). The effect of forest insects on the global forest resource is discussed in further detail in Chapter 6. Reporting on forest health indicators is commonly completed following monitoring by aircraft and/or by ground crews who recognize and report the geographic location of disturbances. The data generated from aircraft surveys are generally of a broad scale, providing rough locations of infested forest stands and some measure of the severity to provide annual guidance and predictions regarding jurisdiction-wide trends. This information is passed onto finer scale surveys, or to field crews, who work to prevent the spread of infestations. A full literature review of the techniques used is provided in Chapter 3. Typically, problems associated with these types of surveys include errors of commission (false positives) and omission (missed cases). With commission error the expense of a field survey is increased when crews are sent to investigate trees that are not infested. Furthermore, the extent of infestation is erroneously increased through inclusion of uninfested trees to the area 3 impacted by insects. Errors of omission fail to identify number of attacked trees, due to changes in foliage colour or the lack of damaged foliage required to detect infestations. Variations in the timing of insect feeding can create the impression that forest stands are healthy. Typically, foliage of attack trees fades gradually and approximately 90% of attacked trees will exhibit red foliage one year after attack (White et al. 2004). If foliage has not faded by the time surveys are implemented, trees can be reported to be healthy when in fact they are already infested. Another reason for errors of omission is that single infested trees are easily missed both from the air and on the ground, especially when foliage stress is pre-visual. Infestations often expand rapidly from these isolated, missed trees. Errors of omission especially impact forest health when conducting aerial surveys because the coarse scale monitoring guides finer scale surveys and there is no capacity to detect infested trees missed in large areas. These infested trees remain and contribute to the infestation. Accordingly, it is vital to identify infested trees as accurately as possible to properly guide finer scale surveys. 1.3 Forest inventory Forests are the most widely distributed ecosystem on Earth accounting for approximately 4 billion hectares or 30% of the land surface (United Nations Food and Agriculture Organization 2006). In comparison, British Columbia covers 95 million ha\u00E2\u0080\u0099s, almost two- thirds of which are forested. Forest management regimes seek to enhance and maintain the economic, environmental, social, and cultural benefits (Franklin 2001). Sustainable forest management not only recognises the benefit from economic values provided by timber sales or eco-tourism, but also from intrinsic values such as preservation of 4 biodiversity, increased soil stability, water quality, and spiritual values. In order to maintain services provided by these ecosystems, a fundamental understanding is required to preserve the forest resource for the benefit of current and future generations (Franklin 2001). Information to satisfy reporting for sustainable forest management is typically collected from monitoring programs and retained in inventories from which analyses on the state of forests can be performed (K\u00C3\u00B6hl et al. 2006). Inventories are typically performed at 3 levels, first at broad scale strategic, second at finer-scale tactical that satisfy landscape level reporting, and finally at operational stand level management (Anthony 1965). The level of inventory used typically depends on the information requirement by the end users (Lund 1998). Strategic level inventories are used to satisfy broad scale sustainable forest management objectives (Davis et al. 2001) and report on timber production, pests and diseases (Wulder et al. 2006), fire, preservation of biodiversity, and provide general information on the forest resource such as forest area and tree species (United Nations Food and Agricultural Organization 2006). Tactical level inventories determine when, and where, forest management activities should take place and supply data to support these decisions. These inventories are used to satisfy the broad scale objectives from the strategic level to report the amount of timber harvested and the placement of roads (Davis et al. 2001). They also record more specific information on forest stand and individual tree characteristics (K\u00C3\u00B6hl et al. 2006). Operational inventories are typically concerned with forest stand level information, for example, the total timber volume to be harvested in a block (Gunn 1991). The data collected in these inventories are compiled by the jurisdiction that requires the information, government agencies at the strategic level, and 5 by forest companies and land managers at the tactical and operational levels. The compiled data are archived in a central database and can be used to produce reports based on statistically sound practices that allow for repeatable, unbiased, precise, and credible estimates of forests. National level inventories are typically initiated by generating broad scale estimates over large areas with more detailed forest measurement providing a means of obtaining finer estimates for verification purposes at selected locations. Commonly, aerial photographs are used to stratify land into forest and non-forest areas and to define forest cover and type classes over the land base with many countries routinely acquiring aerial photographs to delineate forest stands and estimate forest attributes using photo interpretation. A brief review of forest inventories by country is provided by Tokola (2006a; 2006b; 2006c) and recognises that photographically derived interpreted estimates are used in North America, Asia, and Europe. Photo interpretation does not yield all the attributes required for a forest inventory however, and after large area stratification is complete, aerial photograph estimates are required to be verified and augmented with ground sampling (Husch 1971). Also, sampling estimates can be used to generate relationships between aerial photograph estimates and forest and tree measurements, to determine photograph attributes by proxy. Sampling provides estimates of a population within an area of interest, rather than a census survey and can be used to determine variables such as: forest area, forest cover, species, tree height, tree volume, forest condition, tree vigour, mortality, removals, trends, and forest health (K\u00C3\u00B6hl et al. 2006; see chapter 5 of this thesis). 6 A specific example of forest inventory that uses a combination of aerial photographs and sampling is the Vegetation Resources Inventory (VRI) in British Columbia. This inventory is specifically designed to answer a wide range of forest management needs and provides wall-to-wall coverage of the province (Leckie and Gillis 1995). Historically, inventory data has been collected in British Columbia since the early 1900\u00E2\u0080\u0099s, starting with rudimentary estimates of the forested area and the amount of timber on the land base. Over the last century inventory data collection and reporting has steadily improved, taking advantage of other technological developments such as, aerial photography and remote sensing. The forest inventory program that started as \u00E2\u0080\u0098forest reconnaissance surveys\u00E2\u0080\u0099 in British Columbia became the VRI in 1988 (Parminter 2000). The VRI is now in the 3rd period of renewal, and is updated every 10 years, with data collection starting in year 1 of the inventory cycle and measurements completed for 8 million hectares per year (Leckie and Gillis 1995). The main objectives are two-fold: 1) to determine where forest is; and 2) to ascertain how much resource is present (Sandvoss et al. 2005). To satisfy these objectives the inventory is completed in two phases: Phase 1: Aerial photographs are acquired over the entire province (94 million hectares) and used to classify land cover types. First, the boundaries of homogeneous forest stands are delineated as polygons. Second, vegetation cover attributes defined in the photograph strata are interpreted to satisfy land cover classification requirements (Sandvoss et al. 2005); Phase 2: Sample plots are randomly located within the stands delineated in the photograph plots. Forest stand and tree measurements are recorded within the plots and 7 used to adjust the Phase 1 estimates. This process is assumed to decrease the bias in photograph estimates and increase the precision within each stratum. Initial attributes estimated during Phase 1 and Phase 2 are projected over time to reduce the need to re-measure field plots and the costs of conducting the inventory. In particular, basal area, stand density (stems per hectare), height, DBH, stand age, and volume at various utilization levels can be projected over time by stand-specific models such as Variable Density Yield Projection. This model uses Phase 2 estimates to project a variety of inventory attributes that update the initial attributes (British Columbia Ministry of Forests 2009). The data generated by this inventory are available as a digital layer that allows integration and rapid analysis with a geographic information system to support tactical level management decisions. Typical management activities involve government level decision processes such as to determine timber supply analysis information for allowable annual cut allocations, or operational level planning to supply accurate information to determine harvesting priorities. 1.4 Remote sensing in forest inventory Digital remote sensing has the potential to augment and update inventories with accurate information about the forest resource (Leckie and Gillis 1995). Advanced remote sensing techniques offer a practical solution to generating data needed to provide a greater quantity of information about the forest resource with satellite instruments (Leckie 1990; McRoberts et al. 2010) and from airborne imagery (King 2000). Holmgren and Thuresson (1998) evaluate the use of satellite remote sensing imagery to extract forest inventory parameters information and describe several limitations to estimate forest 8 inventory data. They focus on technologies available at the time of writing and suggest data derived from imagery are not suitable to detect information for strategic, tactical, and operational inventories. However, the advent of finer scale imagery suggests technologies are improving and data can be used successfully to predict forest attributes (for example Hyde et al. 2006 and Chubey et al. 2006). Techniques to obtain data from digital imagery exhibit distinct advantages over aerial photo interpretation methods. Primarily, photo interpretation requires significant experience and expertise from staff. However, estimates may contain biases as interpretation can be subjective. Furthermore, interpreters are becoming harder to find as the skills used require considerable field experience coupled with the ability to determine forest characteristics on aerial photography (Morgan et al. 2010). Many forestry organisations and researchers recommend the need for new tools and methods to help evolve from aerial photo interpretation to satellite based estimates of forest variables (Leckie 1990; Leckie and Gillis 1995; Wulder 1998; Holopainen and Kalliovirta 2006; McRoberts and Tomppo 2007), and subsequently a number of inventory programs are beginning to incorporate tools to extract data from digital remotely sensed imagery (Tokola 2006a; 2006b; 2006c; Falkowski et al. 2009). Digital imagery has been used to generate forest information since the 1970\u00E2\u0080\u0099s, with the advent of the Landsat series of satellites in 1972 (Langley 1975) and have been shown to provide precise estimates of the forest resource (Wulder 1998) with predictions within acceptable bias and precision limits usually >75% (Hussin and Bijker 2000; McRoberts and Tomppo 2007) in otherwise inaccessible areas where observations would normally be impossible. Broad scale digital remote sensing imagery can produce predictions that provide the 9 location of forest resources, such as forest type and timber volume (Leckie and Gillis 1995), while finer scale information has the potential to estimate stand and tree level attributes. Data derived from digital sources can be immediately integrated with global positioning system data, geographic information systems, and modelling approaches (K\u00C3\u00B6hl et al. 2006) and the data can be analysed with non-subjective computer software such as object-based delineation (a full review of approaches are provided in Flanders et al. 2003; Bunting and Lucas 2006; Chapter 4 of this thesis) or advanced remote sensing techniques such as geometric optical models (Li and Strahler 1985; Chapter 7 of this thesis). These approaches enable automatic interpretation to acquire detailed information over very large areas, either by grouping pixels into objects (Bunting and Lucas 2006) or on a per pixel basis (Li and Strahler 1985). Therefore, digital remotely sensed imagery has been demonstrated to provide a cost efficient method to acquire data with savings of 12:1 when compared to aerial photographs (Franklin 2001). Remote sensing displays advantages over conventional data collection methods for forest inventory purposes. Typically, inventories are updated infrequently (once every 10 \u00E2\u0080\u0093 20 years) due to the costs and time required to complete large area data collection (Leckie 1990). Subsequently, decisions by land managers and policy makers can be subject to errors as data are not current. Updates supplied by conventional models rely on information collected at the beginning of the inventory cycle and on field work to provide calibration and validation. Within these time scales, forests are subject to additions and depletions that are not accounted for until re-inventory (Gillis 2001). Additions take the form of increases in stand volume and dimensions such as basal area and DBH, which occur as trees mature and grow. Conversely, depletions from timber harvesting or from 10 pests and disease should also be considered. Inventory cycles also do not account for advances in technology, in particular remote sensing technologies, which have developed considerably in the last 4 decades. Specific increases have been the advent of high spatial resolution1 sensors in the last 20 years which has enabled finer scale measurements (Wulder 1998; Wulder et al. 2004). When remotely sensed data are utilized it is important to choose an appropriate spatial resolution because this often defines the type and quality of information that may be extracted from an image (Wulder 1998; Wulder et al. 2004). Choosing the appropriate spatial resolution is a balancing act between information requirements, manageable file sizes (see Table 3 in Wulder 1998; Nelson et al. 2001), budget constraints (Leckie 1990; Wulder et al. 2006), and image extent. Generally, as resolution gets finer imagery is more expensive; but has greater detail and can supply rich spatial and contextual information (Leckie 1990). Medium-spatial resolution satellite imagery (i.e., pixels sided 10 \u00E2\u0080\u0093 100 m), such as that acquired from Landsat and SPOT (Satellite Pour l'Observation de la Terre) are unable to detect individual tree crowns because the pixel size is too large. A Landsat Thematic Mapper (TM) pixel size is 30 x 30 m, larger than a tree crown; however, this imagery captures information over large areas (185 x 185 km for Landsat). As a result this type of imagery is used in forest inventory applications in many countries in Europe (Tokola 2006a), Asia (Tokola 2006b), and North America (Tokola 2006c) to classify land cover type. High spatial resolution satellite imagery is a relatively new technology in the remote sensing arena. Imagery is currently available with spatial resolutions as fine as 0.46 m (panchromatic) and 1.84 m (multispectral; DigitalGlobe 2009). The area captured by this imagery is smaller than a Landsat scene (Quickbird for 1 Spatial resolution is defined as the smallest object discernable on an image (Lillesand et al. 2004). 11 example covers 16.5 x 16.5 km), but contains information-rich spatial data that can define small groups of tree crowns within an image. Very fine scale or very high-spatial resolution imagery such as digital aerial imagery (imagery acquired with a digital camera), is providing opportunities to partially replace aerial photography, and can have spatial resolutions as fine as 0.05 m (Jamie Heath, pers. comm2.); however, each scene covers only a few hectares. Once imagery is acquired sampling approaches can be applied to derive information and generate estimates of forest inventory variables (a full review is available in K\u00C3\u00B6hl et al. 2006). Measurements can be generated using regression, ratio estimators, or stratification techniques to extrapolate plot level measurements to strata defined on remote sensing imagery (McRoberts and Tomppo 2007). Ratio estimators assume a linear relationship between two attributes that show positive correlation (for example, crown area and DBH) and a sub sample of both attributes can be measured in field plots and data used to fit a relationship between measurements and this relationship is then applied across the image. Regression techniques are similar to ratio estimators as they also require a linear relationship between an attribute and an auxiliary variable. Stratification techniques first define strata over the landscape and then use sub-sampling to determine conditions within the strata. Considerable research has been devoted to ratio, regression, and stratification for mapping and describing relationships between spectral values and forest observations and includes predictions of forest cover, canopy cover, stand density, mean DBH, mean tree height, basal area, forest biomass, and tree volume [general reviews can be found in Wulder (1998) and McRoberts et al. (2002)]. Application of these techniques can be found in Roy et al. (1996), who used strata to define forests on Landsat TM imagery; 2 Jamie Heath, Terrasaurus Aerial Photography Ltd. June, 2007. 12 Franklin (1986) and Trotter et al. (1997) who used regression analysis with Landsat TM data; and in Danson (1987), who used ratio sampling with SPOT-1 data. In the case of ratio estimators and regression techniques, remotely sensed data are first stratified into classes such as forest/non-forest, urban/non-urban, or equivalent. Once stratification has been implemented, forest attributes are measured within the strata, and estimates of the mean of the populations, the variance, and confidence limits are extrapolated to the strata level (K\u00C3\u00B6hl et al. 2006). Advances in remote sensing technologies have also led to development of sophisticated models that allow predictions of variables from medium spatial resolution imagery (Strahler et al. 1986). Complex algorithms are used to determine estimates of crown dimensions, tree height, and stocking density (Li and Strahler 1985; Scarth and Phinn 2000) which can then also be converted to represent forest inventory variables such as DBH and stand age. In addition to collecting forest inventory variables, remotely sensed imagery can also be used to generate estimates for forest health surveys (Knight 1967). The role of remote sensing in the field of entomology has application in three distinct areas: 1) observation of the insects; 2) detection of the damage caused by insects; and 3) monitoring of environmental factors that affect insect behaviours (Riley 1989). Forest health data can be updated with observations derived from remotely sensed imagery as long as crown foliage in diseased or infested trees is easily distinguished from otherwise healthy trees. This thesis focuses on remote sensing to detect damage to trees caused by insect herbivory, principally on systems that observe foliage colour change following attack by bark beetles. Typically, changes in foliage colour are noticed as beetles burrow into the tree stem and disrupt the nutrient supply to the leaves in the crown. The leaves begin to 13 desiccate and undergo pigment loss which results in fading of the needles from green (healthy) through to red (Hill et al. 1967). In the case of generating forest health data, Landsat imagery can record large infestations (stand level), while high spatial resolution imagery (such as WorldView-2) is better suited to delineating individual tree crowns. Very high spatial resolution imagery can, however, be used to identify individual trees from which forest inventory estimates can be derived and this information has the potential to augment inventory data with forest health information (Coops et al. 2006). As a result, medium spatial resolution imagery could supply estimates on the extent and severity of attack at the stand level while high spatial resolution satellite imagery has utility in estimating finer scale attributes and very high spatial resolution digital aerial imagery can be used to estimate individual tree variables over small specific locations. Using this hierarchical approach, information can be estimated on the number of infested trees, the proportion of the landscape affected, when mortality occurred, and the extent of affected trees (Wulder et al. 2006), as well as individual tree attributes such as estimates of DBH, and stand density. Furthermore, if imagery is captured in a time series over the same region it can also be used in a monitoring approach to determine the rate of spread over a number of years (Skakun et al. 2003). 1.5 Research focus 1.5.1 Research goals This thesis investigates the applicability of a hierarchy of optical remotely sensed imagery to monitor insect infestations and predict a range of forest inventory variables. 14 Estimates of individual tree and forest stand variables, particularly those related to forest health, will be scaled from individual plots to forest stands and across broad landscapes. Traditionally in remote sensing research, a single sensor is commonly used to predict forest attributes which are then validated against measurements provided by aerial photo interpreters or field data (Woodcock and Strahler 1987). However, few studies have used multiple sensors to achieve accurate forest inventory estimates. Scaling with remotely sensed imagery has the potential to predict forest health variables and report on the spread of insect infestations, first from field measurements, then to high spatial resolution imagery, and finally to large areas using medium spatial resolution imagery. Recent advances in remote sensing technologies have the potential to provide data sources which, through implementation of a sampling approach, may increase the precision of forest inventory estimates and demonstrate the compatibility of multi-scale, multi-source data sets. In western Canada remotely sensed data have significant application in detection and mapping of insect infestations, which require accurate timely data to report the spread and extent of attack in large areas of remote and inaccessible forests. 1.5.2 Objectives Based on the goals of this thesis the following objectives have been outlined: 1) to investigate the implications of varying survey detection accuracies (derived from conventional surveys as well as a range of remotely sensed datasets) on mitigating and bringing under control mountain pine beetle populations (Chapters 2 and 3). 15 2) to assess whether a simple insect infestation spread model can be parameterised using tree structure and spatial characteristics derived from high spatial resolution digital aerial imagery (Chapters 4). 3) to demonstrate how current forest inventory data can be augmented with estimates supplied by innovative sampling techniques and advanced remote sensing models (Chapters 5, 6, and 7). 1.5.3 Methodological approaches This thesis provides evidence of the utility of remotely sensed data to detect mountain pine beetle infestations, to provide accurate geographic locations of infested trees, and estimate the severity of attack over very large areas. Based on the objectives above, the methods used in this thesis will: 1) demonstrate that survey detection accuracies derived from remotely sensed data and applied to mitigation activities have the potential to control mountain pine beetle infestations. 2) develop techniques to help assess the impact and severity of infestations and demonstrate the capacity of remotely sensed derived data to model infestations over the landscape. 3) predict both forest stand, and individual tree, attributes from remotely sensed data, validate these estimates against field data and utilise these predictions to parameterize and initiate infestation spread models, 4) Propose, apply, and validate a technique to stratify large area remotely sensed imagery to map the susceptibility of forests to mountain pine beetle attack. 16 Ultimately, these tools can complement conventional survey techniques with information supplied by up-to-date approaches in detection and monitoring programs to aid decisions to prioritise areas and enable efficient resource allocation to control infestations. 1.5.4 Study area 1.5.4.1 Location The focus area is located in western Canada, spanning the border between British Columbia and Alberta, and centered on 54\u00C2\u00B038\u00E2\u0080\u0099 N, 120\u00C2\u00B041\u00E2\u0080\u0099 W. On the British Columbia side it covers approximately 6400 km2 in the Northern Interior Forest Region and is within the eastern portion of the Prince George forest district, and the southern part of the Fort St. John forest district. The land in this area is managed under the Dawson Creek Timber Supply Area. In Alberta, the land is managed under the northwest forest management unit. Twenty percent of the focus area is classed as Provincial Park with Kakwa Park spanning the border between British Columbia and Alberta, known as Kakwa Provincial Park and Kakwa Wildlands Park in each province, respectively. Secondly, the focus area also contains Wilmore Wilderness Park in Alberta which is situated in the south. Several townships and cities are in close proximity, although none are directly within the focus area. In British Columbia, Tumbler Ridge and Dawson Creek are situated to the north, McBride to the south, and Prince George to the west. In Alberta, Grande Prairie is located to the east of the site, with Grande Cache to the south (Figure 1.1). The focus area was primarily chosen due to its proximity to mountain pine beetle attack in British Columbia and increases in observed spot infestations in Alberta. The potential 17 for mountain pine beetle attack to spread into this area has increased as winter weather has become warmer allowing greater brood survival rates as beetles overwinter beneath the bark of trees (Carroll et al. 2004). The forest stands represent high value, productive forest lands, typical of those found on the border of British Columbia and Alberta (Wilson 2004) and are also considered highly susceptible to mountain pine beetle attack due to the abundance of susceptible host trees (Carroll et al. 2004). The area was selected in 2006, when infestation levels were relatively low in the area, although attack had been observed spreading eastwards, and groups of trees had been attacked to the north of the study area (Westfall 2007). Figure 1.1: The location of the study area in western Canada, with surrounding townships, cities, and road network defined. 18 Nine sample plots were established within the focus area from which field measurements were recorded on individual trees and forest stands. The sample plots covered an area of 64km2 each, approximately the same extent as a high spatial resolution digital remotely sensed satellite image. Plots were chosen that share similar attributes, slope, aspect, mean stem diameter, stocking density, and proportion of host trees. The sample plots were positioned in areas deemed susceptible to mountain pine beetle, where attack was determined to be more likely in stands with a high proportion of lodgepole pine (Pinus contorta Dougl. ex. Loud var. latifolia Engl.), in stands with a mean diameter at breast height greater than 15 cm at breast height, greater than 80 years old, growing at a density between 750 to 1500 trees per ha, and at lower elevations (Shore and Safranyik 1992). To identify areas suitable for sample plot placement, a pine mask was generated from forest inventory data for Alberta and British Columbia and stands considered susceptible if forest greater than 50% pine. Using a digital elevation model, areas at less than 1500 m elevation were determined to be acceptable for sample site selection. Six of the sites were positioned in valleys along the western portion of the focus area, which based on beetle pressure from the southwest and northwest and were highly likely to be attacked. The remaining plots were positioned in the parks, as no infestation had been recorded in these areas as of 2006 (Figure 1.2). 1.5.4.2 Ecology The focus area is situated around the Rocky Mountains and consists of high-elevation mountainous regions, mid-elevation forests, and valley basins. The British Columbia Biogeoclimatic Ecosystem Classification (BEC) system defines 4 zones in the focus area, Boreal White and Boreal Spruce (BWBS), Englemann Spruce-Subalpine Fir (ESSF), 19 Boreal Altai Fescue Alpine (BAFA), and Interior Mountain Heather Alpine (IMA). The BWBS zone is indicative of low lying topography in valley bottoms, existing below the ESSF. The climate in this zone is typically cold, with long winters and a short Figure 1.2: The sample plots within the focus area, plots 3, 7. 8, and 9 were utilized in the study as access to these areas was easiest. growing season. Annual precipitation ranges between 330 to 570 mm with a mean temperature of -2.9\u00C2\u00B0C to -2\u00C2\u00B0C (Meidinger and Pojar 1991). The BWBS is represented by a single subzone in the focus area, BWBSwk, wet cool. These forests consist of black spruce (Picea mariana (Mill.) BSP), white spruce (Picea glauca (Monech) Voss), and subalpine fir (Abies lasiocarpa (Hook.) Nutt). The ESSF signifies high elevation forest 20 (900 \u00E2\u0080\u0093 1700 m) and typically experiences long cold winters with deep snowpacks. Annual precipitation can exceed 2000 mm, up to 70% of which falls as snow. The mean annual temperature ranges between -1\u00C2\u00B0C and +2\u00C2\u00B0C. The forests typically consist of Engelmann spruce (Picea engelmannii Parry ex. Engelm.) and lodgepole pine, both of which regenerate freely in this zone (Meidinger and Pojar 1991). Five subzones are present in the focus area including two wet: wet cool, wet cool parkland, and three moist: moist very cold, moist mild, and moist mild parkland. The remaining biogeoclimatic zones are former Alpine Tundra zones occurring at high elevations and experience harsh conditions with temperatures remaining below 0\u00C2\u00B0C for 7 to 11 months of the year, and the majority of the precipitation falling as snow. Most of the ground cover are heathers, grass, and alpine shrubs (Meidinger and Pojar 1991). Overall, the lodgepole pine in the area naturally regenerated following fires in the early part of the 20th Century, which resulted in even-aged, pine dominated stands that grow to uniform dimensions (Moir 1965) as is typical of many lodgepole pine forests. In western Canada historical information suggests fires were the dominant natural disturbance in the forest resulting in large burned areas that were naturally repopulated with lodgepole pine (Logan and Powell 2001). 1.5.4.3 Field work The sample plots were first visited in 2006, 2007 and 2008 to determine presence of mountain pine beetle attack and measure calibration and validation data from trees and forests stands. Field work was conducted in late August of each year, after attacking beetles had dispersed to assess the level of new attack each year. Four of the nine plots were visited over the course of the study: 3 and 7 in Alberta, and 8 and 9 in British 21 Columbia. Field data were not collected from the remaining sample plots due to lack of access and distance from other sites. Admission to Kakwa Wildland Park (plot 3) was granted for research purposes by Alberta Parks and required travel by helicopter, whereas plots 7, 8, and 9 were accessed by road. These plots offered an opportunity to observe low level mountain pine beetle damage and study the impact and expansion of infestations. The sample plots also offered the opportunity to examine the effects of mitigation tactics. In Alberta, infested trees had been removed since 2005 in aggressive mitigation programs enforced by the province to control mountain pine beetle outbreaks at the leading edge (Ono 2004). Conversely, in British Columbia the sample plots had not been subject to recent mitigation and infestations were observed to be expanding rapidly. In 2006, the plots in British Columbia were considered to be within the edge of the infestation, and field observations determined spot infestations were expanding. In Alberta the plots were at the edge of the infestation with attack limited to one or two infested trees. Beetles life stages excavated from beneath the bark in late-August were typically under-developed and it was presumed cold winter weather would cause mortality to life stages that had not reached the 3rd instar. Beetle larvae need to develop to at least the 3rd instar as this lifestage offers high glycerol reserves to survive winter temperatures (Wygant 1940). As such, in 2006 high mortality during the winter is assumed to have kept beetle populations low and prevent further infestation. However, in the following years the number of trees attacked and killed by beetles had increased and by 2008 large areas of forest in the sample plots experienced mortality (Westfall and Ebata 2009). 22 1.5.5 Digital aerial imagery To meet the aims of this study very high spatial resolution digital airborne imagery was used to detect and locate mountain pine beetle infestations. Imagery was acquired over forests in the sample plots with a Canon EOS-1Ds Mark II camera, fitted with an f1.8 Canon lens and a Bayer pattern filter, mounted on a fixed wing aircraft. The camera uses a complementary metal\u00E2\u0080\u0093oxide\u00E2\u0080\u0093semiconductor (CMOS) sensor which provides an effective resolution of 16.7 megapixels. To estimate the expansion of insect infestations, imagery must be acquired in two time-steps, ideally a year apart, and under similar viewing conditions, at the same time of year. Imagery was acquired near-nadir during August, 2006 and 2007 from a flying height of 1100 m, producing imagery with a 10 cm spatial resolution, and used for the research described in Chapters 2, 4 and 5. Each 10 cm spatial resolution image covered an area of approximately 0.14 km (0.44 x 0.31 km or 4850 x 3110 pixels). Imagery was also acquired during August 2008 from a height of 2200 m, with a focal length of 85 mm, to provide 20 cm spatial resolution, and used in Chapter 6 of this thesis. Relative to the 10 cm imagery, the 20 cm spatial resolution images were acquired to provide larger areal coverage to study mountain pine beetle expansion in the sample plots. The images acquired in 2008 covered an area of 40 km2 (10 x 4 km or 50,000 x 20,000 pixels) and were mosaicked together to form a continuous image. Imagery was acquired in 3 channels representing the spectral ranges which approximate to: 0.4 \u00E2\u0080\u0093 0.5 \u00C2\u00B5m (blue), 0.5 \u00E2\u0080\u0093 0.6 \u00C2\u00B5m (green), and 0.6 \u00E2\u0080\u0093 0.7 \u00C2\u00B5m (red) as close to solar noon as possible resulting in reduced illumination variation over the imagery. Imagery was georectified to a QuickBird multispectral (2.44 m spatial resolution) image projected to UTM North American Datum 1983. Image coordinates were supplied by an 23 onboard GPS coupled with an inertial navigation system to assist accurate georectification. 1.5.6 Document structure This thesis is divided into two sections (Figure 1.3); the first contains two chapters, while the second is separated into four chapters. The first section provides an overview of the impact mountain pine beetle, Dendroctonus ponderosae (Hopkins), infestations have on forests in western Canada, with a focus on British Columbia. It also outlines conventional methods of surveying and discusses methods to mitigate attacked trees to prevent further infestation. Current methods to mitigate insect infestations are then reviewed and the impact of these methods to drive persistent mitigation on the spread of attack is examined (Coggins et al. 2011; Chapter 2). Digital remotely sensed data have the potential to augment conventional survey techniques, and in the second paper in this first section the accuracies of digital remotely sensed imagery to detect attacked trees and stands is reviewed and the implications of these accuracy statements on bark beetle infestations investigated (Coggins et al. 2008a; Chapter 3). The second section outlines methods to locate infestations and estimate forest inventory variables such as the status of mountain pine beetle attack, individual tree (diameter at breast height, crown dimensions, tree height) and forest stand measurements (stand density) from digital remote sensing technology. These data are scaled in a hierarchy from field plots to fine scale imagery to broad scale using sampling techniques and remote sensing models, and are statistically validated against field measurements. This second section begins by generating estimates of individual tree characteristics from 24 digital aerial imagery and statistically compares these against measurements recorded on trees in field plots. Estimates are then used to initiate spread models to demonstrate how Figure 1.3: The linkages between ground data and a hierarchy of multi-source, multi-scale remotely sensed imagery to determine the severity and extent of insect infestations. The thesis is separated into two sections, the first discusses current practices (Chapter 2) and demonstrates how remotely sensed data provide additional information to guide these surveys (Chapter 3). The second section first locates and then determines the presence of attacked trees. It then creates relationships between ground data and high spatial resolution imagery (Chapter 4). Estimates of beetle attack are the extrapolated to the per hectare level (Chapters 5 and 6), and finally these are used for stratification at the landscape level (Chapter 7). infestations expand on the landscape using measurements from field plots rather than from generalised data for areas in western Canada (Coggins et al. 2008b, Chapter 4). Next, a sampling approach known as adaptive cluster sampling was used in a novel 25 application to determine estimates of the numbers and location of mountain pine beetle killed trees on high-spatial resolution imagery and to generate estimates of the mean number of attacked trees, the variance, and expected range around the mean (Coggins et al. 2010; Chapter 5). With confidence in the adaptive cluster sampling approach the technique was applied across a time series of high spatial resolution imagery to monitor the rate of expansion and the location of rapidly expanding infestations.. The mean number of trees per hectare, variance, and confidence limits are calculated, and the means compared to generate the rate of spread over the image (Coggins et al. 2011b); Chapter 6). The fourth research chapter in section two is a culmination of the data and techniques used throughout this thesis. It combines the use of the field data, tree and stand estimates, and high and medium spatial resolution imagery into an advanced remote sensing model. Model output is used to determine the susceptibility of forest stands over a landscape to infestations by the mountain pine beetle. The susceptibility model requires input of proportion of pine in forest stands, stand age, stand density, and a location factor (Shore and Safranyik 1992). Typically, forest information is derived from aerial photographs or from model projections. However, by using advanced remote sensing models, imagery can be acquired over a large area and can provide up-to-date information on key forest attributes which are most susceptible to infestation by mountain pine beetle, and enable forest managers to assign appropriate resources and prioritise mitigation activities (Coggins et al. In press; Chapter 7). Finally, recommendations are provided on the use of a range of remote sensing datasets and innovative data collection methods to update forest inventory information. Very fine scale estimates can be extrapolated from high spatial resolution digital aerial image 26 samples to broad scale stand level strata determined from medium spatial resolution digital remotely sensed imagery. The results from the research and the implications of the findings are presented in the concluding chapter (Chapter 8). 27 1.6 References Anthony, R.N. 1965. Planning and control systems: A framework for analysis. Harvard University Graduate School of Business Administration, Boston Massachusetts. 180 p. British Columbia Ministry of Forests. 2009. Variable density yield projection. (online). http://www.for.gov.bc.ca/hts/vdyp/user_guides/volume1_vdyp_overview_revised _april2010.pdf. Accessed 14th October 2010. Bunting, P. and Lucas, R. 2006. The delineation of tree crowns in Australian mixed species forests using hyperspectral Compact Airborne Spectrographer Imager (CASI) data. Remote Sensing of Environment. 101(2): 230 \u00E2\u0080\u0093 248. Carroll, A.L., Taylor, S.W., R\u00C3\u00A9gni\u00C3\u00A8re, J., Safranyik, L., 2004. Effects of climate and climate change on the mountain pine beetle. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Proceedings of the mountain pine beetle symposium: challenges and solution. pp. 223 \u00E2\u0080\u0093 232. October 30-31, 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Information Report BC-X-399. 298 p. Chubey, M.S., Franklin, S.E., Wulder, M.A. 2006. Object-based analysis of Ikonos-2 imagery for extraction of forest inventory parameters. Photogrammetric Engineering and Remote Sensing. 72(4): 383 \u00E2\u0080\u0093 394. Coggins, S.B., Coops, N.C., Wulder, M.A., Bater, C.W., Ortlepp, S.O. 2011. Comparing the impacts of mitigation and non-mitigation on mountain pine beetle populations. Journal of Environmental Management. 92(1): 112 \u00E2\u0080\u0093 120. DOI: 10.1016/j.jenvman.2010.08.016. Coggins, S.B., Coops, N.C., Wulder, M.A., White, J.C. 2008a. Linking survey detection accuracy with ability to mitigate populations of mountain pine beetle. The Forestry Chronicle. 84(6): 900 \u00E2\u0080\u0093 909. Coggins, S.B., Coops, N.C., Wulder, M.A. 2008b. Initialisation of an insect infestation spread model using tree structure and spatial characteristics derived from high spatial resolution digital aerial imagery. Canadian Journal of Remote Sensing. 34(6):485\u00E2\u0080\u0093502. Coggins, S.B., Coops N.C., Wulder M.A. 2010. Improvement of low level bark beetle damage estimates with adaptive cluster sampling. Silva Fennica. 44(2): 289 \u00E2\u0080\u0093 301. Coggins, S.B., Coops, N.C., Wulder, M.A. 2011b. Estimates of bark beetle expansion factors with adaptive cluster sampling. International Journal of Pest Management. 57(1): 11 \u00E2\u0080\u0093 21. Coops, N.C., Johnson, M., Wulder, M.A., White, J.C. 2006. Assessment of QuickBird high spatial resolution imagery to detect red attack damage due to mountain pine beetle infestation. Remote Sensing of Environment. 103(1): 67 \u00E2\u0080\u0093 80. 28 Danson, F.M. 1987. Preliminary evaluation of the relationships between SPOT-1-HRV data and forest stand parameters. International Journal of Remote Sensing. 8(10): 1571 \u00E2\u0080\u0093 1575. Davis, L.S.., Johnson, K.N., Bettinger, P.S., Howard, T.E. 2001. Forest management: To sustain ecological, economic, and social values. 4th Edition. McGraw Hill, New York. 804 p. DigitalGlobe. 2009. WorldView-2. (online). http://www.digitalglobe.com/digitalglobe2/file.php/786/WorldView2-DS- WV2.pdf. Accessed 12th October 2010. Falkowski, M. J., Wulder, M. A., White, J. C., and Gillis, M. D. 2009. Supporting large- area, sample-based forest inventories with very high spatial resolution satellite imagery. Progress in Physical Geography. 33(3): 403 \u00E2\u0080\u0093 423. Flanders, D., Hall-Beyer, M., Pereverzoff, J. 2003. Preliminary evaluation of eCognition object-based software for cut block delineation and feature extraction. Canadian Journal of Remote Sensing. 29(4): 441 \u00E2\u0080\u0093 452. Franklin, J. 1986. Thematic mapper analysis of coniferous forest structure and composition. International Journal of Remote Sensing. 7(10): 1287 \u00E2\u0080\u0093 1301. Franklin, S.E. 2001. Remote sensing for sustainable forest management. CRC Press. 424 p. Gillis, M. 2001. Canada\u00E2\u0080\u0099s national forest inventory (responding to information needs). Environmental Monitoring and Assessment. 67(1-2): 121 \u00E2\u0080\u0093 129. Gunn, E.A. 1991. Some aspects of hierarchical planning in forest management. In Buford, M.A. (Ed.). Proceedings of the 1991 Symposium on Systems Analysis in Forest Resources. pp. 54 \u00E2\u0080\u0093 62. South East Forest Experimental Station, General Technical Report SE-74. 423 p. Hill, J.B., Popp, H.W., Grove, A.R. Jr. 1967. Botany: A textbook for colleges. 4th Edition. McGraw-Hill Book Co., Toronto, Ontario. 634 p. Holmgren, P. and Thuresson, T. 1998. Satellite remote sensing for forestry planning: A review. Scandinavian Journal of Forest Research. 13(1-4): 90 \u00E2\u0080\u0093 110. Holopainen, M. and Kalliovirta, J. Modern data acquisition for forest inventories. In: Kangas, A. and Maltamo, M. (Eds.). Forest Inventory \u00E2\u0080\u0093 Methodology and Applications. pp. 343 \u00E2\u0080\u0093 362. Springer, Netherlands. 364 p. Husch, B. 1971. Planning a forest inventory. Food and Agriculture Organization of the United Nations, Rome. 120 p. Hussin, Y.A. and Bijker, W. 2000. Inventory of remote sensing applications in forestry for sustainable management. In: Remetey-F\u00C3\u00BCl\u00C3\u00B6pp, G., Clevers, J., Jan Beek, K. (Eds.). ISPRS 2000 congress : geoinformation for all. Part B7. pp. 575 \u00E2\u0080\u0093 587. Amsterdam, the Netherlands, 16-23 July, 2000. 