Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Using RapidEye satellite imagery to detect forest disturbances in British Columbia Arnett, John Thomas Te Reo 2014

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2014_september_arnett_john.pdf [ 2.08MB ]
Metadata
JSON: 24-1.0167248.json
JSON-LD: 24-1.0167248-ld.json
RDF/XML (Pretty): 24-1.0167248-rdf.xml
RDF/JSON: 24-1.0167248-rdf.json
Turtle: 24-1.0167248-turtle.txt
N-Triples: 24-1.0167248-rdf-ntriples.txt
Original Record: 24-1.0167248-source.json
Full Text
24-1.0167248-fulltext.txt
Citation
24-1.0167248.ris

Full Text

  Using RapidEye Satellite Imagery to Detect Forest Disturbances in British Columbia  by John Thomas Te Reo Arnett  B.Sc., University of British Columbia, 2012  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in  The Faculty of Graduate and Postdoctoral Studies (Forestry) THE UNIVERSITY OF BRITISH COLUMBIA  (Vancouver) July 2014 © John Thomas Te Reo Arnett, 2014 ii  Abstract  Improving our ability to track and monitor changes on Earth’s surface will inevitably enhance our ability to manage and monitor the biosphere. Remote Sensing technologies developed to monitor the Earth’s surface have already improved our understating of dynamic land cover change at a variety of scales. Fundamental to the identification of land cover change is the detection of abrupt disturbance events. These events constitute direct changes to the composition and structure of ecological systems and may have long lasting effects. In a forestry context it is important to identify disturbances in a timely manner in order to inform management decisions.  The RapidEye constellation is a series of five identical Earth orbiting optical sensors capable of achieving five meter spatial resolution imagery with a daily return time. In this thesis we present two studies which assess the capacity of RapidEye to detect (1) stand replacing disturbances and (2) non-stand replacing disturbances in British Columbia.    In the first study we develop a robust method to identify stand-replacing disturbances across seven regions in British Columbia. Overall accuracy for the classification of forest disturbance ranged from 83.65 ± 0.77% to 97.65 ± 0.25% for individual 25 X 25 km test locations. In the second study the utility of the RapidEye constellation to detect and characterize a low severity fire in a dry Western Canadian Forest was examined. Estimates of burn severity from field data were correlated with a selected suite of common spectral vegetation indices. All correlations between the ground estimates and vegetation indices produced significant results (p < 0.01). Consumption estimates of woody surface fuels ranged from 3.38 ± 1.03 Mg ha-1 to 11.72 ± 1.84 Mg ha-1 across four extrapolated severity classes. iii  The results of this thesis demonstrate the capacity of the RapidEye constellation to accurately detect stand replacing and non-stand replacing disturbances. We conclude by recommending the use of RapidEye to assess forest disturbances, with an emphasis on the spatial detail and temporal availability of imagery captured by the RapidEye sensors.    iv  Preface  This thesis is the combined work of two scientific papers of which I am the lead author. I preformed the primary research, data analysis, interpretation of the results and prepared the final manuscripts. Dr. Nicholas Coops project oversight as well as editorial assistance. Dr. Sarah Gergel provided guidance on ecological principals and statistical approaches. Dr. Robert Falls provided insight into the practical application of the research as well as editorial assistance. Dr. Lori Daniels provided guidance on ecological principals and fire ecology. Russell Baker provided guidance on the practical and applied aspects of the research.  Publications resulting from this thesis thus far include (reprinted with permission of the publishers): A Version of Chapter 2 has been published in:  Arnett, J.T.T.R., Coops, N.C., Gergel, S.E., Falls, R.W., Baker, R.H.  2014. Detecting Stand-Replacing Disturbance using RapidEye Imagery: a Tasseled Cap Transformation and Modified Disturbance Index. Canadian Journal of Remote Sensing 40: 1-14. A Version of Chapter 3 has been submitted for publication in:   Arnett, J.T.T.R., Coops, N.C., Daniels, L.D., Falls, R.W. 2014.  Detecting Forest Damage After a Low-Severity Fire using Remote Sensing at Multiple Scales. Submitted for publication.   v  Table of Contents  Abstract .......................................................................................................................................... ii Preface ........................................................................................................................................... iv Table of Contents .......................................................................................................................... v List of Tables ............................................................................................................................... vii List of Figures ............................................................................................................................... ix Acknowledgments ........................................................................................................................ xi Dedication .................................................................................................................................... xii 1. Introduction ........................................................................................................................... 1 1.1 Remote Sensing ................................................................................................................ 4 1.2 Thesis Objectives ............................................................................................................. 8 1.3 Remote Sensing of Disturbance ....................................................................................... 9 1.3.1 Sensor Development ................................................................................................. 9 1.3.2 Multispectral Vegetation Indices ............................................................................ 11 1.3.3 Change Detection Techniques ................................................................................ 13 1.4 Significance of Research ................................................................................................ 15 2 Detecting Stand Replacing Disturbance using RapidEye Imagery: a Tasseled Cap Transformation and Modified Disturbance Index................................................................... 17 2.1 Introduction .................................................................................................................... 17 2.2 Methods .......................................................................................................................... 21 2.2.1 Dataset 1: MoFLNRO VRI Disturbance Polygons ................................................. 22 2.2.2 Dataset 2: RapidEye Bitemporal Image Pairs ......................................................... 22 2.2.3 Dataset 3: RapidEye TCT Images........................................................................... 23 2.2.4 Dataset 4: EOSD LCC Derived Mask .................................................................... 23 2.2.5 Tasseled Cap Transformation ................................................................................. 25 2.2.6 Disturbance Index ................................................................................................... 26 2.2.7 Threshold Development .......................................................................................... 27 2.2.8 Verification ............................................................................................................. 29 2.3 Results ............................................................................................................................ 30 2.4 Discussion ...................................................................................................................... 37 vi  2.4.1 Accuracy Interpretation .......................................................................................... 38 2.4.2 Assessment of Burned Areas .................................................................................. 41 2.4.3 Impact of Atmospheric Distortion .......................................................................... 41 2.4.4 Threshold Selection ................................................................................................ 42 2.4.5 Future Applications ................................................................................................. 