1760 p. Hyde, P., Dubayah, R., Walker, W., Blair, J.B., Hofton, M., Hunsaker, C. 2006. Mapping forest structure for wildlife habitat analysis using multi-sensor (LiDAR, 29 SAR/lnSAR, ETM+, Quickbird) synergy. Remote Sensing of Environment. 102(1- 2): 63 \u00E2\u0080\u0093 73. King, D.J. 2000. Airborne remote sensing in forestry: Sensors, analysis and applications. The Forestry Chronicle. 76(6): 859 \u00E2\u0080\u0093 876. Knight, F.B. 1967. Evaluation of forest insect infestations. Annual Review of Entomology. 12: 207 \u00E2\u0080\u0093 228. K\u00C3\u00B6hl, M., Magnussen, S.S., Marchetti, M. 2006. Sampling methods, remote sensing and GIS multiresource forest inventory. Springer, Berlin. 373 p. Langley, P.G. 1975. Multistage variable probability sampling: theory and use in estimating timber resources from space and aircraft photography. PhD Dissertation. Berkeley, California. 101 p. Leckie, D.G. 1990. Advances in remote sensing technologies for forest surveys and management. Canadian Journal of Forest Research. 20(4): 464 \u00E2\u0080\u0093 483. Leckie, D.G. and Gillis, M.D. 1995. Forest inventory in Canada with emphasis on map production. The Forestry Chronicle. 71(1): 74 \u00E2\u0080\u0093 88. Li, X. and Strahler A.H. 1985. Geometric-optical modeling of a conifer forest canopy. IEEE Transactions on Geoscience and Remote Sensing. GRS23: 705 \u00E2\u0080\u0093 721. Lillesand, T.M., Kiefer, R.W., Chipman, J.W. 2004. Remote sensing and image interpretation. 5th Edition. John Wiley and Sons, New York. 763 p. Lund, H.G. 1998. IUFRO guidelines for designing multi-purpose forest resources inventories. IUFRO world series, vol 8. IUFRO, Vienna. 216 p. McRoberts, R.E., Wendt, D.G., Nelson, M.D., Hansen, M.H. 2002. Using land cover classification based on satellite imagery to improve the precision of forest inventory estimates. Remote Sensing of Environment. 81(1): 36 \u00E2\u0080\u0093 44. McRoberts, R. E. and Tomppo, E.O. 2007. Remote sensing support for national forest inventories. Remote Sensing of Environment. 110(4): 412 \u00E2\u0080\u0093 419. McRoberts, R.E., Cohen, W.B., N\u00C3\u00A6sset, E., Stehman, S.V., Tomppo, E. 2010. Using remotely sensed data to construct and assess forest attribute maps and related spatial products. Scandinavian Journal of Forest Research. 25(4): 340 \u00E2\u0080\u0093 367. Meidinger, D.V. and Pojar, J. 1991. Ecosystems of British Columbia. (online). http://www.for.gov.bc.ca/hfd/pubs/Docs/Srs/Srs06.htm. Accessed 13th December 2010. Moir, W.H. 1965. The lodegpole pine zone in Colorado. American Midland Naturalist. 81(1): 87 \u00E2\u0080\u0093 98. Morgan, J.L., Gergel, S.E., Coops, N.C. 2010. Aerial photography: A rapidly evolving tool for ecological management. Bioscience. 60(1): 47 \u00E2\u0080\u0093 59. Nelson, T., Wulder, M., Niemann, K.O. 2001. Spatial resolution implications of digitizing aerial photography for environmental applications. The Imaging Science Journal. 49(4): 223 \u00E2\u0080\u0093 232. 30 Ono, H. 2004. The mountain pine beetle problem: scope of the problem and key issues in Alberta. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Mountain pine beetle symposium: challenges and solutions. pp. 62\u00E2\u0080\u009366. 30\u00E2\u0080\u009331 October 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Victoria, British Columbia, Information Report BC-X-399. 298 p. Parminter, J. 2000. From timber to biodiversity: the evolution of British Columbia\u00E2\u0080\u0099s forest inventory program. Forest History Today. Fall: 12 \u00E2\u0080\u0093 19. Roy, P.S., Sharma, K.P., Jain, A. 1996. Stratification of density in dry deciduous forest using satellite remote sensing digital data \u00E2\u0080\u0093 An approach based on spectral indices. Journal of Bioscience. 21(5): 723 \u00E2\u0080\u0093 734. Riley, J.R. 1989. Remote sensing in entomology. Annual Review of Entomology. 34: 247 \u00E2\u0080\u0093 271. Sandvoss, M., Clymont, B., Farnden, C. 2005. User\u00E2\u0080\u0099s guide to the vegetation resources inventory. (online). http://www.for.gov.bc.ca/hfd/library/documents/bib106996.pdf. Accessed 16th October 2010. Scarth, P. and Phinn, S. 2000. Determining forest structural attributes using an inverted geometric-optical model in mixed Eucalypt forests, Southeast Queensland, Australia. Remote Sensing of Environment. 71(2): 141 \u00E2\u0080\u0093 157. Shore, T.L. and Safranyik, L. 1992. Susceptibility and risk-rating systems for the mountain pine beetle in lodegpole pine stands. Pacific Forestry Centre, Canadian Forestry Service, Pacific and Yukon Region. Information Report BC-X-336. 12 p. Skakun, R.S., Wulder, M.A., Franklin, S.E. 2003. Sensitivity of the thematic mapper enhanced wetness difference index to detect mountain pine beetle red-attack damage. Remote Sensing of Environment. 86(4): 433 \u00E2\u0080\u0093 443. Strahler, A.H., Woodcock, C.E., Smith, J.A. 1986. On the nature of models in remote sensing. Remote Sensing of Environment. 20(2): 121 \u00E2\u0080\u0093 139. Tokola, T. 2006a. Europe. In: Kangas, A. and Maltamo, M. (Eds.). Forest Inventory \u00E2\u0080\u0093 Methodology and Applications. pp. 295 \u00E2\u0080\u0093 308. Springer, Netherlands. 364 p. Tokola, T. 2006b. Asia. In: Kangas, A. and Maltamo, M. (Eds.). Forest Inventory \u00E2\u0080\u0093 Methodology and Applications. pp. 309 \u00E2\u0080\u0093 324. Springer, Netherlands. 364 p. Tokola, T. 2006c. North America. In: Kangas, A. and Maltamo, M. (Eds.). Forest Inventory \u00E2\u0080\u0093 Methodology and Applications. pp. 325 \u00E2\u0080\u0093 340. Springer, Netherlands. 364 p. Trotter, C.M., Dymond, J.R., Goulding, C.J. 1997. Estimation of timber volume in a coniferous plantation forest using Landsat TM. International Journal of Remote Sensing. 18(10): 2209 \u00E2\u0080\u0093 2223. United Nations Food and Agriculture Organization. 2006. Global forest resources assessment 2005: Progress towards sustainable forest management. (online). ftp://ftp.fao.org/docrep/fao/008/A0400E/A0400E00.pdf Accessed 9th January 2008. 31 United Nations Food and Agricultural Organisation. 2009. Global review of forest pests and diseases. Rome, Italy. 235 p. Westfall, J. 2007. 2006 Summary of forest health conditions in British Columbia. British Columbia Ministry of Forests and Range, Forest Practices Branch, Victoria, BC. 73 p. Westfall, J. and Ebata, T. 2009. 2008 Summary of Forest Health Condition in British Columbia. Pest Management Report Number 15. British Columbia Ministry of Forests and Range, Forest Practices Branch, Victoria, BC. 85 p. White, J.C., Wulder, M.A., Brooks, D., Reich, R., Wheate, R.D. 2004. Mapping mountain pine beetle infestation with high spatial resolution satellite imagery. The Forestry Chronicle. 80(6): 743 \u00E2\u0080\u0093 745. Woodcock, C.E. and Strahler, A.H. 1987. The factor of scale in remote sensing. Remote Sensing of Environment. 21(3): 311 \u00E2\u0080\u0093 332. Wulder, M.A. 1998. Optical remote-sensing techniques for the assessment of forest inventory and biophysical parameters. Progress in Physical Geography. 22(4): 449 \u00E2\u0080\u0093 476. Wulder, M., Hall, R., Coops, N., Franklin, S. 2004. High spatial resolution remotely sensed data for ecosystem characterization. Bioscience. 54(6): 1 \u00E2\u0080\u0093 11. Wulder, M.A., White, J.C., Bentz, B.J., Ebata, T. 2006. Augmenting the existing survey hierarchy for mountain pine beetle red-attack damage with satellite remotely sensed data. The Forestry Chronicle. 82(2): 187 \u00E2\u0080\u0093 202. Wygant, N.D. 1940. Effects of low temperature on the Black Hills beetle (Dendroctonus ponderosae Hopkins). Ph.D. dissertation summary, State University of New York, College of Environmental Science and Forestry, Syracuse, NY. 57 p 32 2 COMPARING THE IMPACTS OF MITIGATION AND NON-MITIGATION ON MOUNTAIN PINE BEETLE POPULATIONS3 2.1 Introduction 2.1.1 The impact of mountain pine beetle infestations The mountain pine beetle, Dendroctonus ponderosae (Hopkins), is a bark beetle that aggressively attacks and causes mortality of pine trees (Pinus spp.). The natural range of the insect extends from northern Mexico, through the western United States and into western Canada. Infestations are of particular concern in British Columbia, Canada, where in 1999 beetle damage was observed over an area of 164 000 hectares of lodgepole pine (Pinus contorta Dougl. ex. Loud var. latifolia Engl.), which spread to 13 million hectares in 2008 (Westfall and Ebata 2009). Mountain pine beetle infestations have increase rapidly due to increasingly warm winters (Stahl et al. 2006; Macias Fauria and Johnson 2009) and the availability of a susceptible host. Susceptibility increases in stands with a mean stem diameter larger than 15 cm, greater than 80 years of age, and with a stocking density between 750 and 1000 stems per hectare (Shore and Safranyik 1992). Due to aggressive fire suppression in British Columbia, large areas of even-age mature lodgepole pine forests are present, which coupled with a warming climate provide a suitable environment for mountain pine beetle infestations to initiate, expand, and cause extensive levels of tree mortality (Safranyik 1978; Taylor et al. 2006). Under regular climate conditions, larvae overwintering beneath bark experience mortality when 3 A version of this chapter was published as: Coggins, S.B., Coops, N.C., Wulder, M.A., Bater, C.W., Ortlepp, S.M. 2011. Comparing the impacts of mitigation and non-mitigation on mountain pine beetle populations. Journal of Environmental Management. 92(1): 112 \u00E2\u0080\u0093 120. 33 temperatures drop below -40\u00C2\u00B0C (Safranyik 1978; Wygant 1942) which diminishes the potential number of attacking beetles. Temperatures across western Canada have increased in the last century (1895 \u00E2\u0080\u0093 1995) by as much as 1.7 \u00C2\u00B0C (British Columbia Ministry of Water, Land, and Air Protection 2010) enabling a higher proportion of beetles to survive and attack susceptible host trees (Raffa et al. 2008). As a result of increasing temperatures, infestations occur at higher elevations and in areas that have no historical record of mountain pine beetle attack (Logan and Powell 2001). For instance, attack is thought to have expanded as small pockets of infestation in areas previously kept stable by cold weather (Macias Fauria and Johnson 2009) and dispersing beetles are thought to have traversed the Rocky Mountains to impact the pine forests of western Alberta, and could potentially spread further eastwards and infest the Canadian boreal forest, an area where mountain pine beetle infestations have not been previously recorded (Carroll et al. 2004). The capacity for mountain pine beetle to thrive and subsequently spread in the boreal remains to be determined. Jack pine, the dominant pine species of the boreal, has been shown in some circumstances to support mountain pine beetles (Cerezke 1995). Further, the jack pine of the boreal is not as spatially contiguous as the pine forests of British Columbia and western Alberta also favouring a situation where spread will not be so rapid or upon such a high proportion of forests present. Although population development and subsequent spread has not been demonstrated in jack pine, mortality has been caused to lodgepole pine x jack pine hybrids near Grande Prairie, Alberta (Rice and Langor 2009). If infestation were to spread to jack pine it would be disastrous for the ecological, economical, and social values in the boreal (Ono 2004) as it has been in the central interior of British Columbia. 34 Attack by mountain pine beetles on trees causes the crown foliage to fade from green (green attack) to red (red attack) over a period of approximately 12 months (Wulder et al. 2006). Approximately 12 months after attack 90% of trees will have red needles, which eventually fall from the tree. Approximately three years after attack trees are often completely defoliated (Wulder et al. 2006). Areas undergoing the early stages of infestation are characterized by the presence of red crowns and are likely associated with green attack trees (Wulder et al. 2009a), as when beetles emerge from previously attacked hosts they typically disperse less than 30 m (Safranyik et al. 1992) before attacking adjacent trees (Mitchell and Preisler 1991). Up to 0.2% of beetles disperse from the stand and travel on warm air thermals hundreds of metres, sometimes, hundreds of kilometres to infest stands (Safranyik et al. 1992). The association between green and red attacked trees is commonly expressed as a ratio or an expansion factor, for example, 2:1 (green:red) or an expansion factor of 2. Common expansion factors during the current outbreak reach 4 in southern British Columbia and 2 in northern British Columbia (British Columbia Ministry of Forests and Range 2002). Historically, the area infested by mountain pine beetle has been limited by successive years of cold temperatures during winter. Mortality can also be expected during unseasonably cold weather prior to the overwintering stage larvae that have not developed sufficient cold-hardiness can be killed during a cold snap. Similarly, mortality may also be experienced during spring, prior to emergence, should temperatures be unseasonably cold (Wygant 1942). To instigate effective control of mountain pine beetle infestations, hazard and risk rating systems can be used to determine which stands are susceptible to attack (Fettig et al. 2007). Alternatively, in stands experiencing attack, 35 mitigation tactics are implemented. Trees are best removed while beetle populations are low, and attack is infrequent and has not begun to rapidly expand. Areas where mitigation work has been completed should be monitored annually to ensure all attacked trees were removed (i.e. no red attack is visible) and infestations levels remain low. 2.1.2 Objectives To better understand mountain pine beetle attack the impact of infestations was examined in a network of mitigated and non-mitigated plots in western Canada using field-based observations. The efficacy of mitigation is demonstrated using a suite of simple population models (described by Carroll et al. 2006) and mitigation is applied by removing a proportion of the infested trees. First a summary is provided of the types of mitigation available to forest managers. Then population models that use field measured expansion factors demonstrate how rapidly infestations expand, predict the number of trees affected, estimate the proportion of trees required to be treated to control infestations, and allow calculation of the length of time mitigation must persist to be effective. The models are used to demonstrate the differences in the spread of infestation under different mitigation scenarios using stand conditions derived from field data. In this chapter three modelling scenarios are used. The first simulates the spread of infestation in mitigated and non-mitigated plots over time, the second demonstrates the mitigation levels required to control the current infestation in the non-mitigated plots, and the third examines the duration that mitigation must continue to ultimately halt the infestation. The effectiveness of current mitigation activates is also examined. Finally, recommendations are provided that improve the efficacy of future activities and examine the consequences of leaving infested trees in a given stand. 36 2.2 Summary of mitigation techniques Mitigation techniques aim to remove either selected infested trees or all trees attacked during an infestation (Carroll et al. 2006). Mitigation strategies can be either indirect or direct; the former is also known as preventive management (Whitehead et al. 2006) and consists of silvicultural techniques that create unfavourable conditions for beetle attack, or employ prescribed burns to reduce stand susceptibility to infestation (Shore et al. 2006). Direct control consists of removing green attack trees to decrease the number of beetles that are available to infest trees in successive years, and is initiated after beetles have infested a given stand. For direct and indirect mitigation tactics to be effective timely detection, accurate susceptibility and risk assessment, and access to infestations are important considerations when selecting the appropriate mitigation tactic or combination of tactics (Coops et al. 2008; Shore et al. 2006). Six direct and indirect tactics are used to manage mountain pine beetle infestations (defined by Maclauchlan and Brooks 1998), tactics 1-3 are considered direct control and 4-6 are indirect. 1) Survey and assessment determines where infestations are in the landscape using fixed wing aircraft, aerial photography, and ground surveys to locate infestations and then provide an estimate of susceptibility to attack using a hazard and risk rating (Shore and Safranyik 1992); 2) Harvesting, which aims to reduce infestations by removing attacked trees (sanitation), provide revenue by logging dead trees (salvage), reduce attack hazard by removing high hazard host trees where pine trees older than 80 years with a stem diameter greater than 15 cm are most susceptible, and priority is assigned to harvest stands, where beetle attacked stands are most important; 3) single tree treatments remove individual trees or small groups (<2 hectares) where newly infested trees are treated with 37 monosodium methane arsenate or can be removed using fall and burn where trees are felled and then burned. Alternatively bark is removed from the trees to expose the life stages beneath the bark and cause mortality, insecticide can be used to prevent trees from being attacked, but is usually restricted to campsites, urban areas, and other specialized circumstances as it needs to be repeated annually and would be prohibitively expensive if applied to forest stands; 4) Baiting techniques, where pheromone baits are used to attract attacking beetles to trees which can then be removed or repel beetles from valuable stands. Baiting is often used in conjunction with single tree treatments, baited trees are attacked and then felled to cause mortality to the beetles beneath the bark; 5) Beetle proofing (stand thinning) reduces susceptibility to attack by removing basal area or trees with thick phloem, which are thought to be selected preferentially so life stages may survive cold winters. This tactic also prevents beetles from attacking forest stands by spacing the trees to increase wind speed and temperature in the stand, and increasing tree vigour where vigorous trees are thought to be more resistant to attack (Whitehead et al. 2006). The minimum tree spacing recommended is 3.5 m with a maximum of 5 m, while optimal spacing is between 4 and 4.5 m; 6) Silvicultural treatments which include: species manipulation, where a mosaic of tree species is encouraged to grow to lessen the proportion of host tree in stands (Fettig et al. 2007); and, age class manipulation, by dividing the stands into age classes the long-term susceptibility of the forest will be decreased. In British Columbia, in order to implement control and to measure the efficacy of mitigation, forest health surveys are conducted to assess the severity and extent of infestations. Typically, infestations are detected and monitored using a top-down 38 hierarchy consisting of aerial and ground surveys. In British Columbia, annual province- wide aerial overview surveys are used to assess the severity and extent of infestations, and are intended to provide a broad overview of forest health conditions and aid in wide- area strategic decision making. Areas of interest identified during aerial overview surveys can later be targeted by helicopter surveys to record the location and number of red attack trees. Lastly, in areas deemed at highest risk, ground crews may be dispatched to locate, fell, and burn newly attacked trees (Maclauchlan and Brooks 1998). Given the nature of newly infested trees close to those that were previously attacked, it is possible that trees missed by aerial surveys will be detected on the ground, reducing the potential for future infestation expansion (Carroll et al. 2006; Coggins et al. 2008). Mitigation of attacked trees is often not fully effective as infested trees can be difficult to detect, necessitating ongoing annual monitoring to track the severity and extent of infestations (Coops et al. 2008). If a low proportion of trees are mitigated (i.e., less than 50%) infestations continue to expand to cause mortality to trees (Carroll et al. 2006). Mitigation of infested trees should remove a proportion that is equal to, or higher than, the rate of infestation to cause a beetle population to decline or remain stable. If mitigation consistently causes a decline in population numbers, populations could be extirpated and attack could eventually be halted (Carroll et al. 2006). For mitigation to be fully effective it should be rapid and continuous (Carroll et al. 2006; Coggins et al. 2008). Trees remaining after mitigation that contain viable populations of beetles support infestation development and spread. If mitigation is effective, attack from within the stand should not occur because the attacking population is removed when the trees are felled and burned. 39 The most important step when beginning a mitigation program is to detect all the infested trees and provide the locations of these trees to ground crews who can remove attacked trees and reduce the insect population (Carroll et al. 2006). Mitigation carried out with ground surveys must be completed at the appropriate time of year, after beetles have dispersed, and should be completed as the larvae overwinter beneath the bark (Maclauchlan and Brooks 1998). Flight times are influenced by temperature as beetles rely on heat accumulation to develop (Safranyik 1978) and the approximate flight time can be estimated in day degrees calculated from weather data collected on site and this should be accounted for when planning mitigation surveys (Macias Fauria and Johnson 2009). 2.3 Methods 2.3.1 Study plots This research was conducted using field plots in forests situated in the focus area described in Chapter 1. A network of 28 field plots was established in the area and were centered on forest stands known to contain mountain pine beetle attack (n = 18; located at 122\u00C2\u00B0 0\u00E2\u0080\u0099 32.505\u00E2\u0080\u009D W; 54\u00C2\u00B0 49\u00E2\u0080\u0099 10.618\u00E2\u0080\u009D N) or where mitigation of infested trees had been completed (n = 10; located at 118\u00C2\u00B0 37\u00E2\u0080\u0099 10.067\u00E2\u0080\u009D W; 54\u00C2\u00B0 10\u00E2\u0080\u0099 39.927\u00E2\u0080\u009D N). Each plot had a 30 m radius, which corresponds to the expected distance beetles have been observed to disperse (Safranyik et al. 1992). Plots were chosen that share similar attributes, similar slope, aspect, stem diameter, stocking density, and proportion of host trees (Table 2.1). 40 Table 2.1. Stand characteristics and measurements within the study sites. Site A Site B Aspect South East North East Mean slope (degrees) 8.2 7.5 Mean stocking density (trees per plot) 213 245 Mean stem diameter (cm) 25.3 20.8 Proportion of pine within stands 76% 86% Proportion of other species within stands 23% 14% In the plots subject to mitigation activity, trees undergoing attack have been removed since 2005 using the fall and burn strategy, where green attacked trees are felled and burned on site. Field work for all plots was undertaken in August 2008, and at each plot the total number of trees was counted and the mountain pine beetle attack status recorded according to foliage colour (healthy, green attack, and red attack). In all plots, infestation expansion factors were calculated for 2008 using the attack status of trees. In the mitigated plots, the health status of attacked trees and the number of trees within each attack class were recorded along with the number of tree stumps, which indicated the number that were felled and burned. The total infestation (red attack plus stumps) was calculated for each plot to determine the number of infested trees prior to mitigation. The number of stumps was divided by the total number of trees infested to generate the proportion of trees mitigated. For this research I assumed the inverse indicates the proportion of infested trees remaining undetected during mitigation in each plot or trees that had been attacked after mitigation took place. The average mitigation was calculated for all plots to determine the efficacy of mitigation. 41 2.3.2 Modelling The models applied in this study build upon the work in Coggins et al. (2008; Chapter 4), who examined the impact of mountain pine beetle infestations across a range of forest stands under differing infestation intensities using three population modelling scenarios with hypothetical expansion factors. The population-scale modelling scenarios adapted from Carroll et al. (2006) were again utilized in this research to assess the impacts of mountain pine beetle infestations on forest stands with expansion factors generated from field observations. These models are stand level and, therefore, I assume migration is from the trees infested within the stands only and do not account for long-range dispersal by beetles into these stands. Further, long range dispersal is highly stochastic and not appropriately added to such models at this time. A non-zero infestation level can likely be assumed for any susceptible pine stand known to be in an eligible catchment for long- range dispersal. The first model investigated the potential spread of infestation in the mitigated and non-mitigated plots. Infestation expansion factors were calculated for each of the mitigated and non-mitigated plots by calculating the ratio of green attack to red attack trees observed during the 2008 field campaign. The number of green attack and red attack trees were established in each plot and the number of red attacked trees was backcast to become the number of green attacks in the previous year (2006), following Wulder et al. (2009b). The attacked trees in 2007 were then forward cast to 2008, whereby the red attacks became grey attacked and the green attacks became red attack. The expansion factor for each plot in each year was then calculated by dividing the number of green attacked trees by the number of red attacked trees averaged for the mitigated and non-mitigated plots. To model the potential infestation, the number of 42 green attacked trees in 2006 was projected forwards annually using the average infestation expansion factors estimated from the field data. In 2009 and thereafter, an expansion factor of 2 was utilized because this has been determined to represent the common rate of infestation expansion experienced in the study area (British Columbia Ministry of Forests, 2002; Carroll et al. 2006; Wulder et al. 2009b). In the second scenario, I utilized the average mitigation efficacy from field data to determine whether removing attacked trees in a single time step will impact a population of attacking beetles in the long term. The proportion (P) of trees requiring mitigation according to the rate of population increase is defined as (Carroll et al. 2006): P = 1-1/R (3.1) where R is the rate of infestation expansion. In this scenario, the average mitigation accuracy was calculated from the mitigated field plots where the number of infested trees removed compared to those remaining defined mitigation accuracy in each plot, and was then used to determine whether suppression of the beetle population was possible at the current rate of mitigation. If all attacked trees are removed, the forest stand should experience no further infestation in subsequent years, with the exception of immigration from long-range dispersal. If, however, attacked trees remain in the stand, the infestation may continue and should infestation expansion rates increase, attack will become more severe. Conversely, the level of detection required can also be determined if the expansion factor is known. The required detection levels for the mitigated and non- mitigated plots are derived based on average infestation expansion levels in 2007. The final scenario calculates the length of time required for ongoing detection, monitoring, and mitigation to bring an infestation under control. In the long-term, 43 persistent mitigation may be required to reduce infestations. Low-level attack may be controlled in a single time step; however, larger infestations are more difficult to control even when doubling (R = 2), and are dependent on persistent removal of a proportion of the attacked trees. For example, Carroll et al. (2006) describe an infestation covering 300 000 hectares with R = 2, where 150 000 hectares of trees must be removed each year to ensure the infestation remains stable. To remove such a large proportion of trees would be impossible if attempted in a single mitigation event (Carroll et al. 2006). The number of trees infested in a given year can be estimated using: N = N0[R(1-P)] t (3.2) where the number of trees initially infested (N0), the yearly rate of increase (R), the proportion of trees treated each year (P), and the number of years (t). Estimates of R and P can determine the number of years for persistent direct mitigation as defined in Equation 3.1 (Carroll et al. 2006). The concept is explored to determine the time required to suppress the infested trees within the field plots. The total number of attacked trees remaining in the mitigated and non-mitigated plots were used as a baseline. The number of years required to bring the infestation under control is then estimated using the current mitigation efficacy. 2.4 Results In both 2007 and 2008 the non-mitigated plots experienced higher levels of tree mortality attributed to mountain pine beetle attack than the mitigated plots, and infestations expanded more rapidly. The first scenario was run for a period of ten years, starting in 2006 with green attack trees, to provide a realistic demonstration of the progression of an 44 infestation when mitigation is utilized versus when no control is applied to a stand. The expansion factors calculated from the 2007 and 2008 field observations were used to estimate the amount of new infestation in each of those years. In 2007 infestation expanded at a rate of 5.09 in the non-mitigated plots, and 1.35 in the mitigated plots. Attack decreased in 2008, with average expansion rates of 0.29 and 0.12 in the non- mitigated and mitigated plots, respectively (Table 2.2). Some plots listed as mitigated had no evidence of mitigation activities indicating they were either of low priority and had not been completed or had been assessed and were not seen as a threat to surrounding forest because there had been no expansion from these plots or because expansion was so small they were not worth mitigating at this time. An expansion factor of 2 was used to determine the rate of infestation spread, which is similar to rates found in other studies conducted within the study area (Wulder et al. 2009b), and which was recommended by Carroll et al. (2006) and the British Columbia Ministry of Forests and Range (2002) in northern British Columbia. After 10 years, the non-mitigated plots had an average of 125 infested trees, ranging in number from 0 to 768 (Figure 2.2). If infestation levels increased, more trees would be attacked over the time period, and to examine these effects the proportional increase from the average amount of infestation to the maximum in each year was calculated and used to provide a range between which infestation might fluctuate. If infestation increased there would be 146 attacked trees; however, if infestation decreased by the same proportion there would be 105 infested trees to mitigate after a period of 10 years. The mitigated plots all had 0 infested trees after 10 years because the rate of infestation after 2008 did not produce sufficient numbers of infested trees within the stand to continue attack. 45 Table 2.2. Estimated expansion factors in 2007 and 2008 for each non-mitigated (top) and mitigated plot (bottom), where GA = green attack, RA = red attack, and GR = grey attack. 2008 2007 2006 GA RA GR Expansion GA RA GR Expansion GA RA GR 0 53 22 0.00 53 22 0 2.41 22 0 0 6 14 7 0.43 14 7 0 2.00 7 0 0 12 16 1 0.75 16 1 0 16.00 1 0 0 1 3 2 0.33 3 2 0 1.50 2 0 0 2 2 1 1.00 2 1 0 2.00 1 0 0 0 1 0 0.00 1 0* 0 1.00 0 0 0 0 2 1 0.00 2 1 0 2.00 1 0 0 0 14 14 0.00 14 14 0 1.00 14 0 0 3 21 8 0.14 21 8 0 2.63 8 0 0 3 15 3 0.20 15 3 0 5.00 3 0 0 1 17 3 0.06 17 3 0 5.67 3 0 0 0 25 5 0.00 25 5 0 5.00 5 0 0 11 18 0 0.61 18 0 0 18.00 6 11 4 0.55 11 4 0 2.75 4 0 0 5 19 2 0.26 19 2 0 9.50 2 0 0 4 21 3 0.19 21 3 0 7.00 3 0 0 1 3 0 0.33 3 0 0 3.00 0 0 0 0.29 5.09 2008 2007 2006 GA RA GR Expansion GA RA GR Expansion GA RA GR 0 8 3 0.00 8 3 0 2.67 3 0 0 0 0 4 0.00 0 4 0 0.00 4 0 0 0 5 4 0.00 5 4 0 1.25 4 0 0 1 0 4 1.00 0 4 0 0.00 4 0 0 1 3 0 0.33 3 0 0 3.00 0 0 0 0 8 4 0.00 8 4 0 2.00 4 0 0 0 1 3 0.00 1 3 0 0.33 3 0 0 0 4 2 0.00 4 2 0 2.00 2 0 0 0 4 5 0.00 4 5 0 0.80 5 0 0 0 4 5 0.00 4 5 0 0.80 5 0 0 0 4 2 0.00 4 2 0 2.00 2 0 0 0.12 1.35 * A score of 0 indicates no infestation was present in plot. However, infestation increased to 1 in the following year, indicating an expansion of 1. 46 Figure 2.1: The number of infested trees expected to be present without mitigation using the number of infested trees in the non-mitigated plots to initiate the model and using expansion factors derived from field-based observations in 2007 and 2008, with an expansion factor of 2 used in 2009 onwards. The average infestation per year is shown by the thick black line, with the range of infestation that could be expected in each year in dark grey, and the minimum and maximum amount of infestation shown in light grey (after Carroll et al. 2006). In the second scenario, the average mitigation accuracy was used to determine how effective the current level of mitigation will be (Table 2.3) with an average efficacy of 43% determined (Figure 2.3). The average infestation expansion factor in the non- mitigated plots in 2007 was 5.1, requiring an 80% detection accuracy to maintain a static population (Figure 2.3). However, the uppermost range of infestation expansion in the plots was 18, which requires a detection rate of 94% (Figure 2.3). The average infestation expansion factor in 2007 in the mitigated plots was 1.1, which requires a minimum detection rate of less than 10% (Figure 2.3), with a range between 0 and 2.67. 47 Table 2.3: Mitigation efficacy determined from field observations in the mitigated plots, where GA = green attack and RA = red attack. Plot GA RA STUMP Total RA % mitigated % missed 1 0 4 6 10 60% 40% 2 1 0 0 0 0% 100% 3 0 4 8 12 67% 33% 4 0 1 0 1 0% 100% 5 0 0 1 1 100% 0% 6 0 5 4 9 44% 56% 7 0 4 4 8 50% 50% 8 1 3 0 3 0% 100% 9 0 8 3 11 27% 73% 10 0 4 14 18 78% 22% The final scenario estimated the length of time that is required to conduct ongoing detection, monitoring, and mitigation activities. The average number of red attack trees remaining in the non-mitigated stands after 10 years was 273 (approximately 15% of the total number of trees measured during the field work). The minimum and maximum numbers of attacked trees found in the plots were 64 and 768, respectively. These numbers were used as a baseline to determine the number of years required to provide control to forest stands. If the average mitigation efficacy of 43% calculated for scenario 2 is utilized, none of the infestation in the plots will be controlled effectively. If the average number of red attack trees remaining after the plots were mitigated were used as a baseline, with a 70% detection accuracy mitigation will take 11 years, at 80% 48 Figure 2.2: The minimum proportion of infested trees requiring treatment during mitigation can be calculated for the average, minimum, and maximum expansion factors in the non-mitigated plots, and for the average mitigation efficacy in the mitigated plots. The minimum proportion of infested trees requiring treatment to control an infestation expansion of 5.1 (the average expansion factor calculated) is approximately 80%. The maximum expansion rate was 18 which required a minimum of 93% of the infested trees to be treated to provide control. Infestation in the mitigated plots requires a minimum of 10% of infested trees to be treated to control current levels of infestation. The average mitigation efficacy of 43% is ineffective at controlling a doubling population (after Carroll et al. 2006). mitigation will take 6 years, and at 90% mitigation will take 3 years. If the number of red attack trees started at the maximum number of red attacks found in a plot (768 trees), mitigation will take 13 years with a detection accuracy of 70%, 7 years with 80%, and 4 years with 90%. Finally, with the lowest number of infested trees found in a plot mitigation with a 70% detection accuracy will take 8 years, 5 years with 80%, and 3 years with 90% (Figure 2.4 and Table 2.4). 49 Figure 2.3: Number of years required to suppress infestations of mountain pine beetle using the number of red attack trees was derived from those remaining after mitigation and the expected infestation after 10 years in the non-mitigated plots. The average number of years required to complete mitigation is shown by the thick black line, with the range of infestation that could be expected in each year in dark grey, and the minimum and maximum amount of infestation shown in light grey (after Carroll et al. 2006). Table 2.4: The number of years (t) required to suppress infested stands of mountain pine beetle using a range of initial infestation (N0) and a range of detection accuracies where the number of red attack trees was derived from those remaining after mitigation and the expected infestation after 10 years in the non-mitigated plots. Detection accuracy Number of red attack 70% 80% 90% Average 11 6 3 768 (High) 13 7 4 64 (Low) 8 5 3 50 2.5 Discussion In British Columbia and Alberta, it is difficult to halt mountain pine beetle infestations because beetles have infested over 13 million hectares of forest (Westfall and Ebata 2009). However, with persistent detection, monitoring, and mitigation, forest managers can reduce attacking beetle populations at the infestation\u00E2\u0080\u0099s edge and strive for control. Infestations can be halted by removing all infested trees; however, attacked trees that remain after mitigation is completed will extend the duration of the local infestation. Therefore, continued monitoring in subsequent years can detect and reduce the impact of mountain pine beetles by removing infested trees. The three modelling scenarios employed utilise field data to demonstrate how infestations could behave if mitigation is not completed, whether mitigation is possible with current detection capabilities, and how long mitigation will be required to be completed. The first scenario suggests that where mitigation was conducted, the infestation appears to have been controlled, considering no infestation exists in 2014. However, compared to the infestation expansion in 2007, the 2008 attack appears to be minimal, which may be the result of environmental conditions. Extreme cold in the winter (temperatures < -40\u00C2\u00B0C), as is common in the study area, may have caused extensive beetle mortality and decreased attack in subsequent years. Differences in temperature as well as the amount of infestation already present near plots influenced the levels of infestation. The maximum number of infested trees after the 10 year modelling period was 768 and an average over all the non-mitigated plots was 125 infested trees. This range demonstrates the influence by external factors such as climate, stand and tree characteristics, and beetle pressure on stands should the average number of infested trees increase or decrease by 50%. If beetle 51 pressure is high and winter temperatures are suitable for insect development, infestation could be expected to be above average, whereas if the opposite external influences were experienced infestation levels may be expected to be lower than the average. If a proportion of trees are felled and burned, trees are removed that contain beetles that may otherwise cause infestation. In the second scenario, the mitigated plots require a minimum of less than 10% of trees to be removed to control infestations. If more than 10% of trees are removed mitigation will be more effective at controlling attack and will lead to a shorter infestation period and consequently, over the long term, fewer healthy trees will be attacked. This scenario relies on expansion factors estimated from previous studies and does not allow for increases or decreases in population survival and lacks the ability to incorporate immigration. It is important to continue monitoring these plots to ensure mitigation has been successful and to monitor for future attacks. The third scenario suggests that by using the average mitigation efficacy beetle infestations will not be controlled. However, if a detection accuracy of approximately 70% is achieved, infestations at all levels of severity are controlled within 15 years (Table 2.3). Shorter periods of time are required to control infestations if less attack is present and if the detection accuracy achieved is higher. Such information can be used to guide the need to monitor areas, although expansion factors should be taken into account, as these can fluctuate widely from year to year resulting in vastly different detection rates being required to control attack. Infestations are likely to increase rapidly if monitoring is not completed, especially if temperatures during the winter allow beetle populations to increase. 52 In scenarios 1 and 3, the expected variation around the average amount of infestation is given to demonstrate how infestations may change over an area. Variation may exist in some areas within a region because they will experience differences in climate or biophysical characteristics of the forest stand and trees, which influences beetles as they develop, with colder climates causing greater mortality of a brood as they overwinter beneath the bark of trees (Macias Fauria and Johnson 2009). Other areas may be closer to infested trees and the pressure from attacking beetles causes great infestation than in areas further away. There are also the physical properties of forests stands to consider, such as the presence of suitable host trees to support attacking beetles, and the proportion of host trees within the stand (Fettig et al. 2007). If forest stands contain a small proportion of pine trees compared to other species and are of small diameter they are less likely to be attacked than stands of large diameter pure pine. Whitehead et al. (2004) reviewed a number of manipulation studies describing the effects on forest stands by thinning trees. Beetle attack was significantly less if trees were spaced, than if they remained at the same stocking density because both wind speed and within-stand temperatures were increased. These effects are debated, Waring and Pitman (1985) posit that as stands are thinned, tree vigour increases and are able to more effectively resist beetle attack. Both sets of effects are outcomes of thinning and contribute towards reducing stand susceptibility and mortality due to mountain pine beetle attack (Coops et al. 2009). The infestation expansion rates in the field plots are highly variable, suggesting that at some locations mitigation was successful and all infested trees were removed. In others mitigation was partially complete or had not been completed, which left a proportion or 53 all previously infested trees in the plots. In some areas it may not be feasible to deploy crews until the infestation has reached a certain size, and only possible when the benefit of removing the trees is commensurate with the cost of removal. In some plots it is possible mitigation was prioritised, with higher priority given to developing infestations, once infested trees are removed mitigation is completed in areas thought to be less susceptible to attack. The biophysical characteristics of trees within the stands could also explain the variability of expansion factors. Sites with desirable features for mountain pine beetle attack (as listed by Shore and Safranyik 1992), large stem diameter, optimal stocking, greater than 80 years old, on north facing versus south facing slopes all influence the development of beetles beneath bark. Forest stands with preferred characteristics will increase the probability that beetles survive the winters and will provide a higher population of attacking beetles than other sites with less suitable host (Shore and Safranyik 1992). A second influence on the amount of expansion is temperature, some years stands experience higher temperatures than others, as a result less mortality is caused to beetle populations beneath the bark and after beetles disperse more previously unattacked trees are colonised, and infestation increases at a greater rate than in colder years. Temperature fluctuations over the year can explain some of the temporal variability found in our study sites where beetles emerge and disperse late in the year in these areas due to colder weather which slows development. In previous years, winter temperatures had not become cold enough to cause mortality and supported the development of life stages, allowing infestations to increase once adult beetles emerged, dispersed, and colonised. However, it appears that life stages from the 2008 brood did not become cold hardy to winter temperatures and were killed beneath the bark due to a 54 colder winter which caused a decrease in the level of infestation. If subsequent years experience warmer winter weather infestations are expected to increase, but will decrease if winters continue to be cold. Mitigation should concentrate on areas with higher infestation severity in forest stands where healthy trees have susceptible characteristics, such as stands older than 80 years with large diameters. The average mitigation efficacy is currently too low to provide control of doubling infestations, and will not control population expansions experienced in the non-mitigated field plots with a single treatment. I found a range of infestation expansion factors. At the uppermost expansion factor of 18, over 90% of trees currently infested by mountain pine beetle need to be removed to bring a population under control. Detection accuracy to this extent is not likely; however, expansion factors of this level appear to be extreme, with factors of near 3 being more common and more easily controlled. Once beetle pressure builds in an area, mitigation intensity will have to increase to ensure the attacking population is controlled effectively and economically. In other areas outside the study area, expansion factors may be greater and will give greater chance for the infestations to expand because beetle pressure will be higher. Persistent monitoring and detection is required to provide continued management to mountain pine beetle infestations, unless mitigation is 100% effective (Carroll et al. 2006; Coggins et al. 2008). Even so, dispersing beetles from nearby (or distant) infestations may attack trees within the stand and cause further damage. Furthermore, if susceptible trees remain within the stand (pine trees, with a stem diameter greater than 15 cm and older than 80 years) a higher likelihood of being attacked exists than for stands that do not share these known susceptibility characteristics. Mitigation is required not only on 55 Crown land in British Columbia, but also on private land, in parks, and remote areas where infestations may flare up unnoticed or may be left uncontrolled. The aerial overview surveys performed each year monitor the amount of infestation over the province (Wulder et al. 2009c), however mitigation activities are subject to certain constraints. The first being the financial constraint to private land owners, locating and removing trees before infestations become too large to control can be costly; secondly, access to infestations within parks has been restricted until recently allowing infestations to build and populations to increase unabated and provide a source of beetles to stands outside the park where land owners are attempting to control beetles; lastly, infestations are not easily detected in remote areas due to the sheer size of land to be covered; further research is required to provide accurate geographic locations of small infestations before they become larger. Infestations in each of these areas provide a viable source of beetles which given adequate climatic conditions and suitable host will disperse and expand infestations. Therefore, it is necessary to monitor these areas also, to determine how much infestation is present and to ascertain the risk to surrounding forest from dispersing beetles. 