43 2.5 Conclusion ...................................................................................................................... 45 3 Detecting Forest Damage after a Low-Severity Fire using Remote Sensing at Multiple Scales ............................................................................................................................................ 47 3.1 Introduction .................................................................................................................... 47 3.2 Site Description .............................................................................................................. 52 3.3 Methods .......................................................................................................................... 53 3.3.1 Field Data ................................................................................................................ 53 3.3.2 Satellite Imagery ..................................................................................................... 55 3.3.3 Pre Processing ......................................................................................................... 55 3.3.4 Derivation of Vegetation Indices ............................................................................ 55 3.3.5 Exploring Spatial Variability and Image Semi-variance ........................................ 57 3.3.6 Relating field data to Vegetation Indices ................................................................ 57 3.4 Results: ........................................................................................................................... 58 3.5 Discussion ...................................................................................................................... 64 3.5.1 Assessment of Spatial Variability ........................................................................... 64 3.5.2 Relating Field Data to Vegetation Indices .............................................................. 65 3.5.3 Estimation of Sub-canopy Fuel Consumption ........................................................ 66 3.5.4 Monitoring Implications ......................................................................................... 68 3.6 Conclusion ...................................................................................................................... 69 4 Conclusion ........................................................................................................................... 70 4.1 Key Findings .................................................................................................................. 71 4.2 Future Work ................................................................................................................... 73 References .................................................................................................................................... 75  vii  List of Tables Table 1.1 An overview of different remote sensing platforms and resolutions. *MS = multispectral, *SWIR = shortwave infrared, pan = panchromatic.......................................... 5 Table 1.2 RapidEye spectral characteristics (modified from Tyc et al. 2005) ............................. 10 Table 1.3 Examples of RapidEye applications ............................................................................. 11 Table 1.4 describes commonly used vegetation indices ............................................................... 12 Table 2.1 A summary of MoFLNRO harvest indicated in RapidEye 25 X 25 km tiles, “-“ indicates measure was not calculated, *note tile 1063414 contains both fire and harvest ... 23 Table 2.2 A summary of polygons representing confirmed harvest indicated to have occurred between RapidEye image aqusition dates ............................................................................. 28 Table 2.3 A Summary of burn scar and stratified sampling areas indicated to have occurred between image acquisition dates ........................................................................................... 29 Table 2.4 Eigenvalue loadings derived from Principal Components Analysis of 50 images from British Columbia ................................................................................................................... 30 Table 2.5 Disturbance accuracies for individual tiles used in disturbance classification, D = Disturbance, N = No Change classes. Columns D and N represent sample proportions of map class (n = 10,000 samples/tile) ...................................................................................... 36 Table 2.6 Manually verified accuracy for overall disturbance classification, D = Disturbance, N = No Change classes. Columns D and N represent sample proportions of map class (n = 100 samples/tile) .......................................................................................................................... 37 Table 3.1 A summary of Landsat and RapidEye images selected in this study ........................... 55 viii  Table 3.2: Common vegetation indices and their formulas, * L refers to a constant used to shift the relationship between NIR and red wavelengths, in this study we used the conventional value of ‘1’ (see Huete 1988) ................................................................................................ 56 Table 3.3 Estimate of woody surface fuels consumed grouped by burn severity class, error is estimated as standard error of the mean. ............................................................................... 64 ix  List of Figures Figure 2.1 Processing flow for change detection using RapidEye imagery ................................. 25 Figure 2.2 Visual comparisons of three derived tasseled cap indices for an area containing various stages of forest succession, approximately 55 km southeast of Vanderhoof, BC, RE Tile ID 1063813. From left to right: TCT1, TCT2, TCT3 ................................................... 31 Figure 2.3 2D scatter plots depicting density slices of RapidEye tile 1063813, an area approximately 55 km southeast of Vanderhoof, BC ............................................................. 32 Figure 2.4 a comparison of TOA RapidEye (RGB) image and Tasselled Cap image, approximately 55 km southeast of Vanderhoof, BC. Recently cleared areas appear red, recovering regions are orange, stable areas are in blue. Image size is approximately 4 X 4 km.  TCT1, T TCT2, TCT3 are displayed using red, green, and blue colour guns respectively............................................................................................................................................... 32 Figure 2.5 A comparison of random stratified disturbance index values for burned and un-burned areas (sampled at a density of 10pts/ha). Unburned areas were sampled at random from a 500 m buffer around the indicated disturbance (‘Fire’ = from within burned area, ‘NC’ = from within no-change area). ................................................................................................ 33 Figure 2.6 A comparison of random stratified disturbance index values for harvested and non-harvested areas (sampled at a density of 10pts/ha). Non-harvested areas were sampled at random from a 500 m buffer around the indicated disturbance (‘Cut’ = from within har harvested area, ‘NC’ = from within no-change area). .......................................................... 34 Figure 2.7 Harvest activities near Burns Lake, BC. Areas indicted as harvest by MoFLNRO data are outlined in red. Left to right: disturbance index for image 1; disturbance index for image 2; difference image; and extracted disturbance using threshold >0.79 ................................. 35 x  Figure 2.8 User's and Producer's accuracy for a disturbance threshold gradient using RapidEye tile 964123 (Burns Lake, BC) ............................................................................................... 43 Figure 3.1 A map depicting the location of the Vasuex lake Study Area ..................................... 53 Figure 3.2: Field data for individual tree damage estimates for the 115 trees sampled. The combined burn percentage ranges from 0-300%, as three variables were summed. ............ 54 Figure 3.3 Semi-variograms comparing RapidEye and Landsat imagery for visible and near infrared bands........................................................................................................................ 59 Figure 3.4 Comparison of crown scorch severity estimates for Landsat and RapidEye Sensors. * indicates indices are not comparable across sensors............................................................. 