56 2.6 References British Columbia Ministry of Land, Water, and Air Protection. 2002. Indicators of Climate Change for British Columbia 2002.. (online). Available at http://www.env.gov.bc.ca/cas/pdfs/indcc.pdf. Accessed 14th December 2010. British Columbia Ministry of Forests and Range. 2002. What is the theoretical Maximum Green:Red? (online). Available at http://www.for.gov.bc.ca/hfp/health/fhdata/maxbeetles.htm. Accessed 12th July 2010. Carroll, A.L., Taylor, S.W., R\u00C3\u00A9gni\u00C3\u00A8re, J., Safranyik, L., 2004. Effects of climate and climate change on the mountain pine beetle. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Proceedings of the mountain pine beetle symposium: challenges and solution. pp. 223 \u00E2\u0080\u0093 232. October 30-31, 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Information Report BC-X-399. 298 p. Carroll, A.L., Shore, T.L., Safranyik, L., 2006. Direct Control: Theory and Practice. In Safranyik, L. and Wilson, B. (Eds.). The Mountain Pine Beetle: A Synthesis of Biology, Management, and Impacts on Lodgepole Pine. pp. 155\u00E2\u0080\u0093172. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC, Canada. 317 p. Cerezke, H.F. 1995. Egg gallery, brood production, and adult characteristics of mountain pine beetle, Dendroctonus ponderosae Hopkins (Coleoptera: Scolytidae), in three pine hosts. Canadian Entomologist. 127(6): 955 \u00E2\u0080\u0093 965. Coggins, S.B., Coops, N.C., Wulder, M.A. 2008. Initialisation of an insect infestation spread model using tree structure and spatial characteristics derived from high spatial resolution digital aerial imagery. Canadian Journal of Remote Sensing. 34(6): 485 \u00E2\u0080\u0093 502. Coops, N.C., Timko, J.A., Wulder, M.A., White, J.C., Ortlepp, M.A., 2008. Investigating the effectiveness of mountain pine beetle mitigation strategies. International Journal of Pest Management. 54(2): 151 \u00E2\u0080\u0093 165. Coops, N.C., Waring, R.H., Wulder, M.A., White, J.C. 2009. Prediction and assessment of bark beetle-induced mortality of lodgepole pine using estimates of stand vigor derived from remotely sensed data. Remote Sensing of Environment. 113(5): 1058 \u00E2\u0080\u0093 1066. Fettig, C.J., Klepzig, K.D., Billings, R.F., Munson, A.S., Nebeker, T.E., Negr\u00C3\u00B3n, J.F., Nowak, J.T., 2007. The effectiveness of vegetation management practices for prevention and control of bark beetle outbreaks in coniferous forests of the western and southern United States. Forest Ecology and Management. 238(1-3): 24 \u00E2\u0080\u0093 53. 57 Logan, J. and Powell, J. 2001. Ghost forests, global warming, and the mountain pine beetle (Coleoptera: Scolytidae). American Entomologist. Fall: 162\u00E2\u0080\u0093172. Macias Fauria, M. and Johnson, E.A. 2009. Large-scale climatic patterns control area affected by mountain pine beetle in British Columbia, Canada. Journal of Geophysical Research. 114: DOI:10.1029/2008JG000760. Maclauchlan, L.E. and Brooks, J.E. 1998. Strategies and tactics for managing the mountain pine beetle, Dendroctonus ponderosae. British Columbia Forest Service, Kamloops Region Forest Health, Kamloops, BC. 55 p. Mitchell, R.G. and Preisler, H.K. 1991. Analysis of spatial patterns of lodgepole pine attacked by outbreak populations of the mountain pine beetle. Forest Science 37(5): 1390\u00E2\u0080\u00931408. Ono, H. 2004. The mountain pine beetle problem: scope of the problem and key issues in Alberta. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Mountain pine beetle symposium: challenges and solutions. pp. 62\u00E2\u0080\u009366. 30\u00E2\u0080\u009331 October 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Victoria, British Columbia, Information Report BC-X-399. 298 p. Raffa, K.F., Aukema, B.H., Bentz, B.J., Carroll, A.L., Hicke, J.A., Turner, M.G., Romme, W.H. 2008. Cross-scale drivers of natural disturbances prone to anthropogenic amplification: The dynamics of bark beetle eruptions. BioScience 58(6): 501\u00E2\u0080\u0093517. Rice, A.V. and Langor, D.W. 2009. Mountain pine beetle-associated blue-stain fungi in lodgepole x jack pine hybrids near Grande Prairie, Alberta (Canada). Forest Pathology. 39(5): 323 \u00E2\u0080\u0093 334. Safranyik, L. 1978. Effects of climate and weather on mountain pine beetle populations. In: Berryman, A.A., Amman, G.D., Stark, R.W. (Eds.). Theory and practice of mountain pine beetle management in lodegpole pine forests. pp. 79 \u00E2\u0080\u0093 86. 25-27 April 1978, University of Idaho, Moscow, Idaho. Symposium Proceedings. 224 p. Safranyik, L., Linton, D.A., Silversides, R., McMullen, L.H. 1992. Dispersal of released mountain pine beetles under the canopy of a mature lodgepole pine stand. Journal of Applied Entomology. 113(5): 441 \u00E2\u0080\u0093 450. Shore, T.L. and Safranyik, L. 1992. Susceptibility and risk-rating systems for the mountain pine beetle in lodegpole pine stands. Pacific Forestry Centre, Canadian Forestry Service, Pacific and Yukon Region. Information Report BC-X-336. 12 p. Shore, T.L., Safranyik, L., Whitehead, R.J. 2006. Priniciples and concepts of management. In: Safranyik, L. and Wilson, B. (Eds.). The Mountain Pine Beetle: A Synthesis of Biology, Management, and Impacts on Lodgepole Pine. pp. 117\u00E2\u0080\u0093 122. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC, Canada. 317 p. Stahl, K., Moore, R.D., McKendry, I.G. 2006. Climatology of winter cold spells in relation to mountain pine beetle in British Columbia, Canada. Climate Research. 32(1): 13 \u00E2\u0080\u0093 23. 58 Taylor, S.W., Carroll, A.L., Alfaro, R.I., Safranyik, L. 2006. Forest, climate and mountain pine beetle outbreak dynamics in western Canada. In: Safranyik, L. and Wilson, B. (Eds.). The Mountain Pine Beetle: A Synthesis of Biology, Management, and Impacts on Lodgepole Pine. pp. 67\u00E2\u0080\u009394. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC, Canada. 317 p. Waring, R.H. and Pitman, G.B. 1985. Modifying lodgepole pine stands to change susceptibility to mountain pine beetle attack. Journal of Ecology. 66(3): 27 \u00E2\u0080\u0093 41. Westfall, J. and Ebata, T. 2009. 2008 Summary of Forest Health Condition in British Columbia. Pest Management Report Number 15. British Columbia Ministry of Forests and Range, Forest Practices Branch, Victoria, BC. 70 p. Whitehead, R.J., Safranyik, L., Russo, G.L., Shore, T.L., Carroll, A.L. 2004. Silviculture to reduce landscape and stand susceptibility to the mountain pine beetle In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Mountain pine beetle symposium: challenges and solutions. pp. 223\u00E2\u0080\u0093232. 30\u00E2\u0080\u009331 October 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Victoria, British Columbia, Information Report BC-X-399. 298 p. Whitehead, R.J., Safranyik, L., Shore, T.L. 2006. Preventive management. In: Safranyik, L. and Wilson, B. (Eds.). The Mountain Pine Beetle: A Synthesis of Biology, Management, and Impacts on Lodgepole Pine. pp. 173\u00E2\u0080\u0093192. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC, Canada. 317 p. Wulder, M.A., Dymond, C.C, White, J.C., Leckie, D.G., Carroll, A.L. 2006. Surveying mountain pine beetle damage of forests: A review of remote sensing opportunities. Forest Ecology and Management. 221(1-3): 27 \u00E2\u0080\u0093 41. Wulder, M.A., White, J.C., Carroll, A.L., Coops, N.C. 2009a. Challenges for the operational detection of mountain pine beetle green attack with remote sensing. The Forestry Chronicle. 85(1): 32 \u00E2\u0080\u0093 38. Wulder, M.A., Ortlepp, S.M., White, J.C., Coops, N.C., Coggins, S.B. 2009b. Monitoring the impacts of mountain pine beetle mitigation. Forest Ecology and Management. 258(7): 1181 \u00E2\u0080\u0093 1187. Wulder, M.A., White, J.C., Grills, D., Nelson, T., Coops, N.C., Ebata, T. 2009c. Aerial overview survey of the mountain pine beetle epidemic in British Columbia: Communication of impacts. BC Journal of Ecosystems and Management. 10(1): 45\u00E2\u0080\u009358. Wygant, N.D. 1942. Effects of low temperature on the Black Hills beetle (Dendroctonus ponderosae). PhD thesis. U.S. Department of Agriculture Forest Service, Rocky Mountain Forest and Range Experimental Station, Fort Collins, Colorado. 57 p. 59 3 LINKING SURVEY DETECTION ACCURACY WITH ABILITY TO MITIGATE POPULATIONS OF MOUNTAIN PINE BEETLE4 3.1 Introduction The natural range of the mountain pine beetle occupies much of the province of British Columbia. Overview surveys indicate that the beetle affected 164,000 hectares of lodgepole pine forest (Pinus contorta Dougl. ex. Loud var. latifolia Engl.) in 1999 (British Columbia Ministry of Forests and Range 2000), which increased to 10.1 million hectares by 2007 (Westfall 2007). The area infested has approximately doubled each year between 1999 and 2004, with the highest increase in infested area experienced in 2000 (British Columbia Ministry of Forests and Range 2001). Large tracts of susceptible forests and a lack of sufficiently cold winters, among other factors, have favoured population growth and resulted in an expansion in the mountain pine beetle's range northward to higher elevations in British Columbia (Safranyik and Carroll 2006; Taylor et al. 2006), and in increasing numbers, eastward into Alberta, where 2.8 million trees were surveyed as attacked by the beetle in 2006 (Alberta Sustainable Resource Development 2007). Infestations occur largely in lodgepole pine forests, but there is increasing concern beetles may successfully reproduce in jack pine (Pinus banksiana Lamb) forests (Logan and Powell 2001; Carroll et al. 2004), enabling further eastward expansion into the vast Canadian boreal forest. 4 A version of this chapter was published as: Coggins, S.B., Wulder, M.A., Coops, N.C. 2008. Linking survey detection accuracy with ability to mitigate populations of mountain pine beetle. Forestry Chronicle. 84(6): 900 \u00E2\u0080\u0093 909. 60 Mountain pine beetle attack is commonly described as occurring in three stages that sequentially manifest (albeit at a variable rate) over a three year period. Infestation of trees initially occurs between late July and early August when adult beetles emerge from beneath the bark of attacked trees and commence flight in search of new host trees. Initially, attacked trees are referred to as green attack because beetles have attacked the stem of trees and have begun to form egg-laying galleries beneath the bark. However, at this stage, the tree's foliage remains green. The following year, foliage becomes desiccated and fades from green to yellow to red. The fade is the result of a combination of factors, including gallery development and fungal inoculation that impacts the ability of the tree to translocate nutrients (Safranyik and Carroll 2006). While fade rates are variable, it can be expected that 90% of attacked trees will have red needles one year after attack (Wulder et al. 2006a), followed by shedding of needles from the branches, resulting in the final stage known as grey attack (Wulder et al. 2006b). Usually, beetles exist as endemic populations (Table 3.1) attacking small groups of trees, and are controlled by lethal cold winter temperatures which cause mortality to beetles (Maclauchlan and Brooks 1998) acting to either maintain or decrease population levels. The lethal low temperature changes according to the time of year, whereby if beetle larvae have not fully developed mortality is caused at less extreme temperatures. Maximum cold-hardiness is achieved in December-January and large larvae can potentially survive short exposures to -38\u00C2\u00B0C during this period. During the fall and spring, when beetles are more susceptible to extreme cold, unseasonably low temperatures (< -26\u00C2\u00B0C) can cause widespread mortality (Safranyik and Linton 1991). Indeed, the last extensive infestation in British Columbia occurred on the Chilcotin 61 Plateau during the early 1980\u00E2\u0080\u0099s and was eventually halted by low temperature events in 1984 and 1985 where sustained low temperatures caused significant mortality and halted the infestation in 1987. In 1984, temperatures of -26\u00C2\u00B0C were experienced on October 31st Table 3.1: Characteristics associated with the population states of mountain pine beetle and the likely rate of population expansion associated with each population state (adapted from Wulder et al. 2006b). Population state Population characteristics Likely population expansion rate Endemic \u00E2\u0080\u00A2 Widespread in mature pine forests; however they are restricted to weakened and decadent trees. \u00E2\u0080\u00A2 Frequently found in trees attacked by secondary bark beetle species. Trees containing mountain pine beetles can be very difficult to locate on the ground and even from the air since many of the trees will be in the intermediate to suppressed crown classes, the faded crowns of which are partially hidden below the crowns of taller, uninfested trees. \u00E2\u0080\u00A2 Currently attacked trees are often not located near brood trees. \u00E2\u0080\u00A2 There is no obvious relationship between the probability of attack and tree diameter. \u00E2\u0080\u00A2 Yearly tree mortality is normally less than volume growth. R = < 2 Incipient- epidemic \u00E2\u0080\u00A2 Most infested trees are in the larger diameter classes. \u00E2\u0080\u00A2 Clumps of infested trees are scattered and confined to some stands. \u00E2\u0080\u00A2 The infested clumps vary considerably in size and number from year to year but tend to grow over time. \u00E2\u0080\u00A2 Frequently, the groups of infested trees first appear in the following situations; draws and gullies, edges of swamps or other places with wide fluctuations in the water table; places where lodgepole pine is growing among patches of aspen, perhaps indicating the presence of root disease; dry, south and west-facing slopes. R = > 2 Epidemic \u00E2\u0080\u00A2 Resilient to large proportional losses through natural mortality. \u00E2\u0080\u00A2 Generation mortality is usually in the range of 80-95%, corresponding to potential rates of population increase of two-eight fold. The usual rate of increase, however, is two- four fold when measured over the entire epidemic area. \u00E2\u0080\u00A2 Infestations are widespread and exist at the landscape-level. \u00E2\u0080\u00A2 There are usually large annual increases in both infested areas and numbers of infested trees. R = 4 or 5 * Table is adapted from Wulder et al. (2006b). 62 which further decreased to -43\u00C2\u00B0C by December 30th (Safranyik and Linton 1991). In 1985, lethal low temperatures ranged from -27.5\u00C2\u00B0C on November 11th decreasing to - 43\u00C2\u00B0C on November 27th and remaining below -30\u00C2\u00B0C until December 2nd (Safranyik and Linton 1991). Recent climate records for British Columbia indicate temperatures have been warming steadily over the past decade, enabling beetle populations to survive winters (Carroll et al. 2004) and increase to epidemic population levels (British Columbia Ministry of Forests and Range 2006). In instances where natural control of beetle populations is ineffective, anthropogenic controls are typically initiated. A range of survey methods are used to detect mountain pine beetle infestations and these methods can be considered in the context of an information hierarchy, whereby each method provides greater detail over a smaller spatial extent (Wulder et al. 2006c). There are generally three scales of observation for surveying mountain pine beetle damage: regional (coarse), landscape (medium), and local (fine) (Wulder et al. 2006b). Regional- scale surveys are conducted on an annual basis by trained specialists in aircraft, who delineate broad areas of infestation on topographic maps (e.g., aerial overview surveys or sketch-mapping) (Wulder et al. 2006a). Landscape-scale surveys typically utilize data generated by digital remote sensing (e.g., Landsat Thematic Mapper sensor) that can provide large spatial coverage with low per unit area image costs (Franklin et al. 2003) and can supply quantitative information on the extent of infestations (Wulder et al. 2004). Local-scale surveys are used to detect single infested trees or small groups of infested trees by helicopter-GPS surveys (heli-GPS), 1:30 000 scale (or larger) colour aerial photographs, and high-spatial resolution satellite imagery (<1 m pixel size) and are augmented and validated with targeted field observations. Each scale of observation will 63 satisfy a specific information need; for example, regional-scale surveys are able to capture provincial-level infestation status, whereas local-scale surveys capture fine detail in specific areas. The survey intensities and the information required differ and consequently a given survey may cover large or small areas and the information returned is governed by the scale of observation (Wulder et al. 2004). Ground crews locate red attack trees and associated green attack trees by conducting walkthrough surveys or \u00E2\u0080\u0098probes\u00E2\u0080\u0099. These surveys obtain data on the proportion of green attack in forest stands and the amount of remaining susceptible host material (Maclauchlan and Brooks 1998). Walkthroughs are ground-based surveys completed prior to probes to locate attack and provide general estimates of the severity and extent of infestations. Probes are systematic, strip type surveys that provide highly detailed information on the extent of the infestation, the susceptibility of stands to further attack, and identify potential constraints to harvesting (British Columbia Ministry of Forests 1995; Maclauchlan and Brooks 1998). This hierarchy of survey data sources can be used to reduce the cost and time required to collect detailed information on beetle infestations; regional-scale data may be used to identify broad areas of infestation and guide the acquisition of landscape-scale data, which in turn may be used to guide local-scale data acquisition and the deployment of field crews. Each survey method has limitations relating to the timeliness of data acquisition and the results provided, data cost, capacity to detect red attack damage, the spatial extent of the survey, and the speed with which the survey can be completed and information provided to forest managers (Wulder et al. 2006b). For example, it is critical to deploy ground crews to conduct mitigation based on data collected during recent 64 surveys so that the full extent of the infestation is realised and appropriate control methods can be used. In general, it is not economically feasible to use local-scale surveys over large areas and therefore regional-scale and landscape-scale surveys have an important role to play to efficiently and cost effectively stratify the focus area and identify general locations of infested forest stands. Local-scale surveys are preferred for the detection of individual infested trees; however, high spatial resolution imagery typically has a limited spatial extent and is significantly more expensive than regional- or landscape-scale surveys (Wulder et al. 2006c). This is problematic when small, isolated groups of infested trees (which are not detectable with coarser data sources) are scattered over large areas, as these spot infestations may remain undetected until they expand and coalesce with other similar spots and cause extensive mortality. Furthermore, surveys are subject to errors of commission (trees incorrectly classified as red attack) and omission (red attack trees missed during image interpretation), with higher error rates associated with coarse-scale survey approaches and lower error rates with local-scale approaches (Nelson et al. 2006). Detection of red attack stage trees is used as a primary means for locating green attack stage trees (Wulder et al. 2004). Red attack trees may be identified relatively consistently and accurately, following both digital or manual approaches (Wulder et al. 2006b). Based upon an understanding of mountain pine beetle spread, green attack stage trees are known to be found in close proximity to existing red attack trees. Thus, the use of remotely sensed data to map green attack trees in the proximity of red attack trees provides little operational gain over existing practices. Locating green attack trees not in proximity to red attack stage trees (i.e., seeking locations of infestation due to long-range transport) is 65 limited by biological, environmental, geographic, temporal, economic, and technological considerations which limit the operational utility of remotely sensed data for green attack detection. Therefore, locating green attack trees using remotely sensed data, from an operational point of view, is currently not realistic (Wulder et al. 2006b). Green attack trees are the primary target of mitigation activities. Several mitigation methods exist which can be implemented either on individual trees or small groups of trees (Maclauchlan and Brooks 1998). In large infestations, mitigation tactics include: sanitation harvesting, where standard silvicultural practices such as clearcutting, shelterwood, and selective cutting are utilized to remove infested trees; salvage logging, where dead trees are removed and processed for lumber; high-hazard host removal, where the stands deemed at highest risk from mountain pine beetle attack are removed to decrease the risk of attack to the forest; and a harvest priority rating system, where forest stands with the heaviest beetle concentrations are removed first. Mitigation tactics for smaller infestations can be grouped into direct control methods that involve removing infested trees to reduce current and future population levels, and indirect methods, which attempt to prevent trees from becoming infested (i.e., stand thinning, pesticides). Common mitigation tactics include: felling and removing single trees or patches of trees (i.e., infestations less than 1 ha in size are removed); felling and burning of individual trees; applying pesticides to individual infested trees to either prevent beetles from boring beneath the bark or to cause mortality of beetles already beneath the bark; and baiting areas with synthetic pheromones to aggregate beetles into areas that are subsequently harvested or treated. There has only been one documented incident of successful mitigation, which occurred near Banff, Alberta in the early 1940\u00E2\u0080\u0099s 66 (Hopping and Mathers 1945); with the use of aggressive and persistent mitigation tactics, an outbreak of mountain pine beetle was declared extinct 3 years after initial infestation (Carroll et al. 2006). Persistent mitigation, completed annually, has the potential to slow infestations or even cause localised extirpation; however, if the accuracy of red attack detection is limited, infestations may continue to spread. 3.2 Objectives In this chapter I present the implications of survey detection accuracies on mitigation activities designed to control mountain pine beetle populations. First, I provide background on the use of conventional forest health survey data and digital remotely sensed imagery for the detection of mountain pine beetle infestations. Each of the scales of observation currently used for detecting mountain pine beetle red attack in British Columbia are linked to the scale of measurement and described according to the inherent accuracy of the data source at detecting the status of mountain pine beetle infestations. Second, I follow with a complementary discussion of how detection accuracies of a range of survey techniques can be utilized to provide mitigation data to slow the progress of infestations. The ramifications of undetected infested trees on forest stands are then presented. Finally, I conclude by calculating the number of trees that must be removed to maintain populations at endemic levels and prevent an infestation from spreading, and estimating the period of time required to monitor infested stands, as a function of detection accuracy, to ensure all infested trees are removed. 67 3.3 Review of current methods for red attack detection Following the nomenclature of Wulder et al. (2006c), three scales of observation of mountain pine beetle detection can be defined: regional, landscape, and local (Table 3.2). A literature search was conducted to compile studies investigating mountain pine beetle infestation that had a reported a range of red attack detection accuracy and which were conducted in a similar manner (Table 3.2). From these studies I produced an assessment of the documented range from each of the three scales and used two studies to demonstrate the efficacy of detection data to drive mitigation activities. I provide examples of recent studies that provide red attack detection accuracy statements in the sections below. 3.3.1 Regional scale Aerial overview surveys, also known as sketch mapping, are primarily used for regional- scale detection of infestations across federal, provincial, and state government jurisdictions (Wulder et al. 2004). In British Columbia, the main objective of aerial overview surveys is to monitor and record the extent, severity, and general location of insect infestations in forests on an annual basis (Westfall 2007). The surveys provide a broad overview at map scales between 1:100 000 to 1:250 000 (Wulder et al. 2006b). In comparison to other survey methods, data acquired from aerial overview surveys is considered to be an effective, low-cost detection and mapping method (Wulder et al. 2006c). To date, no rigorous accuracy assessment has been conducted for this data 68 Table 3.2: Data sources for remote sensing of mountain pine beetle red attack mapping including sources and detection accuracies. Scale Technology Source (Date) Reported detection accuracy Regional Aerial overview survey To date, no rigorous accuracy assessment has been conducted for this data. N/A Landscape Single date Landsat TM Multiple date Landsat TM (TCT and EWDI) Landsat ETM+ (TCT and EWDI) Franklin et al. (2003) Skakun et al. (2003) Wulder et al. (2006d) 73% (\u00C2\u00B1 7%) 76% (\u00C2\u00B1 12%) (groups of 10-29 RA trees) 81% (\u00C2\u00B1 11%) (groups of 30-50 RA trees) 86% (\u00C2\u00B1 7%) Local Heli-GPS IKONOS Nelson et al. (2006) White et al. (2005) 92.6% (\u00C2\u00B1 10 infested trees) 71% (\u00C2\u00B1 8%) (lightly infested stands; 1-5% RA) 92% (\u00C2\u00B1 5%) (moderately infested stands; 5-20% RA) source; however, there are issues related to positional and attribution accuracy (Aldrich et al. 1958; Nelson et al. 2004). Errors which may occur in aerial overview survey data have been attributed to off-nadir viewing, distortions due to lighting conditions, interpreter fatigue, and level of experience (Aldrich et al. 1958, Leckie et al. 2005, Wulder et al. 69 2006a, Wulder et al. 2006b). Interpreter variability has been tested by conducting a comparison between the numbers of infested trees observed from the aerial overview surveys against the number of infested trees counted on aerial photographs (Harris and Dawson 1979). Estimates were found to vary widely, with interpreters varying by -42% to 73%, with a mean deviation of 7% for experienced interpreters and an 8% deviation for inexperienced interpreters. 3.3.2 Landscape scale Detailed surveys conducted at the landscape-scale can provide spatially explicit estimates of the number of infested trees and the volume of timber affected. The data generated from these surveys provides information for tactical plans, which provide methods to implement the broad objectives outlined in a forestry strategic plan and specify areas that require more detailed ground surveys. Recently, landscape-scale information of forest condition has been derived from the Landsat satellite which has shown utility in detecting mountain pine beetle infestations (Skakun et al. 2003, Franklin et al. 2003, Wulder et al. 2006d). For example, Franklin et al. (2003) found that when using single-date Landsat TM imagery with a 30 m pixel size over areas of 0.2 ha, it was possible to generate a red attack detection accuracy of 73.3% \u00C2\u00B1 6.7%, p = 0.05 (Franklin et al. 2003). Multi-date imagery, one prior to, and one following, attack can also be used to monitor forest change due to mountain pine beetle infestations. These approaches typically include image transformations which utilise specific Landsat band combinations designed to enhance a range of forest conditions. The Tasselled Cap Transformation (TCT) was used to process multi-date imagery to obtain wetness indices and the Enhanced Wetness Difference Index (EWDI) used in turn to interpret spectral patterns in stands with 70 confirmed red attack (Skakun et al. 2003). This approach produced an accuracy of 76% (\u00C2\u00B1 12%, p < 0.05) for groups of 10 to 29 infested trees, and 81% (\u00C2\u00B1 11%, p < 0.05) for groups of 30 to 50 infested trees. To reduce the reliance upon the application of change thresholds, Wulder et al. (2006d) used the Enhanced Wetness Difference Index (EWDI) in conjunction with slope and elevation surfaces. These data were analyzed using a logistic regression approach to map red attack damage in the Lolo National Forest in Montana, USA. With this method, red attack was mapped with 86% accuracy (\u00C2\u00B1 7%). In addition, new generation satellite sensors such as Hyperion, onboard EO-1, offer improved spectral sensitivity over existing systems (White et al. 2007). A single date of Hyperion imagery was used to generate six moisture indices and then compared to the proportion of each Hyperion pixel that was independently surveyed as red attack stage. Results indicated that moisture indices incorporating the shortwave infrared and near infrared regions were significantly correlated to levels of damage (r2 = 0.51; p < 0.001). This study demonstrated that Hyperion data may be used to map low-level infestations of mountain pine beetle red attack at the landscape scale. 3.3.3 Local scale Local-scale surveys detect low infestation levels and have traditionally included surveys conducted with a helicopter where the geographic location of infestation centres are recorded with a global positioning system (a process otherwise known as heli-GPS) and fine-scale aerial photographs (1:30 000). Heli-GPS points are a rapid and accurate means for detailed mapping of mountain pine beetle red attack. Using this method, tree clusters or individual tree counts, and damage severity are identified. Nelson et al. (2006) assessed the spatial accuracy of 100 heli-GPS points delineated during aerial surveys and 71 compared them to ground data (Nelson et al. 2006). Results were calculated for several error ranges and indicated that 17.4% of points correctly identified the number of infested trees when comparing heli-GPS estimates to field data, but contained high errors of commission and omission (Nelson et al. 2006). Heli-GPS data were shown to be 92.6% accurate with an error of \u00C2\u00B110 trees, which provides sufficient detail and accuracy for regional planning and management purposes. An additional survey method used to provide local-scale data is aerial photography. Fine- scale aerial photographs, acquired at a scale of 1:30 000, enable identification of areas with low infestation levels and provide data which support mitigation activities by supplying geographic locations and number of trees infested (Wulder et al. 2006c). The photographs are digitized (scanned) and infested trees are visually interpreted using digital photogrammetric software. To date, no objective accuracy assessment has been conducted for this data source. Detailed surveys at the stand level can also be undertaken using high-spatial resolution remotely sensed imagery with a pixel size of <5 m. High-spatial resolution digital satellite imagery is available to supply detection information for insect infestations. IKONOS is a high-spatial resolution satellite which can be used to detect mountain pine beetle infested trees, with a multispectral pixel size of 4 m (White et al. 2005). An unsupervised clustering of image spectral values was used to detect mountain pine beetle infestations in lightly or moderately infested trees near Prince George, British Columbia. Light infestations were categorized when 1% to 5% of a forest stand contained infested trees. Similarly, moderate infestations were defined where >5% to <20% of the forest stand contained attacked trees. To account for positional error a 4 m (one pixel) buffer 72 was applied to red attack control data for comparison with the image classification. When compared to independent validation data collected from aerial photography, the accuracy of red attack detection from IKONOS imagery was shown to be 71.0% for lightly infested stands and 92.5% for moderately infested stands. 3.4 Effects of mitigation on mountain pine beetle infestations To facilitate discussion on how mitigation can affect mountain pine beetle infestations, I used population-scale modelling scenarios (Carroll et al. 2006) to predict the impact of management tactics on beetle populations. In this discussion I suggest a theoretical framework for the suppression of mountain pine beetle infestations using mitigation driven by detection accuracy statements. This framework uses mountain pine beetle population dynamics in a population scale model to predict the level of mitigation required to bring populations under control, the extent of mountain pine beetle damage following a single mitigation event, and the number of years mitigation must continue to suppress infestations. Population-scale models consider the interactions between a forest disturbance agent (e.g. mountain pine beetle) and its host species to predict the spread of infestation over time. For instance, models can estimate the current population size based on the previous year\u00E2\u0080\u0099s population (Barclay et al. 1985, Thompson 1991) and use mathematical functions to predict population fluctuations, based on numerous parameters, such as weather and host conditions (Mitchell and Preisler 1991; Thompson 1991; Beukema et al. 1997; Malmstr\u00C3\u00B6m and Raffa 2000). Modelling scenarios allow us to examine the effects of mitigation driven by detection accuracies to suppress beetle populations; assess the extent 73 of an infestation following a single mitigation event; and finally, estimate the length of time required to suppress infestations given differing levels of detection accuracy. It is important to consider the role played by detection accuracies in an ongoing mitigation program where the objective is to halt the infestation or stabilize beetle populations. Table 3.1 provides examples of the likely rates of population expansion associated with various population states of the mountain pine beetle. In northern British Columbia it is common for the rate of beetle population increase to double on an annual basis (British Columbia Ministry of Forests and Range 2002) and then remain at a constant rate for a number of years. To maintain a static population level when the population is doubling, at least 50% of infested trees must be detected and removed before the annual flight period. To reduce the population when the rate of growth is 2 (population is doubling), more than 50% of infested trees must be removed, provided the rate of population increase remains constant. However, populations regularly increase 4- fold, and as much as 5-fold, in areas of southern British Columbia, but rarely exceed this level (British Columbia Ministry of Forests and Range 2002). Under these growth scenarios a high proportion of infested trees must be treated in order to reduce the population. The proportion (P) of trees requiring mitigation according to the rate of population increase is defined as (Carroll et al. 2006): P = 1-1/R (3.1) where (R) is the rate of population expansion. Detection accuracy data can be incorporated into Equation 1 when used as a surrogate for the proportion of trees requiring mitigation (P). I utilise three detection accuracies in three population-based modelling scenarios to demonstrate mitigation efficacy (assuming mitigation is 100% 74 effective) and estimate the subsequent effects on mountain pine beetle populations. For the modelling scenarios it is pertinent to include only survey methods that are suited to spatially detailed detection of mountain pine beetle infestations. It is unlikely that regional-scale data would be used to guide mitigation because although these survey methods are critical for broad classification of the severity and extent of infestations on the landscape, they lack the spatial and attribute accuracy necessary to guide mitigation efforts. Therefore, finer-scale data sources, such as those available at the landscape and local scales, are used to supply spatial and attribute data to guide mitigation crews. Landscape-scale data such as Landsat has a 30 m by 30 m pixel, which may contain an amalgamation of several forest elements typical of a pine stand, i.e. trees, shadowing, and understorey (Wulder et al. 2006b). This amalgamation may dilute the appearance of red attack tree crowns in each pixel and make detailed mapping of red attack difficult because patches of infestation will become clear only when the majority of trees within a pixel become infested. Local-scale infestations do detect individual infested trees and also a range of infestation severity. In order to complete the modelling I used generic detection accuracies at 70%, 80%, and 90% (Table 3.3), which represent the range of accuracies reported in Table 3.2. The results of Equation 3.1 are graphically represented in Figure 3.1. The modelling demonstrates the potential effect that detection accuracies have on a beetle population assuming a single detection activity over the duration of an infestation. Carroll et al. (2006) indicate that if mountain pine beetle populations increase by a factor of 2 (R = 2), starting with one undetected infested tree in year 1, then by year 10 that stand will contain 512 infested trees, which is estimated to be 2% of the trees contained within a 20 75 Table 3.3: Number of years (t) required to suppress infested stands of mountain pine beetle using a range of initial infestation (N0) and a range of detection accuracies representative of the survey methods discussed. Scale of observation Detection accuracy 1000 2000 5000 10000 100000 70% 13.5 14.9 16.7 18.0 22.5 80% 7.5 8.3 9.3 10.1 12.6 90% 4.3 4.7 5.3 5.7 7.2 hectare area (Carroll et al. 2006). Hypothetically, infestations of this magnitude could escape detection for some time depending on the scale of observations utilized and whether the infestation is scattered throughout the stand or is clumped in a single infestation. However, if in year 4, a proportion of infested trees is removed, infestations expand to lower levels than previously experienced. This implies that when mitigation is completed on 70% of infested trees, 5 trees are correctly identified, resulting in the remaining population expanding to 154 infested trees by year 10. If 80% of infested trees are removed, 3 infested trees would remain in the stand and by year 10, the infestation will have expanded to cause mortality of 102 trees. Lastly, if 90% of infested trees are correctly identified in year 4, 7 trees would be detected and by year 10, 51 trees would be infested (Figure 3.1). Two examples of survey methods that detect moderate to low-levels of infestation are provided and include accuracies generated from high-spatial resolution satellites and heli- GPS data (Figure 3.2). First, the satellite IKONOS was used to detect mountain pine beetle red attack in forest stands with low (1% to 5% of stand infested) and moderate levels (>5% to 20% of stand infested) of infestation (White et al. 2005). The study was 76 designed to examine the potential of 4 m multispectral IKONOS imagery to detect red attack and map areas suitable for suppression. Results show that imagery detected 93% (\u00C2\u00B1 5%) of red attack trees when identifying >5% to 20% of the stand as infested and 71% (\u00C2\u00B1 8%) of attacked trees when 1% to 5% of the stand is infested (White et al. 2005). Figure 3.1: Prediction of the number of trees killed using a range of detection accuracies. In incipient-epidemic conditions, populations triple annually (R = 3) and can be suppressed using a detection accuracy above 71% (Figures 3.2). In epidemic conditions where population expansion rates can be R = 5, mitigation driven by detection accuracies of 92% (Figure 3.2) is able (assuming 100% mitigation success) to reduce 77 Figure 3.2: Generalized prediction of the effect of detection accuracies given by two detection accuracies provided by recent research; 92% and 71%. 78 the population to less than populations observed in previous years. Similarly, using a detection accuracy of 71% results in population reduction at expansion levels of R = 4 (Figure 3.2B). These examples indicate that a decrease in detection accuracy causes a decrease in the proportion of attacked trees removed from an infestation. If undertaking mitigation with an accuracy lower than 70%, suppression becomes even less possible. When considering the margin of error associated with accuracy statements (as shown in Figures 3.2A and 3.2B), the efficacy of resulting mitigation may be decreased further. The lower error bar in each of the examples provided in Figure 3.2 infers a decrease in the proportion of attacked trees detected and subsequently removed. When providing mitigation for a doubling population (R = 2) detection accuracies of 50% are adequate; however, if it less than 50% mountain pine beetle populations will be able to rapidly increase each year when residual infestation is fuelling future expansion. The final scenario calculates the length of time required for ongoing detection, monitoring, and mitigation to bring a given infestation under control. In order to effectively control and reduce populations a proportion of infested trees should be removed each year as part of a persistent mitigation program (Carroll et al. 2006). During an outbreak, the number of trees killed annually can often be in the millions and may cover hundreds of thousands of hectares. Due to the magnitude of these infestations, management tactics to reduce the entire population remain futile even though populations may only increase at the same rate as low-level expansion (R = 2) (Carroll et al. 2006). For example, Carroll et al. (2006) describe an outbreak of 300 000 hectares with R = 2, where 150 000 hectares of infested trees must be mitigated each year to ensure the infestation remains stable. In this situation, removing such a large number of infested 79 trees would be impossible if attempting a single mitigation activity (Carroll et al. 2006). Further, existing rules regarding sustainable forest management, including limitations to the land-base eligible for harvest (e.g., operability, road access, land-use, etc.) and annual allowable cuts, must also be considered. Large area infestations require persistent mitigation for several years (equation 3.2): N = N0[R(1-P)] t (3.2) where the number of trees initially infested (N0), the yearly rate of increase (R), the proportion of trees treated each year (P), and the number of years (t) (equation 3.2). N is therefore an estimate of the number of trees infested in any given year. Knowledge of R and P can determine the number of years needed for continuous direct suppression as defined in equation 3.1 (Carroll et al. 2006). To further explore this concept, five examples were developed using initial infestations of 1000, 2000, 5000, 10 000, and 100 000 trees (Figure 3.3). For each example it is possible to calculate the number of years needed to suppress a mountain pine beetle infestation using the data from each scale of observation (Table 3.3). Mitigation using a detection accuracy of 90% is estimated to bring infestations under control in the shortest time. If the initial infestation is limited to individual infestations on the landscape, mitigation may be completed within 4 years (N0 = 1000); however, for larger infestations persistent mitigation may be required for 7 years (N0 = 100 000). A detection accuracy of 80% provides data to estimate that under an intensive mitigation program, infestation are controlled within 8 years (N0 = 1000) extending to 13 years for larger infestations (N0 = 100 000). Mitigation took longest when guided by the 70% detection accuracy to bring infestations under control, 80 infestations are estimated to be controlled within 14 years (N0 = 1000) and large-scale infestations within 23 years (N0 = 100 000). Figure 3.3: Number of years required to suppress infestations of mountain pine beetle using detection accuracies (\u00C2\u00B1 known error) generated from remotely sensed and conventional data sources (after Carroll et al. 2006). 3.5 Use of survey data to monitor mountain pine beetle infestations In this study, I have assumed that mitigation is 100% effective; however, trees exhibiting evidence of infestation may be overlooked during surveys, because foliage appears to be healthy and there is no capacity to detect damage caused by attacking beetles to the bole of the tree. Mitigation efficacy has not been thoroughly assessed, although Fettig et al. (2007) report results from a mitigation study on the southern pine beetle. This beetle 81 shares similar traits to the mountain pine beetle in that it is a bark beetle that aggressively attacks pine trees causing scattered, low-level mortality of forest stands which, given favourable conditions, can lead to severe infestations on the landscape. Mitigation of detected infestations reduced population expansions by 77.4% when compared to untreated controls (Fettig et al. 2007). In addition Coops et al. (2008) reviewed mountain pine beetle mitigation strategies and the impact of these approaches on insect population levels. In British Columbia mitigation tactics typically employ fall and burn strategies, which require ground crews to identify attacked trees, fall and burn trees to kill beetles beneath the bark. Other suggested techniques to control beetles include prescribed burning (landscape-level), silvicultural treatments (forest stand level), and altering chemical cues received and emitted by beetles (individual tree level). The effectiveness of mitigation approaches may be monitored using a similar hierarchical approach. Forest managers face the task of deciding which treatment best suits the scale of infestation, according to the rate of population expansion and the cost associated with each type of survey method. It therefore follows that fine-scale, high accuracy data from local-scale data are critical in individual tree level identification whereas landscape-scale surveys link to silvicultural-based approaches, and finally coarse-scale mitigation techniques such as fire associated with regional-scale data. When deciding on an appropriate survey method to guide mitigation it is important to consider the limitations associated with each survey method, the information required from the survey, and the time it takes to receive the survey data. For example, aerial overview surveys provide a broad measure of the severity and extent of an infestation over large areas, but can take approximately 6 months to plan, collect, and prepare data 82 for use. Exceptions to this include cases where data are intended to be used to aid planning of more detailed surveys and can be made available within 2 to 3 months. Furthermore mountain pine beetle information is usually available at the beginning of November proceeding survey completion. Heli-GPS surveys, however, provide data on individual trees in small forest stands which can be available upon completion of survey. Therefore, the type of survey utilized depends on the last known extent of an infestation or whether new infestation is being detected. If an infestation is known to be established it is more likely broad-scale technologies will be used to monitor the annual spread. If a new infestation is discovered and rapid data acquisition is important, surveys such as heli-GPS should be utilized which rapidly acquire highly accurate data to guide mitigation efforts. The type of survey selected is also subject to financial limitations: aerial overview surveys have been estimated to cost $0.01 per hectare, whereas heli-GPS costs $0.15 per hectare (Wulder et al. 2006c). Survey costs increase according to the spatial resolution of the data used, where high-spatial resolution satellite imagery can cost up to $28.60 / km2 over a minimum area of 64 km2 (Wulder et al. 2006c) for a cost per hectare of $0.26. Overall, the greater the detection accuracy utilized the more successful mitigation will be, albeit subject to the constraints mentioned above. Either funds should be allocated to supply high accuracy detection in the short term or allow for persistent mitigation in the long term. However, long-term monitoring is needed on a continuous basis, not just when populations begin to reach incipient-epidemic levels and end-users must possess the skills and capacity to utilize data when received. 83 3.6 References Alberta Sustainable Resource Development. 2007. Beetle Bulletin. (online). Available at http://srd.alberta.ca/forests/pdf/Beetle_Bulletin_May_July.pdf. Accessed 16th October 2007. Aldrich, R.C., Heller, R.C., Bailey, W.F. 1958. Observation limits for aerial sketch- mapping southern pine beetle in the southern Appalachians. Journal of Forestry. 56: 200 \u00E2\u0080\u0093 202. Barclay, H.J., Otvos, L.S., Thomson, A.J. 1985. Models of periodic inundation of parasitoids for pest control. Canadian Entomologist. 117: 705 \u00E2\u0080\u0093 716. Beukema, S.J., Greenough, J.A., Robinson, D.C.E., Kurz, W.E., Smith, E.L., Eav, B.B. 1997. The westwide pine beetle model: a spatially-explicit contagion model. In: Teck, R., Moeur, M., Adams, J. (Eds.). Proceedings of the Forest Vegetation Simulator Conference. 3\u00E2\u0080\u00937 February, Fort Collins, CO. pp. 126 \u00E2\u0080\u0093 130. Department of Agriculture, Forest Service, Intermountain Research Station, Gen. Tech. Rep. INT-GTR-373. Ogden, UT. 222 p. British Columbia Ministry of Forests and Range. 1995. Bark beetle management guidebook (online). Available at http://www.for.gov.bc.ca/tasb/legsregs/fpc/fpcguide/beetle/chap3a.htm. Accessed 16th October 2007. British Columbia Ministry of Forests and Range. 2000. 1999 Aerial overview survey summary (online). Available at http://www.for.gov.bc.ca/hfp/health/overview/1999table.htm. Accessed 16th October 2007. British Columbia Ministry of Forests and Range. 2001. 2000 Aerial overview survey summary (online). Available at http://www.for.gov.bc.ca/hfp/health/overview/2000table.htm. Accessed 16th October 2007. British Columbia Ministry of Forests and Range. 2002. What is the theoretical Maximum Green:Red? (online). Available at http://www.for.gov.bc.ca/hfp/health/fhdata/maxbeetles.htm. Accessed 16th October 2007. British Columbia Ministry of Forests and Range. 2006. Preparing for climate change: adapting to impacts on British Columbia\u00E2\u0080\u0099s forest and range resources (online). Available at http://www.for.gov.bc.ca/mof/Climate_Change/Preparing_for_Climate_Change.p df. Accessed 16th October 2007. Carroll, A.L., Shore, T.L., Safranyik, L. 2006. Direct Control: Theory and Practice. In: Safranyik, L. and Wilson, B. (Eds.). The Mountain Pine Beetle: A Synthesis of Biology, Management, and Impacts on Lodgepole Pine. pp. 155 \u00E2\u0080\u0093 172. Natural 84 Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC, Canada. 317 p. Carroll, A.L., Taylor, S.W., R\u00C3\u00A9gni\u00C3\u00A8re, J. Safranyik L. 2004. Effects of climate change on range expansion by the mountain pine beetle in British Columbia. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Mountain pine beetle symposium: challenges and solutions. 30\u00E2\u0080\u009331 October 2003, Kelowna, British Columbia, Canada. pp. 223 \u00E2\u0080\u0093 232. Canadian Forest Service, Pacific Forestry Centre, Victoria, British Columbia, Information Report BC-X-399. 298 p. Coops, N.C., J.A. Timko, M.A. Wulder, J.C. White, and S.O. Ortlepp. 2008. Investigating the effectiveness of mountain pine beetle mitigation strategies. International Journal of Pest Management. 54(2): 151 \u00E2\u0080\u0093 165. Fettig, C.J., Klepzig, K.D., Billings, R.F., Munson, A.S., Nebeker, T.E., Negr\u00C3\u00B3n, J.F., Nowak, J.T. 2007. The effectiveness of vegetation management practices for prevention and control of bark beetle infestations in coniferous forests of the western and southern United States. Forest Ecology and Management. 238(1-3): 24 \u00E2\u0080\u0093 53. Franklin, S.E., Wulder, M.A., Skakun, R.S., Carroll, A.L. 2003. Mountain pine beetle red-attack forest damage classification using stratified Landsat TM data in British Columbia, Canada. Photogrammetric Engineering and Remote Sensing. 69(3): 283 \u00E2\u0080\u0093 288. Harris, J.W.E. and Dawson, A.F. 1979. Evaluation of aerial forest pest damage survey techniques in British Columbia. Canadian Forestry Service, Victoria, Information Report, BC-X-198. 22 p. Hopping, G.R. and Mathers W.G. 1945. Observations on outbreaks and control of mountain pine beetle in the lodegpole pine stands of western Canada. The Forestry Chronicle. 21(2): 98 \u00E2\u0080\u0093 108. Logan, J. and Powell, J. 2001. Ghost forests, global warming, and the mountain pine beetle (Coleoptera: Scolytidae). American Entomologist. Fall: 162 \u00E2\u0080\u0093 172. Leckie, D.G., Clooney, E., Joyce, S.P. 2005. Automated detection and mapping of crown discolouration caused by jack pine budworm with 2.5 m resolution multispectral imagery. International Journal of Applied Earth Observation and Geoinformation. 7(1): 61 \u00E2\u0080\u0093 77. Maclauchlan, L.E. and Brooks, J.E. 1998. Strategies and tactics for managing the mountain pine beetle, Dendroctonus ponderosae. British Columbia Forest Service, Kamloops Region Forest Health, Kamloops, BC. 55 p. Malmstr\u00C3\u00B6m, C.M. and Raffa, K.F. 2000. Biotic disturbance agents in the boreal forest: considerations for vegetation change models. Global Change Biology. 6(1) (Supplement): 35 \u00E2\u0080\u0093 48. Mitchell, R.G. and Preisler, H.K. 1991. Analysis of spatial patterns of lodgepole pine attacked by outbreak populations of the mountain pine beetle. Forest Science. 37(5): 1390 \u00E2\u0080\u0093 1408. 85 Nelson, T., Boots, B., Wulder, M.A. 2004. Spatial-temporal analysis of mountain pine beetles infestations to characterize pattern, risk, and spread at the landscape level. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Mountain pine beetle symposium: challenges and solutions. 30\u00E2\u0080\u009331 October 2003, Kelowna, British Columbia, Canada. pp. 164 \u00E2\u0080\u0093 173. Canadian Forest Service, Pacific Forestry Centre, Victoria, British Columbia, Information Report BC-X-399. 298 p. Nelson, T., Boots, B., Wulder, M.A. 2006. Rating the susceptibility of forests to mountain pine beetle infestations: the impact of data. Canadian Journal of Forest Research. 36(11): 2815 \u00E2\u0080\u0093 2825. Safranyik, L. and Linton, D.A. 1991. Unseasonably low fall and winter temperatures affecting mountain pine beetle and pine engraver beetle populations and damage in the British Columbia Chilcotin region. Journal of the Entomological Society of British Columbia. 88: 17 \u00E2\u0080\u0093 21. Safranyik, L. and Carroll, A.L. 2006. The biology and epidemiology of the mountain pine beetle in lodgepole pine forests. In: Safranyik, L. and Wilson, B. (Eds.). The Mountain Pine Beetle: A Synthesis of Biology, Management, and Impacts on Lodgepole Pine. pp. 3-66. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC, Canada. 317 p. Skakun, R.S., Wulder, M.A. Franklin, S.E. 2003. Sensitivity of the Thematic Mapper Enhanced Wetness Difference Index (EWDI) to detect mountain pine beetle red- attack damage. Remote Sensing of Environment. 86(4): 433 \u00E2\u0080\u0093 443. Taylor, S.W., Carroll, A.L., Alfaro, R.I. Safranyik, L. 2006. Forest, climate and mountain pine beetle outbreak dynamics in western Canada. In: Safranyik, L. and Wilson, B. (Eds.). The Mountain Pine Beetle: A Synthesis of Biology, Management, and Impacts on Lodgepole Pine. pp. 67-94. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC, Canada. 317 p. Thompson, A.J. 1991. Simulation of mountain pine beetle (Dendroctonus ponderosae Hopkins) spread and control in British Columbia. Pacific Forestry Centre, Canadian Forestry Service, Pacific and Yukon Region, B.C. Information Report BC-X-329. 18 p. Westfall, J. 2007. 2006 Summary of forest health conditions in British Columbia. British Columbia Ministry of Forests and Range, Forest Practices Branch, Victoria, BC. 73 p. White, J.C., Wulder, M.A., Brooks, D., Reich, R., Wheate, R.D. 2005. Detection of red attack stage mountain pine beetle infestation with high spatial resolution satellite imagery. Remote Sensing of Environment. 96(3-4): 340 \u00E2\u0080\u0093 351. White, J.C., Coops, N.C., Hilker, T., Wulder, M.A., Carroll, A.L. 2007. Detecting mountain pine beetle red attack damage with EO-1 Hyperion moisture indices. International Journal of Remote Sensing. 28(10): 2111 \u00E2\u0080\u0093 2121. Wulder, M.A., Dymond, C.C., Erickson, B. 2004. Detection and monitoring of the mountain pine beetle. Natural Resources Canada, Canadian Forest Service, 86 Pacific Forestry Centre, Victoria, British Columbia, Information Report BC-X- 398. 24 p. Wulder, M.A., Dymond, C.C., White, J.C., Erickson, B. 2006a. Detection, mapping, and monitoring of mountain pine beetle. In: Safranyik, L. and Wilson, B. (Eds.). The mountain pine beetle: A synthesis of biology, management, and impacts on lodegpole pine. pp. 123-154. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, British Columbia. 317 p. Wulder, M.A., Dymond, C.C., White, J.C., Leckie, D.G., Carroll, A.L. 2006b. Surveying mountain pine beetle damage of forests: a review of remote sensing opportunities. Forest Ecology and Management. 221(1-3): 27 \u00E2\u0080\u0093 41. Wulder, M.A., White, J.C., Bentz, B.J. Ebata, T. 2006c. Augmenting the existing survey hierarchy for mountain pine beetle red-attack damage with satellite remotely sensed data. The Forestry Chronicle. 82(2): 187 \u00E2\u0080\u0093 202. Wulder, M.A., White, J.C., Bentz, B., Alvarez, M.F., Coops, N.C. 2006d. Estimating the probability of mountain pine beetle red-attack damage. Remote Sensing of Environment 101(2): 150 \u00E2\u0080\u0093 166. 87 4 INITIALISATION OF AN INSECT INFESTATION SPREAD MODEL USING TREE STRUCTURE AND SPATIAL CHARACTERISTICS DERIVED FROM HIGH-SPATIAL RESOLUTION DIGITAL AERIAL IMAGERY5 4.1 Introduction Very high spatial resolution digital aerial imagery (pixel size <1 m) has the potential to become a new data source for forest managers to predict characteristics of forest stands and individual trees (Brandtberg 1999; Tuominen and Pekkarinen 2005) due to its high quality, cost effectiveness, and timely acquisition (Wulder et al. 2004). In comparison to conventional aerial photography, the digital format enables the imagery to be viewed soon after collection to provide quality assurance and ensure proper image capture (Wulder et al. 2006a). Digital imagery can also produce data over large areas of remote, inaccessible land (Ciesla 2000) with the potential to obtain census surveys of large geographic areas, rather than samples traditionally collected during ground surveys, and is easily integrated with other data sources (Wulder et al. 2006a). As a result, the imagery can provide essential information to investigate the spatial dynamics of forest stands (Bone et al. 2005) and has been shown to be able to identify individual tree crowns (Bunting and Lucas 2006), supply individual tree locations (Brandtberg 2002), enable the detection and classification of tree health status (Sims et al. 2007), and assess damage caused by insects (White et al. 2005; Wulder et al 2006b; Coops et al 2006; White et al (2007). 5 A version of this chapter was published as: Coggins, S.B., Coops, N.C., Wulder, M.A. 2008. Initialisation of an insect infestation spread model using tree structure and spatial characteristics derived from high- spatial resolution digital aerial imagery. Canadian Journal of Remote Sensing. 34(6): 485 \u00E2\u0080\u0093 502. 88 In the province of British Columbia, Canada, lodgepole pine forests (Pinus contorta Dougl. ex. Loud var. latifolia Engl.) are experiencing severe mortality caused by mountain pine beetles, Dendroctonus ponderosae (Hopkins) (Westfall and Ebata 2008). This bark beetle is indigenous to the region where cold winters (indicated when temperatures fall below -40\u00C2\u00B0C) traditionally maintained endemic (low-level) populations (Carroll and Safranyik 2004). However, the onset of warmer annual temperatures and abundance of suitable host tree species (Taylor and Carroll 2004) has allowed the current infestation to reach a size of 10.1 million hectares in 2007 (Westfall and Ebata 2008). Beetles have been found in increasing numbers in the neighbouring province of Alberta and there is concern that further climatic warming may cause the infestation to continue further eastwards to infest the boreal jack pine (Pinus banksiana Lamb.) forests of eastern Canada and the north-eastern United States of America (Logan and Powell 2001, 2003; Carroll et al. 2004, 2006). Early detection and monitoring at the leading edge will enable attacked trees to be controlled and help mitigate the impacts of the infestation (Carroll et al. 2006). Beetles are thought to visually select host trees using tree stem diameter (Carroll and Safranyik 2004) where mature (approximately 80 years of age), large diameter (>12.5 cm stem diameter) trees are preferred (Shepherd 1966; Safranyik et al. 1999; Nelson et al. 2006) because they present a larger silhouette (Shepherd 1966) and have thicker bark which provides insulation against cold temperatures as larvae develop (Amman 1972; Taylor and Carroll 2004). Other studies have shown that host selection can also be related to tree age (Shore et al. 2000), lack of tree vigour (Mitchell et al. 1983; Waring and Pitman 1985), and stand density (Safranyik et al. 1999). The nature of the spread by 89 mountain pine beetles causes infestations to radiate to surrounding trees from previously infested trees provided the quantity of attacking beetles are numerous and a suitable amount of host trees are present (Shore et al. 2000) in close proximity to previously attacked trees. This behaviour, given suitable host and climatic conditions, enables populations to rapidly progress from endemic to epidemic proportions. As a result, infestation epicentres gradually increase in size and coalesce with neighbouring infestations until the landscape is dominated by attacked trees (Mitchell and Preisler 1991; Robertson et al. 2007). Therefore, the spatial arrangement of forest stands, and individual tree characteristics within these stands, are important parameters to consider during an insect infestation (Geiszler and Gara 1978) and when implementing modelling simulations of beetle attack (Mitchell and Preisler 1991). Typically for modelling purposes, estimates of forest stand and individual tree parameters such as tree species, stem diameter, stocking density, individual tree locations, and crown areas are collected during ground surveys. These surveys are, however, expensive, time consuming, and are sample based (Wulder et al. 2006a) As an alternative, high-spatial resolution digital aerial imagery can potentially provide surrogates for these variables, albeit with varying degrees of precision (Ahern and Leckie 1987; Brandtberg and Walter 1998; Gougeon and Leckie 2002; Tuominen and Pekkarinen 2005), which can then be used in models to predict infestation spread. Modelling beetle movement at the forest stand level considers the interactions between a forest pathogen (e.g., mountain pine beetles) and its host species, using population dynamics which incorporate individual tree data, and other variables, to predict the spread of infestation within a forest stand. These models first predict the current population size based on the previous years\u00E2\u0080\u0099 population (Barclay et al. 90 1985; Thompson 1991) and secondly, apply mathematical functions to describe the infestation probability based on the trees within the stand while including other additional parameters describing secondary effects on beetle populations, such as weather and host conditions (Mitchell and Preisler 1991; Thompson 1991; Beukema et al. 1997; Malmstr\u00C3\u00B6m and Raffa 2000). Models generally estimate which individual tree will be selected by beetles, to predict the result of attack, and use the rate of beetle mortality to determine the extent of infestation within a stand. These range from complex models, such as the Western Pine Beetle Model (Beukema et al. 1997) which has a number of input parameters that cannot be rapidly or readily measured, to more simple models which have fewer parameters and can be more easily initialised (Mitchell and Preisler 1991; Thompson 1991). Output can be incorporated into a visual interface to demonstrate how infestations increase within forest stands and how they manifest in previously un- attacked stands. The aims of this chapter are two-fold: the first aim is to demonstrate how spatial data derived from high-spatial resolution digital aerial imagery can be used to predict a range of plot level, and individual tree level attributes. Once derived the second aim is to populate an infestation spread model with these data and, in turn, use these remote sensing based inputs to predict more accurately the extent and spread of mountain pine beetle infestation on the landscape when compared to conventional approaches. Ground survey data and high spatial resolution digital aerial imagery were acquired over areas of the western Canadian provinces of British Columbia and Alberta where mountain pine beetle infestations have become more prominent in the last 5 years. First parameters derived from high-spatial resolution digital aerial imagery were selected that were most 91 applicable to parameterise a mountain pine beetle infestation spread model. An object- based classification procedure was applied to derive parameters relating to the spatial characteristics at the plot and individual tree level. Secondly, accuracy statements for each parameter were developed using ground survey measurements. Finally, these parameters were entered into a spread model to demonstrate remotely sensed data can describe the multi-year spread characteristics of mountain pine beetle infestations. 4.2 Materials and methods 4.2.1 Survey data This research was conducted in the focus area described in Chapter 1 using a network of variable radius plots. First in 2006, 10 cm spatial resolution digital aerial imagery (also described in Chapter 1) was acquired near-nadir over the centre point of eight of the variable radius plots. Ground data was collected shortly after image acquisition over a two-year period from twenty-three variable radius plots (also known as variable area plots, relascope plots, prism plots and angle-count samples), 8 in 2006 and 15 in 2007. Due to the relatively light infestation in 2006 and 2007, plots were selectively placed in stands exhibiting mountain pine beetle infestation using the following approach derived from polygon-based forest inventory data (British Columbia Vegetation Resource Inventory). First a stratification was developed to define a mature pine mask (defined as any stand with > 50% pine, and a mean diameter of 12.5 cm) to exclude all stands with a high proportion of non-pine species, and young stands that contained trees with diameters smaller than 12.5 cm. A digital elevation model was then used to remove all stands on steep topography to ensure imagery was acquired over modelled flat terrain, thus 92 minimising image distortions which could decrease the ability to identify tree locations. Potential stands were then selected if located near identifiable landscape features such as rivers and road edges to ensure ease of location in the field and on the corresponding digital aerial imagery. Potential stands were then visually examined on the imagery and if trees exhibited symptoms of mountain pine beetle infestation, plots were placed at these locations. At each of the variable radius plots the geographic centre was recorded using a global positioning system (GPS) which was post-differentially corrected to an estimated accuracy of 2.2 m. A basal area sweep defined trees within the variable radius plot and subsequently all trees were identified and numbered on paper copies of the digital aerial imagery (n = 364), enabling direct reference of tree measurements to tree crowns on the imagery. The species and stem diameter (diameter at 1.3 m up the stem) for each tree identified as within the plot was then measured. Crown area (m2) measurements were recorded for at least two dominant trees in the plot by measuring the width of the crown at the widest diameter and then taking a second measurement perpendicular to the first. Crown area was calculated by assuming a circular crown of a radius equal to the average of the diameter measurements. Based on these data, average stocking density over the plots was 2139 stems per hectare, typical of mature age lodgepole pine stands in the area (Kranabetter et al. 2005). Stem diameters in the plots ranged between 5.0 - 44.9 cm confirming that the trees measured in the stand are likely susceptible to attack by mountain pine beetles (Table 4.1). Finally, mountain pine beetle attack status of each tree was assessed according to a 4 category classification system where 0 is healthy, 1 is green attack, 2 is red attack, and 3 is grey attack (Table 4.2). 93 Table 4.1: Summary of ground survey data collected from twenty-three variable radius plots during the summers of 2006 and 2007. Year Plot Infested trees Mean stem diameter (\u00C2\u00B1 SD) Number of trees Stems per hectare 2006 1 0 19.7 2.7 7 921 2006 2 0 21.0 1.7 5 576 2006 3 0 21.6 4.7 7 764 2006 4 0 26.4 4.4 7 512 2006 5 0 19.5 3.1 10 1337 2006 6 0 21.8 1.9 10 1068 2006 7 0 27.3 3.0 6 410 2006 8 0 20.6 3.3 8 959 2006 9 0 15.8 2.6 5 1023 2007 10 12 16.4 4.0 16 3028 2007 11 7 19.0 5.1 30 4218 2007 12 10 30.0 8.0 16 908 2007 13 11 26.9 5.4 13 914 2007 14 21 14.8 3.2 49 11405 2007 15 8 14.8 3.0 37 8624 2007 16 7 19.1 6.0 20 2802 2007 17 12 22.0 4.7 15 1580 2007 18 8 23.6 5.9 17 1557 2007 19 12 23.7 4.0 17 1542 2007 20 11 26.4 6.0 15 1100 2007 21 13 22.9 5.3 23 2233 2007 22 14 28.0 6.2 16 912 2007 23 12 29.7 6.0 14 807 94 Table 4.2: Assessment criteria of mountain pine beetle infestation status on individual lodgepole pine trees. Mountain pine beetle infestation status Code Description Healthy 0 Tree is unattacked. Green-attack 1 Foliage remains green immediately after attack. Pitch tubes surrounding beetle entrance holes. Pitch tubes are cream to pinkish coloured mixtures of boring dust and resin extruding from the entrance of beetle galleries. They are sticky and pliable to the touch. Boring dust will be deposited around the base of the stem and trapped in bark crevices. Eggs, larvae, pupae, live and dead adult beetles will be present under the bark. Galleries that extend vertically up the stem will be present beneath the bark. Approximately two weeks after attack the wood surrounding galleries will be stained blue by fungi inoculated by the beetles. Red attack 2 Foliage on most trees will become red by spring of the year following infestation. Pitch tubes are coloured dark yellow, have become dried and hardened. Boring dust will be deposited around the base of the stem and trapped in bark crevices Circular emergence holes (approximately 1 mm in diameter) are present on the stem. Dead adult beetles are present beneath the bark in galleries. The colour of the wood beneath the bark has become dark blue or black. Bark on the stem has been removed by woodpeckers. Secondary bark beetle may be present and forming galleries in the wood. Grey attack 3 Foliage has fallen from the branches. Most of the bark on the stem is missing. 95 4.2.2 Individual tree crown delineation Methods for extracting individual trees from high spatial resolution imagery are reviewed in Wang et al. (2004), Pouliot et al. (2002), and Bunting and Lucas (2006). Common individual tree crown delineation methods may be grouped by five major types: 1) enhancement and thresholding, also known as local maxima filtering (Culvenor 2001, 2002; Pinz et al. 1993), where a high pass filter is applied to imagery and the brightest values are extracted as estimates of tree locations; 2) template matching, which when used for individual tree crown recognition was based on the matching of a synthetic tree crown model with an aerial photograph (Pollock 1996); 3) multi-scale analysis, where the occurrence of edges over several image scales are examined to define a region in which the brightest pixel value is taken as the tree apex (Brandtberg and Walter 1998); 4) valley following or contour based method (e.g., Gougeon and Leckie 2002), whereby tree crowns are automatically delineated according to the presence of shaded materials between tree crowns; and 5) spatial clustering (also known as region growing), which was designed to delineate tree crowns automatically in high-spatial resolution digital aerial imagery by assuming the centre of a crown is brighter than the edge and then finds the boundary between crowns (Brandtberg and Walter 1998; Culvenor 2002). Other approaches do exist; for example, Strand et al. (2006) demonstrated a wavelet estimation technique, and Pinz et al. (1993) applied a neural network approach. Based upon the spatial and spectral characteristics of the digital aerial imagery, I determined that a spatial clustering type approach would be most appropriate, and use Definiens software (Definiens GmbH, Munchen, Germany, formerly known as eCognition) which utilizes a modified spatial clustering, or object-based region growing, 96 technique. This software uses fuzzy logic to enable efficient inclusion of spatial concepts by segmenting an image into multi-pixel objects according to both spatial and spectral features (Flanders et al. 2003; Bunting and Lucas 2006). These objects are defined to maximise between-object variability and minimize within-object variability for user- chosen inputs, principally spectral values, shape, proximity to other objects, and the relative border within and between different hierarchical levels (Bunting and Lucas 2006). Rule-based decisions may then be applied to assign a class to each object (Flanders et al. 2003) which can then be exported to GIS software. Previous approaches have shown that object-oriented techniques can be used to successfully delineate individual tree crowns (Greenberg et al. 2005; Bunting and Lucas 2006). Bunting and Lucas (2006) successfully utilised an object-based classification approach on Compact Airborne Spectrographic Imager (CASI) imagery defining individual tree crowns in eucalypt forests in Australia. To do so they used a forest mask to define the outer boundaries of tree crowns, followed by segmentation within the forest area to form objects representative of trees or clusters of trees. To delineate tree crowns for the purposes of this study a similar approach to Bunting and Lucas (2006) was applied (Figure 4.1) and the subsequent procedure followed: 1) Differentiation of forest and non-tree vegetation, bare ground, and roads; 2) Segmentation of forest and non-forest; 3) Classification of forest within the image; 4) Segmentation of forest to form individual tree crowns; 5) Classification of forest to give polygons of individual tree crowns; and 6) Export of resultant polygons to a GIS. 97 The first step in crown delineation of individual trees involved smoothing the image to enhance the perimeter of the tree crown to ensure crown dimensions are adequately delineated and to ensure fair comparison with tree crowns measured during the ground Figure 4.1: Processing steps taken to complete the object-based classification to delineate individual tree crowns on digital aerial imagery. Processing involved: 1) Differentiation of forest and non-tree vegetation, bare ground, and roads; 2) Segmentation of forest and non-forest; 3) Classification of forest within the image; 4) Segmentation of forest to form individual tree crowns; and 5) Classification of forest to give polygons of individual tree crowns; and 6) Export of resultant polygons to a GIS. 98 surveys. A median filter, which replaces the central pixel within a 5 x 5 kernel with the median value of the surrounding pixels was applied to each digital aerial image. Using this type of filter smoothed internal crown structure while preserving edges (Lillesand et al. 2004). The size of the filter was chosen based on iteratively varying the kernel size from no filter to using a kernel size of 11 x 11 and then comparing tree shapes with those observed in the field. As the filter size increased, the tree crowns merged into clumps on the image, the 5 x 5 filter most adequately defined the tree crowns because inter-crown variation was removed while maintaining the integrity of the crown edge of each tree in the image (Figure 4.2). Figure 4.2: Demonstration of the effects of iteratively changing the filter kernel size to identify tree crown edges prior to crown delineation on 10 cm spatial resolution digital aerial imagery. Filter settings used ranged from no filter (a), 5 x 5 filter (b), and 7 x 7 filter (c), and 11 x 11 filter (d). 99 Following smoothing, a mask was created to differentiate between forest and non-tree vegetation such as bare ground and roads. The role of the mask was critical as it defined the outer boundaries of tree crowns and aimed to remove shadowing and ground vegetation from the segmentation procedure (Gougeon and Leckie 1999; Pouliot et al. 2002; Bunting and Lucas 2006). The mask was created from a density slice classification of the 0.4 \u00C2\u00B5m to 0.5 \u00C2\u00B5m (blue) wavelengths, which was then combined with the original spectral bands of the image as a fourth analytical band. The thresholds applied in the classification were determined by iteratively increasing and decreasing the cut-off values to choose the amount of forest covered by the mask. The most appropriate value was selected which maximised crown area of trees while causing minimal commission of tree crown pixels into the mask. Following masking, each digital aerial image was then imported into object-based Definiens classification software where a multi-resolution segmentation algorithm was applied to create segments of forest and non-forest. Segments were generated with all three multispectral bands and the mask using a multi- resolution segmentation algorithm. The algorithm can be adjusted using: 1) a scale parameter, which is an arbitrary term that determines the maximum allowed heterogeneity for the resulting image objects, with larger scale parameters defining larger image objects; 2) a shape : colour parameter, which assigned a relative weight to be applied to spectral versus shape criteria (Flanders et al. 2003); and 3) a compactness : smoothness parameter, which emphasised objects based on the smoothness of object borders and provides distinction from non-compact image objects with relatively weak spectral contrast (Flanders et al. 2003). Once objects are defined, a nearest neighbour classification algorithm was used to define image objects which requires the operator to 100 choose representative samples of objects to define classes within the image. The segmentation process was designed to remove all objects within the non-forest mask. The first pass used weights to define the mask and used: scale = 5; shape= 0; colour= 1. To classify the mask, groups of pixels within the masked portion of the image were selected. During classification, objects similar to those specified during the sample selection process, based on texture, shape, and layer values (e.g. brightness, colour, and relationship to neighbouring objects) were assigned to a class. Once the portions of imagery covered by the mask were classified as non-forest, the remaining objects were used for further delineation and classification. The second pass used increased weighting applied to the 0.4 \u00C2\u00B5m to 0.5 \u00C2\u00B5m (blue) and the 0.5 \u00C2\u00B5m to 0.6 \u00C2\u00B5m (green) bands to delineate objects comparable in size to tree crowns (scale = 20; shape = 0.9; colour = 0.1; compactness = 0.4; smoothness = 0.6), after which the results of the classification were exported to a GIS (Figure 4.2). 4.2.3 Assessment of the crown delineation A previously published approach was used to assess the tree crown delineation. Individual tree locations measured in the field were overlaid on the delineated tree crowns and an association classification methodology (Culvenor 2001) was applied to assess the crown delineation. This method counted the number of individual tree locations within tree crown polygons delineated by the object-based classification. When a tree position was located within a delineated crown, it was given a score of 1 for a 1:1 relationship. A score of 0 was given where either two or more tree positions fell within one delineated crown or when individual tree locations did not fall within crowns. The 101 scores for 1:1 relationships were reported as a percentage to generate an overall assessment of the tree crown delineation. Once the tree crown delineation was assessed, the spatial and structural attributes of trees (crown area) and plots (stem diameter and stocking density) were evaluated by comparing image-derived measurements with field-based measurements. To accurately compare crown areas a two-fold process was used. First, crowns impacted by beetle infestation and therefore likely to have greyer and less distinct appearance were removed from the analysis. Secondly, trees with stem diameters smaller than 12.5 cm were removed due to the minimum size criterion in the cluster analysis. Finally, all 1:1 crown associations were visually assessed as either good, medium, or poor based on the extent of background observable within each tree cluster, and tree crowns where a complete crown delineation (i.e. partial crowns were delineated) did not occur. The delineated crowns assigned \u00E2\u0080\u0098medium\u00E2\u0080\u0099 and \u00E2\u0080\u0098good\u00E2\u0080\u0099 ranks were used to compare the crown areas because they represented the crowns most correctly in each plot. Subsequently crown delineations assigned a rank of poor were removed because they poorly represented crowns in the plots. Predictions of stem diameter were generated by developing a relationship between field measured crown area and stem diameter (Figure 4.3). The resulting equation was applied to image predicted crown areas generated during the segmentation process to define stem diameters for each tree delineated by the object-based classification. Stocking density was calculated from the number of trees correctly delineated on the image. The field measured stocking density was calculated using a standard variable radius plot equation: densityStockingtrees MStDiap BAF =\u00C3\u0097 \u00C3\u0097 \u00C3\u0097 # 40000 (4.1) 102 where BAF is the basal area factor, a constant used to determine which trees are represented within the variable radius plot (in this study, a BAF of 4 m /hectare/tree was used), \u00E2\u0080\u0098MStDia\u00E2\u0080\u0099 is the mean stem diameter, and \u00E2\u0080\u0098#trees\u00E2\u0080\u0099 is the number of trees within the plots or those trees derived from the imagery with a 1:1 relationship between tree location and crown polygon. Figure 4.3: The relationship between crown area (m 2 ) and stem diameter of lodgepole pine trees measured during ground surveys which was used to develop stem diameter measurements from crowns delineated on digital aerial imagery. An estimate of stocking density based on the imagery was determined using the same equation however, substituting the number of trees extracted on the imagery with a 1:1 match. 103 4.2.4 Mountain pine beetle infestation simulations In order to demonstrate the applicability of data derived from digital aerial imagery to populate spread models of mountain pine beetle infestations I applied a model developed by Mitchell and Preisler (1991). The model required information on tree diameter and distance from previously attacked trees and modelled the likelihood pine trees are attacked in a given year. The model calculated the probability a tree will be attacked as a function of the distance between previously attacked trees to recipient healthy trees and stem diameter (Mitchell and Preisler 1991). Predictably, large diameter trees in close proximity to previously attacked trees are more likely to become infested than smaller trees further away. The model used three regression equations to predict the likelihood (L) that a flying beetle will intercept a recipient pine tree during dispersal. Equations were assigned to a range of stem diameters which predict the likelihood of attack, where small diameters include stems \u00E2\u0089\u00A4 20 cm (equation 4.2), medium diameters are > 20 cm and \u00E2\u0089\u00A4 26 cm (equation 4.3), and large diameters are > 26 cm (equation 4.4) such that: L = 0.0059 \u00C3\u0097 ln(Ak) + 1.738 (4.2) L = 0.3957 \u00C3\u0097 ln(Ak) - 0.1115 (4.3) L = 0.2225 \u00C3\u0097 ln(Ak) + 0.5021 (4.4) where: Ak = 2*arcsin \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB + kdr r and r = radius of each tree, dk = the distance from the nearest attacked pine tree to a recipient pine tree where k corresponds to an individual tree. 104 Conventionally, mountain pine beetle infestations are modelled using randomly generated data or within small plots. However, remotely sensed data have potential to provide parameters to more accurately predict the spread of infestations. The simulations used in this study are designed to demonstrate how the output from conventional models may be affected when data derived from remotely sensed imagery are used. \u00E2\u0080\u00A2 Simulation 1: Random stem diameter and random tree positions; \u00E2\u0080\u00A2 Simulation 2: Random stem diameters and tree positions derived from digital aerial imagery; and \u00E2\u0080\u00A2 Simulation 3: Stem diameters derived from digital aerial imagery tree positions derived from digital aerial imagery. The simulations each covered a ground area of approximately 0.5 hectares and were run for 10 years. They used a two-fold population increase (Carroll and Safranyik 2004) and commenced by selecting an equivalent level of infestation as found during the ground surveys. Simulations 1 and 2 indicated that the use of random data can provide vastly different results when compared with data derived from digital aerial imagery. In the first simulation, randomized data were generated for both stem diameters and distances between trees. Stem diameters were created randomly between the ranges of 12.5 cm and 44.9 cm to mimic the largest pine tree diameter measured during the ground survey. Tree locations were created by a random point generation tool, which gave a different spatial pattern than the tree positions used in simulations 2 and 3 (Figure 4.4). The second simulation used random stem diameters and tree locations derived from digital aerial imagery. It is anticipated that varying the stem diameter values should have a significant impact on the simulations as the beetles are more likely to fly to larger diameter trees. 105 Figure 4.4: Spatial distribution of random tree positions, Simulation 1 (a) and positions derived from digital aerial imagery, Simulations 2 and 3 (b). Therefore, increases in diameter size will vary and the pattern of attack within the forest stand will change. The third simulation utilised stem diameter and distance data derived from the digital aerial imagery to assess the spread of an infestation and demonstrated the impact of using actual data to simulate mountain pine beetle infestation rather than using random data. The stem diameter distribution determined from the imagery in this simulation is skewed with many more smaller diameter trees present in the stand, when compared to the even distribution of diameters in simulation 1 (Figure 4.5). These modelling simulations demonstrated that utilizing actual tree locations (simulation 3) can predict the extent of beetle infestations when analysing the behaviour of beetles such as mountain pine beetles, which kill trees in groups (Mitchell and Preisler 1991). 106 Figure 4.5: Stem diameter distributions of simulations 1 and 2 (a), using random number values for diameter, and the distribution of diameters derived from digital aerial imagery used in simulation 3 (b). 