60 Figure 3.5:Commparisons of Coefficients of Determination (R2) for correlations between SBI and vegetation indices derived from Landsat and RapidEye Imagery ................................. 61 Figure 3.6 Best fit between Landsat Soil Adjusted Vegetation Index and Simple Burn Index, p<0.05, R2 = 0.56, RE SAVI p<0.05, R2 = 0.51 ................................................................... 62 Figure 3.7 Map of dSAVI fire severity classes using transferred dNBR classes described by Key and Benson (2006) ................................................................................................................ 63   xi  Acknowledgments  This research was supported by Biosphere Management System Inc. with a components funded by a Natural Sciences and Engineering Research Council (NSERC) engage grant to Dr. Nicholas Coops. Special thanks to Russ Baker, Dr. Robert Falls, Dr. Sarah Gergel, and Dr. Lori Daniels for their co-authorship and insight from both research and industry perspectives.  I would like to thank the staff and researchers at the British Columbia Ministry of Forests, Lands, and Natural Resource Operations for their support and provisioning of data spatial data; Kat Gunion and Onat Yazir for their technical support and computer programming knowledge; Andrew Pylypchuk, Lee Smith and Jamin Doherty for providing RapidEye imagery and support; Canadian Wildlife Service for permission to conduct field work; G. Greene for assistance in the field, GIS collection and insights into the project; S. Lavallee, H. Erasmus, B. Forge, M. Labonté and J. Elleouet for assistance in the field.  Finally I would like to thanks the IRSS lab -I’m sure you’ve all helped me at some point along the way!     xii  Dedication I dedicate this thesis to my parents Barb and Chris Arnett for their support, my Grandparents John and Norma Arnett getting me there, and my Grandma Judy Anderson for getting me started.  1  1. INTRODUCTION  Identifying forest disturbances is a fundamental objective of forest management. Disturbance events are key drivers of forest structure, function, and composition (Franklin et al., 2002) and occur throughout a range of temporal and spatial scales. Globally, forest disturbances may have cumulative effects on species diversity and distribution (Defries et al. 1999, Gutschick and Bassirirad 2003, Loboda et al. 2012) as well as geophysical and landscape processes (e.g. Sidle 1992, Samonil et al. 2010, Pohl et al. 2012). Disturbance may also affect climate through structural and biochemical changes of forest vegetation; this can include change in land use and cover, surface albedo (e.g. Davin and de Noblet-Ducoudré 2010), evapotranspiration (Maness et al. 2013), and carbon flux (Kurz and Apps 1999). Consistent and timely identification of forest disturbances offers many benefits to managers and stakeholders concerned with monitoring various forest attributes.  Sustainable forest management often relies on the ‘natural range of variability’ (NRV) concept as a fundamental guide to managing disturbance (Landres et al. 1999). Within the proposed NRV, persistent small-scale disturbances are important for the maintenance of diverse habitat types and eventually contribute to the complexity of old-growth forest types (Oliver 1980). Emerging understanding of fine-scale disturbances suggests that low frequency, spatially discrete disturbances contribute to the maintenance of biodiversity (e.g. Picket and White 1985). Disturbances encompassed by the NRV are typically non-catastrophic in scale and tend to present a ‘no net loss’ scenario with respect to forest biomass and biodiversity. For the purpose of this research we refer to these as ‘non-stand replacing disturbances’.  2  Anthropogenic agents are arguably the main drivers of forest disturbance in many areas on Earth (Vistosek 1997, Marsh and Grossa 1996). Unlike disturbances encompassed by the NRV, anthropogenic disturbance events often result in permanent changes to the landscape (Foster 1992, Franklin et al. 2002). Furthermore anthropogenic disturbances tend to exceed the frequency and intensity of those occurring within the NRV. For example, Pearson (2010) found that during a 140-year period the extent of areas affected by logging was ten times that of disturbances attributed to non-anthropogenic agents. Anthropogenic disturbances such as harvest and land conversion are often abrupt, directly changing structural aspects of forests (Franklin et al. 2002). Although forest management paradigms have begun to emulate natural disturbance as a component of management practice, clear-cutting is nonetheless a common method of forest harvest (Bouchard et al. 2008).  Globally, the cumulative contribution of ecological services (such as maintenance of a suitable climate and biodiversity) may fluctuate in response to changes in the forest disturbance regime (Davin and de Noblet-Ducoudré 2010, Sunderlin et al. 2005). Changes to natural disturbance regimes may have undesirable consequences for people that rely on the ecological services provided by forests (e.g. Maruyama and Morioka 1998, Vedeld et al. 2007). Concerns over forest disturbance at a variety of scales have led to initiatives aimed at monitoring the Earth’s 3.8 billion hectares of forest area (FAO 2012). Programmes such as the United Nations Framework Convention on Climate Change (UNFCCC) and Reducing Emissions from Deforestation and Forest Degradation (REDD+) aim to reward the conservation of forests through the payment for ecological services (PES) by large carbon emitters (e.g. Mahanty et al. 2013). Identification of key disturbances is crucial for adopting management strategies best suited to meet the needs of a particular region. Attributes such as disturbance type, extent, and duration are all important 3  considerations for government and stakeholders wishing to assess and manage disturbances (e.g. Coops et al. 2009).  Timely, cost effective monitoring of disturbances across large areas is best achieved through the implementation of remote sensing technologies.  Remote sensing has aided implementation of forest disturbance monitoring initiatives (Cohen et al. 1996, Rogan and Miller 2006). For example, Brazil’s PRODES project, aimed at mapping annual deforestation in the Amazon, collects monthly data on abrupt changes in forest cover to estimate deforestation (Shimabukuro et al. 2006). The PRODES programme has been widely successful despite the relatively coarse grain size of 25 ha. Initiatives such as REDD+ have begun to develop global frameworks for assessment of anthropogenic disturbances and deforestation. The REDD+ initiative, aimed primarily at tropical forests located in developing countries, attempts to reward nations based on their ability to maintain the integrity of local forests (Miles and Kapos 2008, Pistorius 2012). Penman et al. (2003) suggest that monitoring for REDD+ must include (1) spatial extent of deforested area and (2) estimates of carbon stock densities, in order successfully address management concerns.  In Canada, the Canadian Forest Service’s National Forest Carbon Monitoring initiative (NFCM) was developed to meet the 1997 Kyoto protocol’s concerns over carbon uptake. The NFCM utilizes spatial forest inventory data collected at the national, provincial/regional, operational, and stand level to model deforestation, afforestation, and reforestation over the entire country (Kurz and Apps 2006). Currently, spatial aggregation of disturbance within the NFCM bounds is generated from aerial photography and remote sensing.  British Columbia is an ideal location to develop remote sensing technologies aimed at monitoring and tracking forest disturbances. In British Columbia, an economic reliance on forest 4  products has helped foster some of the world’s most comprehensive forest practices. As part of their commitment to monitoring forest resources, British Columbia’s Ministry of Forests, Lands and Natural Resources Operations (FLNRO) maintains a relatively comprehensive forest inventory database. Spatial data pertaining to disturbance type, forests management unit, harvest activity, and species composition are included in the inventory, which is derived from a combination of aerial photography and interpretation, as well as field observations.  Three characteristics make the Province particularly suitable to conduct remote sensing disturbance research.  These are summarized as follows:  1. The Province contains a variety of diverse forest types 2. The Province is host to a large range of forest disturbances types occurring at multiple spatial and temporal scales 3. The Provincial government arguably maintains an extensive and detailed forest inventory which could be used to test and calibrate disturbance detection routines 1.1 Remote Sensing  Remote sensing is increasingly used to assess ecological change and thus has become a key data source for forest and land managers (Joseph et al. 2010). Remotely sensed imagery can provide a suite of information pertaining to forest characteristics and spatial attributes of vegetation cover (Wulder and Franklin 2007). Furthermore remote sensing provides valuable information for ecological monitoring and is a cost-effective alternative to field and aerial surveys (e.g. Mumby et al. 1999).   Earth observing missions such as the National Aeronautics and Space Administration’s (NASA) Moderate Resolution Imaging Spectroradiometer (MODIS) and National Oceanic and 5  Atmospheric Administration’s (NOAA) Advanced Very High Resolution Radiometer (AVHRR) offer long-term continual data archives with applications in global forest monitoring. These satellites carry sensors directed at capturing Earth surface phenomena encompassing a range of spatial, temporal, radiometric, and spectral resolutions. Table 1.1 reviews some contemporary satellite sensors commonly used for earth surface observations.   Table 1.1 An overview of different remote sensing platforms and resolutions. *MS = multispectral, *SWIR = shortwave infrared, pan = panchromatic SATELLITE (SENSOR) SWATH WIDTH (km) SPATIAL RESOLUTION (m) (*) SPECTRAL RESOLUTION (nm) POSSIBLE APPLICATIONS LOW SPATIAL RESOLUTION SENSORS  NOAA  (AVHRR) 3000 1100 500-1250 Regional Forest Cover (Patopov et al. 2011) ENVISAT  (MERIS)  1150  300-1200  390-1040  Oceanic productivity (Moore et al. 2012)  Terra/Aqua (MODIS)  2330  250-1000  366-14385  Global fire occurrence (Hantson et al. 2013) SeaStar (SeaWiFS) 1500 1100 402-885 Oceanic Observations (Karabashev and Evdoshenko 2013) MEDIUM SPATIAL RESOLUTION SENSORS  Landsat 5,7,8 (TM, ETM+, OLI) 185 30 450-2350 Large area land cover mapping (Wulder et al. 2008)  SPOT 5 (HRG) 60 10 (MS); 20 (SWIR) 500-1730 Terrain Mapping (Wallerman et al. 2010) EO-1 (HYPERION) 7.5 30 400-2500 Coastal classification 6  HIGH SPATIAL RESOLUTION SENSORS  QuickBird-2 16.8 2.44 (MS); 0.8 (pan) 450-900 Disaster response (Meslem et al. 2011) IKONOS  13.8 4 (MS); 1 (pan) 450-850 Urban planning/mapping (Tack et al. 2012) RapidEye 77 6.5 (MS) 440-850 Agricultural monitoring (Tapsall et al. 2010)  Until recently, most satellites providing publically available data offered imagery of relatively low spatial and temporal resolution (Fensholt et al. 2011).  Higher spatial resolution platforms such as Ikonos and Quickbird sensors are capable of resolving fine scale land cover patterns; however, these sensors are limited in swath width and temporal resolution, and are generally considered expensive on an area basis. High temporal resolution sensors such as MODIS are capable of a two-day return time; however, its 250-meter grain size is unable to resolve smaller scale disturbance and land change events (Townsend and Justice 1988). Similarly, AVHRR offers a global revisit time every six hours but provides a course spatial resolution of 1.09 km/pixel (at nadir).   Both AVHRR and MODIS have been used to produce maps depicting changes in global forest cover for over a decade; however, their relatively low spatial resolution makes these sensors less useful for mapping change at the local or regional scale. Local mapping of forest is typically carried out using the NASA’s Landsat sensor (Cohen et al. 2002). Landsat data has been largely successful in assessing land cover trends over large areas and relatively long temporal intervals; however, with a 30-m resolution and 16-day return time, Landsat may be unable to resolve important abrupt or fine-scale changes (e.g. Gong and Xu 2003, Xie et al. 2008), especially in 7  areas with persistent cloud cover. This time delay can mean that events such as forest fires, illegal harvesting, and insect infestation may have long since occurred before they are identified using satellite imagery.  A study using Landsat images gathered over a 29-year period in the Brazilian Cerrado region concluded that the probability of acquiring a cloud free image during the wet season was <10% (Sano et al. 2007). A similar study conducted by Asner (2001) found that the probability of acquiring a cloud-free image of the northern Amazon was nearly zero for any given season. Although these studies are presented in a Brazilian context, global cloud cover maps developed by the University of Manitoba suggest that Brazilian annual cloud cover is comparable to the west coast of North America (Anderson 2014).  Landsat’s 16-day return interval allows for a maximum of 22 images to be acquired over the same area in a given year. An increase in the temporal frequency at which images are recorded would increase the probability of achieving cloud free scenes necessary for detection of abrupt changes in land cover. Recent advancements in remote sensing technologies have attempted to solve dilemmas surrounding the spatial and temporal limitations of older remote sensing platforms. Satellite constellations composed of multiple sensors can improve temporal coverage while maintaining finer spatial resolution. The RapidEye Constellation (RapidEye) is a network of five identical satellites working in unison to capture 5-meter multispectral resolution imagery with a 1-day return interval over most areas of Earth. RapidEye’s ability to combine attributes of high spatial and temporal resolution sensors make it unique amongst other Earth observation platforms. Automating basic change detection procedures using RapidEye could provide stakeholders with a framework to routinely and rapidly identify land cover change and disturbances across large areas with minimal user intervention. 8  1.2 Thesis Objectives This thesis consists of two main objectives. The first is to develop a robust routine for detecting stand-replacing disturbances in British Columbia using high-spatial and temporal resolution RapidEye remote sensing imagery. The second objective is to assess the capacity of RapidEye to map and evaluate less severe, smaller, non-stand replacing disturbances. To achieve these objectives four research questions have been proposed:  1. Can a modified disturbance index be derived from imagery of British Columbian forests provided by the RapidEye sensor?  2. How accurately can we identify stand-replacing disturbances from areas containing a high level of recent forest disturbances in British Columbia?  3. Can we use RapidEye to assess stand damage caused by a low severity fire in a dry Western Canadian forest? 4. How well can vegetation indices derived from RapidEye correlate with ground measures of burn severity – and how do these measures compare to indices derived from conventional Landsat imagery?   Chapter 2 derives a modified disturbance index using RapidEye imagery based on the methods of Healey et al. (2005). In this study a tasselled cap transformation (TCT) is derived from RapidEye imagery using forested regions across British Columbia. Components of the TCT are used to derive a disturbance index which is then assessed on its ability to delineate stand-replacing disturbance from seven 625 km2 RapidEye tiles. The accuracy of the modified disturbance index is tested using Provincial forest inventory data from manually verified ground locations.  9  Chapter 3 evaluates the ability of RapidEye to detect fine-scale forest damage associated with a non-stand-replacing fire in a dry Western Canadian forest. This study makes use of detailed ground data, which is used to assess stand damage on a tree-by-tree basis. Estimates of tree damage are correlated with conventional vegetation indices derived from RapidEye, Landsat TM, and Landsat OLI. Three commonly used aspects of imagery associated with the detection of disturbance are explored. Firstly, metrics of semi-variance are assessed and interpreted to provide an indication of the level of detail provided by the imagery from both sensors. Secondly, several common vegetation indices are correlated with tree damage on pixel-by-pixel basis. Finally, the consumption of woody surface fuel by fire is estimated using the magnitude of change detected by the sensors.   Finally Chapter 4 provides a synthesis of the overall research, a summary of the two studies, and provides guidance for future research.  The remainder of Chapter 1(section 1.3) describes the technical components of the RapidEye Constellation as well as change detection techniques explored in Chapters 2 and 3.   