4.3 Results Tree crown delineation was validated using field tree locations overlaid on the tree crown polygons and produced a delineation accuracy of 80.2% (with a range of 50% to 100%) for 1:1 relationships. Tree crown polygons that contained two or more tree positions constituted 16.5% of the total, and 3.3% of tree locations were positioned outside tree crown polygons. With confidence in the tree identification procedure, the predicted tree and stand level parameters were evaluated. Crown area measurements from the field and digital aerial image were compared and found to be related (r2 = 0.55, se = 1.19, p<0.001, n = 59), although image derived crown areas were under predicted when compared to those measured in the field (Figure 4.6). However, the comparison is skewed by one observation that is greatly underestimated by the image object procedure and possibly decreases the fit of the regression line. 107 Figure 4.6: Comparison of crown area (m 2 ) measurements derived from the field and digital aerial imagery using simple linear regression (r 2 = 0.55, se = 1.19, p <0.001, n = 57, digital aerial image crown = 0.5208x + 1.4065). Using the field derived stem diameter and crown area relationship, stem diameters were produced ranging between 15.4 \u00E2\u0080\u0093 39.1 cm (Table 4.3) and was highly significant (r2 = 0.50, se = 2.63, p<0.001 n = 23) (Figure 4.7). However, the image derived measurements over-estimated smaller diameter trees and under-estimated larger diameters. Stocking density was also highly significant (r2 = 0.91, se = 506.65, p<0.001, n = 23); however, densely stocked stands are under-estimated when measured on digital aerial imagery (Figure 4.8). The modelling simulations commenced at year 1 with either the randomly located, or image derived stem maps. In all simulations, four trees were infested in the centre of the 108 Table 4.3: Summary of data derived from digital aerial imagery, which corresponds to ground survey data. Year Plot Mean stem diameter (\u00C2\u00B1 SD) Number of trees Stems per hectare 2006 1 21.0 1.4 7 811 2006 2 22.9 1.7 5 486 2006 3 26.8 1.2 5 355 2006 4 28.1 1.9 7 452 2006 5 28.5 2.1 8 501 2006 6 26.8 1.0 8 566 2006 7 30.1 2.4 6 337 2006 8 24.9 1.3 6 491 2006 9 19.9 1.2 3 384 2007 10 22.1 2.7 8 830 2007 11 24.3 2.3 24 2064 2007 12 26.8 4.1 12 850 2007 13 27.5 2.7 13 874 2007 14 15.5 0.8 38 8082 2007 15 20.4 2.0 33 4031 2007 16 19.5 0.7 13 1735 2007 17 26.4 0.8 13 947 2007 18 21.0 1.9 12 1383 2007 19 23.8 0.2 14 1262 2007 20 24.9 1.7 13 1072 2007 21 24.1 0.5 19 1660 2007 22 24.2 1.1 13 1132 2007 23 30.0 4.9 12 680 stand, reflecting the infestation conditions observed in the field during the ground survey. By year 10, the number of trees affected by mountain pine beetles in each simulation differed by 634 trees (Figure 4.9) where simulation 1 (random locations, random diameters) contained 1024 trees affected by mountain pine beetles (Figure 4.10), simulation 2 (digital aerial image derived locations, random diameters) provided the same 109 Figure 4.7: Comparison of stem diameters (cm) measured during ground surveys and stem diameters (cm) estimated from digital aerial imagery using simple linear regression (r 2 = 0.51, se = 2.63, p <0.001 n = 23). The 1: 1 relationship between parameters is indicated by the dashed line. results with 1024 trees affected. However, simulation 3 (digital aerial image derived locations and diameters) predicted only 390 trees were affected (Figure 4.11, Table 4.4). In all cases infestation centred around initially attacked trees then spreads outwards in a radial fashion provided there were large diameter pine trees in close proximity to support a mountain pine beetle population. Infestations that begin separately eventually coalesce to form large areas of infestation, supporting the findings of Robertson et al. (2007). The simulations produced similar results to Mitchell and Preisler (1991) whereby large 110 diameter trees are preferentially attacked and smaller diameter trees remain uninfested until larger diameter trees are no longer available for infestation. Figure 4.8: Comparison of stocking density (stems per hectare) measured during ground surveys and stocking density (stems per hectare) estimated from digital aerial imagery using simple linear regression (r 2 = 0.90, se = 506.65, p<0.001, n = 23). The 1: 1 relationship between parameters is indicated by the dashed line. 111 Figure 4.9: The scale of mountain pine beetle infestations over a 10 year period depicted by the number of trees infested. The sensitivity of the model to the quality of data entered is depicted as the difference in the number of trees infested in simulation 1 and 2 (solid line) and 3 (dashed line). 112 Figure 4.10: Output from simulation 1 illustrating the number of trees infested in each stage of attack. The simulation used randomized stem diameters and tree locations to predict the number of trees infested by mountain pine beetles over a 10 year period. 113 Figure 4.11: Output from simulation 3 illustrating the number of trees infested in each stage of attack. The simulation used stem diameters and tree locations derived from the digital aerial imagery to predict the number of trees infested by mountain pine beetles over a 10 year period. 114 Table 4.4: Number of trees infested by mountain pine beetles over a 10 year period in each simulation. Simulation 1 Year 1 Year 2 Year 3 Year 4 Year 5 Year 6 Year 7 Year 8 Year 9 Year 10 Green-attack 0 4 8 16 32 64 128 256 512 1024 Red-attack 0 0 4 8 16 32 64 128 256 512 Grey-attack 0 0 0 4 12 28 60 124 252 508 Total 0 4 12 28 60 124 252 508 1020 2044 Simulation 2 Year 1 Year 2 Year 3 Year 4 Year 5 Year 6 Year 7 Year 8 Year 9 Year 10 Green-attack 0 4 8 16 32 64 128 256 512 1024 Red-attack 0 0 4 8 16 32 64 128 256 512 Grey-attack 0 0 0 4 12 28 60 124 252 508 Total 0 4 12 28 60 124 252 508 1020 2044 Simulation 3 Year 1 Year 2 Year 3 Year 4 Year 5 Year 6 Year 7 Year 8 Year 9 Year 10 Green-attack 0 4 4 4 6 12 24 48 96 192 Red-attack 0 0 4 4 4 6 12 24 48 96 Grey-attack 0 0 0 4 8 12 18 30 54 102 Total 0 4 8 12 18 30 54 102 198 390 4.4 Discussion 4.4.1 Information derived from ground surveys and digital aerial imagery This study has demonstrated that crown area, stem diameter, and stocking density can be estimated from digital aerial imagery when compared to ground survey data and can, therefore, be used as surrogates for data collected from ground surveys. Obtaining these parameters from digital aerial imagery is particularly important when areas of forest are inaccessible to ground survey crews yet susceptible to infestation by mountain pine beetles. It is possible to obtain these data by acquiring high-spatial resolution digital aerial imagery over these forest stands. Data provided by ground surveys and estimates of tree structure and forest stand parameters from digital aerial imagery confirm that forest stands within the study area are susceptible to attack by mountain pine beetles. The crown comparison is skewed by one 115 observation that appears to shift the regression line down. However, if this point was identified as an outlier and subsequently removed from the analysis the power of the regression actually decreases. Stem diameter values derived from crown area measurements from digital aerial imagery indicated that trees are on average, larger than the threshold value of 12.5 cm (Nelson et al. 2006) and stocking density of pine is high, which suggests infestation by mountain pine beetles can increase and cause widespread mortality if they remain undetected. When using digital aerial imagery only the dominant canopy layers can be mapped. This is problematic because pine trees susceptible to mountain pine beetle infestation in forest stands will remain undetected by optical remote sensing instruments. However, trees in the upper canopy are commonly attacked first because they are usually associated with larger stem diameters and can be located on digital aerial imagery. Furthermore, lodgepole pine stands typically contain only small proportions of pine in the understorey because they require hot fire to regenerate. After fires, lodgepole pine stands in British Columbia grow to uniform dimensions and, therefore usually contain a pine overstorey of uniform height. In this study area, an understorey of spruce and fir was observed which are unsuitable hosts for mountain pine beetle infestations. Therefore, trees in the understorey, undetected by digital remotely sensed imagery, will probably not be of importance when infestations initiate in a stand. 4.4.2 Issues affecting the tree crown delineation Tree crown delineation algorithms supply data that identify individual tree locations, which is considered an important variable when providing parameters for infestation spread models (Mitchell and Preisler 1991). The evaluation of the tree crown delineation 116 and, therefore, tree locations were assessed using an association classification, which indicated the majority of tree crowns were correctly delineated by the object-based classification. The accuracy of the delineation could potentially be reduced because portions of tree crowns (i.e. crown edge) could be covered by the forest mask. A similar problem was identified by Bunting and Lucas (2006) where errors of commission caused inclusion of non-forest pixels in the forest mask. Furthermore, crown polygons containing two or more trees were most likely influenced by the median filter approach to smooth imagery, which smoothed some of the within crown pixel values, removing some of the subtleties required to accurately delineate some tree crowns. Therefore, some crowns were enlarged, resulting in crown polygons containing one or more field tree locations. Other issues which affected the delineation of tree crowns and the association classification were the illumination conditions and structural differences between forest stands. Due to the northern latitude of the study site, some imagery was captured during periods of low sun angles resulting in some shadowing, especially notable in complex, multi-storey stands. Low sun angles can hide small crowns by creating long shadows for tall trees, an effect which becomes pronounced when a few tall individual trees are present within a stand which otherwise consists of trees of uniform height. Tall trees with wide crowns cause extensive shadowing, which darken and obscure adjacent smaller crowns forcing them to become undetectable to the object-based classification. The stands used in this study also exhibited large gaps in the canopy, which combined with low sun angles caused extensive shadowing within the stand. Furthermore, crowns higher in the canopy will cause shadow over crowns lower in the canopy and suppressed 117 trees are likely to be occluded by shadow. To resolve the issue of shadowing within the forest stand, guidelines were set for the time of day used for image acquisition. However, it was not always possible to completely limit flight times because the window of opportunity to acquire data over the study plots is restricted to approximately 4 months, to coincide with the foliage colour change caused by mountain pine beetle infestation and appropriate weather conditions. To effectively cope with shadowing issues it is recommended that imagery be captured in overcast conditions, which decreases the amount of shadowing in the stand, but only if aircraft are able to complete the flight below cloud level. Imagery is best acquired when overcast conditions cast the ground vegetation in the instantaneous field of view in shadow. These conditions clearly define tree crowns which enable the object-based classification to more accurately delineate tree crowns in a digital aerial image. 4.4.3 Mountain pine beetle infestation simulations Extensive tree mortality is experienced in each of the simulations albeit with different levels of severity and extent. After the final year, modelling simulations 1 and 2 indicated that 2% of the stand suffered mortality, which is similar to the observed infestation in similar stands (Carroll et al. 2006). However, mortality is much lower in simulation 3 which utilised digital aerial image derived tree locations and stem diameters. In simulation 3, the infestation has focussed on approximately 270 trees near the centre of the infestation. The diameter distribution of the stand in simulations 1 and 2 indicated even distribution across a range of diameter classes with numerous larger diameter trees and an average stem diameter of 18.5 cm. In simulation 3 the diameter distribution is heavily skewed and features many smaller tree diameters, where the average stem 118 diameter is 14.9 cm (Figure 4.5). The spatial and individual tree attributes featured in these simulations indicate that simulation 1 is most susceptible to infestation due to large diameter trees positioned closer to one another (Shore et al. 2000). In simulation 3, stem diameters are smaller and trees are widely spaced in comparison to simulation 1 causing less infestation after 10 years. The modelling simulations illustrated that the number of trees affected by mountain pine beetles can differ greatly when the structural and spatial attributes of forest stands are altered. They demonstrated that data derived from digital aerial imagery entered into simulations generated output and predicted the severity and extent of infestations. This has implications when using these models to predict the effects of mitigation. Spread following mitigation efforts could be over predicted which would indicate that mitigation is only partially successful and will need to be completed in the years following initial attack should infestation continue to spread. The simulations illustrate how digital aerial imagery provides data to control beetle populations at the outset of an infestation. Individual tree crowns are detected by high-spatial resolution digital aerial imagery which enables early detection of mountain pine beetle infestations and can be used to halt the spread of attack across a landscape if infested trees are removed. Future research will incorporate non-pine species into the modelling simulations and will identify and delineate red attack trees only. The effect of incorporating a proportion of non-pine is likely to affect how infestation spreads in the stand. Distance from previously attacked trees has a significant influence on spread rate and pattern and if non-pine species are present in the stand beetles may have to fly for longer distances to reach fresh hosts. 119 4.5 References Ahern, F.J. and Leckie, D.G. 1987. Digital remote sensing for forestry: requirements and capabilities, today and tomorrow. Geocarto International. 2(3): 43 \u00E2\u0080\u0093 52. Amman, G.D. 1972. Mountain pine beetle brood production in relation to thickness of lodgepole pine phloem. Journal of Economic Entomology. 65(1): 138 \u00E2\u0080\u0093 140. Barclay, H.J., Otvos, L.S., and Thomson, A.J. 1985. Models of periodic inundation of parasitoids for pest control. Canadian Entomologist. 117(6): 705 \u00E2\u0080\u0093 716. Beukema, S.J., Greenough, J.A., Robinson, D.C.E., Kurz, W.E., Smith, E.L., Eav, B.B. 1997. The westwide pine beetle model: a spatially-explicit contagion model. In: Teck, R., Moeur, M., Adams, J. (Eds.). Proceedings of the Forest Vegetation Simulator Conference. 3\u00E2\u0080\u00937 February, Fort Collins, CO. pp. 126\u00E2\u0080\u0093130. Department of Agriculture, Forest Service, Intermountain Research Station, Gen. Tech. Rep. INT-GTR-373. Ogden, UT. 222 p. Bone, C., Dragicevic, S., Roberts, A. 2005. Integrating high resolution remote sensing, GIS and fuzzy set theory for identifying susceptibility areas of forest insect infestations. International Journal of Remote Sensing. 26(21): 4809 \u00E2\u0080\u0093 4828. Brandtberg, T. 1999. Automatic individual tree based analysis of high spatial resolution aerial images on naturally regenerated boreal forests. Canadian Journal of Forest Research. 29(10): 1464 \u00E2\u0080\u0093 1478. Brandtberg, T. 2002. Individual tree-based species classification in high spatial resolution aerial images of forests using fuzzy sets. Fuzzy Sets and Systems. 132(3): 371 \u00E2\u0080\u0093 387. Brandtberg, T. and Walter, F. 1998. Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis. Machine Vision and Applications. 11(2): 64 \u00E2\u0080\u0093 73. Bunting, P. and Lucas, R. 2006. The delineation of tree crowns in Australian mixed species forests using hyperspectral Compact Airborne Spectrographic Imager (CASI) data. Remote Sensing of Environment. 101(2): 230 \u00E2\u0080\u0093 248. Carroll, A.L. and Safranyik, L. 2004. The bionomics of the mountain pine beetle in lodegepole pine forests: establishing a context. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Proceedings of the mountain pine beetle symposium: challenges and solution. pp. 21 \u00E2\u0080\u0093 32. October 30-31, 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Information Report BC-X-399. 298 p. Carroll, A.L., Taylor, S.W., R\u00C3\u00A9gni\u00C3\u00A8re, J., Safranyik, L. 2004. Effects of climate and climate change on the mountain pine beetle. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Proceedings of the mountain pine beetle symposium: challenges and solution. pp. 21 \u00E2\u0080\u0093 32. October 30-31, 2003, Kelowna, British Columbia, Canada. 120 Canadian Forest Service, Pacific Forestry Centre, Information Report BC-X-399. 298 p. Carroll, A.L., Shore, T.L., Safranyik, L. 2006. Direct control: theory and practice. In: Safranyik, L. and Wilson, B. (Eds.). The mountain pine beetle \u00E2\u0080\u0093 a synthesis of biology, management, and impacts in lodgepole pine. pp. 155 \u00E2\u0080\u0093 172. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC. 317 p. Ciesla, W. M. 2000. Remote sensing in forest health protection. USDA Forest Service, Fort Collins, CO., FHTET; Report No. 00-03. 266 p. Coops, N.C., Johnson, M., Wulder, M.A., White, J.C. 2006. Assessment of QuickBird high spatial resolution imagery to detect red attack damage due to mountain pine beetle infestation. Remote Sensing of Environment. 103(1): 67 \u00E2\u0080\u0093 80. Culvenor, D.A. 2001. Evaluation of the tree identification and delineation algorithm (TIDA) using a forest canopy simulation model. IGARSS 2001: Scanning the present and resolving the future. IEEE International symposium on Geosciences and Remote Sensing. Conference proceedings. 2: 777 \u00E2\u0080\u0093 779. Culvenor, D.A. 2002. TIDA: an algorithm for the delineation of tree crowns in high spatial resolution remotely sensed imagery. Computers and Geosciences. 28(1): 33 \u00E2\u0080\u0093 44. Flanders, D., Hall-Beyer, M., Pereverzoff, J. 2003. Preliminary evaluation of eCognition object-based software for cut block delineation and feature extraction. Canadian Journal of Remote Sensing. 29(4): 441 \u00E2\u0080\u0093 452. Geiszler, D.R. and Gara, R.I. 1978. Mountain pine beetle attack dynamics in lodgepole pine. In: Berryman, A.A., Amman, G.D., Stark, R.W. (Eds.). Theory and practice of mountain pine beetle management in lodegpole pine forests. pp. 182 \u00E2\u0080\u0093 187. 25- 27 April 1978, University of Idaho, Moscow, Idaho. Symposium Proceedings. 224 p. Gougeon, F.A. and Leckie, D.G. 2002. Individual tree crown image analysis \u00E2\u0080\u0093 a step towards precision forestry. In: O\u00E2\u0080\u0099Shea, M. (Ed.). First International Precision Forestry Symposium. pp. 43 \u00E2\u0080\u0093 50. 17-20 June, 2001, Seattle, Washington. 193 p. Greenberg, J.A., Dobrowski, S.Z., Ustin, S.L. Shadow allometry: estimating tree structural parameters using hyperspatial image analysis. Remote Sensing of Environment. 97(1): 15 \u00E2\u0080\u0093 25. Kranabetter, J.M., Friesen, J., Gamiet, S., Kroeger, P. 2005. Ectomycorrhizal mushroom distribution by stand age in western hemlock \u00E2\u0080\u0093 lodgepole pine forests of northwestern British Columbia. Canadian Journal of Forest Research. 35(7): 1527 \u00E2\u0080\u0093 1539. Lillesand, T.M., Kiefer, R.W., Chipman, J.W. 2004. Remote sensing and image interpretation. Fifth edition. John Wiley and Sons, United States of America. 763 p. 121 Logan, J. and Powell, J. 2001. Ghost forests, global warming, and the mountain pine beetle (Coleoptera: Scholytidae). American Entomologist. Fall: 162 \u00E2\u0080\u0093 172. Logan, J. and Powell, J. 2003. Modelling mountain pine beetle phenological response to temperature. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Proceedings of the mountain pine beetle symposium: challenges and solution. pp. 221 \u00E2\u0080\u0093 222. October 30-31, 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Information Report BC-X-399. 298 p. Malstr\u00C3\u00B6m, C.M. and Raffa, K.F. 2000. Biotic disturbance agents in the boreal forest: considerations for vegetation change models. Global Change Biology. 6(1) (Supplement): 35 \u00E2\u0080\u0093 48. Mitchell, R.G. and Preisler, H.G. 1991. Analysis of spatial patterns of lodgepole pine attacked by outbreak populations of the mountain pine beetle. Forest Science. 37(5): 1390 \u00E2\u0080\u0093 1406. Mitchell, R.G., Waring, R.H., Pitman, G.B. 1983. Thinning lodgepole pine increase tree vigor and resistance to mountain pine beetle. Forest Science. 29(1): 204 \u00E2\u0080\u0093 211. Moir, W.H. 1965. The lodegpole pine zone in Colorado. American Midland Naturalist. 81(1): 87 \u00E2\u0080\u0093 98. Nelson, T., Boots, B., Wulder, M.A., Shore, T., Safranyik, L., Ebata, T. 2006. Rating the susceptibility of forests to mountain pine beetle infestations: The impact of data. Canadian Journal of Forest Research. 36(11): 2815 \u00E2\u0080\u0093 2825. Pinz, A., Zaremba, M.B., Bischof, H., Gougeon, F.A., Locas, M. 1993. Neurological methods for recognition of compact image objects. Machine Graphics and Vision. 2(3): 209 \u00E2\u0080\u0093 229. Pollock, R. 1996. The automatic recognition of individual trees in aerial images of forest based on a synthetic tree crown model. Vancouver, Canada: PhD. Thesis Department of Computer Science. University of British Columbia. p 172. Pouliot, D.A., King, D.J., Bell, F.W., and Pitt, D.G. 2002. Automated tree crown detection and delineation in high-resolution digital camera imagery of coniferous forest regeneration. Remote Sensing of Environment. 82(2-3): 322 \u00E2\u0080\u0093 334. Robertson, C., Nelson, T.A., and Boots, B. 2007. Mountain pine beetle dispersal: the spatial-temporal interaction of infestations. Forest Science. 53(3): 395 \u00E2\u0080\u0093 405. Safranyik, L., Barclay, H., Thomson, A., Riel, W.G. 1999. A population dynamics model for the mountain pine beetle, Dendroctonus ponderosae Hopk. (Coleoptera : Scolytidae). Pacific Forestry Centre, Canadian Forest Service, Victoria, B.C. Information Report BC-X-386, 41 p. Shepherd, R.F. 1966. Factors influencing the orientation and rates of activity of Dendroctonus ponderosae Hopkins (Coleoptera: Scolytidae). The Canadian Entomologist. 98(5): 507 \u00E2\u0080\u0093 518. Shore, T.L. and Safranyik, L. 1992. Susceptibility and risk-rating systems for the mountain pine beetle in lodegpole pine stands. Pacific Forestry Centre, Canadian Forestry Service, Pacific and Yukon Region. Information Report BC-X-336. 12 p. 122 Shore, T.L., Safranyik, L., and Lemieux, J.P. 2000. Susceptibility of lodgepole pine stands to the mountain pine beetle: testing of a rating system. Canadian Journal of Forest Research. 30(1): 40 \u00E2\u0080\u0093 44. Sims, N.C., Stone, C., Coops, N.C., and Ryan, P. 2007. Assessing the health of Pinus radiata plantations using remote sensing data and decision tree analysis. New Zealand Journal of Forestry Science. 37(1): 57 \u00E2\u0080\u0093 80. Strand, E.K., Smith, A.M., Bunting, S.C., Vierling, L.A., Hann, D.B., Gessler, P.E. 2006. Wavelet estimation of plant spatial patterns in multi-temporal aerial photography. International Journal of Remote Sensing. 27(10): 2049 \u00E2\u0080\u0093 2054. Taylor, S.W. and Carroll, A.L. 2004. Disturbance, forest age, and mountain pine beetle outbreak history dynamics in BC: a historical perspective. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Proceedings of the mountain pine beetle symposium: challenges and solution. pp. 41 \u00E2\u0080\u0093 51. October 30-31, 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Information Report BC-X-399. 298 p. Thompson, A.J. 1991. Simulation of mountain pine beetle (Dendroctonus ponderosae Hopkins) spread and control in British Columbia. Pacific Forestry Centre, Canadian Forestry Service, Pacific and Yukon Region, B.C. Information Report BC-X-329. 18 p. Tuominen, S. and Pekkarinen, A. 2005. Performance of different spectral and textural aerial photograph features in multi-source forest inventory. Remote Sensing of Environment. 94(2): 256 \u00E2\u0080\u0093 268. Wang, L., Gong, P, Biging, G.S. 2004, Individual tree-crown delineation and treetop detection in high-spatial resolution aerial imagery. Photogrammetric Engineering and Remote Sensing. 70(3): 351 \u00E2\u0080\u0093 357. Waring, R.H. and Pitman, G.B. 1985. Modifying lodgepole pine stands to change susceptibility to mountain pine beetle. Ecology. 66(3): 889 \u00E2\u0080\u0093 897. Westfall, J. and Ebata, T. 2008. 2007 summary of forest health conditions in British Columbia. Pest Management Report Number 15. Forest Practices Branch, British Columbia Ministry of Forests and Range. 81 p. White, J.C., Wulder, M.A., Brooks, D., Reich, R., and Wheate, R.D. 2005. Detection of red attack stage mountain pine beetle infestation with high spatial resolution satellite imagery. Remote Sensing of Environment. 96(3-4): 340 \u00E2\u0080\u0093 351. White, J.C., Coops, N.C., Hilker, T., Wulder, M.A., and Carroll, A.L. 2007. Detecting mountain pine beetle red attack damage with EO-1 Hyperion moisture indices. International Journal of Remote Sensing. 28(10): 2111 \u00E2\u0080\u0093 2121. Wulder, M.; Hall, R.; Coops, N., and Franklin, S. 2004. High spatial resolution remotely sensed data for ecosystem characterization. BioScience. 54(6): 1 \u00E2\u0080\u0093 11. Wulder, M.A., White, J.C., Bentz, B.J., and Ebata, T. 2006a. Augmenting the existing survey hierarchy for mountain pine beetle red-attack damage with satellite remotely sensed data. The Forestry Chronicle. 82(2): 187 \u00E2\u0080\u0093 202. 123 Wulder, M.A., Dymond, C.C., White, J.C., Leckie, D.G., and Carroll, A.L. 2006b. Surveying mountain pine beetle damage of forests: a review of remote sensing opportunities. Forest Ecology and Management. 221(1-3): 27 \u00E2\u0080\u0093 41. 124 5 IMPROVEMENT OF LOW LEVEL BARK BEETLE DAMAGE ESTIMATES WITH ADAPTIVE CLUSTER SAMPLING6. 5.1 Introduction 5.1.1 Mountain pine beetle Infestation by the mountain pine beetle, Dendroctonus ponderosae (Hopkins) is of particular importance in western Canada due to the widespread damage to its pine forests and continues to be the leading cause of mortality across the region (Westfall and Ebata 2008). Infestations typically initiate in individual trees or small groups on the landscape that expand rapidly to large areas. In 1999, attack was estimated to cover an area of 164,000 hectares (Westfall and Ebata 2008), and by 2008 this area increased to over 13 million hectares (Westfall and Ebata 2009). In British Columbia, beetles have attacked the lodgepole pine (Pinus contorta Dougl. ex. Loud var. latifolia Engl.) forests that dominate much of the southern and central interior region of the Province. Infestation has continued to spread east into the pine forests of Alberta, some of which historically have been unaffected by the mountain pine beetle. In Alberta, the beetles have the potential to transition from lodgepole pine to jack pine (Pinus banksiana Lamb.) and infest the boreal forest should annual temperatures remain favourable for colonisation, emergence, and dispersal (Logan and Powell 2001, 2003; Carroll et al. 2004, 2006). Expansion has occurred because previous limitations to infestation have relaxed, allowing large populations of mountain pine beetles to affect areas with no historical 6 A version of this chapter was published as: Coggins, S.B., Coops, N.C., Wulder, M.A. 2010. Improvement of low level bark beetle damage estimates with adaptive cluster sampling. Silva Fennica. 44(2): 289 \u00E2\u0080\u0093 301. 125 record of attack. Infestation has spread rapidly due to two factors, the first being favourable periods of weather sustained over long periods of time (Safranyik 1978) and more recently, alterations in climatic thresholds (Raffa et al. 2008) that historically caused mortality of beetles (minimum temperatures less than -40\u00C2\u00B0C) which have enabled larvae to survive cold winters, therefore increasing the size of the attacking population. Secondly, the abundant pine forests in the interior forests of British Columbia and western Alberta provide large areas of highly suitable host material for attack by the beetles (Safranyik 1978; Taylor and Carroll 2004). 5.1.2 Forest health monitoring In western Canada, forest health surveys locate trees attacked by forest pests and monitor the spread of diseases and insect damage, and provide information to guide mitigation activities. Typically, control of infestations is implemented by detecting mountain pine beetle killed trees. Approximately 1 year after attack trees exhibit red foliage (known as red attack) which indicates the locations of infestation. Ground crews are dispatched to these locations and the infested trees in close proximity to the red attacks are located, felled, and burned (Maclauchlan and Brooks 1998). By removing infested trees the beetle population is decreased and future infestations will decline or remain stable because the number of attacking beetles available the following year is reduced. Given the nature of mountain pine beetle infestations to infest trees close to previously attacked trees it is possible that trees missed during surveys will be detected on the ground and the potential for future infestation expansion is further reduced (Carroll et al. 2006; Coggins et al. 2008). 126 Surveys record the cause of the damage, and assess the severity and extent of mortality within forest stands (Westfall and Ebata 2008). Mountain pine beetle attack information is collected using a variety of survey techniques, ranging from coarse (regional) to fine- scale (operational), with each used differently depending on the survey scale and the requirements of the end-user (Wulder et al. 2006a). Aerial overview surveys provide regional data and are completed by flying over the Province in fixed-wing aircraft to identify forest stands affected by pests and diseases, this regional information then guides finer-scale surveys over select portions of the land-base which record the damaging agent, number, and geographic location of infested trees. Digital remotely sensed data can also be used to identify areas of forest pest and disease (Ciesla 2000). Historically, Landsat imagery (30 m spatial resolution) has been used to identify mountain pine beetle infestations, with detection accuracies ranging between 70% and 85% (Franklin et al. 2003; Skakun et al. 2003; Wulder et al. 2006b). Franklin et al. (2003) identified infestations within a 2 hectare area on a single image acquired from the TM sensor at an overall red attack detection accuracy of 73.3% \u00C2\u00B1 6%, p = 0.05 (Franklin et al. 2003). Skakun et al. (2003) processed a time series of Landsat TM data to identify and confirm red attack damage in forest stands. This approach produced an accuracy of 76% (\u00C2\u00B1 12%, p < 0.05) for groups of 10 to 29 infested trees, and 81% (\u00C2\u00B1 11% for groups of 20 to 50 infested trees). Multi-date Landsat scenes were also utilised by Wulder et al. (2006c) to monitor forest change due to mountain pine beetle infestation and reports an 86% accuracy (\u00C2\u00B1 7%). High spatial resolution imagery has also shown ability to detect infestations. White et al. (2005) utilised IKONOS imagery (4 m multispectral spatial resolution) with an unsupervised clustering approach to identify 127 infestations near Prince George, British Columbia. Light infestations (1% to 5% of the trees infested within a forest stand) were detected with an accuracy of 71% and moderate infestations (>5% to <20% of a forest stand) with 92.5%. Coops et al. (2006) used imagery from the QuickBird satellite (2.44 m multispectral spatial resolution) to detect red attack damage. The imagery was classified into attacked trees and healthy trees and the number of red pixels counted. the relationship between the number of red attack pixels and red attack crowns observed in forest health surveys was found to be significant (r2 = 0.48, p<0.001, standard error = 2.8 crowns). Very high spatial resolution digital aerial imagery (as fine as 5 cm) also has the potential to identify mountain pine beetle attack. Imagery is usually acquired in the visible portion of the electromagnetic spectrum (e.g. blue, green, red, approximately 0.4 - 0.7 \u00C2\u00B5m) and has similar characteristics to aerial photographs. Coggins et al. (2008) utilised 10 cm spatial resolution digital aerial imagery to extract information including mountain pine beetle red attack, which was defined with an accuracy of 80.2% when compared to field plots. 5.1.3 Role for sampling A limitation of high spatial resolution satellite and digital aerial imagery is the small image extent, causing large area acquisition to be costly and resulting in the need for much image processing prior to analysis. The limited extent of very high spatial resolution airborne imagery is however, well suited to a sampling approach where imagery can be acquired over several smaller areas and integrated into a sampling scheme, from which forest health variables can then be defined. This technique offers a lower-cost solution to obtain accurate data over large areas in a statistically sound manner. Sampling for infestation in its simplest form can consist of conducting a simple 128 random sample on a remotely sensed image with observations recorded in sample plots selected at random locations over the entire area of the image. Estimates of the mean, variance, and confidence limits for the number of red attacked trees are determined using simple random sample estimates and then scaled up by strata. This method, however, can provide high variability and a wide confidence range. Adaptive cluster sampling has been demonstrated to determine rare and elusive populations that are spatially clustered (Thompson 1990) and can provide estimates of population densities over large areas. Previous studies have utilised adaptive cluster sampling for a variety of applications including, for example, providing estimates of low density mussel populations (Smith et al. 2003), estimating the density of wintering waterfowl (Smith et al. 1995), and estimating stock size of fish in estuarine rivers (Conners and Schwager 2002). In a forestry context this adaptive cluster sampling approach has also been utilised to assess the presence of rare tree species in Nepal (Acharaya et al. 2000), in combination with probability proportional to size sampling to predict forest inventory variables in the United States (Roesch 1993), and to inventory sparse forest populations in Finland (Talvitie et al. 2006). 5.1.4 Objectives The goal of this chapter is to demonstrate an approach for using samples of airborne imagery to produce robust estimates of population wide estimates of low level mountain pine beetle attack. To meet this goal, the primary objective is to determine the location and number of individual red attack trees within large areas by utilising an adaptive cluster sampling in a line transect design. To define areas of infestation an automated object-based classification system (Bunting and Lucas 2006) was employed, and sites 129 were located along the transect lines. The mean number of infested trees and the variance was then calculated and compared to estimates of statistics derived from a conventional non-adaptive approach. A relative efficiency estimator was used to demonstrate the utility of the adaptive cluster approach to determine the number of mountain pine beetle killed trees over the landscape. 5.2 Materials and methods 5.2.1 Phase 1: Individual tree crown delineation on 10 cm imagery This research was conducted in forests situated in the focus area described in Chapter 1 and used 20 cm digital aerial images (as described in Chapter 1) to form the basis of the sampling design. Imagery was acquired over an area of 40 km2 (10 x 4 km or 50,000 x 20,000 pixels) and mosaicked together to form a continuous image. Individual tree crowns can be delineated on high-spatial resolution imagery using object-based classification techniques and can be further classified according to species or health status. Bunting and Lucas (2006) successfully utilised Compact Airborne Spectrographic Imager remotely sensed data to define individual tree crowns in Australian forests with accuracies of approximately 70% (range 48% - 88%) for clusters and individual trees. Tree crowns were also successfully delineated on 10 cm spatial resolution digital aerial imagery in forests in western Canada with accuracies between 50% and 100% (mean 80.2%) when trees delineated on the imagery were correctly identified and compared with field measured trees (Coggins et al. 2008). Following crown delineation, stem diameter and stocking density were estimated from the image derived crowns and also 130 compared to field measurements using t-tests (stocking density: r2 = 0.91, se = 506.65, p <0.001; stem diameter: r2 = 0.51, standard error (se) = 2.63, p <0.001). Both these studies provide significant confidence in the approach and demonstrate that object-based classification has the ability to accurately define individual trees on remotely sensed data. With the methodology previously demonstrated I applied the same technique to delineate individual red attack tree crowns on the 20 cm spatial resolution imagery using Definiens Developer version 7 (Definiens AG 2007). The object-based classification algorithm (Figure 5.2) first identified individual trees within the image; secondly, determined the number of red attack trees; and finally, generated estimates of the total number of all trees and calculate crown areas. A mask was first created to differentiate between forest and non-tree vegetation such as bare ground and roads. The role of the mask was critical as it defined the outer boundaries of tree crowns and aimed to remove shadowing and ground vegetation from the segmentation procedure (Gougeon and Leckie 2002; Pouliot et al. 2002; Bunting and Lucas 2006). Secondly, all non-forested areas in the image were classified by identifying features with bright pixels, e.g. roads, recent clearcuts, and oil and gas landings. Thirdly, all remaining objects were classified as forest and a delineation algorithm was created to define individual tree crowns. To begin the delineation process the brightest objects in the forest class were used to identify as individual tree crowns (Bunting and Lucas 2006). Following identification, bordering objects with similar features were defined and the algorithm was programmed to merge and reclassify these objects into individual tree crowns. Following delineation, tree crowns were classified using four shape criteria, area, roundness, elliptical fit, and the ratio of object length to width, each of which has been 131 Figure 5.1: Flow chart of the crown delineation and individual tree object-based classification algorithm. proven to be useful when used to classify tree crowns (Bunting and Lucas 2006). Red attack trees were distinguished from healthy trees by applying thresholds to the mean of the red band, the mean of the green band, and red ratio criteria7. Every red tree was identified and was used to provide an estimate of the population of mountain pine beetle attacked trees over the area in the image. 7 The red ratio computed as the number of red pixels to all pixels within an object, where redness is defined by the operator (Definiens AG 2009). 132 5.2.2 Phase 2: Adaptive cluster sampling Of the possible sampling options (e.g. simple random, systematic, stratified), adaptive cluster sampling with a line transect approach was utilised in this study. The adaptive cluster sampling is initiated by placing a sample grid over the area of interest (Figure 5.3a) from a random starting point. With adaptive cluster sampling the area sampled is increased, from the initial sample unit containing one or more objects of interest (yi) by adding additional units at the cardinal directions around the initial sample unit (Figure 5.3b). The object of interest in this study is the number of red attack trees present in each of the networks. Sample units continue to be added according to a predetermined condition of interest C, if for example C > 1 then all units adjacent to the initial unit in the cardinal directions are added to sample and the number of units increases in a similar fashion until C is no longer satisfied. The final collection of sample units is known as the sample network (Figure 5.3c). The units at the periphery of the sample network which do not satisfy C are also included and are colloquially referred to as edge units (Figure 5.3d; Thompson 1990) so the sample has a number of units at the centre that contain the object of interest and are surrounded by a number of blank units. With adaptive cluster sampling the initial sample size is determined using a simple random sample estimator (Thompson 1990), and units are placed at random on a square grid throughout an image. The line transects were chosen at random and placed within the grid using a simple random sample estimator: M E t sizesample / var 2 2 \u00C3\u0097 = (5.1) 133 where t is the t-value for a 98% confidence level, E is the acceptable error, in this case 5%, M is the number of grid squares in each transect line (secondary units), and the variance (var) was taken from a study performed within the area (Wulder et al. 2009). Figure 5.2: An example of an initial sample unit located within a grid square in an adaptive cluster sampling design (a). Additional sample units positioned at the cardinal directions of the initial sample unit (b). The final sample network (c) and the edge units which contain no instances of the object of interest (d). The presence of red attack in cells is indicated by RA. 134 Variance is calculated using the equation: ( ) 4 var 2 minmax RARAiance \u00E2\u0088\u0092 = (5.2) where RAmax is the highest number of red attack trees that exist within the study area, and RAmin the lowest number of red attack trees. To initiate the adaptive cluster sampling approach a square grid comprising of grid squares 60 x 60 m was overlaid on the digital aerial image (Figure 5.4a). The grid squares correspond to the size of field plots used during a reconnaissance of the study area in 2008. Furthermore, mountain pine beetles are known to disperse within a 30 m radius from previously attacked trees (Safranyik et al. 1992). Therefore, this plot size was thought to be suitable to locate mountain pine beetle infestation over the landscape. Transect lines (primary units) were positioned at random intervals on the sample grid after which mountain pine beetle damage was located within each line (Figure 5.4a). Initial sample units were located at each point where mountain pine beetle attack occurred, following which sample networks were built around each sample unit (Figure 5.4b). The number of red attack trees in each sample network was obtained using an object based classification technique to first delineate all tree crowns within the sample network and then was trained to focus on the red attack trees only. Estimates of the mean, variance, and confidence limits were calculated using the number of red attacked trees in each network. To estimate the mean and the variance a Horvitz-Thompson estimator (Horvitz and Thompson 1952) is used, which provides an unbiased estimate by dividing each y-value by the probability that unit is included in the sample (Thompson 1991a). For the line transect method this probability is estimated by determining which primary units are likely to intersect network k in the initial sample. This probability is given by: 135 \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092 \u00E2\u0088\u0092= n N n xN k k /1\u00CF\u0080 (5.3) where N is the number of primary units available within the sample grid, n is the number of sample transects used for the study and xk is the width of the network at the point where the initial sample unit is located within the line transect sample. This probability is calculated for each network over the sample area, following which the probability that on or more of the primary units that intersect network k and j is included in the initial sample (Thompson 1991a): \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00EF\u00A3\u00BA \u00EF\u00A3\u00BB \u00EF\u00A3\u00B9 \u00EF\u00A3\u00AF \u00EF\u00A3\u00B0 \u00EF\u00A3\u00AE \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092\u00E2\u0088\u0092\u00E2\u0088\u0092 \u00E2\u0088\u0092\u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092 +\u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092 \u00E2\u0088\u0092= n N n xxxN n xN n xN kjjkjk kj /1\u00CF\u0080 (5.4) where xk and xj refer to the width of each network in a pair, and xkj refers to the number of primary units that intersect both networks k and j (Thompson 1991a). The probabilities calculated by the equations are used to provide unbiased estimates of the mean and variance: \u00E2\u0088\u0091 = = K k k k acs y MN 1 1 \u00CF\u0080 \u00C2\u00B5 (5.5) \u00E2\u0088\u0091\u00E2\u0088\u0091 = = \u00EF\u00A3\u00B7 \u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC \u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092= K k K j jk kj kj jk acs yy NM Var 1 1 22 1 1 \u00CF\u0080\u00CF\u0080 \u00CF\u0080 \u00CF\u0080 (5.6) where all variables remain the same as previously described, k is any given network within the population and K is the total number of networks. The variance estimator can be used to provide estimates of the standard deviation by calculating the square root of the variance estimator. The standard deviation can be used to calculate a confidence interval around the mean. These equations are the basis which 136 provides estimates of the number of red attack trees within a landscape using an adaptive cluster sampling approach. 5.2.3 Phase 3: Non-adaptive approach In order to assess the efficiency of the adaptive cluster sampling a non-adaptive approach was also utilised, whereby the sample size corresponded to the number of sample units used for the adaptive cluster sampling approach. The sample units were also 60 x 60 m in size which were randomly placed throughout the sample grid and all these units were run through an object-based classification algorithm to extract the number of red attack trees. To calculate the number of red attacked trees within the landscape using a non-adaptive sampling technique an unbiased estimator of the mean was used: \u00E2\u0088\u0091 = = n i iY Mn 1 1 \u00C2\u00B5 (5.7) where Yi is the number of red attack trees in the sampling unit and all other variables are described previously. An unbiased estimator of the variance is: 2 12 var s NnM nN SRS \u00E2\u0088\u0092 = (5.8) where ( )\u00E2\u0088\u0091 = \u00E2\u0088\u0092 \u00E2\u0088\u0092 = n i i MY n s 1 22 1 1 1 \u00C2\u00B5 As for the adaptive sampling technique, the mean, variance, standard deviation and a confidence interval were calculated for the non-adaptive approach. 137 Figure 5.3: The initial sample grid of 60 x 60 m overlaid on the digital remotely sensed imagery with the transect lines shown in solid colour (a). The randomly placed transect lines positioned within the sample area with the sample networks (b) and the resulting red attack crown delineation within the sample networks and transect lines (c). 138 5.2.4 Phase 4: Relative efficiency Lastly, the relative efficiency of adaptive cluster sampling compared with a non-adaptive approach was calculated. The relative efficiency is calculated by comparing the variance estimates of one sampling technique to the other (K\u00C3\u00B6hl et al. 2006). In this study, the variance of the adaptive cluster sampling approach (varACS) was compared to the variance of the non-adaptive approach (varSRS): SRS ACSRE var var = (5.9) High values indicate the numerator is more efficient than the sampling technique used for the denominator. Comparatively, a value close or equal to 1 suggests there is little difference between one sampling method over the other (K\u00C3\u00B6hl et al. 2006). 