1.3 Remote Sensing of Disturbance 1.3.1 Sensor Development The RapidEye constellation was originally proposed in 1996 as a means of providing frequent multispectral imagery to augment a variety of Earth observing applications, including the monitoring of agricultural, forestry, disturbances, and natural disasters. The constellation was launched in 2008 and began providing commercial imagery in early 2009. The RapidEye mission has been designed to last at least seven years.  10  The RapidEye constellation is comprised of five sun-synchronous satellites orbiting the Earth at an altitude of 630 km. Each satellite carries an identical pushbroom scanner capturing multispectral imagery at a spatial resolution of 6.5 meters (at nadir). Imagery provided by RapidEye undergoes routine radiometric calibration to provide interchangeable imagery amongst the other sensors (Naughton et al. 2011). The RapidEye sensors are able to capture imagery covering approximately 5 million km2 of the Earth in a 24 hour period, as well as a daily return time (off-nadir) and a 5.5-day return time at nadir. Each sensor stores imagery onboard until the satellite passes the receiving station in Svalbard, Norway. RapidEye’s five band multispectral configuration, which includes a unique red-edge band, makes it an ideal platform for vegetation monitoring (Bindel 2011). In addition RapidEye’s high frequency return time allows for much more subtle changes in land cover to be detected, however, in order for these changes to be detected images must undergo substantial processing. Table 1.2 modified from Tyc et al. (2005) provides a summary of RapidEye’s spectral band configuration.  Table 1.2 RapidEye spectral characteristics (modified from Tyc et al. 2005) Band Description  Spectral range (nm) Specified center wavelength (nm) 1 Blue 440–510 475.0 2 Green 520–590 555.0 3 Red 630–685 657.5 4 Red-edge 690–730 710.0 5 NIR 760–850 805.0  Table 1.3 provides a summary of existing studies where RapidEye has been implemented as a successful remote sensing platform. 11  Table 1.3 Examples of RapidEye applications Application  Findings  Benefits  Author  Satellite image calibration  RapidEye sensor measures radiance within 4% of predicted values RapidEye sensor demonstrated to provide precise radiometric resolution  Naughton et al. (2011) Agricultural Crop Monitoring  Mapping agricultural land cover Potential for improved crop monitoring  Tapsall et al. (2010) Ecosystem Classification  Combined results of red-edge and NIR bands may be used for increase accuracy for separation vegetation classes Improved mapping for conservation and ecological classification  Bindel and Hese (2011) Grassland Monitoring  Differences among grassland habitats Additional data for spatial habitat modeling  Franke et al. (2012)  RapidEye imagery is available through BlackBridge, which owns and operates the RapidEye constellation. Commercial imagery is readily available through BlackBridge’s web-based ‘EyeFind’ tool where specific orders can be queried by entering known coordinates, selecting a specific tile of interest, or by uploading a KML, KMZ or SHP file. Processed imagery is available in either geometrically uncorrected, or as an orthorectified tile product. 1.3.2 Multispectral Vegetation Indices  Simplification of multiband remote sensed data into interpretable indices allows powerful conclusions to be made regarding vegetation condition (Jackson and Huete 1991). By performing various arithmetic operations on pixel reflectance values we can derive indices that may be more useful for assessing and quantifying Earth surface phenomena than simply using the original bands. The Normalized Difference Vegetation Index (NDVI) is one such index commonly derived from remote sensed data. The NDVI takes advantage of a large increase in reflectance 12  between the visible red and NIR wavelengths as exhibited by healthy vegetation. Other indices, such as the enhanced vegetation index (EVI), the soil adjusted vegetation index (SAVI), the normalized burn ratio (NBR) also take advantage of similar differences in wavelengths reflected by vegetation. Table 1.4 provides a summary of common derived vegetation indices.  Table 1.4 describes commonly used vegetation indices Indices Used   Formula  Reference  Enhanced Vegetation Index (EVI) 2.5 * ((NIR – Red)/(NIR+6 * Red -7.5 * Blue + 1)) Gao et al. (2000) Normalized Difference Vegetation index (NDVI) (NIR – Red) / (NIR + Red) Tucker (1979)  Soil Adjusted Vegetation Index (SAVI) [(NIR – Red) / (NIR + Red + 0.5)] *(1 + 0.5) Huete (1988) Normalized Burn Ratio (NBR) (NIR – SWIR) / (NIR + SWIR) Key and Benson (1999) Normalized Difference Moisture Index (NDMI) (NIR – MIR) / (NIR + MIR) Wilson and Sader (2002) Tasseled Cap Transform 1 (TC1) Linear weighting of all bands emphasizing overall pixel brightness (depends on sensor) Originally described by Kauth and Thomas (1976)  Tasseled Cap Transform 2 (TC2) Linear weighting of all bands emphasizing differences between NIR and RED wavelengths  (depends on sensor) Originally described by Kauth and Thomas (1976) Tasseled Cap Transform 3 (TC3) Linear weighting of all bands (depends on sensor) Originally described by Kauth and Thomas (1976)  Another set of indices commonly used in image processing are the indices derived from the Tasseled Cap Transformation (TCT) (Kauth and Thomas 1976, Crist and Kauth 1986, Horne 2003). The TCT calculates linear uncorrelated combinations of bands, reducing redundancy in the data; typically undertaken using principal components analysis. Originally developed for the 13  four Landsat MSS bands (Kauth and Thomas 1976), TCTs have since been produced for many other sensors such as Landsat Thematic Mapper (TM), ASTER, Ikonos, and Quickbird (Yarbrough et al. 2005, Horne 2003). The original TCT developed for Landsat MSS described two main indices: brightness and greenness, as well as a third “yellowness”; however the addition of shortwave infrared (SWIR) band on Landsat TM (as well as additional spectral bands) allowed a third ‘wetness’ index to be interpreted (Healey et al. 2005). As part of the processing of the RapidEye data, a sub-objective of this research will be to derive a TCT for RapidEye. The TCT has been shown to be more successful at detecting change between image dates than the NDVI as it is derived from all spectral bands and allows a comparison of both changes in the greenness and the overall brightness spectral responses (Healey et al. 2005). 1.3.3 Change Detection Techniques  Assessment of ground cover change from remotely sensed imagery typically relies on identification of changed areas from images collected on two or more dates (Coppin et al. 2004). Techniques for identifying change within digital imagery have been developed within the disciplines of remote sensing, photogrammetry, computer vision, and image processing. Singh (1989) defines change detection as the identification of difference in an objects state by observing it at different times. Lu et al. (2004) and Singh (1989) provide comprehensive reviews of change detection using remote sensing beyond the scope of this thesis. Prior to implementation of successful change detection various stages of preprocessing must first be applied to satellite data.  (1) Geometric fidelity is a major consideration of change detection; this implies that images pixels within multitemporal will be co-registered to the same geographic location. (2) DN numbers of similar feature must be represented by similar values across images -this done through radiometric normalization of the images during preprocessing stages.  14  Once the selected imagery has been radiometrically and geometrically corrected a number of change detection techniques can be applied to extract and map regions of change. Two basic categories of change detection exist in the literature; these are (1) classification based and (2) radiometric based approaches (Johnson and Kasischke 1998). Classification based approaches rely on delineating land cover classes then spatially assessing how these classes have changed over time. The second radiometric based approach assesses the magnitude of radiometric change between corresponding image pixels. Perhaps the most basic method of radiometric change detection involves image differencing; a process where pixel values from one image are subtracted from that of another, as follows:                     Where T1 is the first image and T2 represents the second image in a stack; the result is known as a difference image. In an n-band image a simple image differencing will result in n-bands representing vectors of change. Change can be attributed to the difference image by setting thresholds for pixels constituting change, or by using a probability distribution (Singh 1989).  Indices derived from linear combinations of band values may also be used in more robust methods of change detection (Lyon et al. 1998). The utilization of vegetation indices, or transformed band differencing, relies on the derivation of vegetation indices, which are then differenced to produce a change map (e.g. Nelson 1983).  The disturbance index (DI) proposed by Healey et al. (2005) is a method of transformed band differencing that was developed to be specifically sensitive to loss of forest vegetation. This approach takes advantage of linear combinations of Landsat TM or ETM+ bands resulting in a single band image. The single band image for one date can be subtracted from another to 15  produce a bi-temporal difference image, depicting disturbances. Healy et al.’s (2005) DI is summarized as follows:              [    (       )]  [    (       )] where B, G, and W, represent Landsat Brightness, Greenness, and Wetness values. A threshold for extracting areas of change can then be applied to the single band DI. Healey et al. (2005) suggest using an arbitrary value of ‘2’ to discriminate between disturbed and no change pixels.   More complex statistical procedures such as multivariate alteration detection (MAD) (Canty and Nielsen 2008) rely on information derived from the covariance matrices of from both images to identify change. The MAD procedure used by Canty and Neilsen (2008) may also be used as a means of radiometric normalization for two or more images.   1.4 Significance of Research Remote sensing provides a cost-effective, viable solution to begin to assess a multitude of metrics associated with forest disturbance and mapping. In order to better understand the dynamic state of the world’s forests, improve inventory assessments, and provide better information to managers and policy makers it is necessary to provide up-to-date and detailed products derived from satellite imagery. RapidEye and other constellations can potentially provide a solution to regional and local scale disturbance monitoring needs due to the sensors fine spatial and temporal resolutions and their ability to recapture imagery at frequent intervals.  By providing tools and methods to support the need of annual disturbance detection this project will contribute to large-scale mapping frameworks to assess changes in forested ecosystems. The 16  results of this study will be especially pertinent to regions that may lack sufficient resources and infrastructure to monitor forest disturbances.    17  2 DETECTING STAND REPLACING DISTURBANCE USING RAPIDEYE IMAGERY: A TASSELED CAP TRANSFORMATION AND MODIFIED DISTURBANCE INDEX  2.1 1Introduction Understanding changes in vegetated land cover is a central objective in the sustainable management of forests and is vital in developing effective forest policies (Pitt and Pineau 2009). Currently, much of the information used to assess land cover change is generated through optical remote sensing platforms or through the interpretation of aerial photography (Rogan and Miller 2006).  Monitoring land cover change using remote sensing consists of three distinct categories. These are summarized by Verbesselt et al. (2012) as (1) seasonal changes associated with vegetation phenology and climate, (2) gradual changes or trends in ecosystems, and (3) abrupt or immediate disturbances.  Abrupt disturbance events such as those caused by fire, harvest, and flooding, are key drivers of forest structure, function, and composition (Franklin et al. 2002). These events constitute perturbations to the normal state of an ecosystem relative to the spatial and temporal scale at which that particular ecosystem is studied (Picket and White 1985). Disturbances occur globally at a range of spatial scales and may have cumulative effects on species diversity and distribution (Defries et al. 1999; Gutschick and Bassirirad, 2003; Loboda et al. 2012) as well as geophysical and landscape processes (e.g., Sidle, 1992; Samonil et al. 2010; Pohl et al. 2012). Disturbance events also affect climate through structural and biochemical changes of forest vegetation. These                                                  1 A version of this chapter has been published. Arnett, J.T.T.R., Coops, N.C., Gergel, S.E., Falls, R.W., Baker, R.H. 2014. Detecting Stand Replacing Disturbance using RapidEye Imagery: a Tasseled Cap Transformation and Modified Disturbance Index. Canadian Journal of Remote Sensing 40: 1-14.  18  can include changes in land use and cover, surface albedo (e.g., Davin and de Noblet-Ducoudré, 2010), evapotranspiration (Maness et al. 2013), and carbon flux (Kurz and Apps 1999). Although abrupt disturbance events occur under natural conditions, the rate at which they occur tends to be intensified by human activity (Pearson 2010; Vistosek et al. 1997). Additionally many disturbances are often small and scattered; requiring detailed spatial information for their identification (Townshend and Justice 1988).  Consistent and timely identification of forest disturbances offers many benefits to managers and stakeholders concerned with monitoring various forest attributes. However, feasible and timely monitoring of forest disturbance over large areas is only achievable through the use of remote sensing platforms (e.g., Kerr and Ostrovsky 2003). The ability to automate or semi-automate detection of abrupt disturbances across the landscape can provide vital information to compliment forest inventories at a range of administrative levels.   A variety of methods exist for the detection of change within satellite imagery. Coppin et al. (2004) provides an overview of nine distinct methods of change detection using data derived from remotely sensed imagery. One of the simplest and most widely used methods is univariate image differencing, a process whereby pixel values from one image date are subtracted from that of another coregistered image from another image date. The resulting image depicts changes that have occurred between image acquisition times. A simple threshold can be applied to the univariate difference image whereby values above the threshold can be attributed to change. Univariate image differencing is commonly applied to transformed data such as vegetation indices (e.g., Lyon et al. 1998; Lu et al. 2004).   19  A commonly used set of vegetation indices is the Tasseled Cap Transformation (TCT). The TCT derives new linear uncorrelated combinations of bands, typically using principle components analysis (PCA), which reduces redundancy in the dataset (Yarbrough et al. 2005). Originally derived using the four Landsat Multispectral Scanner (MSS) bands (Kauth and Thomas 1976), TCT transformations have been calculated for ASTER, IKONOS, and Quickbird data (Yarbrough et al. 2005; Horne 2003). The original TCT described the three indices, ‘brightness’, ‘greenness’ and ‘yellowness’ (Kauth and Thomas 1976); however the addition of shortwave infrared (SWIR) band on Landsat TM allowed for the third index to be interpreted as ‘wetness’ (Healey et al. 2005). While the use of remote sensing data to monitor disturbance has been successful in identifying and mapping disturbances at a range of scales (Coppin and Bauer 1996), consistent frameworks aimed at ongoing monitoring of forest disturbance are limited (Mayaux et al. 2005; Xin et al. 2013). Monitoring forests using remote sensing often results in a compromise between high spatial and temporal resolution imagery, with high temporal resolution imagery often being limited to coarse spatial resolutions and fine spatial resolution imagery limited by swath width and sensor return time (Chambers et al. 2007). The Landsat series is arguably the most frequently used remote sensing platform (Wulder et al. 2012), however  with 30 m spatial resolution and a 16 day return time, Landsat may miss important change events, especially in areas where cloud obscuration is common (Asner 2001; Sano et al. 2007). Increasing the return time of a satellite sensor inherently increases the probability of acquiring cloud-free imagery; however, many high-temporal resolution sensors such as the Advanced Very High Resolution Radiometer (AVHRR), and the Moderate Imaging Spectrometer (MODIS), are limited to spatial resolutions of 1 x 1 km and 250 X 250 m respectively and may not be suitable for monitoring and 20  mapping local and regional forest disturbances. Frequently updated mapping of disturbance is fundamental to any forest inventory; however, achieving an appropriate balance between the spatial and temporal resolutions of a sensor remains a challenge. Satellite constellations can potentially provide a solution to the spatial/temporal compromise associated with other sensors. For example, a collaborative effort between the United Kingdom, Algeria, Nigeria, Thailand, and Turkey has led to the creation of the Disaster Monitoring Constellation (DMC). The