5.3 Results The adaptive cluster sampling approach was conducted on a 20 cm digital aerial image mosaic covering an area of 40 km2. With a 60 x 60 m sample plot size the total number of primary units (N) available was 162, with 69 secondary units (M) contained within each transect line (Table 5.1). The total number of sample units possible for the area is N \u00C3\u0097 M = 11,178 sample units. To obtain a sample size using equation 1, the maximum number of red attack trees was 155 and the minimum was assumed to be 0, which estimated the variance (from equation 5.2) to be 1501.56. The number of transect lines (primary units) estimated to provide accurate results was 5 (n). The number of sample units used for the non-adaptive approach was 192, the same number of units utilised in all networks in the adaptive approach. 139 Table 5.1: A summary of input variables and estimates provided by adaptive cluster sampling and the non-adaptive approach. Variables Adaptive cluster sampling Non-adaptive approach N 162 192 M 69 N/A n 5 N/A Number of red attack trees located 29,635 164 Networks 34 N/A Mean 7.36 61.56 Variance 18.34 41.43 Standard deviation 4.28 6.44 Confidence limit -12.45, 27.18 40.53, 82.59 The object-based classification algorithm indicated red attack tree locations on each transect line. Initial sample units were positioned over each occurrence of mountain pine beetle damage, in total 37 initial sample units were positioned within the transect lines. Sample networks were then built around the initial sample units and the red attack trees were identified within each network by the object-based classification algorithm (Figure 4c). The total number of red attack trees defined in the networks was 29,635. The mean number of red attacked trees per hectare located using adaptive cluster sampling was 7.36 trees. The variance was 18.34, and the standard deviation was 4.28 trees per hectare. The confidence limit at the 95% level ranged from -12.45 to 27.18 with t 0.05/2, 5-1 = 2.776. The non-adaptive approach had a mean of 61.56 red attack trees per hectare, with a variance of 41.43, and a standard deviation of 6.44 red attack trees. The confidence interval ranged from 40.53 to 82.59 using a t-value of 1.96 (t 0.05/2 192-1). Only 164 red attack trees were delineated in the sample units for the non-adaptive approach. The relative efficiency of the non-adaptive approach compared with the adaptive cluster 140 sampling approach demonstrates the latter gives (varSRS / varACS = 2.26) more than twice the efficiency when estimating the number of red attack trees on the landscape. 5.4 Discussion Adaptive cluster sampling is well suited to locate low level infestations and estimate the number of mountain pine beetle attacked trees over large areas. Results indicate the mean and variance for the adaptive technique (7.36 mean and 18.34 variance) are considerably smaller than those estimated by the non-adaptive technique (61.56 mean and 41.43 variance). Similar results were found by Thompson (1991a), who used adaptive cluster sampling with line transects. The high relative efficiency value is caused by the low number of red attack trees determined within the sample units in the non-adaptive technique. The random placement of sample units resulted in areas that were sampled without red attack damage, or were very close to red attack trees but did not encapsulate them and therefore, out of 192 sample units only 27 contained red attack trees. Compared to a non-adaptive approach, once initial sample units were determined for the adaptive approach, sampling was concentrated over areas containing mountain pine beetle attack. Therefore, many red attack trees were defined and estimates from these sample networks are less variable than from the non-adaptive approach. Despite the apparent advantages of cluster sampling to provide estimates of low-level mountain pine beetle attacks there a number of caveats. First, the final sample size cannot be fully determined prior to sampling because networks are grown during the sampling process. Second, due to the nature of the calculations it is difficult to perform adaptive cluster sampling over very large areas if small area sample units are required. Therefore, 141 the sample unit size must be chosen carefully before sampling is initiated. If, however, sample unit sizes are too large, an object of interest will always be contained with the unit, consequently very large areas are sampled and there would be little benefit from conducting adaptive cluster sampling. Adaptive cluster sampling can be easily applied in combination with most conventional sampling designs. For example, this chapter used adaptive cluster sampling in conjunction with line transect sampling, where the initial sample points (primary units) are lines. Each line is equally divided into square secondary units and sampling starts with all squares that contain the object of interest (Thompson 1991a). Other examples of variations on adaptive cluster sampling have been used in combination with conventional sampling techniques. Systematic adaptive sampling, where the primary sample plots are placed throughout an image or area at a fixed distance apart (Thompson 1991a). Double sampling with adaptive cluster sampling where samples are selected in two phases, first an inexpensive first phase sample is selected using adaptive cluster sampling design, then the networks are used to select an ordinary one- or two-phase subsample of units (Felix- Medina and Thompson 2004). Finally, stratified adaptive cluster sampling has been used whereby the population is stratified and then networks containing a object of interest are built in each strata following sample plot placement (Thompson 1991b). The ease by which adaptive techniques are used in conjunction with other sampling designs suggests it would be relatively simple to scale the number of red attack trees from very high spatial resolution imagery to larger areas, using a 2-phase stratified sampling design. This approach could be employed to predict the number of red attack trees over very large areas. Very fine scale (i.e., < 20 cm spatial resolution) imagery could be used 142 as sample plots within strata in a much larger area and adaptive cluster sampling performed in these images and then extrapolated up to the strata level and finally to the landscape level. Thereby, accurate estimates of the number of infested trees could be provided over very large areas. At the landscape level, inferences could be made regarding the location of red attacked trees and the severity of the attack over the landscape. The information provided by adaptive cluster sampling can be utilised to provide additional data for mitigation crews, the results from this approach have the potential to provide an approximate number of infested trees per hectare that can be expected. For the purpose of this discussion, detection of infested trees by surveys or through sampling methods implies these trees will be removed during ground surveys. The mean number of infested trees per hectare provides an estimate of the severity and extent of the infestation. The variance around the mean, however, provides an indication of the number of trees per hectare that are potentially infested. In this study, the adaptive cluster sampling approach generated a variance of 19 trees per hectare, which indicates that a further 11 infested trees per hectare could exist. If results from adaptive cluster sampling were utilised, mitigation would be completed on 8 infested trees per hectare if strictly following the mean. The trees left undetected and unmitigated would provide a source of beetles to attack and continue infestation the following year. Ground surveys would lessen the potential for infestation to continue. However, forests should be monitored in subsequent years to ensure infestations are detected and controlled to keep populations stable or in decline. 143 Adaptive cluster sampling has the potential to be beneficial when estimating small clusters of mountain pine beetle damage at the leading edge of the infestation. In areas such as western Alberta, where the beetle occurs in small clusters, adaptive cluster sampling could be used to identify areas of special concern where attack is starting to expand and return statistically sound estimates of the levels of attack and their locations. The spatial locations of attack are especially important as mitigation crews can be guided by this information to help slow the eastward spread of attack through mitigation. Other advantages to consider when using remotely sensed data in conjunction with this sampling technique include digital processing of remotely data to enhance and locate all red attack trees, and the ability to extract other data, such as the volume of timber attacked. In areas other than the leading edge, the aerial overview surveys currently utilised are sufficient to gather data on the progress of the infestation. Besides which, these areas generally contain high levels of infestation which would preclude the use of adaptive cluster sampling and also do not require fine-scale estimates of forest health data. Lastly, adaptive cluster has the potential to determine other rare and clustered events. This approach has been used to identify rare tree species (Acharaya et al. 2000), and sparse forest populations (Talvitie et al. 2006) and to predict forest inventory variables (Roesch 1993). The methodology used in this study is applicable to forests globally, to detect rare and clustered populations on the landscape that may be easily identified on remotely sensed imagery. The object of interest could be defined as windblow, root disease, old growth forest, or insect infestations, all of which can be defined on remotely sensed imagery and statistics generated from adaptive cluster sampling to define their 144 populations. Adaptive cluster sampling also has the potential to be combined with conventional sampling schemes, such as stratified sampling. The United States and Canada have large areas of forest cover which are generally homogeneous. However, forests in Europe are distinctly more fragmented and therefore, the landscape could be stratified into land use classes and adaptive cluster sampling used on forested areas to generate information. 145 5.5 References Acharaya, B., Bhattarai, G., de Gier, A., Stein, A. 2000. Systematic adaptive cluster sampling for the assessment of rare tree species in Nepal. Forest Ecology and Management. 137(1-3): 65 \u00E2\u0080\u0093 73. Bunting, P. and Lucas, R. 2006. The delineation of tree crowns in Australian mixed species forests using hyperspectral Compact Airborne Spectrographic Imager (CASI) data. Remote Sensing of Environment. 101(2): 230 \u00E2\u0080\u0093 248. Carroll, A.L., Taylor, S.W., R\u00C3\u00A9gni\u00C3\u00A8re, J., Safranyik, L. 2004. Effects of climate and climate change on the mountain pine beetle. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Proceedings of the mountain pine beetle symposium: challenges and solutions. pp. 223 \u00E2\u0080\u0093 232. October 30-31, 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Information Report BC-X-399. 298 p. Carroll, A.L., Shore, T.L. & Safranyik, L. 2006. Direct control: theory and practice. In: Safranyik, L, Wilson, B. (Eds.). The mountain pine beetle \u00E2\u0080\u0093 a synthesis of biology, management, and impacts in lodgepole pine. pp. 155 \u00E2\u0080\u0093 172. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC. 317 p. Ciesla, W. M. 2000. Remote sensing in forest health protection. USDA Forest Service, Fort Collins, CO., FHTET; Report No. 00-03. 266 p. Coggins, S.B., Coops, N.C. & Wulder, M.A. 2008. Initialisation of an insect infestation spread model using tree structure and spatial characteristics derived from high spatial resolution digital aerial imagery. Canadian Journal of Remote Sensing. 34(6): 485 \u00E2\u0080\u0093 502. Conners, M.E. and Schwager, S.J. 2002. The use of adaptive cluster sampling for hydroacoustic surveys. ICES Journal of Marine Science. 59(6): 1314 \u00E2\u0080\u0093 1325. Coops, N.C., Johnson, M., Wulder, M.A., White, J.C. 2006. Assessment of Quickbird high spatial resolution imagery to detect red attack damage due to mountain pine beetle infestation. Remote Sensing of Environment. 103(1): 67 \u00E2\u0080\u0093 80. Definiens AG. 2007. Definiens Developer. Version 7. M\u00C3\u00BCnchen, Germany. Definiens AG. 2009. eCognition Developer 8: user guide. Definiens AG, Munich, Germany. Felix-Medina, M.H., Thompson, S.K. 2004. Adaptive cluster double sampling. Biometrika. 91(4): 877 \u00E2\u0080\u0093 891. Franklin, S.E., Wulder, M.A., Skakun, R.S., Carroll, A.L. 2003. Mountain pine beetle red attack forest damage classification using stratified Landsat TM data in British Columbia, Canada. Photogrammetric Engineering and Remote Sensing. 69(3): 283 \u00E2\u0080\u0093 288. 146 Gougeon, F.A. and Leckie, D.G. 2002. Individual tree crown image analysis \u00E2\u0080\u0093 a step towards precision forestry. In: O\u00E2\u0080\u0099Shea, M. (Ed.). First International Precision Forestry Symposium. pp. 43 \u00E2\u0080\u0093 50. 17-20 June, 2001, Seattle, Washington. 193 p. Horvitz, D.G. and Thompson, D.J. 1952. A generalization of sampling without replacement from a finite universe. Journal of the American Statistical Association. 47(3): 663 \u00E2\u0080\u0093 685. K\u00C3\u00B6hl, M., Magnussen, S., Marchetti, M. 2006. Sampling methods, remote sensing and GIS multiresource forest inventory. Springer, New York. 373 p. Logan, J. and Powell, J. 2001. Ghost forests, global warming, and the mountain pine beetle (Coleoptera: Scholytidae). American Entomologist. Fall: 162 \u00E2\u0080\u0093 172. Logan, J. and Powell, J. 2003. Modelling mountain pine beetle phenological response to temperature. In: Shore, T.L., Brooks, J.E., & Stone, J.E. (Eds.). Proceedings of the mountain pine beetle symposium: challenges and solutions. pp. 210 \u00E2\u0080\u0093 222. October 30-31, 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Information Report BC-X-399. 298 p. Maclauchlan, L.E. and Brooks, J.E. 1998. Strategies and tactics for managing the mountain pine beetle, Dendroctonus ponderosae. British Columbia Forest Service, Kamloops Region Forest Health, Kamloops, BC. 55 p. Pouliot, D.A., King, D.J., Bell, F.W. & Pitt, D.G. 2002. Automated tree crown detection and delineation in high-resolution digital camera imagery of coniferous forest regeneration. Remote Sensing of Environment. 82(2-3): 322 \u00E2\u0080\u0093 334. Raffa, K.F., Aukema, B.H., Bentz, B.J., Carroll, A.L., Hicke, J.A., Turner, M.G., Romme, W.H.. 2008. Cross-scale drivers of natural disturbances prone to anthropogenic amplification: The dynamics of bark beetle eruptions. BioScience. 58(6): 501 \u00E2\u0080\u0093 517. Roesch, F.A. 1993. Adaptive cluster sampling for forest inventories. Forest Science. 39(4): 655\u00E2\u0080\u0093669. Safranyik, L. 1978. Effects of climate and weather on mountain pine beetle populations. In: Berryman, A.A., Amman, G.D., Stark, R.W. (Eds.). Theory and practice of mountain pine beetle management in lodegpole pine forests. pp. 79 \u00E2\u0080\u0093 86. 25-27 April 1978, University of Idaho, Moscow, Idaho. Symposium Proceedings. 224 p. Safranyik, L., Linton, D.A., Silversides, R., McMullen, L.H. 1992. Dispersal of released mountain pine beetles under the canopy of a mature lodgepole pine stand. Journal of Applied Entomology. 113 (5): 441 \u00E2\u0080\u0093 450. Skakun, R.S., Wulder, M.A., Franklin, S.E. 2003. Sensitivity of the thematic mapper enhanced wetness difference index to detect mountain pine beetle red attack damage. Remote Sensing of Environment. 86 (4): 433 \u00E2\u0080\u0093 443. Smith, D.R., Villella, R.F., Lemarie, D.P. 2003. Application of adaptive cluster sampling to low-density populations of freshwater mussels. Environmental and Ecological Statistics. 10(1): 7 \u00E2\u0080\u0093 15. 147 Smith, D.R., Conroy, M.J. & Brakhage, D.H. 1995. Efficiency of adaptive cluster sampling for estimating density of wintering waterfowl. Biometrics. 51(2): 777 \u00E2\u0080\u0093 788. Talvitie M., Leino O., Holopainen M. 2006. Inventory of sparse forest populations using adaptive cluster sampling. Silva Fennica. 40(1):101\u00E2\u0080\u0093108. Taylor, S.W. and Carroll, A.L. 2004. Disturbance, Forest Age, and Mountain Pine Beetle Outbreak Dynamics in BC: A Historical Perspective. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Mountain Pine Beetle Symposium: Challenges and Solutions. p. 41 \u00E2\u0080\u0093 51. October 30-31, 2003, Kelowna, British Columbia. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Information Report BC-X-399, Victoria, BC. 298 p. Thompson, S.K. 1990. Adaptive cluster sampling. Journal of the American Statistical Association. 85(412): 1050 \u00E2\u0080\u0093 1059. Thompson, S.K. 1991a. Adaptive cluster sampling: Designs with primary and secondary units. Biometrics. 47(3): 1103 \u00E2\u0080\u0093 1115. Thompson, S.K. 1991b. Stratified adaptive cluster sampling. Biometrika. 78(2): 389 \u00E2\u0080\u0093 397. Westfall, J. and Ebata, T. 2008. 2007 Summary of Forest Health Condition in British Columbia. Pest Management Report Number 15. British Columbia Ministry of Forests and Range, Forest Practices Branch, Victoria, BC. 81 p. Westfall, J. and Ebata, T. 2009. 2008 Summary of Forest Health Condition in British Columbia. Pest Management Report Number 15. British Columbia Ministry of Forests and Range, Forest Practices Branch, Victoria, BC. 85 p. White, J.C., Wulder, M.A., Brooks, D., Reich, R., Wheate, R.D. 2005. Detection of red attack stage mountain pine beetle infestation with high spatial resolution satellite imagery. Remote Sensing of Environment. 96(3-4): 340 \u00E2\u0080\u0093 351. Wulder, M.A., Dymond, C.C., White, J.C., Leckie, D.G., Carroll, A.L. 2006a. Surveying mountain pine beetle damage of forests: a review of remote sensing opportunities. Forest Ecology and Management. 221(1-3): 27 \u00E2\u0080\u0093 41. Wulder, M.A., White, J.C., Bentz, B.J., & Ebata, T. 2006b. Augmenting the existing survey hierarchy for mountain pine beetle red attack damage with satellite remotely sensed data. The Forestry Chronicle. 82(2): 187 \u00E2\u0080\u0093 202. Wulder, M.A., White, J.C., Bentz, B., Alvarez, M.F., Coops, N.C. 2006c. Estimating the probability of mountain pine beetle red attack damage. Remote Sensing of Environment. 101(2): 150 \u00E2\u0080\u0093 166. Wulder, M.A., Ortlepp, S.M., White, J.C., Coops, N.C., & Coggins, S.B. 2009. Monitoring the impacts of mountain pine beetle mitigation. Forest Ecology and Management. 258(7): 1118 \u00E2\u0080\u0093 1187. 148 6 ESTIMATES OF BARK BEETLE INFESTATION EXPANSION FACTORS WITH ADAPTIVE CLUSTER SAMPLING8 6.1 Introduction 6.1.1 Global forest resource Forests are complex ecosystems that provide a variety of goods and services to support economic, environmental, social, and cultural demands. Forestry produces raw fibre and non-timber forest products that provide economic return, and environmental services such as maintenance of biodiversity, increased air and water quality, and enhanced carbon sequestration (United Nations Food and Agriculture Organisation 2009). Forested land also provides recreation, and amenity and aesthetic benefits for users and for cultures with a connection to forests. Therefore, it is important to protect forested ecosystems from damaging agents such as fire, pests and diseases, pollution, adverse weather, and invasive species (United Nations Food and Agriculture Organisation 2009). 6.1.2 Forest insects Insects are reported to be the largest threat to forest resources (United Nations Food and Agricultural Organisation 2009). Between 1998 and 2002 insects damaged an estimated 37,134,000 hectares or 1.4% of the World\u00E2\u0080\u0099s forested area (United Nations Food and Agricultural Organisation 2006). Surveys to detect and monitor infestations are required to guide mitigation measures and prevent further spread. Data from these surveys are 8 A version of this chapter was published as: Coggins, S.B., Coops, N.C., Wulder, M.A. 2011. Estimates of bark beetle infestation expansion factors with adaptive cluster sampling. International Journal of Pest Management. 57(1): 11 \u00E2\u0080\u0093 21. 149 used to satisfy international reporting requirements, in particular for the Montr\u00C3\u00A9al Process which recognises the importance of forest health and vitality (Montr\u00C3\u00A9al Process Working Group 2001). Consequently, forest agencies perform surveys to detect and monitor insect infestations to determine the type of damaging agent, the amount of damage caused by them, and the location of affected forest stands. Most insects reported to cause damage to forests belong to the orders Coleoptera and Lepidopetra, with bark beetles being the most important with respect to reducing forest health. A common symptom of bark beetle attack is the appearance of red foliage (known as red attack). Green attacked trees have been infested by beetles but the foliage has yet to change colour, are strongly associated with red attack trees (Wulder et al. 2009a) and occur typically in close proximity. In terms of reducing the spread of infestations the green attack stage is most important because the population is still present beneath the bark, and if these trees are removed so are the beetles that cause further infestation. Members of the Dendroctonus genus cause large areas of tree mortality throughout Europe, China, and the Americas (United Nations Food and Agriculture Organisation 2009). Other Coleoptera beetles that cause damage to trees include Ips, Hylurgus, Pseudohylesinus, Scolytus, and Xylosandrus (United Nations Food and Agriculture Organisation 2009). Tree species commonly affected by Coleopeterans include, pine (Pinus spp.), firs (Abies spp.), and spruce (Picea spp.). The host symptoms associated with attack can include pitch tubes on the bark surface, frass (boring dust) that collects in bark fissures and on the forest floor around the tree. Typically, the foliage may remain green and appear healthy up to one year following attack; however, the burrowing action of the beetles beneath the bark constrains translocation of nutrients and water from the 150 roots to the foliage. The needles of the trees desiccate slowly over time, becoming red, usually one year after attack. The most aggressive bark beetle reported to occur is the mountain pine beetle, Dendroctonus ponderosae, Hopkins (United Nations Food and Agriculture Organisation 2009). The native range of the beetle extends from western Canada throughout the western United States to Mexico. The forests of western Canada in particular have experienced significant beetle damage due to steadily increasing annual temperatures and a readily available susceptible host, lodgepole pine (Pinus contorta Douglas. ex. Loud var. latifolia Englemann) In 1999, the mountain pine beetle had damaged a forest area of 164,000 hectares which increased to over 13 million hectares by 2008 with the potential to migrate south into the United States (United Nations Food and Agriculture Organisation 2009) and east across Canada (Raffa et al. 2008). Monitoring infestations is important for managing and controlling insect populations and for determining how quickly infestations spread. A common method used to calculate the rate at which infestations expand each year is to count both the number of newly attacked trees and those trees previously attacked, and express their numbers as a ratio (British Columbia Ministry of Forests and Range 2002). For example, if 2 newly infested trees exist for every previously attacked tree the ratio is 2:1. This ratio can also be reported as an expansion factor, in this case, 2. 6.1.3 Sampling to detect and quantify insect infestations Common techniques for monitoring and detecting forest insect pests involve a combination of aerial and ground surveys, or informal surveys that consist of observations from foresters and forest workers (United Nations Food and Agriculture Organisation 2009). Aerial surveys are implemented over very large areas of the 151 landscape to provide strategic information for informing forest management. These surveys detect and monitor a range of forest pests and are typically performed annually using fixed-wing aircraft to record the approximate location and severity of infestations. Finer-scale surveys then record the number and exact location of attacked trees; helicopters are guided by information recorded during the aerial surveys. Where possible, ground crews are dispatched to remove green attacked trees. While this combination of surveys is vital to the monitoring and detection of infestations, they are governed by the data quality provided by aerial overview surveys which do not easily detect individual or small groups of infested trees. Therefore, infestations occurring in areas that have not been attacked previously may not be detected until they have become large and therefore will not be controlled as easily as smaller infestations. With this survey hierarchy, an opportunity exists for using high spatial resolution airborne imagery to inform infestation dynamics and guide mitigation activities. Digital aerial imagery can also be used to identify, detect, and locate infestations and provide information for mitigation activities (Ciesla 2000). Very high spatial resolution digital aerial imagery can be acquired over small areas, providing estimates of forest structure and tree characteristics (Coggins et al. 2008a). However, the area captured on a single digital aerial image is relatively small compared to other imagery and is expensive to acquire over the very large areas required to monitor the leading edge of the infestation. Such imagery can, however, be used in a sampling context allowing accurate estimates of the location and number of infested trees to be obtained. This sampling technique offers a lower-cost solution to obtaining accurate data over large areas in a statistically sound manner. 152 Sampling for detection and location of insect infestations should aim to provide estimates of the mean and variance within defined confidence limits. Techniques such as simple random sampling can be used, but may not capture the full extent of infestations, thus resulting in high variability and wider confidence limits due to infestations being rare at the leading edge and, therefore, not easily detected. In contrast, adaptive cluster sampling has been demonstrated to more accurately characterise rare populations that are spatially clustered (Thompson 1990) and can provide estimates of population densities over large areas. Thompson (1991) compared adaptive cluster sampling to a non-adaptive approach and estimated the adaptive approach to be 5 times more efficient. Several previous studies have used adaptive cluster sampling for a variety of applications including estimating low density mussel populations (Smith et al. 2003), density of wintering waterfowl (Smith et al. 1995), and stock size of fish in estuarine rivers (Conners and Schwager 2002). Within forestry, applications of adaptive cluster sampling have included assessment of rare tree species in Nepal (Acharaya et al. 2000) inventorying sparse forest populations (Talvitie et al. 2006) and improving estimates of the location of insect infestations and providing estimates of the severity of insect attack (Coggins et al. in press). 6.1.4 Objectives The goals of this chapter were two-fold: first an adaptive cluster sampling approach was applied using very high spatial resolution digital remotely sensed data to detect low level bark beetle infestations within forests in western Canada; secondly, the expansion of insect populations over a two year period was examined and population estimates were obtained. 153 To meet these aims, the location and number of individual trees attacked by mountain pine beetles within our study sites was determined using adaptive cluster sampling, in a line transect design, with imagery acquired in both 2007 and 2008. Secondly, an automated object-based classification system was employed to determine the location and number of infested trees along the transect lines. Thirdly, estimates of the mean number of trees infested, the expected variance, and confidence intervals, and the rate of infestation expansion on the landscape were calculated. It is intended that these expansion rates be used by managers to determine the spread of infestations. Finally a sensitivity analysis was conducted to estimate the impact of commission and omission errors on the adaptive cluster sampling results. 6.2 Methods 6.2.1 Study area The research was conducted in the focus area described in Chapter 1 and used two study sites. Site A is in British Columbia, in the leading edge, close to the current infestation, whereas Site B is further east in Alberta, at the periphery of the infestation. Given the location of the sites in relation to the infestation, Site A was predicted to have a larger number of infested trees than Site B. To meet the aims of this study I used high spatial resolution digital airborne imagery to detect and locate mountain pine beetle infestations. To estimate the expansion of insect infestations, imagery must be acquired in two time- steps, ideally a year apart, and under similar viewing conditions, at the same time of year. High spatial resolution digital aerial images were acquired, as described in Chapter 1. Imagery was acquired near-nadir during August, 2007 from a flying height of 1100 m 154 and during August 2008 from a height of 2200 m, with a focal length of 85 mm, producing imagery with a 10 cm and 20 cm spatial resolution, respectively. Imagery acquired in 2007 and 2008 for both sites, each covered an area of 35 km2 (10.0 x 3.5 km for Site A and 7 x 5 km for Site B). Individual images were mosaicked to form a continuous image for each site in each year. The 2007 imagery was resampled to 20 cm using a nearest neighbour algorithm to ensure that I obtained similar resolutions in subsequent image processing. 6.2.2 Tree crown delineation Object-based classification techniques can be used to delineate individual tree crowns on high spatial resolution imagery which can then be further classified according to species or health status. Eucalypt trees in Australia were successfully delineated on Compact Airborne Spectrographic Imager remotely sensed data with accuracies of approximately 70% (range 48% - 88%) for clusters and individual trees (Bunting and Lucas 2006). Coggins et al. (2008a) successfully delineated individual tree crowns on 10 cm spatial resolution digital aerial imagery in forests in western Canada with accuracies between 50% and 100% (mean 80.2%). Estimates of stem diameter and stocking density were calculated for delineated tree crowns and compared with field measured trees using t- tests. Prediction of stocking density was highly significant (r2 = 0.91, SE = 506.65, p <0.001), while stem diameter was significant (r2 = 0.51, SE = 2.63, p <0.001). Both studies demonstrate that object-based crown delineation approaches are appropriate for reliably delineating individual trees on high spatial resolution remotely sensed image data. 155 A similar technique to those previously described (Bunting and Lucas 2006; Coggins et al. 2008a) was applied to delineate individual red attack tree crowns within the imagery of both sites. The object-based classification algorithm firstly identified individual trees within the image; and secondly determined the number of red attack trees; finally, it generated estimates of the total number of all trees and calculated crown areas. A mask was created to differentiate between forest and non-tree vegetation such as bare ground and roads (Gougeon and Leckie 2002; Pouliot et al. 2002; Bunting and Lucas 2006). Next, all non-forested areas in the image were classified to remove bright pixels such as roads, recent clear cuts, and smaller harvested areas. Finally, all remaining objects were classified as forest and a delineation algorithm was created to define individual tree crowns. To begin the delineation process the brightest objects in the forest class were used to identify individual tree crowns (Bunting and Lucas 2006). Following identification, bordering objects with similar features were defined and the objects merged and reclassified into individual tree crowns. After delineation was complete, tree crowns were classified using four shape criteri: area, roundness, elliptical fit, and the ratio of object length to width, each of these has been proven to be useful in classifying tree crowns (Bunting and Lucas 2006). Red attack trees were distinguished from healthy trees by applying thresholds to the mean of the red band, the mean of the green band, and red ratio computed as the number of red pixels to all pixels within an object (similar to Coops et al. 2006). Each identified red attack tree was used to provide an estimate of the population density of mountain pine beetle attacked trees in the imagery. 156 6.2.3 Adaptive cluster sampling Adaptive cluster sampling is an innovative approach which defines rare and clustered populations on a landscape. It is known to produce more accurate estimates than other methods such as simple random, systematic, stratified, and cluster sampling. To initiate sampling, a grid of sample cells is placed over an area, and within the grid, transect lines are placed at random. The number of transect lines (primary units) is determined using a simple random sample size estimator (equation 6.1): ( )[ ] M E RARAt sizesample / 4/ 2 2 minmax 2 \u00E2\u0088\u0092\u00C3\u0097 = (6.1) where t is the t-value for a 98% confidence level, E is the acceptable error (in this case 5%), M is the number of grid squares in each transect line (secondary units). Once transect lines are identified, initial sample units are determined where an object of interest exists (in this case, trees infested by mountain pine beetles; Figure 6.2a). The initial sample units are then expanded to form networks whereby, for each initial sample unit, the grid cells at the cardinal directions are examined to ascertain whether the object of interest is contained within the cells (Figure 6.2b). Sample cells are added to the initial sample unit when the object of interest is detected within them. Cells continue to be added at cardinal directions until the object of interest is no longer detected (Figure 6.2c); consequently, the sample network are represented by a core that contains the objects of interest and are surrounded by blank cells at the periphery, colloquially known as edge units (Figure 6.2d; Thompson 1990). Adaptive cluster sampling was first performed on the 2007 imagery at each site, where transect lines were positioned, the locations of red attack trees were determined, and 157 sample networks were defined. The transect lines and networks remained in the same locations for the 2008 image data and were reassessed. Initial sample units were added where new infestation existed along the transect lines, and networks were established where infestation had expanded since 2007. To obtain a sample size for the number of transect lines using equation 6.1, the maximum number of red attack trees (RAmax) was 155 and the minimum (RAmin) was assumed to be 0 (Wulder et al. 2009b). The number of transect lines estimated to provide accurate results in Site A was 5 (Figure 6.3) and 3 in Site B. Adaptive cluster sampling was initiated at both study sites with a grid of 60 x 60 m squares overlaid on the digital aerial imagery (Figure 6.3). The size of the grid was based on previous field studies (Wulder et al. 2009b) where tree-level mountain pine beetle infestation data were collected, as well as being the approximate dispersal distance of mountain pine beetles from previously attacked trees to susceptible hosts (30 m radius; Safranyik et al. 1992). The total number of primary units (N) in Site A was 158, with 63 secondary units (M) contained within each transect line. The total number of sample units possible for the area was N \u00C3\u0097 M = 9954. While both study sites were the same area, the dimensions of each site were different, in Site B the total number of primary units was 74 with 130 secondary units with a total number of sample units of 9620. Object-based classification determined the presence of mountain pine beetle damage in the initial sample units and then in the grid cells at each cardinal direction and defined each network based on the presence of red attacked trees. This process was performed first on the 2007 imagery and then on the 2008 so that some networks remained the same size or became larger as the infestation expanded from the previous year (Figure 6.2e). 158 Figure 6.1: Progression of network in adaptive cluster sampling, starting with the initial sample plot overlaid on 2007 imagery (a), sample units are added at the cardinal directions (b), the final sample network (c) and the edge units which contain no mountain pine beetle attack (d). Finally, the 2008 imagery is examined and the network is extended to account for infestation spread (e). 159 Figure 6.2: The 60 x 60 m sample grid (top) on which the adaptive cluster sampling was implemented in Site A. Five randomly placed transect lines are shown by black lines. The object-based classification defined each network placed along the transect lines in 2007 (black boxes). The transect lines were re-examined in 2008, new networks were established and old networks were extended to capture instances of mountain pine beetle attack. The 2008 networks are shown by the grey shading. The mean numbers of red attack trees found in 2007 and in 2008 were calculated and the rate of expansion was defined by dividing the mean number of trees attacked in 2008 by those attacked in 2007. To estimate the mean number of red attack trees and the variance, a Horvitz-Thompson estimator (Horvitz and Thompson 1952) was used, which provides an unbiased estimate 160 by dividing each y-value by the probability that a unit is included in the sample (Thompson 1991). For the line transect method this probability is estimated by determining which primary units are likely to intersect network k in the initial sample. This probability is given by: \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092 \u00E2\u0088\u0092= n N n xN k k /1\u00CF\u0080 (6.3) where N is the number of primary units possible in the sample area, n is the number of secondary units within each line, and xk is the width of the network at the point where the initial sample unit is located within the line transect sample. This probability is calculated for each network over the sample area, and then the probability that one or more primary units are intersected by each pair of networks is calculated using the equation: \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00EF\u00A3\u00BA \u00EF\u00A3\u00BB \u00EF\u00A3\u00B9 \u00EF\u00A3\u00AF \u00EF\u00A3\u00B0 \u00EF\u00A3\u00AE \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092\u00E2\u0088\u0092\u00E2\u0088\u0092 \u00E2\u0088\u0092\u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092 +\u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092 \u00E2\u0088\u0092= n N n xxxN n xN n xN kj /1 122121\u00CF\u0080 (6.4) where x1 and x2 refer to the width of each network in a pair, and x12 refers to the combination of a pair of networks. The probabilities calculated by the equations are used to provide unbiased estimates of the mean and variance, with all variables as described above: \u00E2\u0088\u0091 = = K K k k acs y MN 1 1 \u00CF\u0080 \u00C2\u00B5 (6.5) \u00E2\u0088\u0091\u00E2\u0088\u0091 = = \u00EF\u00A3\u00B7\u00EF\u00A3\u00B7 \u00EF\u00A3\u00B8 \u00EF\u00A3\u00B6 \u00EF\u00A3\u00AC\u00EF\u00A3\u00AC \u00EF\u00A3\u00AD \u00EF\u00A3\u00AB \u00E2\u0088\u0092= K K K j acs yy NM Var 1 1 21 12 12 21 22 1 1 \u00CF\u0080\u00CF\u0080 \u00CF\u0080 \u00CF\u0080 (6.6) A variance estimator can be used to provide estimates of a confidence interval around the mean. 161 6.2.4 Sensitivity analysis A sensitivity analysis was undertaken to assess the impact of mis-classification by the object-based approach to classify trees as attacked when they are not actually attacked (error of commission) and where attack was overlooked (error of omission). These errors determine the variation experienced when an object-based classification is conducted, and it is especially important in terms of pest management, as missed trees can result in infestation spread. By removing attacked trees the population is depleted and less infestation will occur in future years (Carroll et al. 2006; Coggins et al. 2008b). Previously, I estimated the accuracy of the object-based classification to be 80.2% (Coggins et al. 2008a). That accuracy statement was derived from 23 research plots and was the average value from within a range of 50% to 100%. This range in accuracy provides the boundary conditions for the sensitivity analysis. If the number of trees identified in each network is assumed to be 80% accurate, then the sensitivity around this value would be represented by the range. To represent the lower range, the number of red trees delineated by the object-based classification was reduced by 30% and for the higher range the number of trees was increased by 20%. 6.3 Results The adaptive cluster sampling approach was conducted on two digital aerial image mosaics, each covering an area of 32 km2. The results are summarized in Table 6.1. In 2007, a total of 39 initial sample units were positioned within transect lines in Site A and 17 in Site B. In 2008, the number of infested trees detected in the transect lines increased 162 Accordingly, initial sample plots were added and some networks were extended. The number of initial sample units increased to 54 in Site A and 39 in Site B. Higher numbers of infested trees/ha were delineated in Site A, as would be expected from a site closer to the current infestation. In Site B, at the extremity of the leading edge, attacked trees were less common and therefore mean numbers of trees were lower. Both sites experienced an increase in the number of red attacks between 2007 and 2008. The number of trees initially attacked in 2007 was higher in site A because it was closer to the \u00E2\u0080\u0098current\u00E2\u0080\u0099 infestation. Site B had less initial attack, and infestation expanded at a slower Table 6.1: Inputs and estimated variables provided by adaptive cluster sampling in 2007 and 2008 for Sites A and B. Site A Site B 2007 2008 2007 2008 N 158 74 M 63 130 N 5 3 Number of red attacks located 744 2267 41 89 Networks 30 36 14 26 Number of sample grid cells 403 766 17 208 Mean (trees/ha) 5.22 11.02 0.25 0.47 Variance (trees/ha) 10.65 24.83 0.02 0.08 Standard deviation (trees/ha) 3.27 4.98 0.14 0.28 Confidence interval (95%) -9.88, 20.32 -12.04, 34.07 -0.82, 1.32 -1.53, 4.01 t-value 2.076 4.302 Rate of expansion 2.11 1.87 rate than in site A. The adaptive cluster sampling results for each site are similar to these initial observations. In 2007, the mean number of infested trees per hectare, and the variance, in site A was higher than site B. The infestation expanded in 2008 and the 163 means and variances increased for both sites, with site A increasing more than site B. More variation was present in Site A as more trees became newly attacked and existing spot infestations spread to a greater degree than in Site B. Wider confidence intervals in 2008 at both sites indicated lower precision in the estimate of the mean (Figure 6.4). Confidence intervals are governed by variation in the infestations and as attack spreads in 2008 more networks are introduced and greater variation in the number of infested trees were found. The mean number of trees in 2007 and 2008 was calculated by adaptive cluster sampling and for each site were divided to determine the infestation expansion factor. The number of infested trees approximately doubled, in site A infestation expanded by 2.11 trees/ha and in site B, 1.87 trees/ha. -10 0 10 20 30 40 50 Site 8 2007 Site 8 2008 Site 7 2007 Site 7 2008 M e a n n u m b e r o f in fe s te d t re e s p e r h e c ta re Figure 6.3: The mean number of infested trees/ha with confidence interval for each site in 2007 and 2008. The sensitivity analysis indicated that if all infested trees were detected by the object- based classification up to 5 infested trees/ha could be present in Site A in 2008, with 2 164 trees present in 2007. The total number of trees delineated at the upper range in 2008 was 2720, with 893 trees in 2007. In Site B, infestation was considerably lower with 0.21 infested trees/ha (total trees delineated was 107) in Site B in 2008 and 0.11 trees/ha in 2007 (total trees delineated was 49). If in 2008, 50% of the trees were detected Site A would have contained 3 infested trees/ha (total number of trees identified was 55) with 1.32 trees/ha in 2007 with 521 trees delineated within the networks. In 2008, with the same rate of detection Site B would have contained 0.12 infested trees/ha (with a total of 62 infested trees detected) and in 2007, 0.06 infested trees/ha (with a total of 29 infested trees detected in networks). Furthermore, when more trees were accurately delineated the confidence limits become wider. Subsequently in 2008 in Site A the upper boundary of the confidence limit indicates 41 infested trees/ha were expected to exist within the study area, and in Site B a maximum of 3 infested trees/ha was expected. 