DMC constellation is a series of satellites with sensors acquiring data comparable to imagery produced from the Landsat red, green, and near-infrared (NIR) bands, however the constellation design and wider swath width allows for daily coverage (da Silva-Curiel et al. 2005). The success of the DMC constellation has led to a second generation of DMC satellites launched between 2005 and 2011. The commercially operational RapidEye is another constellation platform that makes use of five identical sensors to provide 5 X 5 m spatial resolution imagery with a daily time of 5.5 days at nadir (daily if considering off-nadir imagery). The RapidEye sensors consist of five bands situated in the visible and NIR portions of the electromagnetic spectrum with a range of 440-850 nm. Cross calibration studies have shown individual RapidEye sensors to be interchangeable, with recorded radiometric differences within 2% amongst sensors (Thiele et al. 2012). RapidEye’s spectral band configuration is ideal for vegetation and ecological monitoring (Bindel et al. 2011), while the temporal coverage may be useful for natural disaster monitoring (Metternicht et al. 2005). In this paper we investigate the detection of stand-level disturbances in annual and bi-annual forest cover in British Columbia, Canada. British Columbia (BC) is a key forested area in Canada and encompasses many climatic gradients and forest types. We use provincial forest inventory data to identify calibration and validation sites to assess disturbance throughout the 21  province. We then derive a set of indices using RapidEye 5 band data to develop a ‘disturbance index’, using the TCT following methods originally proposed by Healey et al. (2005) and assess its accuracy using conventional forest inventory polygons and visual interpretation approaches. 2.2 Methods For our purpose, BC is an ideal region to study forest change. In BC a gradient of forest types exist, ranging from wet temperate rainforests along coastal regions to dry mixed grassland forests in the southern interior. In addition, detailed forest inventory data is routinely collected by the BC Ministry of Forests, Lands and Natural Resources Operations (MoFLNRO). Forest inventory layers catalogued by the MoFLNRO provided valuable information on the location and date of the disturbances being studied. The RapidEye constellation consists of five identical sensors employed to deliver a resampled 5 X 5 m spatial resolution product. The radiometric resolution of RapidEye imagery is provided in a 16-bit range representing scaled radiance values. RapidEye bands are located in the blue, green, red, red edge, and NIR regions of the electromagnetic spectrum, with respective wavelength ranges at 440-510 nm, 520 – 590 nm, 630 – 685 nm, 690 – 730 nm, and 760 – 850 nm.. We used orthorectified (level 3a) 25 X 25 km tile products in this study.  Four datasets were utilised in this study: The MoFLNRO vegetation resources inventory (VRI) disturbance polygons which were used to select the bitemporal image pairs ensuring they covered areas of recent disturbance (2) seven overlaying bitemporal Rapideye image pairs which contained several disturbance events occurring between acquisition times. (3) A set of 50 RapidEye images from across the Province were used to derive a set of tasseled cap coefficients, and (4) Earth Observation for the Sustainable Development of Forests Land Cover Classification 22  (EOSD LCC) was employed to mask non-vegetated areas from the 50 RapidEye images used to derive the tasseled cap coefficients. Each of these datasets is explained in more detail below.  2.2.1 Dataset 1: MoFLNRO VRI Disturbance Polygons In BC, the VRI programme maps and catalogues geospatial information pertaining to BC’s forests. The VRI serves as a continuously updated archive and is implemented using photointerpretation, ground sampling, and satellite imagery. Aerial photos used in the VRI data are typically collected at a 1:10,000 scale prior to ground sampling. Updates of disturbance polygons (fire, harvest, and other events) are submitted to the MoFLNRO by licensees or through a variety of data collection sources (MoFLNRO, 2013) and can include both actual and planned harvests. For our analysis only contemporary VRI data was used to ensure the VRI disturbance polygons coincided with the RapidEye imagery dates. These polygons were indicated to have been gathered principally from aerial photographic interpretation, and ground sampling as well as Landsat satellite imagery and other GIS data provided by forest tenure holders. The VRI data also served as reference data to assess the accuracy of our estimated threshold between disturbed and unchanged forest.   2.2.2 Dataset 2: RapidEye Bitemporal Image Pairs We located RapidEye tiles containing recent disturbances (i.e., since 2010) based on data from the MoFLNRO VRI polygons. The first image dataset consisted of 7 image pairs shown in Table 2.1.  Of the 7 pairs selected, 4 contained only harvest activity, 2 contained only fires, and 1 contained both recent fires and harvest activity. Polygon files from the VRI disturbance database were used to delineate harvest and fire data across BC. In order to assess the capacity of RapidEye to detect these disturbance types, a stratification of disturbance events was undertaken by type. Image pairs were selected to occur within one month of their anniversary date to ensure 23  similar seasonal conditions. A total of 4484 ha of fire and 6114 ha of harvest was identified across the seven scenes.  2.2.3 Dataset 3: RapidEye TCT Images  A second set of 50 level 3a RapidEye images were acquired from five forested regions and used to derive a TCT. Images were selected to have <10% cloud cover (as indicated by the RapidEye cloud detection algorithm) and to have been recorded between June 21 and September 21, 2012. RapidEye digital numbers (DNs) were converted to Top of Atmosphere (TOA) reflectance as specified by the RapidEye product guide (RapidEye 2013). Table 2.1 A summary of MoFLNRO harvest indicated in RapidEye 25 X 25 km tiles, “-“ indicates measure was not calculated, *note tile 1063414 contains both fire and harvest Tile ID T1  T2 Indicated disturbance area (ha) Total polygons (MoFLNRO) Average size (ha) Standard dev. harvest 962221 July 7, 2010 July 18, 2012 983 55 17.9 17.0 964123 July 23, 2010 July 24, 2012 1665 46 36.2 62.8 1062622 Sept. 9, 2011 Sept. 8, 2012 756 9 84.0 75.7 1063413 Aug. 11, 2011 Aug.16, 2012 718 16 44.9 104.5 *1063414 July 24, 2010 Aug. 18, 2012 3058 53 57.7 72.4 fire 963622 July 13, 2010 July 23, 2012 572 1 - - 1063908 July 17, 2010 Aug. 8, 2012 17966 1 - - *1063414 July 24, 2010 Aug. 18, 2012 3298 1 - - 2.2.4 Dataset 4: EOSD LCC Derived Mask The Earth Observation for the Sustainable Development of Forest Land Cover Classification (EOSD LCC) is a Landsat-derived classification scheme used to map Canada’s ecological 24  regions (Wulder and Nelson 2003). Within EOSD data, six nested hierarchical levels exist to classify vegetated regions while five levels are used to classify non-vegetated regions (Wood et al. 2002).  The EOSD LCC was used to mask areas of snow and ice, barren rock, and surface water prior to calculating a TCT from the 50 cloud free RapidEye images. Figure 2.1 describes the data processing flow used in this analysis. 25   Figure 2.1 Processing flow for change detection using RapidEye imagery 2.2.5 Tasseled Cap Transformation  As TCT coefficients for RapidEye data were not readily available, we implement a PCA approach to derive an appropriate transformation. The PCA approach seeks uncorrelated linear 26  combinations of n bands by creating n uncorrelated principal components (e.g., Wold et al. 1987). The derived principal components are listed in decreasing order of explained variance. In general, correlations between band reflectance levels are accounted for in the same component, while uncorrelated data is distributed amongst other components (Crist and Cicone 1984). A set of eigenvalues are derived from the resulting covariance matrix, with each eigenvalue corresponding to a set of eigenvectors (TCT coefficients). By multiplying the derived eigenvalues by their respective band reflectance values results is a set of derived components (Gil Docampo et al. 2004). These components form the basis of the tasseled cap coefficients.  