6.4 Discussion In North America, mountain pine beetles affect forest stands from Mexico in the south up to Western Canada, causing extensive mortality to pine trees. In Western Canada, aggressive control programs have been implemented to manage populations of mountain pine beetles that have increased rapidly in recent years. Pine forests in British Columbia have experienced the highest levels of tree mortality. Commonly reported expansion factors have been 4 in southern British Columbia and 2 in the north (British Columbia Ministry of Forests and Range 2002). The rate of expansion is used to describe the severity of mountain pine beetle infestations in 3 stages (1) endemic, (2) incipient- epidemic, and (3) epidemic, with each stage more severe than the previous (Safranyik 165 2004). Endemic populations generally expand at a rate of less than 2, and are widespread in mature pine forests, but are restricted to weakened trees (Carroll et al. 2006). Incipient- endemic populations expand at a rate greater than 2 and infest trees in larger diameter classes. Beetles attack clumps of trees, which are scattered over the landscape, which tend to expand over time and commonly occur in gullies, and swamp edges. Epidemic infestations impact large areas and have an infestation expansion rate of between 4 and 5 over the entire affected area. In British Columbia, it is difficult to halt infestations as the attacking population of beetles from the current infestation exerts considerable pressure on susceptible forest stands (Shore and Safranyik 1992). However, with persistent detection, monitoring, and mitigation, forest managers can reduce attacking beetle populations at the leading edge and implement control (Carroll et al. 2006; Coggins et al. 2008b). Guided by aerial overview surveys, remote sensing imagery can be used to determine the locations and numbers of infested trees to support mitigation activities. Remotely sensed imagery has the capacity to detect insect infestations in forests. Landsat imagery (30 m spatial resolution) from the Thematic Mapper and Enhanced Thematic Mapper Plus sensors have been used to identify mountain pine beetle infestations, with detection accuracies ranging between 70% and 85% (Franklin et al. 2003; Skakun et al. 2003). Similarly, White et al. (2005) used IKONOS (4 m spatial resolution) to detect mountain pine beetle infestations at varying levels of attack intensity. When compared to field data, moderate infestations (where >5% to <20% of the forest contained attacked trees) were defined with an accuracy of 92%, and light infestations (where <5% of the forest contained attacked trees) with 71%. Hicke and Logan (2009) mapped the mortality of whitebark pine (Pinus 166 albicaulis Engelmann) with Quickbird imagery (spatial resolution of 2.44 m) after infestation by mountain pine beetle. In this study, a maximum likelihood classification was used to determine insect infestation. Accuracies ranging from 86% to 91% were reported when compared to field measurements. High spatial resolution remotely sensed imagery can be used in object-based classifications to automatically generate counts and locations of infested trees (Bunting and Lucas 2006; Coggins et al. 2008a). However, one drawback with object-based classification is the need to process large data sets, as even small areas of high spatial resolution digital aerial imagery require computers powerful enough to handle the large amount of data. Therefore, it is often quicker and more cost effective to implement object-based classifications on samples of imagery. Adaptive cluster sampling is well suited to detect and locate rare and clustered populations on the landscape (Thompson 1990) and has the capacity to detect forest inventory parameters, such as the presence of rare tree species (Acharaya et al. 2000) sparse populations (Talvitie et al. 2006), and to improve estimates of insect infestations (Coggins et al. in press). Adaptive cluster sampling lends itself well to both detecting and locating the red attack stage of mountain pine beetle infestations in a sample-based approach. The combination of high spatial resolution imagery, object-based classification, and sampling allows large area estimates to be made transparently, rapidly, and relatively easily. A study aimed at estimating the amount of vernal pool habitat in the northeastern United States by Van Meter et al. (2008) showed adaptive cluster sampling, used in conjunction with remotely sensed imagery, to be more accurate than simple random sampling over the same area. The approach presented in our study is portable and 167 has the potential to be implemented over larger areas. The transect lines can be extended to cover longer distances and the location and number of infested trees recorded on maps. If repeated annually, this type of study could be used to calculate overall infestation expansion rates over very large areas. Given that infestations are low and clustered at the leading edge, the combination of high spatial resolution digital aerial imagery and adaptive cluster sampling can provide estimates of the number of attacked trees. The estimates are especially important to forest managers for determining the effects on timber yield. Overall, less initial infestation was present in Site B than Site A because the site is situated on the extremity of the \u00E2\u0080\u0098current\u00E2\u0080\u0099 infestation. Attack had not become established in this area, although field observations determined that the area was susceptible to infestations, because it contained a large amount of mature, large diameter pine. In 2007, both the mean number of trees/ha and the variance were considerably higher in Site A when compared to Site B. Site A had a higher initial number of attacked trees than Site B, leading to larger estimates of the mean number of attacked trees and of the variance. Site A contained a larger number of networks and sample grid cells due to the higher incidence of infested trees; these in turn increased the mean and variance estimates. The confidence limits suggest the estimate of the mean in Site B to be more precise than for Site A. The confidence range for Site A was wider in both years when compared to the results for Site B. In both sites the population was doubling annually, similar to infestation expansion factors estimated by Carroll et al. (2006) and Wulder et al. (2009b). Even though Site B was less infested, the infestations were increasing at the same rate. The level of infestation and the expansion 168 indicates that persistent mitigation could had a greater impact in Site B while populations were still sufficiently low to have enabled cost-effective control. The role of remotely sensed data in this study was confined to very high spatial resolution digital aerial imagery, which captures data over small areas and can be expensive to acquire. However, there is the potential to stratify areas of mountain pine beetle damage over large areas on large area, medium spatial resolution imagery such as Landsat, which can be acquired at no cost. The presence of larger groups of red attack trees can be determined on this imagery and can be used as focal points to concentrate very high spatial resolution imagery. Adaptive cluster sampling can be performed in these areas to return results over the entire landscape and estimate the number and location of individual infested trees. Therefore it may be possible to produce very fine scale estimates over very large areas. Furthermore, object-based classification methods can be used to determine the initial strata and, using the methodology described in this chapter, can be used to perform the adaptive cluster sampling. Essentially, the entire process can be automated; thus decreasing the processing time and possible bias encountered by manually defining strata on the medium spatial resolution imagery and delineating mountain pine beetle infestations on the very high spatial resolution imagery. A key limitation of remotely sensed imagery is its inability to detect newly infested trees which maintain green foliage that appears healthy until after beetles have dispersed from the tree and have then colonised previously unattacked trees. However, beetles often disperse within 30 m of previously attacked trees, and so newly infested trees often have a strong association with red attacked trees (Wulder et al. 2009b). Once the red attack trees are detected, the approximate location of green attack trees can be determined, and 169 expansion factors can be used to ascertain the potential number of newly infested trees. A second major limitation of remote sensing data is errors of omission and commission, where green attack trees are omitted from classifications because they are not detected and infestation will increase from these trees if left undetected. The number of newly attacked trees at the leading edge is usually small in comparison to established infestations, and omissions of small groups of trees can be easily controlled when detected. Errors of omission can also occur where individual red attack trees are present on the landscape and are overlooked due to shadowing from surrounding trees or because the spectral reflectance of the crown is not detected by the classification algorithm. In contrast, large groups of trees are detected easily as the foliage discoloration contrasts with healthy trees and the discolouration is spread over a larger area. Errors of commission are more easily detected, so experienced image analysts can identify red attack trees and correctly classify them. When using multiple images acquired over two or more years, adaptive cluster sampling provides not only an estimate of the number of infested trees/ha, but also the location and the rate of infestation expansion. The rate of spread is important for ascertaining the impact beetles have on the landscape. If both the location of the infestations and the direction of spread can be determined, forest managers can prioritise allocation of resources to susceptible or valuable stands in close proximity to mountain pine beetle infestations. Infested trees detected from adaptive cluster sampling can be used to locate trees to guide mitigation and reduce the infestation in following years. Our results indicate that the population of beetles doubled between 2007 and 2008 in both study sites. Should climatic influences remain constant allowing beetles to survive the winters 170 and susceptible host trees are present, beetle populations will continue to grow unless infested trees are removed during ground surveys. Knowing the rate of infestation expansion, especially in sensitive areas, is important for determining the size of populations, whether they be endemic, incipient-epidemic, or epidemic. The methodology presented here has been used in the context of the mountain pine beetle but could be just as easily applied to other insect pests and forest degradation as long as the damage caused is rare and clustered on the landscape and can be recognised on remotely sensed imagery. Other insect species create symptoms that cause foliage degradation and discolouration, and the infestation they cause has the potential to be mapped with remote sensing data and adaptive cluster sampling. 171 6.5 References Acharaya, B., Bhattarai, G., de Gier, A., Stein, A. 2000. Systematic adaptive cluster sampling for the assessment of rare tree species in Nepal. Forest Ecology and Management. 137(1-3): 65 \u00E2\u0080\u0093 73. British Columbia Ministry of Forests and Range. 2002. What is the theoretical Maximum Green:Red? (online). Available at http://www.for.gov.bc.ca/hfp/health/fhdata/maxbeetles.htm. Accessed 14th September 2009. Bunting, P. and Lucas R. 2006. The delineation of tree crowns in Australian mixed species forests using hyperspectral Compact Airborne Spectrographic Imager (CASI) data. Remote Sensing of Environment. 101(2): 230 \u00E2\u0080\u0093 248. Carroll, A.L., Shore, T.L., Safranyik, L., 2006. Direct Control: Theory and Practice. In Safranyik, L. and Wilson, B. (Eds.). The Mountain Pine Beetle: A Synthesis of Biology, Management, and Impacts on Lodgepole Pine. pp. 155 \u00E2\u0080\u0093 172. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC, Canada. 317 p. Ciesla, W.M. 2000. Remote sensing in forest health protection. USDA Forest Service, Fort Collins, CO., FHTET; Report No. 00-03. p. 266. Coggins, S.B., Coops, N.C., Wulder, M.A. 2008a. Initialisation of an insect infestation spread model using tree structure and spatial characteristics derived from high spatial resolution digital aerial imagery. Canadian Journal of Remote Sensing. 34(6): 485 \u00E2\u0080\u0093 502. Coggins, S.B., Coops, N.C., Wulder, M.A., White, J.C. 2008b. Linking survey detection accuracy with ability to mitigate populations of mountain pine beetle. Forestry Chronicle. 84(6): 900 \u00E2\u0080\u0093 909. Coggins, S.B., Coops, N.C., Wulder, M.A. 2010. Improvement of low level bark beetle damage estimates with adaptive cluster sampling. Silva Fennica. 44(2): 289 \u00E2\u0080\u0093 301. Connors, M.E. and Schwager, S.J. 2002. The use of adaptive cluster sampling for hydroacoustic surveys. ICES Journal of Marine Science. 59(6): 1314 \u00E2\u0080\u0093 1325. Coops, N.C., Johnson, M., Wulder, M.A., White, J.C. 2006. Assessment of QuickBird high spatial resolution imagery to detect red attack damage due to mountain pine beetle infestation. Remote Sensing of Environment. 103(1): 67 \u00E2\u0080\u0093 80. Franklin, S.E., Wulder, M.A., Skakun, R.S., Carroll, A.L. 2003. Mountain pine beetle red-attack forest damage classification using stratified Landsat TM data in British Columbia, Canada. Photogrammetric Engineering and Remote Sensing. 69(3): 283 \u00E2\u0080\u0093 288. 172 Gougeon, F.A. and Leckie, D.G. 2002. Individual tree crown image analysis \u00E2\u0080\u0093 a step towards precision forestry. In: O\u00E2\u0080\u0099Shea, M. (Ed.). First International Precision Forestry Symposium. pp. 43 \u00E2\u0080\u0093 50. 17-20 June, 2001, Seattle, Washington. 193 p. Hicke, J.A. and Logan, J. 2009. Mapping whitebark pine mortality caused by a mountain pine beetle outbreak with high spatial resolution satellite imagery. International Journal of Remote Sensing. 30(17): 4427 \u00E2\u0080\u0093 4441. Horvitz, D.G. and Thompson, D.J. 1952. A generalization of sampling without replacement from a finite universe. Journal of the American Statistical Association. 47(3):663 \u00E2\u0080\u0093 685. Montr\u00C3\u00A9al Process Working Group. 2001. The Montr\u00C3\u00A9al Process [online] http://www.rinya.maff.go.jp/mpci/ Accessed 11th June 2010. Pouliot, D.A., King, D.J., Bell, F.W., Pitt, D.G. 2002. Automated tree crown detection and delineation in high-resolution digital camera imagery of coniferous forest regeneration. Remote Sensing of Environment. 82(2-3): 322 \u00E2\u0080\u0093 334. Raffa, K.F., Aukema, B.H., Bentz, B.J., Carroll, A.L., Hicke, J.A., Turner, M.G., Romme, W.H. 2008. Cross-scale drivers of natural disturbances prone to anthropogenic amplification: the dynamics of bark beetle eruptions. BioScience. 58(6): 501 \u00E2\u0080\u0093 517. Safranyik, L. and Carroll, A.L. 2004. The bionomics of the mountain pine beetle in lodgepole pine forests: establishing a context. In: Shore, T.L., Brooks, J.E., Stone, J.E. (Eds.). Mountain pine beetle symposium: challenges and solutions. pp. 223\u00E2\u0080\u0093 232. 30\u00E2\u0080\u009331 October 2003, Kelowna, British Columbia, Canada. Canadian Forest Service, Pacific Forestry Centre, Victoria, British Columbia, Information Report BC-X-399. p. 298. Safranyik, L., Linton, D.A., Silversides, R., McMullen, L.H. 1992. Dispersal of released mountain pine beetles under the canopy of a mature lodgepole pine stand. Journal of Applied Entomology. 113(5): 441\u00E2\u0080\u0093450. Skakun, R.S., Wulder, M.A., Franklin, S.E. 2003. Sensitivity of the Thematic Mapper Enhanced Wetness Difference Index (EWDI) to detect mountain pine beetle red- attack damage. Remote Sensing of Environment. 86(4): 433 \u00E2\u0080\u0093 443. Smith, D.R., Conroy, M.J., Brakhage, D.H. 1995. Efficiency of adaptive cluster sampling for estimating density of wintering waterfowl. Biometrics. 51(2): 777 \u00E2\u0080\u0093 788. Smith, D.R., Villella, R.F., Lemarie, D.P. 2003. Application of adaptive cluster sampling to low-density populations of freshwater mussels. Environmental and Ecological Statistics. 10(1): 7 \u00E2\u0080\u0093 15. Talvitie, M., Leino, O., Holopainen, M. 2006. Inventory of sparse forest populations using adaptive cluster sampling. Silva Fennica. 40(1): 101 \u00E2\u0080\u0093 108. Thompson, S.K. 1990. Adaptive cluster sampling. Journal of the American Statistical Association. 85(412): 1050 \u00E2\u0080\u0093 1059. Thompson, S.K. 1991. Adaptive cluster sampling: Designs with primary and secondary units. Biometrics. 47(3): 1103 \u00E2\u0080\u0093 1115. 173 United Nations Food and Agriculture Organization. 2006. Global forest resources assessment 2005: Progress towards sustainable forest management. (online). ftp://ftp.fao.org/docrep/fao/008/A0400E/A0400E00.pdf Accessed 9th January 2008. United Nations Food and Agricultural Organisation. 2009. Global review of forest pests and diseases. Rome, Italy. 235 p. Van Meter, R., Bailey, L.L., Grant, E.H.C. 2008. Methods for estimating the amount of vernal pool habitat in the northeastern United States. Wetlands. 28(3): 585 \u00E2\u0080\u0093 593. White, J.C., Wulder, M.A., Brooks, D., Reich, R., Wheate, R.D. 2005. Detection of red attack stage mountain pine beetle infestation with high spatial resolution satellite imagery. Remote Sensing of Environment. 96(3-4): 340 \u00E2\u0080\u0093 351. Wulder, M.A., White, J.C., Carroll, A.L., Coops, N.C. 2009a. Challenges for the operational detection of mountain pine beetle green attack with remote sensing. Forestry Chronicle. 85(1): 32\u00E2\u0088\u009238. Wulder, M.A., Ortlepp, S.M., White, J.C., Coops, N.C., Coggins, S.B. 2009b. Monitoring the impacts of mountain pine beetle mitigation. Forest Ecology and Management. 258(7): 1181 \u00E2\u0088\u0092 1187. 174 7 AUGMENTING FOREST HEALTH AND INVENTORY DATA WITH ATTRIBUTES DERIVED FROM LANDSAT THEMATIC MAPPER IMAGERY WITH ADVANCED REMOTE SENSING MODELS9 7.1 Introduction 7.1.1 Forest inventory Forest inventories are a critical tool in sustainable forest management providing important data on stands and individual trees and a mechanism to report information to aid forest management decisions, at international, national, and local levels. The type of data collected depends on the objective of the inventory (Lund 1998) and three levels are typically discussed, strategic, tactical, and operational (Anthony 1965). Strategic inventories report forest conditions over very wide areas with broad scale objectives (Davis et al. 2001) and supply data to assist with sustainable forest management, such as allowable annual cut allocations and the regional impact of insect infestations (Wulder et al. 2006a). Tactical inventories are used to implement the broad scale objectives supplied by strategic inventories and are conducted at the landscape level to determine when and where activities such as road layout or harvesting schedules take place (Davis et al. 2001). Operational inventories are fine scale and provide practical information at the forest stand level concerned with the timing of forest management activities (Gunn 1991). Strategic and tactical inventory data are usually compiled by the government agencies responsible for natural resource management, whereas operational inventories 9 A version of this chapter has been submitted as: Coggins, S.B., Coops, N.C., Hilker, T., Wulder, M.A. In review. Augmenting forest health and inventory data with attributes derived from Landsat Thematic Mapper imagery with advanced remote sensing models. 175 are implemented by, or on behalf of land tenure holders. Once compiled, forest inventories may be used to report estimates of: forest area, species composition, basal area, volume, age and site index, stand density, and growth and yield. Forest inventories are developed with statistically sound practices that allow for repeatable, unbiased, precise, and credible estimates to be made (K\u00C3\u00B6hl et al. 2006). Forest health is one key data attribute which is compiled in many forest inventories and used to report for strategic level planning purposes and for national schemes (Stone and Coops 2004; United Nations Food and Agricultural Organization 2009). In western Canada, the mountain pine beetle, Dendroctonus ponderosae (Hopkins) has caused mortality of over 13 million hectares of lodgepole pine forest (Pinus contorta Douglas. var. contorta Englemann; Raffa et al. 2009) in central British Columbia and is spreading north east into forests deemed susceptible to attack (Fall et al. 2006). As a result there is an increasing need for inventory level data to better understand the susceptibility of stands to the infestation, prioritize management activities to control mountain pine beetle outbreaks and to provide data to predictive tools for preventative management (Cole 1978; Maclauchlan and Brooks 1998). Once data is obtained the susceptibility of forest stands can be modelled and maps are created to indicate the likelihood of infestation. Mapping of susceptibility supports spread scenario development to determine which stands are most likely to be attacked, assists infestation management to assess the amount of attacked area in the stand, and is used to guide mitigation activities (Maclauchlan and Brooks 1998). Data collection for forest inventories is expensive, takes time to acquire and process, and requires expertise from field crews and from aerial photograph interpreters (Leckie 1990; 176 McRoberts and Tomppo 2007). As such, inventory data are not updated frequently and are subject to spatial and attribute errors due to inconsistent data collection standards and long time frames involved between data collections (Gillis 2001). Furthermore, it is often not possible to complete inventories everywhere if sites are inaccessible, or not under government jurisdiction or allocated for harvesting activities (i.e., parks and protected areas). In the case of forest health susceptibility maps based on inventory data the results can be misleading because re-inventory does not occur for up to 10 years and degradation to the forest resource by insects and management activities are not recorded within the inventory cycle (Wulder et al. 2006a). Therefore, derivation of an accurate susceptibility map is difficult in areas experiencing high tree mortality rates. 7.1.2 Remote sensing in forest inventory Satellite and airborne digital remotely sensed imagery has demonstrated an ability to augment forest inventories (Wulder et al. 2006a) with measurements recorded from remotely sensed imagery being used to regularly update or acquire data that are missing from forest inventories. Digital remote sensing demonstrates distinct advantages over the traditional methods of forest inventory related data capture. Imagery can be acquired successively over the same geographic location within a short time period, is low cost (in some cases free), and provides spatially accurate information over large areas rather than small sample plots. As a result, remotely sensed imagery is becoming more widely integrated with forest inventory data collection and a number of programs now either integrate digital remote sensing data, or have planned to integrate remote sensing into inventories in the next planning horizon (Lund 1998). Measurements at both strategic and tactical levels can be derived from digital imagery and used to predict forest inventory 177 variables. Strategic level estimates derived from remotely sensed data include variables such as: forest area, with estimates of forest type (Zhu and Evans 1994) provided by Advanced Very High Resolution Radiometer (AVHRR) data with 1 km spatial resolution (Zhu and Waller 2003); and forest disturbances, derived from Landsat Thematic Mapper (TM) imagery over the United States between 1984 and 2007 (Oswalt et al. 2009). Tactical level forest inventory measurements which can be derived from remote sensing include basal area per hectare, stand volume per hectare, age and site index, stand density, growth and yield, and tree health (Hussin and Bijker 2000; K\u00C3\u00B6hl et al. 2006). These variables can be derived with minimum accuracies of 75% (Hussin and Bijker 2000) either by direct measurement or from relationships to other more easily measured variables or estimated by modelling. 7.1.3 Geometric optical modelling Geometric optical modelling offers an advanced remote sensing modelling technique which has potential to provide detailed forest inventory information from moderate scale (30m) spatial resolution imagery. The conceptual model determines forest structural attributes by examining the brightness values of picture elements (pixels) within a remotely sensed image. The pixels of an image are much larger than individual trees but smaller than entire forest stands, therefore, the signal arriving at the sensor can be modelled as a linear combination of the response from ground scene elements, the shade and sunlit portions of trees, and the background (Li and Strahler 1985). The model by Li and Strahler (1985) estimates tree height, crown width, crown height, and stand density from endmembers of sunlit canopy and background, and shaded canopy and background. The shadow fraction in a given pixel is related to the geometry and density of a simplified 178 canopy surface model, which is illuminated to cast shadows (Scarth and Phinn 2000) and then seeks to explain the major portion of the variance in reflected and absorbed electromagnetic radiation in a remotely sensed image (Li and Strahler 1985; Scarth and Phinn 2000). If these parameters are combined with the solar zenith angle, the azimuth, the view zenith, local slope, and aspect, the remaining variations are accounted for by a \u00E2\u0080\u0098treeness\u00E2\u0080\u0099 parameter which describes the stand density. Attributes estimated by geometric optical models can be validated using the same field-based surveys as used for forest inventories and then adjusted to most accurately match field measurements. It may also be possible to estimate surrogate data such as DBH when field-based relationships are applied to model output. Geometric optical models have been utilized on large area medium-spatial resolution satellite imagery, such as the Landsat Multi-Spectral Scanner (MSS) and TM, and SPOT (Satellite Pour l\u00E2\u0080\u0099Observation de la Terre) and have proven application to forests in the northern (Li and Strahler 1985) and southern hemispheres (Scarth and Phinn 2000). The attributes predicted by these models can be compared to forest inventory data currently collected and used in susceptibility models (Table 7.1). A number of studies have used geometric optical modelling, the most notable being Li and Strahler (1985) who used Landsat MSS and SPOT in red fir (Abies magnifica Murray) and mixed conifer stands in north eastern California and estimated crown apex angle, crown height, crown radii, and number of tree crowns with between 64% and 90% accuracy. A sensitivity analysis was used to examine the model and demonstrated the model was not overly sensitive to error when determining the component signatures. 179 Scarth and Phinn (2000) used the Li and Strahler geometric optical model on a Landsat TM image of Eucalyptus forests in south east Queensland, Australia to generate estimates of crown cover, canopy size, and stand density. Results were reported to agree well with Table 7.1: Attributes required in forest inventories and susceptibility models with similar attributes provided by geometric optical models. Forest inventory attribute GOM prediction Reference Basal area Crown area Li and Strahler (1985); Scarth and Phinn (2000) Crown closure Crown closure Peddle et al. (2003) DBH Crown area Li and Strahler (1985); Scarth and Phinn (2000) Stand density Treeness Li and Strahler (1985); Scarth and Phinn (2000); Peddle et al. (2003) estimates of the same measurements collected during field surveys. A final layer was generated from the model and then smoothed with a 5 x 5 modal filter to interpret classes more easily. In application Scarth and Phinn (2000) found that crown size and number of tree crowns were the most accurately predicted using the technique with changes in crown height and tree heights more poorly predicted, especially in more uniform forest stands (Scarth and Phinn 2000). Other studies display similar results, Peddle et al. (2003) used a model to produce structural lookup tables for Landsat TM images for pre- and post-harvests in New Brunswick. Model results were validated for stand density, crown radius, and stem counts, and enabled simple forest structural change detection. 7.1.4 Aims This paper aims to extract key attributes from remotely sensed imagery to augment and update forest inventories. To fulfill these objectives, I apply the Li and Strahler geometric 180 optical model to medium spatial resolution Landsat imagery to estimate biophysical attributes of forests on a per pixel basis. The output from the model is then used to parameterize a susceptibility model and determine the likelihood of attack by the mountain pine beetle over large areas. Susceptibility will be derived from the model posited by Shore and Safranyik (1992). Geometric optical models are applied to Landsat data and used to determine a number of susceptibility variables including estimates of DBH, forest age, and stand density which are coupled with location information and the proportion of pine. The susceptibility of each pixel is calculated from these attributes and used to estimate the likelihood of attack by mountain pine beetle attack. 7.2 Material and methods 7.2.1 Remotely sensed data A single Landsat-5 Thematic Mapper image was acquired over the focus area (as described in Chapter 1) in August 2009 and a top of atmosphere correction was completed for all six optical channels using the QUick Atmospheric Correction (QUAC) model to remove the atmospheric distortions from the image (Bernstein et al. 2005). Earth Observation for Sustainable Development (EOSD) data, which identify 23 broad scale land cover classes from Landsat ETM+ imagery (Wulder et al. 2008), were used to create a forest mask to remove mountain tops, roads, clearcuts, and urban areas from the image. The forest mask also aided in the determination of the proportion of pine in each stand, increasing from 0% to 100%. A 30 m digital elevation model of the area was derived from the ASTER sensor aboard the Terra satellite (ASTER GDEM Validation Team 2009). 181 7.2.2 Susceptibility modelling Shore and Safranyik (1992) define susceptible lodgepole pine as trees with a diameter at breast height greater than 15 cm, over 80 years of age, and a stand density of approximately 1200 trees per hectare, that grow in low elevations at southern latitudes. The authors developed a model that used these characteristics to define the susceptibility of forest stands as a continuous variable (Shore et al. 2000) between 0 and 1, where 0 was not susceptible and 1 was highly susceptible (Shore and Safranyik 1992). The attributes in the model are the proportion of pine (P), the age of the stand (A), the stand density (D), and a location factor (L), and are multiplied together to give a total score. Evaluation of this model indicated that it can accurately determine the susceptibility of forest stands For example, Shore et al. (2000) compared the percent of basal area killed in 41 stands by MPB against the stand susceptibility index and found 40 stands fell within a 95% prediction interval of the model data (Shore et al. 2000). The Shore and Safranyik susceptibility model has become widely used in British Columbia and was adopted by the British Columbia Ministry of Forests to determine the hazard and risk of forests stands throughout the province. Susceptibility mapping typically generated using two methods; first, tree and forest stand measurements recorded from sample plots are entered into a susceptibility model (Shore and Safranyik 1992). However, this approach limits the formulation of susceptibility maps to small areas and does not account for variation within forest stands in areas surrounding the sample plots. The second method used data supplied in a forest inventory to generate estimates of susceptibility (Wulder et al. 2004). In this case, susceptibility 182 mapping from forest cover data can be calculated on a polygon basis to indicate the likelihood of attack across large forest areas. 7.2.3 Application of the geometric optical model To apply the model, I used two data sets, first pure pixels which contained 100% of the desired endmembers were identified within the image. Endmembers represent pure elements within an image and are represented in spectral space as a point cloud (Scarth and Phinn 2000). In a process called spectral unmixing the point cloud is rotated in spectral space to find apexes which are used to define the pixels that are most discrete as the endmembers in an image. The endmembers must satisfy a positivity constraint so none of them contain values less than zero (Gruninger et al. 2004). The Sequential Maximum Angle Convex Cone (SMACC) unmixing algorithm was then applied to define the endmembers over the entire scene (Gruninger et al. 2004). Second, initial values of the range of height from the base of the stem to mid-crown, radius of the vertical crown, and radius of the horizontal crown are required for model initialisation and were obtained from field observations (Table 7.2). The solar and view zeniths and azimuth were obtained from the Landsat metadata and assumed to be constant over the entire scene (Li and Strahler 1985). The geometric optical model can be run in either forward or inverse mode, where forward mode provides pixel brightness values of given biophysical attributes, whereas the inverse mode computes biophysical attributes from imagery. While there is no mathematical solution for inverting non-linear models, such as the Li and Strahler approach, one possibility to infer h, b, r, and m is through iterative adjustment of input 183 Table 7.2: Field data input for the geometric optical model, vertical crown height, stand density, crown radius, and tree height. Average Stdev SE CI m 1177 671.0626 76.976 150.871 b 3.623 1.830 0.120 0.236 h 14.271 3.173 0.209 0.409 r 2.121 0.872 0.040582 0.07954 parameters by running the model in forward mode to find a least squares solution that best matches the given endmember proportions. In this study, model inversion was done using the Trust region reflective algorithm of Coleman and Li (1996, 1999) and model outputs were matched to endmember proportions derived from spectral mixture analysis using the SMACC algorithm (Gruninger et al. 2004). In order to allow faster processing of the image data, a lookup table was produced in advance that contains a list of endmember proportions for all possible combinations of input parameters within a given range. The lookup table was then searched using a k-means nearest neighbour algorithm every time an iteration of the trust-region algorithm was made. Based on the previous application of the modelling approach by Scarth and Phinn (2000) I focused the analysis and validation on the estimation of two of the geometric optical model outputs; crown radius and stand density. The variation over these layers was calculated, and then pixels were grouped into forestry inventory polygons where the mean, maximum, minimum, and range were calculated for each polygon. 184 7.2.4 Model validation Validation was performed to determine the model accuracy with output compared to existing forest inventory data. The model results were filtered to remove areas within inventory polygons which contained non- forest, non-productive forest, not sufficiently restocked stands, and areas with sparse tree cover. In absence of crown radius estimates in the forest inventory data the modelled crown radius was compared to inventory crown closure estimates. Model derived stand density was compared to stand density in the vegetation resource inventory and further compared to quadratic mean diameter at breast height. Once confidence in the geometric optical model outputs was achieved I then applied the susceptibility model as described below. 7.2.5 Susceptibility model development Predicted crown radius was converted to DBH in a two step process: 1) crown radius was converted to crown area; and 2) a relationship between field measured DBH and crown area was created from the field data and used to determine mean DBH per pixel from the geometric optical model (Coggins et al. 2008a). Stem diameter = 9.09*log(crown area) + 12.87 (3) A mask was then created to remove all pixels with a DBH less than 15 cm. The mean basal area per pixel and the proportion of pine was calculated on a per pixel basis and used to determine the proportion of pine basal area in the pixel by multiplying the basal area by the proportion of pine. The proportion of pine was divided by the mean basal area to give the percent of susceptible pine in the pixel whereby the most susceptible stands 185 will have a higher proportion of pine with larger DBH, and less susceptible trees will have smaller proportion of pine. The geometric optical model height was converted to stand age using standard lodgepole pine inventory tables (Thrower 1994). The stand density layer (D) was taken directly from the model output of m, and reclassified to reflect stand density factors given by the susceptibility model. Finally a location factor was determined for each pixel, with elevation extracted from the digital elevation model, and the latitude and longitude determined for each pixel in a GIS. The equation given by the susceptibility model to calculate the location factor is (Shore and Safranyik 1992): Location = (24.4 \u00C3\u0097 longitude) - (121.9 \u00C3\u0097 latitude) - (elevation (m)) + (4545.1) (4) Finally, the four layers were then multiplied to determine the susceptibility of each pixel to attack, with values in the final layer between 0 and 1. The layer was reclassified to four categories using four evenly spaced breaks to determine the estimated susceptibility to mountain pine beetle attack giving categories that ranged from 0 to 0.249, 0.25 to 0.499, 0.50 to 0.749, and 0.75 to 100. These categories were assigned labels, where the lowest values were low susceptibility, progressing through medium, high and very high susceptibility. To provide further validation the susceptibility of each forest inventory polygon was calculated using the approach employed by Wulder et al. (2004). The inventory data was then cropped to the same extent as the susceptibility layer estimated from the geometric optical model. The area of forest susceptibility in low, medium, high and very high categories was then calculated for both layers and compared. 186 7.3 Results The Li and Strahler geometric optical model considers tree canopies as a collection of individual objects that cast shadows within the forest (Li and Strahler 1985) to create endmembers of sunlit canopy, sunlit background, shaded canopy, and shaded background. The endmembers representing shaded canopy and shaded background were merged to create a shadow image (Figure 7.2) and are shown with the sunlit canopy and sunlight background images and were input for the geometric optical model. Output from the geometric optical model consisted of four layers, each of which described biophysical attributes (m, r, b, and h) of the trees within each pixel (Figure 7.3). The mean (\u00C2\u00B1 standard deviation), minimum and maximum values from each output layer were calculated and are shown in Table 7.3. To validate the model results, the layers were compared to forest inventory data within the study area. The relationship between the predicted crown radius and crown closure derived from the inventory is shown in Figure 7.4 and demonstrates that that as crown closure increases crown radius decreases. Therefore trees in more open forest stands are predicted to have wider crowns than trees growing in denser forest stands. The standard error around the mean in each crown closure class is also displayed and shows distinct separation between the classes, indicating the crown radius estimates are reasonably predicted by the geometric optical model. The density variable for the model used the treeness layer from the model output (m) and represented the stand density variable (D) in the model. The mean stand density per forest inventory polygon was compared to stand density from the forest inventory data and showed an increasing trend (Figure 7.5) albeit 187 that class 1 appears to be over predicted by the model. Stocking density predicted by the geometric optical model was also compared to quadratic mean diameter at breast height which demonstrated a strong negative trend and distinct separation between classes (Figure 7.6). The validation of crown radius and stand density are considered to accurately reflect expected trends and demonstrate the model can be used to determine forest attributes. Finally, the location variable was generated from equation 4 (L). The layers were multiplied by each other to produce a susceptibility layer over the entire Landsat study area (Figure 7.7). The area of forest susceptibility in low, medium, high and very high categories as derived from the geometric optical model was then compared to the inventory derived approach and some notable differences are obvious (Table 7.4). The geometric optical model output estimated more forest with low susceptibility, but indicates decreases in both the medium and high susceptibility classes, while more area is estimated to be very highly susceptible. Table 7.3: Variation for the model output layers, measured by mean (\u00C2\u00B1 standard deviation), minimum and maximum values. Output Mean \u00C2\u00B1SD Maximum Minimum Treeness (m) 2753 7044 72208 381 Crown radius (r) 2.09 0.34 5.22 0.11 Crown height (b) 7.70 1.39 14.55 2.00 Tree height (h) 12.52 1.41 24.52 5.22 188 Figure 7.1: Endmembers from the SMACC algorithm that represent sunlit canopy (top left), sunlit background (top right), and shadowed (bottom) pixel fractions in the study area. 189 Figure 7.2: Output from the geometric optical model that represent biophysical attributes of the forest stands within Landsat pixels. 190 Figure 7.3: The relationship between crown closure provided by forest inventory data and crown radius (r) estimated by the geometric optical model with standards error around the mean. 191 100.00 1317.25 2534.50 3751.75 4969.00 Stocking density from forest inventory data (stems per ha) 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400 S to c k in g d e n s it y p re d ic te d b y t h e g e o m e tr ic o p ti c a l m o d e l (s te m s p e r h a ) Figure 7.4: The relationship between stocking density (stems per hectare) provided by forest inventory data and stand density (treeness; m) estimated by the geometric optical model with standard errors around the mean. 192 10.3 20.9 31.5 42.1 Quadratic mean diameter at breast height from the forest inventory (cm) 1420 1440 1460 1480 1500 1520 1540 1560 1580 1600 1620 1640 1660 1680 1700 S to c k in g d e n s it y p re d ic te d b y t h e g e o m e tr ic o p ti c a l m o d e l (S te m s p e r h a ) Figure 7.5: The relationship between quadratic mean diameter at breast height (cm) by forest inventory data and stand density (treeness; m) estimated by the geometric optical model with standard errors around the mean. 193 Figure 7.6: Susceptibility of the forest to mountain pine beetle attack in the study area estimated from model output. 194 Table 7.4: Comparison of susceptibility estimated by forest inventory data and by geometric optical models given as the percent of the forest land base. Susceptibility rating Forest inventory data % Geometric optical model data % Low 39% 58% Medium 31% 8% High 28% 7% Very high 2% 27% 195 7.4 Discussion The geometric optical modelling was initiated by supplying endmembers that described pure elements within the image and the SMACC algorithm appeared to successfully defined, sunlit canopy, sunlit background, and shadow across the image. Low levels of sunlit background occurred in shaded areas, such as protected mountain slopes and within the forests in valley bottoms. Sunlit canopy was also well defined considering it is more difficult to determine pure pixels within an image. Results indicate that pixels containing elements of sunlit canopy and shadow, are generally less pure, than the pixels selected for sunlit background. Shadow is also well represented, with low values on snow covered mountain tops and in clearcuts. Due to the complexity of forest stand structure, shadow also occurs in forest stands where shading is present between tree crowns and in small gaps. The geometric optical model output consisted of four layers, stand density (m), crown radius (r), crown height (b), and tree height (h). In this study stand density, and crown radius were compared to equivalent forest inventory variables. While reasonable estimates of forest age were generated by the model, generally age varied only slightly across the study area and therefore did not strongly influence the susceptibility predictions as much as stand density and crown radius. In addition there was no estimate of age available in the inventory data making a comparison with inventory data not possible. Overall results indicate a positive relationship between observed and predicted stand density. Where modelled stand density was high, the stand density in the forest inventory 196 polygons were also high, and similarly agreed when stand density from the inventory was approximately 1300 stems per hectare. However, results for low stand densities were less distinct and over predicted by the geometric optical model, a possible cause for confusion may arise from detection of ground vegetation, which is spectrally similar to tree crowns, and included in estimates of the treenees parameter. Given the uncertainty in the stand density relationship, I compared quadratic mean diameter at breast height from inventory data to stand density predicted from the geometric optical model and determined as tree diameters increased the stand density became lower. This trend is common in forests of similar species and is represented in stand density management diagrams (Smith et al. 1997). Furthermore, the standard error around the mean in each class indicates the geometric optical model output is distinctly separated by quadratic mean diameter at breast height. Crown radius was also compared to forest inventory data, and indicated as crown radius decreased, crown closure increased. As such, open forest had wider crowns, with closed canopies being associated with narrow crowns in dense forests. The modelled crown radius ranged from 2.7 m to 1.8 m, which according to field measurements on tree crowns in the area is an acceptable reduction as forests transition from open to dense forests stands. When these layers were entered into the susceptibility model, it is immediately apparent that forests most susceptible occur on the valley bottoms, with susceptibility decreasing at higher elevation as previously reported in the study area (Fall 2006). Finally, the percent of susceptible forest in the forest inventory was estimated and compared to the amount of susceptible area determined by the geometric optical model. The susceptibility of stands in the study area is believed to be increasing due to the maturing forest, with large proportions of densely stocked pine, which are ideal to 197 support mountain pine beetle populations available as well as increasing external pressure from attacking populations to the west also increasing the likelihood of attack (Fall et al. 2006). This area was previously unsuitable for attack because cold weather in the winter caused mortality to beetles overwintering beneath the bark of trees. However, recent research into climate change in the area suggests forests are becoming more susceptible as the area is becoming climatically suitable to attack (Fall et al. 2006; Shore et al. 2008). The susceptibility derived from geometric optical modelling showed differences when compared to the forest inventory data estimates. These differences are principally due to the time of data acquisition. The Landsat imagery was acquired in 2009, while the majority of the stand information in the forest inventory data has not been updated since 1991. The inventory is updated every 10 years; however, attributes such as growth and yield projections and tree characteristics rely on model projections to provide appropriate data. Therefore, the inventory does not directly account for tree growth or increases in forest age. Whereas the Landsat data is relatively current and produces estimates that reflect current stand conditions by accounting for the current density and DBH measurements which will increase susceptibility ratings. Likewise forest disturbances such as, recent harvesting, damage by fire or insect attack will decrease susceptibility ratings. Harvesting in particular has increased in the area in response to mountain pine beetle threat, and it was posited that tree removal in the study area would significantly affect the spread and extent of mountain pine beetle populations (Fall et al 2006). Both logging and fires have occurred significantly throughout the study area since 1991 and would account for the increase in area considered least likely to be attacked. As such, cohorts of trees which have been harvested from the high susceptibility stand classes 198 from 1991 to 2009 will result in differences in the two methods and could account for differences between the forest inventory data and the layer predicted by the geometric optical model. Geometric optical modelling produces data that can map the susceptibility of forests from medium spatial resolution satellite imagery over large areas. Landsat imagery has increased spatial accuracy over other data sources and is able to provide estimates of biophysical attributes of forest stands on a per pixel basis. For Landsat TM data, attributes are given in a 30 x 30 m pixel, whereas other data sources are polygon-based, covering (in some cases) several hectares. Therefore, data supplied by geometric optical modelling approaches can more accurately reproduce the variability of forest stands rather than average values over a large polygon. Also, given the temporal availability of medium spatial resolution satellite data, forest attributes can be recalculated regularly, without the need for extensive field work to validate output. Therefore, it is relatively simple to map areas that have undergone forest operations or areas affected by natural disturbance. Attacked forest can be delineated on remotely sensed imagery, where the biophysical attributes of attacked stands (DBH, stand density, and type of insect) can be used to augment forest inventory to aid with management and mitigation of insect infestations. The approach demonstrates that geometric optical modelling can also be used in conjunction with other models to stratify the landscape. In this paper I determine strata using geometric optical modelling output in a susceptibility model by Shore and Safranyik (1992) to determine the likelihood forests will be attacked by mountain pine beetles. Medium spatial resolution imagery can also be used to detect the red crowns 199 caused by the beetle infestation; however, low-level infestations are not easily detected (Wulder et al. 2006b; Coggins et al. 2008b). The highly susceptible strata produced by this method provide locations that warrant further investigation by aerial surveys or with additional high spatial resolution imagery to provide the locations and number of infested trees and produce estimates of the spread and extent of infestations. The highly susceptible strata could also be used to initiate secondary sampling schemes. For example, in previous studies, high spatial resolution imagery was acquired over highly susceptible areas and fine-detail forest attributes of stocking density and diameter at breast height were derived from imagery and accurately compared to field data (Coggins et al 2008a). The attributes were then used to initiate a forest infestation model to determine how infestation can spread and to determine which trees are likely to be attacked (Coggins et al 2008a). In addition advanced sampling techniques such as adaptive cluster sampling can be used to determine rare and clustered populations on a landscape (Thompson 1990) and are easily integrated within a stratified sampling approach (Thompson 1991). The location and estimates of the mean number of attacked trees were determined by Coggins et al. (2010) in transect lines throughout a study area that consisted of a 10 cm spatial resolution digital aerial image. Red attack trees were automatically delineated using object-based classification to determine the locations and numbers of attacked trees. Estimates of the mean, variance and confidence intervals were then calculated. This sampling technique was proven to be 5 times more efficient than a simple random sample. With strata defined by geometric optical modelling, the highly susceptible strata can define where high spatial resolution imagery should be flown to then determine whether stands have been attacked and the severity of attack. 200 Further research with susceptibility modelling could use a second model proposed by Shore and Safranyik (1992) which incorporates a beetle pressure model that predicts the likelihood stands will be attacked if beetles are close to susceptible stands. Therefore, stands close to an expanding beetle population are more likely to be attacked than stands further away from infestations. Previous studies have modified estimates of susceptibility to attack with beetle pressure indices using forest inventory data as a base (Wulder et al. 2006c). Beetle pressure estimates could be used in a geographic information system and combined with results from the geometric optical modelling and estimated locations of beetle attack to give a better indication of which stands are most susceptible to attack. 201 7.5 References Anthony, R.N. 1965. Planning and control systems: A framework for analysis. Harvard University Graduate School of Business Administration, Boston Massachusetts. 180 p. ASTER GDEM Validation Team. 2009. ASTER Global DEM validation: Summary report. (online). https://lpdaac.usgs.gov/lpdaac/content/download/4009/20069/version/3/file/ASTE R+GDEM+Validation+Summary+Report.pdf. Accessed 6th October 2010. Bernstein, L.S., Adler-Golden, S,M., Sundberg, R.L., Levine, R.Y., Perkins, T.C., Berk, A., Ratowski, A.J., Felde, G., Hoke, M.L. 2005. A new method for atmospheric correction and aerosol optical property retrieval for VIS-SWIR Multi- and hyperspectral imaging sensors: QUAC (Quick Atmospheric Correction). IGARSS 2005: IEEE International Geoscience and Remote Sensing Symposium. pp. 3549 \u00E2\u0080\u0093 3552 25th IEEE International Geoscience and Remote Sensing Symposium (IGARSS 2005) Seoul, South Korea, July 25-29, 2005. 359 p. Coggins, S.B., Coops, N.C., Wulder, M.A. 2008a. Initialization of an insect infestation spread model using tree structure and spatial characteristics derived from high spatial resolution digital aerial imagery. Canadian Journal of Remote Sensing. 34(6): 485\u00E2\u0080\u0093502. Coggins, S.B., Coops, N.C., Wulder, M.A., 2008b. Linking survey detection accuracy with ability to mitigate populations of mountain pine beetle. Forestry Chronicle. 84(6): 900\u00E2\u0080\u0093909. Coggins, S.B., Coops, N.C., Wulder, M.A. 2010. Improvement of low level bark beetle damage estimates with adaptive cluster sampling. Silva Fennica. 44(2): 289\u00E2\u0080\u0093301. Cole, D.M., 1978. Feasibility of silvicultural practices for reducing losses to the mountain pine beetle in lodgepole pine forests. In: Kibbee, D.L., Berryman, A.A., Amman, G.D., Stark, R.W. Proceedings, Theory and Practice of Mountain Pine Beetle Management in Lodgepole Pine Forests. pp. 140\u00E2\u0080\u0093147. 25\u00E2\u0080\u009327 April 1978, University of Idaho, Moscow. Pullman, Washington. 224 p. Coleman, T.F. and Li, Y.Y. 1996. An Interior, Trust Region Approach for Nonlinear Minimization Subject to Bounds. SIAM Journal of Optimization. 6: 418 \u00E2\u0080\u0093 445. Coleman, T.F. and Li, Y.Y. 1999. On the Convergence of Reflective Newton Methods for Large-Scale Nonlinear Minimization Subject to Bounds, Mathematical Programming. 67, 189 \u00E2\u0080\u0093 224. Davis, L.S.., Johnson, K.N., Bettinger, P.S., Howard, T.E. 2001. Forest management: To sustain ecological, economic, and social values. 4th Edition. McGraw Hill, New York. 804 p. Fall, A., Shore, T.L., Riel, B. 2006. Dawson Creek Mountain Pine Beetle Spread Analysis: Application of the SELES-MPB Landscape-Scale Mountain Pine Beetle 202 Model in the Dawson Timber Supply Area and Tree Farm License 48. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, B.C. Mountain Pine Beetle Initiative Working Paper 2006-18. 22 p. Gillis, M. 2001. Canada\u00E2\u0080\u0099s national forest inventory (responding to current information needs). Environmental Monitoring and Assessment. 67(1-2): 121 \u00E2\u0080\u0093 129. Gruninger, J, Ratkowski, A.J., Hoke, M.L. 2004. The Sequential Maximum Angle Convex Cone (SMACC) endmember model. Proceedings SPIE, Algorithms for Multispectral and Hyper-spectral and Ultraspectral Imagery. pp. 122 \u00E2\u0080\u0093 135. Vol. 5425-1, Orlando Florida, April, 2004. 396 p. Gunn, E.A. 1991. Some aspects of hierarchical planning in forest management. In: Buford, M.A. (Ed.). Proceedings of the 1991 Symposium on Systems Analysis in Forest Resources. pp. 54 \u00E2\u0080\u0093 62. South East Forest Experimental Station, General Technical Report SE-74. 423 p. Hussin, Y.A. and Bijker, W. 2000. Inventory of remote sensing applications in forestry for sustainable management. In: Remetey-F\u00C3\u00BCl\u00C3\u00B6pp, G., Clevers, J., Jan Beek, K. (Eds.). ISPRS 2000 congress : geoinformation for all. Part B7. pp. 575 \u00E2\u0080\u0093 587. Amsterdam, the Netherlands, 16-23 July, 2000. 1760 p. K\u00C3\u00B6hl, M., Magnussen, S.S., Marchetti, M. 2006. Sampling methods, remote sensing and GIS multiresource forest inventory. Springer, Berlin. 373 p. Leckie, D.G. 1990. Advances in remote sensing technologies for forest surveys and management. Canadian Journal of Forest Research. 20(4): 464 \u00E2\u0080\u0093 483. Li, X. and Strahler, A.H. 1985. Geometric-optical modelling of a conifer forest canopy. IEEE Transactions of Geoscience and Remote Sensing. GE-23(5): 705 \u00E2\u0080\u0093 721. Lund, H.G. 1998. IUFRO guidelines for designing multi-purpose forest resources inventories. IUFRO world series, vol 8. IUFRO, Vienna. 216 p. McRoberts, R.E. and Tomppo, E.O. 2007. Remote sensing support for national forest inventories. Remote Sensing of Environment. 110(4): 412 \u00E2\u0080\u0093 419. Maclauchlan, L.E., Brooks, J.E. 1998. Strategies and tactics for managing the mountain pine beetle, Dendroctonus ponderosae. British Columbia Forest Service, Kamloops Health Region. 58 p. Peddle, D.R., Franklin, S.E., Johnson, R.L., Lavigne, M.B., Wulder, M.A. 2003. Structural change detection in a disturbed conifer forest using a geometric optical reflectance model in multiple-forward mode. IEEE Transactions of Geoscience and Remote Sensing. 41: 163 \u00E2\u0080\u0093 166. Raffa, K.F., Aukema, B.H., Bentz, B.J., Carroll, A.L., Hicke, J.A., Turner, M.G., Romme, W.H. 2008. Cross-scale drivers of natural disturbances prone to anthropogenic amplification: the dynamics of bark beetle eruptions. Bioscience. 58(6): 501 \u00E2\u0080\u0093 517. Scarth, P. and Phinn, S. 2000. Determining forest structural attributes using an inverted geometric-optical model in mixed Eucalypt forests, Southeast Queensland, Australia. Remote Sensing of Environment. 71(2): 141 \u00E2\u0080\u0093 157. 203 Shore, T.L., Safranyik, L. 1992. Susceptibility and risk rating systems for the mountain pine beetle in lodgepole pine stands. Information Report BC-X-336, Canadian Forest Service, Pacific and Yukon Region, 17 p. Shore, T.L., Safranyik, L., Lemieux, J.P., 2000. Susceptibility of lodgepole pine stands to the mountain pine beetle: testing of a rating system. Canadian Journal of Forest Research. 30(1): 44 \u00E2\u0080\u0093 49. Shore, T.L., Riel, W.G., Fall, A. 2008. Incorporating present and future climatic suitability into decision support tools to predict geographic spread of the mountain pine beetle. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, B.C. Mountain Pine Beetle Initiative Working Paper 2008-10. 29 p. Smith, D.M., Larson, B.C., Kelty, M.J., Mark, P., Ashton, S. 1997. The practice of silviculture: Applied forest ecology. 9th Edition. Wiley and Sons, New York. 537 p. Stone, C. and Coops, N.C. 2004. Assessment and monitoring of damage from insects in Australian eucalypt forests and commercial plantations. Australian Journal of Entomology. 43(3): 283 \u00E2\u0080\u0093 292. Thompson, S.K. 1990. Adaptive cluster sampling. Journal of the American Statistical Association. 85(412): 1050 \u00E2\u0080\u0093 1059. Thompson, S.K. 1991. Adaptive cluster sampling: Designs with primary and secondary units. Biometrics. 47(3): 1103 \u00E2\u0080\u0093 1115. Thrower, J.S., Nussbaum, A.F., Lucca, D. 1994. Site curves and tables for British Columbia: Interior species. 2nd edition. British Columbia Ministry of Forests Research Branch. Land Management Handbook. Field Guide Insert 6. 50 p. United Nations Food and Agricultural Organisation. 2009. Global review of forest pests and diseases. Rome, Italy. 235 p. Wulder, M.A., Seemann, D., Dymond, C.C., Shore, T., Riel, B. 2004. Arc/Info Macro Language (AML) scripts for mapping susceptibility and risk of volume losses to mountain pine beetle in British Columbia. Technology Transfer Note 33. Canadian Forest Service, Pacific Forestry Centre, 4 p. Wulder, M.A., White, J.C., Bentz, B.J., Ebata, T. 2006a. Augmenting the existing survey hierarchy for mountain pine beetle red-attack damage with satellite remotely sensed data. The Forestry Chronicle. 82(2): 187 \u00E2\u0080\u0093 202. Wulder, M.A., Dymond, C.C., White, J.C., Leckie, D.G., Carroll, A.L., 2006b. Surveying mountain pine beetle damage of forests: A review of remote sensing opportunities. Forest Ecology and Management. 221(1-3): 27 \u00E2\u0080\u0093 41. Wulder, M.A., White, J.C., Dymond, C.C., Nelson, T., Boots, B., Shore, T.L. 2006c. Calculating the risk of mountain pine beetle attack: a comparison of distance- and density-based estimates of beetle pressure. Journal of Environmental Informatics. 8(2): 58 \u00E2\u0080\u0093 69. Wulder, M.A., White, J.C., Cranny, M., Hall, R.J., Luther, J.E., Beaudoin, A., Goodenough, D.G., Dechka, J.A. 2008. Monitoring Canada\u00E2\u0080\u0099s forests. Part 1: 204 Completion of the EOSD land cover project. Canadian Journal of Remote Sensing. 34(6): 549 \u00E2\u0080\u0093 562. Zhu, Z. and Evans, D.L. 1994. U.S. forest types and predicted percent forest cover from AVHRR data. Photogrammetric Engineering and Remote Sensing. 60(5): 525 \u00E2\u0080\u0093 531. Zhu, Z. and Waller, E. 2003. Global forest cover mapping for the United Nations Food and Agriculture Organization Forest Resources Assessment 2000 program. Forest Science. 49(3): 369 \u00E2\u0080\u0093 380. 205 8 CONCLUSION 8.1 Summary This thesis investigates the applicability of a hierarchy of optical remotely sensed imagery to monitor insect infestations and predict a range of forest inventory variables. To achieve this, three objectives were defined: 1) to investigate the implications of varying survey detection accuracies (derived from conventional surveys as well as a range of remotely sensed datasets) on mitigating and bringing under control mountain pine beetle populations (Chapters 2 and 3). 2) to assess whether a simple insect infestation spread model can be parameterised using tree structure and spatial characteristics derived from high spatial resolution digital aerial imagery (Chapters 4). 3) to demonstrate how current forest inventory data can be augmented with estimates supplied by innovative sampling techniques and advanced remote sensing models (Chapters 5, 6, and 7). The thesis was thematically separated into two sections; the first provided an overall framework for how remote sensing can be incorporated into monitoring programs for bark beetle infestations. The second section developed and validated a range of approaches, applicable within a hierarchy of analysis starting with fine-scale ground- based information extending up to remotely sensed imagery acquired from multiple sources, including very high spatial resolution digital aerial imagery and more broad scale Landsat data. These approaches could be applied in a top-down or bottom-up 206 approach. In top-down approaches, akin to conventional survey techniques, imagery at broader scales can be used to estimate the susceptibility of stands, guide the acquisition of higher spatial resolution imagery, and eventually locate field plots. This approach could be used to provide complimentary data to conventional survey approaches. Remotely sensed images have exhibited the capacity to provide more accurate geographic information on infested trees than some conventional survey approaches leading to increased detection rates over very large areas. By detecting infestations more accurately the impact on forest stands are reduced because attacking beetles are readily located and removed. Conversely, in a bottom-up approach, local knowledge can be used to position high spatial resolution imagery, and provide estimates of attack with subsequent approaches applied to scale these estimates to larger areas. This approach provides very fine scale detail on infestations and enables estimates of the number of attacked trees to be extrapolated from very high-spatial resolution imagery over very large areas. As such, the potential impact of the infestation can be estimated and forest stands prioritized according to their value in the land base, thus allowing appropriate resources to be allocated to stands. The conclusions of this thesis are presented below under the objectives outlined in the introduction. 8.2 To investigate the implications of varying survey detection accuracies on mitigating and bringing under control mountain pine beetle populations. Current mitigation techniques applied within British Columbia and Alberta should provide adequate control to reduce and ultimately maintain infestations to within endemic 207 levels. As such, when infestations double in size annually, detection rates of 50% or greater are required to control mountain pine beetle infestations (Carroll et al. 2006). Before embarking on a study to assess the capacity of digital remotely sensed data to provide complimentary data to conventional techniques, it is important to understand the current situation. When detection rates from conventional surveys were applied to simple infestation models (Carroll et al. 2006) results indicated the average infestation expansion rate across a network of plots located at the edge of the infestation was 3.8 in non- mitigated plots, requiring a detection accuracy of 73% to suppress the infestation. In the mitigated plots, infestations were expanding at an average rate of 1.1, requiring a minimum detection accuracy of approximately 10% to suppress infestation spread. The maximum rate of infestations in the unmitigated plots was 18, which would require mitigation rates of greater than 90% to provide suppression. The results lead to the conclusion that when determining an appropriate detection accuracy, the decision needs to be based on the uppermost range of the observed expansion factors (in this case 18). At that rate small infestations can quickly expand to a point where mitigation becomes infeasible to treat individual groups of trees, instead large scale salvage logging is generally required. The results demonstrate the consequences of not completing mitigation, or using inaccurate data to guide mitigation, resulting in prolonged infestation and increased tree mortality (Wulder et al. 2009). Over all mitigated plots, average mitigation efficacy undertaken using conventional surveying techniques in the study area was 43%, considerably lower than the rate recommended by Carroll et al. (2006). As a result mitigation guided by conventional survey techniques present an overwhelming task with respect to time and resources. 208 In the third chapter, remotely sensed data was integrated with the infestation models provided by Carroll et al. (2006) and was used to provide unique detection and monitoring information for mountain pine beetle attack. In a comprehensive review of remote sensing studies and the bionomics of mountain pine beetle attack it was found that high spatial resolution remotely sensed data has the ability to detect 71% of trees in small infestations (1-5% of the stand is infested), while in larger infestations detection accuracy can be as high as 92% (5-20% of the stand is infested; White et al. 2005). In incipient- epidemic conditions, populations triple annually and can be suppressed using a detection accuracy above 71%. In epidemic conditions, infestation expansion rates can increase by a magnitude of 5, which requires 80% detection to maintain stable conditions. However, mitigation driven by detection accuracies of 92% are able (assuming 100% mitigation success) to reduce the infestations to less than observed in previous years. Results indicate the detection capabilities of remotely sensed data are able to accurately distinguish red attack trees at multiple scales and leads to less infestation when detection rates are applied persistently over long time periods. Typically the goal of forest managers is to stop infestations with fall and burn on individual trees or with salvage logging on larger infestations (Machlauclan and Brooks 1998), but is often limited by access, costs, and markets available for felled trees. Typically, mountain pine beetle attack initiates gradually, suggesting mitigation should be completed before low level infestations begin to spread rapidly (Carroll et al. 2006). Then, managers can use remotely sensed data to provide information to determine the extent of infestations and the optimal mitigation strategy to use. Detection rates will depend on the type of imagery used, where very high spatial resolution imagery can 209 easily determine individual tree crowns to provide control of low level infestations. As spatial resolution decreases it becomes more difficult to determine individual trees because elements within pixels dilute and amalgamate. As such, many trees are represented by a single pixel (Wulder et al. 2006) and attacked trees are not detected until infestation increases to the point all trees within a pixel turn red. At this point the sensor can record the discoloration and infestation is detected. With lower spatial resolution imagery, detections rates become less accurate, but can be used to detect the presence of infestations that cover larger areas in more than one pixel. Results indicate that with ongoing monitoring, detection, and mitigation, infestations detected by remotely sensed data can be reduced. However, the current level of mitigation in some areas is not adequate to provide suppression of infestations and therefore techniques to provide higher rates of detection are required. Persistent mitigation requires ongoing control of green attacked trees, guided by the detection of red attacks and necessitate continued funding for mitigation programs from government agencies. Managers should be aware of the limitations of remote sensing imagery to detect green attack trees. Sensors do not possess the ability to accurately detect green attacked trees. However, research has been conducted to investigate the presence of green attacks in close proximity to red trees (Wulder et al. 2009) because dispersal generally occurs within 30 m to previously attacked trees (Mitchell and Preisler 1991; Safranyik et al. 1992). To aid persistent mitigation, government agencies are required to invest significant resources to maintain this level of support at the beginning of infestations and continue to provide financial aid even when the severity of attack appears to be decreasing. 210 8.3 To assess whether a simple insect infestation spread model can be parameterised using tree structure and spatial characteristics derived from high spatial resolution digital aerial imagery. Chapter 4 develops a new approach that offers new information on the impact of remotely sensed data to detect and monitor infestations. Trees were automatically delineated on digital aerial imagery with an object-based classification algorithm. The red attack trees were classified separately from forest that appeared healthy, and were evaluated to be 80.2% to 96.7% compared with ground survey data. Accuracy assessments were generated for two key variables, stocking density and diameter at breast height, and were found to be accurately predicted, DBH (r = 0.71, p <0.001); stand densities (r = 0.95, p <0.001). These variables were then utilised by a simple infestation model based on Mitchell and Preisler (1991) to predict the spread of mountain pine beetle infestations with results indicating that changes in structural and spatial attributes of forest stands will affect infestation levels. This study develops innovative approaches to process digital aerial imagery, demonstrates the capacity of this type of imagery to perform monitoring and detection over low-severity insect infestations, and can be used to examine the consequences of infestation spread over time. The model can determine which trees are most likely to be attacked based on the distance from previously attacked trees and the DBH of healthy trees in close proximity to infestations (Mitchell and Preisler 1991). Importantly, the results from this study could provide complimentary data on individual trees to parameterize a hazard and risk rating model. The model uses forest attributes usually sampled during field programs to determine the likelihood of attack in forest stands 211 (Shore and Safranyik 1992). Of principal importance to the model are stand density and DBH measurements, and instead of supplying estimates from a representative area of forest in a small sample plot, attributes can be generated with object-based classification techniques over larger areas, more akin to forest stand scales. 8.4 To demonstrate how current forest inventory data can be augmented with estimates supplied by innovative sampling techniques and advanced remote sensing models The final objective uses an established sampling method in a novel approach to determine per hectare level estimates of mountain pine beetle attack. Then an advanced remote sensing model was used to generate valuable insights into susceptibility estimates over the landscape level. An adaptive cluster sampling approach was applied to high spatial resolution digital aerial imagery to provide estimates of the number of mountain pine beetle attacked trees on the landscape and to monitor the spread of infestation over the land base. In this study, the object of interest consisted of red attack trees over large areas on the landscape. Results indicated the mean number of infested trees was 7.36 per hectare with a standard deviation of 4.28. In contrast a non-adaptive approach used the same sample area (192 plots) as the adaptive approach, and estimated the mean number of infested trees in the same area to be 61.56 infested trees per hectare with a standard deviation of 6.44. When the variances of each sampling technique were compared with a relative efficiency estimator, adaptive cluster sampling was twice as efficient to assess the number and location of red attack trees on the landscape. Similar results were found 212 in previous research that compared adaptive approaches with conventional sampling techniques (Thompson 1991). A time series of high-spatial resolution digital aerial imagery were used in conjunction with an adaptive cluster sampling approach to monitor the expansion of insect infestations over two sites. The first site had no recent mitigation activity (Site A) whereas attacked trees had been removed in the second site (Site B). In 2007, Site A contained a total of 39 initial sample units positioned within transect lines, with 17 in Site B. The mean number of attacked trees per hectare detected in Site A was 5.22 [\u00C2\u00B1 3.27 standard deviation (sd), sample plots = 403] and in Site B the mean attacked trees per hectare was 0.25 (\u00C2\u00B1 0.14 sd, sample plots = 17). In 2008, the networks that form the adaptive cluster sampling were reassessed and expanded where attacked trees were detected and statistics recalculated for the study area. Infestation expanded in both sites and in 2008, the number of initial sample units increased to 54 in Site A and 39 in Site B. The mean number of attacked trees in Site A increased to 11.02 (\u00C2\u00B1 4.98 sd, sample plots = 766) and in Site B expanded to 0.47 (\u00C2\u00B1 0.28 sd, sample plots = 208). A relative efficiency estimator was used to determine the rate of expansion and indicated attack had increased by 2.11 trees per hectare in Site A and 1.87 in Site B. As such, the main result from chapter six indicated that infestations in the study area increased by a magnitude of 2. Previous research in the study area (Carroll et al. 2006; Wulder et al. 2009) and field experience (British Columbia Ministry of Forests and Range 2002) suggests expansion of this magnitude can be expected in the study area. The adaptive cluster sampling approach can provide accurate data to help inform forest managers when making decisions for pest and disease management purposes. This unique 213 approach provides estimates of the number of attacked trees in the imagery which can be used to make informed decisions on the rate of expansion. The sample networks also provide the location of mountain pine beetle attacked trees. This approach is an important step to generate information on low level infestations, which usually are not detected and expand. These components indicate the level of attack severity allowing forest managers to prioritize resources to control outbreaks in forest stands to control infestations. Adaptive cluster sampling provides a technique to determine infestations on high spatial resolution imagery. However, landscape level estimates of beetle attack are also required. Advanced remotely sensed models that augment current forest inventory data and provide estimates from other digital data sources can be used a new approach to stratify large areas. The final research chapter applied a geometric optical model (Li and Strahler 1985) to Landsat data to obtain estimates of forest inventory variables at the landscape level. Variables predicted include DBH and stocking density which were then compared to forest inventory data and showed distinct trends that proved the geometric optical model generated accurate predictions. Once proven to accurately reflect inventory information the geometric optical model output was used to parameterize the mountain pine beetle susceptibility model (Shore and Safranyik 1992) to create a susceptibility layer. This layer was used to indicate the likelihood of attack by mountain pine beetles on the landscape. Traditionally, information on the severity of attack and data for susceptibility models are collected from field sample plots (Shore and Safranyik 1992), and more recently from forest inventory data (Wulder et al. 2004). Furthermore, conventional methods and maps derived from forest inventory data are subject to the expense of gathering and 214 maintaining accurate data (Leckie and Gillis 1995); whereas, remotely sensed data can be obtained freely and analysis performed to determine susceptibility over very large areas with small amounts of validation data. These methods are relatively inexpensive and less time consuming than gathering data from the field (Leckie 1990). Furthermore, data from remotely sensed imagery can be updated regularly to provide accurate estimates of the biophysical parameters of forest stands. The final research chapter demonstrated that forest inventory information stored in a GIS can be integrated with attributes generated from remotely sensed data and geometric optical models. When used in combination, this data can determine an area-wide characterization of susceptibility to mountain pine beetle. In so doing, an approach has been developed that can augment forest inventory coverage, address non-operable areas that are not captured with inventory implementations, and offer further capacity to include park lands in susceptibility mapping efforts. Further, the capacity to update and/or audit forest inventory data using independently and consistently produced attribution from remotely sensed data is also enabled. The aim of this portion of the thesis was to demonstrate remotely sensed data can provide accurate estimates to forest managers and government agencies about mountain pine beetle infestations. Disturbances from insect attacks rely on up-to-date information to supply accurate estimates of the location and severity of infestations. Remote sensing imagery can be acquired and analysed relatively quickly, when compared to extensive field campaigns or to aerial surveys (Riley 1989). This imagery provides a snapshot of conditions within an area and generates information that cannot be obtained by simple aerial surveys alone. A suite of novel approaches were developed to provide accurate 215 spatial locations of attacked trees, predict biophysical parameters of individual trees and forest stands, and generate estimates of the severity and extent of infestations using novel sampling techniques and advanced remote sensing models. These data are then used to control infestations and prevent populations of mountain pine beetles and other bark beetles from attacking forests. 8.5 Future research The thesis was separated thematically into two sections, the first section utilized detection accuracies from conventional surveys and then from remotely sensed data in simple population models. The population models used in the first section could be expanded to include a variable for climate change, resulting in more realistic results. Increased temperatures would potentially allow more beetles to survive winters which could cause expansion factors to rise, or in locations that experience cold weather insect development would be slowed. As such, emergence and dispersal would be delayed and the life stages beneath the bark of newly attacked trees would be under developed and would suffer extensive mortality and infestations will remain static. The models could be further adjusted in two ways, first to account for the range of expansion factors, and second to account for beetles that use long range dispersal to travel outside the stands and therefore do not contribute towards mortality of trees in the forest they emerged from. Each of these variables requires the expansion factors be manipulated to reflect the influence of temperature and long range dispersal. The models used are based on current expansion factors, it is hoped the range provided in the first section of this thesis demonstrates the scale infestations may increase to if left 216 uncontrolled. For example, the average expansion factor recorded in non-mitigated plots during the field work in Chapter 1 was approximately 5, but the maximum was 18. If infestations were to expand at the maximum rate, forest stands will experience a greater degree of damage than anticipated. Finally, 0.2% of beetles disperse above the canopy and leave the stand to attack trees, although this proportion may increase when few pines trees remain alive in a stand (Safranyik et al. 1992). At present the population models only account for dispersal from within the stands, an adjustment for long range dispersal by beetles could also be added. Another advantage of using remotely sensed data is the improvement of modelling by adding a spatial component to produce realistic and spatially exhaustive results of infestation expansion and severity. Further research is also needed to provide detection and monitoring over remote areas, park lands, private forest, and other inaccessible land. Remote sensing applications could be used to detect and monitor infestations and can also provide other estimates such as forest composition, stand age, stocking density which together can be used to calculate susceptibility and augment current forest inventory data. The second section of this thesis uses new and innovative approaches to generate precise forest inventory estimates from multi-source, multi-scale remotely sensed data in four chapters. Demand for forest inventory parameters generated in a cost-effective, timely manner will only increase due to the environmental and economic importance of forests. Increased accuracy in prediction of forest inventory variables will be realised as remotely sensed imagery becomes more sophisticated. It is necessary to fully research available remotely sensed imagery to reduce costs and time of completing surveys while also providing accurate data to populate forest inventories. 217 Overall, this research aimed to develop a greater understanding of utilizing remotely sensed data to detect damage caused forest pests and diseases and demonstrated the utility of remotely sensed data in a data hierarchy to provide forest inventory variables, and map susceptibility. The mountain pine beetle was used as an example; however, this research has the potential to be used for any pest or disease that can be detected with remotely sensed imagery. For example, other insects cause damage that can be characterised on remotely sensed imagery, and susceptibility models have been determined for other species (such as Douglas-fir beetle, Dendroctonus pseudotsugae Hopkins). Ultimately this research provides a means to initiate a top down, or bottom up survey strategy using conventional survey techniques and current forest inventory information, integrated with remote sensing information from a range of sources. Potentially, this research could be used to guide mitigation strategies to prevent insect infestations and provides methodology to extract forest inventory variables from remotely sensed data to augment and update current forest inventory information to supply accurate information on the movements of pests and disease that effect forests. 8.6 Recommendations The main innovations of this thesis were to determine the impact of remote sensing data on infestations, to develop methods to derive spatially accurate measurements of forest inventory information, to provide forest inventory information on a per hectare basis, and map the spread of infestation over very large areas. If the methodologies discussed in this thesis were to be used on a newly infested area it is recommended that first the level of mitigation required to suppress infestations is calculated, using the population models and 218 methods given in Chapter 2. The expected rate of infestation can be determined either from previous studies carried out in the area (Wulder et al. 2009), from forest inventory data, or from a few brief field surveys. The methods proposed in this thesis provide wall- to-wall coverage of large areas, with highly precise information on the forest resource. The level of detection required by remotely sensed imagery can be determined using methods developed in Chapter 3 according to the severity of infestation and can be used to determine how long mitigation should persist for based on the estimated infestation expansion factors. When the expansion factors and level of detection have been defined it is necessary to decide whether to use the methodologies presented in the second section of this thesis in a top-down, or bottom-up approach. If the study area is remote, a top down approach may be used where geometric optical modelling can determine which forests are mostly likely to be attacked. High spatial resolution imagery can be acquired in the areas most likely to be attacked and adaptive cluster sampling should be used to determine the severity and extent of infestations. Following which, highly accurate measurements of forest inventory information can be derived from the digital aerial imagery to determine the impact of infestations and the location of attacked trees. If the area has been well studied, a bottom-up approach is likely more useful. High spatial resolution images can be obtained for areas at the edge of current infestations, field work may not be required if tree and forest stand measurements exist. The presence of low- level attack can be delineated in an adaptive cluster sampling approach and will enable forest monitoring if successive years of imagery are acquired. Finally, geometric optical modelling output can be used to provide susceptibility maps over the study area and 219 estimates of traditional forest inventory parameters on a per hectare basis. The susceptibility maps can be used to determine the likelihood of infestation and preventative measures can be taken to reduce the chance of attack. The methodologies used to generate results in this thesis have been obtained from data that is inexpensive and with tools that are readily available. The high spatial resolution digital aerial imagery used in these studies is relatively inexpensive to purchase when compared to data generated from extensive field studies. Other digital data used (such as forestry inventory) have been obtained free of charge or with minimal expense. It is recommended before using the methodologies given in this thesis that a short reconnaissance be completed to measure forest stand attributes in field plots, and to mark the location of red attack trees as forests are typically variable in structure and composition and the formulae pertaining to structural measurements and the initial input for the geometric optical modelling are likely to differ from the forests used in this thesis. However, these measurements are cheap to obtain and easily recorded from a small number of field plots. Finally, the software used to analyse the data is widely available and the models used are becoming relatively simple to program and run. Therefore, it should be easy to reuse the methodologies presented and recreate the results for any forest area, be it in western Canada or elsewhere. 220 8.7 References British Columbia Ministry of Forests and Range. 2002. What is the theoretical Maximum Green:Red? (online). Available at http://www.for.gov.bc.ca/hfp/health/fhdata/maxbeetles.htm. Accessed 18th October 2010. Carroll, A.L., Shore, T.L., Safranyik, L. 2006. Direct control: theory and practice. In: Safranyik, L. and Wilson, B. (Eds.). The mountain pine beetle \u00E2\u0080\u0093 a synthesis of biology, management, and impacts in lodgepole pine. pp. 155 \u00E2\u0080\u0093 172. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC. 317 p. Leckie, D.G. 1990. Advances in remote sensing technologies for forest surveys and management. Canadian Journal of Forest Research. 20(4): 464 \u00E2\u0080\u0093 483. Leckie, D.G. and Gillis, M.D. 1995. Forest inventory in Canada with emphasis on map production. The Forestry Chronicle. 71(1): 74 \u00E2\u0080\u0093 88. Li, X. and Strahler A.H. 1985. Geometric-optical modeling of a conifer forest canopy. IEEE Transactions on Geoscience and Remote Sensing. GRS23: 705 \u00E2\u0080\u0093 721. Maclauchlan, L.E. and Brooks, J.E. 1998. Strategies and tactics for managing the mountain pine beetle, Dendroctonus ponderosae. British Columbia Forest Service, Kamloops Region Forest Health, Kamloops, BC. 55 p. Mitchell, R.G. and Preisler, H.K. 1991. Analysis of spatial patterns of lodgepole pine attacked by outbreak populations of the mountain pine beetle. Forest Science 37(5): 1390 \u00E2\u0080\u0093 1408. Riley, J.R. 1989. Remote sensing in entomology. Annual Review of Entomology. 34: 247 \u00E2\u0080\u0093 271. Safranyik, L., Linton, D.A., Silversides, R., McMullen, L.H. 1992. Dispersal of released mountain pine beetles under the canopy of a mature lodgepole pine stand. Journal of Applied Entomology. 113(5): 441 \u00E2\u0080\u0093 450. Shore, T.L. and Safranyik, L. 1992. Susceptibility and risk-rating systems for the mountain pine beetle in lodegpole pine stands. Pacific Forestry Centre, Canadian Forestry Service, Pacific and Yukon Region. Information Report BC-X-336. 12 p. Thompson, S.K. 1991. Adaptive cluster sampling: Designs with primary and secondary units. Biometrics. 47(3): 1103 \u00E2\u0080\u0093 1115. White, J.C., Wulder, M.A., Brooks, D., Reich, R., Wheate, R.D. 2005. Detection of red attack stage mountain pine beetle infestation with high spatial resolution satellite imagery. Remote Sensing of Environment. 96(3-4): 340 \u00E2\u0080\u0093 351. Wulder, M.A., Seemann, D., Dymond, C.C., Shore, T., Riel, B. 2004. Arc/Info Macro Language (AML) scripts for mapping susceptibility and risk of volume losses to 221 mountain pine beetle in British Columbia. Technology Transfer Note 33. Canadian Forest Service, Pacific Forestry Centre, 4 p. Wulder, M.A., White, J.C., Bentz, B.J., Ebata, T. 2006. Augmenting the existing survey hierarchy for mountain pine beetle red attack damage with satellite remotely sensed data. The Forestry Chronicle. 82(2): 187 \u00E2\u0080\u0093 202. Wulder, M.A., Ortlepp, S.M., White, J.C., Coops, N.C., Coggins, S.B. 2009. Monitoring the impacts of mountain pine beetle mitigation. Forest Ecology and Management. 258(7): 1181 \u00E2\u0088\u0092 1187."@en . "Thesis/Dissertation"@en . "2011-05"@en . "10.14288/1.0071579"@en . "eng"@en . "Forestry"@en . "Vancouver : University of British Columbia Library"@en . "University of British Columbia"@en . "Attribution-NonCommercial-NoDerivatives 4.0 International"@en . "http://creativecommons.org/licenses/by-nc-nd/4.0/"@en . "Graduate"@en . "Integration of multi-source, multi-scale remotely sensed imagery with ground survey information to provide forest health and inventory data"@en . "Text"@en . "http://hdl.handle.net/2429/30610"@en .