The original TCT was conducted using a Gram-Schmidt Orthogonalization procedure, explained by Kauth and Thomas (1976). Generally, the first tasseled cap band is an agglomerative representation of overall image brightness, while the second band tends to emphasize overall ‘greenness’ of vegetation. Although RapidEye does not contain a SWIR band, a unique ‘red-edge’ band may be helpful in creating a third interpretable TCT index. Previous studies have demonstrated the red-edge region of the spectrum to be sensitive to chlorophyll concentration (Filella and Peñuelas 1994) and classification of vegetation type (Schuster et al., 2012). We used ENVI version 4.8 (Exelis Visual Information Solutions) to conduct PCA on covariance matrices to derive initial components. To achieve positive values for all pixels, a constant offset vector ‘r’ was then estimated to ensure TCT values occurred only in positive tasseled cap space. The constant offset vector was calculated from the minimum values across all 50 images. 2.2.6 Disturbance Index The disturbance index (DI) proposed by Healey et al. (2005) takes advantage of the TCT for Landsat TM, whereby standardized greenness (Gr) and wetness (Wr) values are subtracted from standardized scene brightness (Br). The resulting index enhances differences between regions 27  that exhibit overall high values in all TCT indices from regions that contain only high brightness values, summarized as follows: Br = (B - Bµ)/Bơ Gr = (G - Gµ)/Gơ Wr = (W - Wµ)/Wơ DI = Br – (Gr + Wr) The derivation of DI bands allows for single band comparisons to be made between multitemporal imagery. To ensure transferability across images and dates, our analysis differs slightly from the original proposed by Healey et al. (2005). We standardized our images using z-scores calculated from all 50 images, whereby µ and ơ denote the mean and standard deviation calculated from all images. For our analysis, we subtracted T1 from T2 to create univariate difference images depicting probable change between image acquisitions. We anticipated that within each univariate difference image, negative values would indicate increased vegetation, positive values would indicate vegetation loss, and near-zero values would indicate no change.  2.2.7 Threshold Development A simple threshold value was required to discriminate between disturbance and no-change cover types. As the VRI harvest polygons contain both actual and planned harvest schedule events, there were a number of polygons delineated for future harvest which had not yet been undertaken between the image dates. To detect and remove these events prior to any sampling, we used an interim threshold value of one standard deviation above the difference image’s mean pixel value. If the harvest polygon’s mean pixel value fell within one standard deviation of the difference image mean (indicative of no change) the polygon was designated as incomplete or erroneous 28  and was removed from the sampling process. The remaining polygons are summarized in Table 2.2. Following the methods of Coops et al. (2006), we randomly extracted 1/3 of the remaining harvest polygons for threshold development, leaving 2/3 for accuracy assessment purposes.   Table 2.2 A summary of polygons representing confirmed harvest indicated to have occurred between RapidEye image aqusition dates tile ID confirmed harvest polygons confirmed harvest area during interval (ha)  average confirmed polygon size (ha) standard deviation (ha) randomly selected polygons  randomly selected polygon area (ha) 962221 41 737 17.98 16.30 14 272 964123 41 1105 26.95 36.16 14 295 1062622 9 701 77.89 72.51 3 255 1063413 11 365 33.18 40.75 4 202 1063414 40 1576 39.40 39.76 13 526  To derive robust threshold value of disturbance across all images and disturbance types, we randomly selected pixels within, and outside, the confirmed MoFLNRO disturbance polygons. Pixels were sampled at random at an average density of 10 pixels/ha within each disturbance event to a maximum of 10,000 pixels/polygon. An equal number of pixels were then selected at random from a 500 m buffer around the disturbance to represent no-change values (following Lee et al. 2013). Due to the limited number of fire events, fire polygons were divided into 1 ha cells, of which 1/3 were randomly selected and sampled at an average point density of 10 pixels/ha. An equal number of unburned pixels from a 500 m buffer around the entire burn area were sampled to represent no-change. A summary of burn scar areas is provided in Table 2.3.  29  Table 2.3 A Summary of burn scar and stratified sampling areas indicated to have occurred between image acquisition dates Tile ID total fire area   stratified fire samples (1 ha) 963622 572 162 1063908 17966 4947 1063414 3298 1009 A universal threshold value was then calculated as the difference between the mean disturbance and no-change pixel values from the seven difference images. The threshold was applied to the univariate difference images to create a set of seven binary threshold images (BTIs). Similar to Hill and Foody (1994), we then applied a 3 x 3 smoothing low pass filter to remove issues associated with single pixel misregistration.   2.2.8 Verification  Two approaches were used to estimate the accuracy of the disturbance detection method. The first approach utilised the VRI as reference data while the second approach relied on a random stratified sample of manually interpreted points. At each of the seven scenes a reference dataset was created using the VRI polygons. Each map comprised four classes depicting (1) disturbance agreement for VRI and BTIs, (2) no-change agreement between VRI data and BTIs, (3) change detected by BTI and not VRI, and (4) change indicated by VRI and not BTI. The EOSD dataset and RapidEye unusable data mask (UDMs) were also used to mask areas of cloud and non-vegetated cover, resulting in a total mapped area of 338,645 ha across the 7 tiles. Measures of pooled and individual accuracy were assessed for each map. The second verification process used a random stratified sample to select 100 random points from the disturbance class and 400 points from the no-change class. Each point was manually interpreted as “disturbed” or “no-change” by visually comparing the original T1 and T2 RapidEye images.    30  Olofsson et al. (2013a) described in detail the calculation of critical metrics which can be applied when assessing the accuracy of change detection maps; these were used as a basis for both accuracy assessments. The proportion of each class was estimated using an unbiased estimation matrix (Olofsson et al. 2013a). Confidence intervals depicting 95% confidence were also estimated using Olofsson et al. (2013b). Estimates of class accuracy using the VRI reference data relied on a random sample of 10,000 pixels/tile, while manually interpreted approach relied on a smaller sample of 500 pixels. 2.3 Results The results of the PCA transformation on the 50 near-cloud free TOA RapidEye images are shown in Table 2.4 indicating the individual band loadings by component.  Table 2.4 Eigenvalue loadings derived from Principal Components Analysis of 50 images from British Columbia component band blue  green  red  red edge  NIR Variance Explained (%)  1 -0.293 -0.354 -0.372 -0.440 -0.676 75.66 2 -0.406 -0.367 -0.446 -0.128 0.697 23.01 3 0.572 0.402 -0.494 -0.498 0.138 1.04 4 0.131 -0.309 0.632 -0.671 0.195 0.20 5 0.636 -0.695 -0.141 0.302 -0.030 0.09  As anticipated, the first three PCs explained over 99% of the variance within the image data. Figure 2.2 provides a visual comparison of these first three components. As expected, the first component (TCT1) represents overall scene brightness with recently harvested areas expressed as lighter grey and mature forest stands expressed as darker grey. The second component (TCT2) appears to be sensitive to vegetation cover, with darker values representing recently cut 31  vegetation and lighter values representing recovering forest and young vegetation. Despite the third component (TCT3) accounting for less than 1% of total image variability, areas of recent harvest are clearly distinguished by dark pixels and as a result this component remained in the analysis. To achieve positive space for all pixel values and complete the transformation, TCT1 values were multiplied by -1 and constant offset vectors were estimated to be 0, 0.93, and 0.74 for TCT1, TCT2, and TCT3, respectively.  Figure 2.2 Visual comparisons of three derived tasseled cap indices for an area containing various stages of forest succession, approximately 55 km southeast of Vanderhoof, BC, RE Tile ID 1063813. From left to right: TCT1, TCT2, TCT3  Two-dimensional scatter plots depicting the first 3 TCTs are shown in Figure 2.3. Several common land cover types are outlined depicting their relative location within the tasseled cap space.  32   Figure 2.3 2D scatter plots depicting density slices of RapidEye tile 1063813, an area approximately 55 km southeast of Vanderhoof, BC  Figure 2.4 provides visual comparison of TOA reflectance values for blue, green, and red bands compared to TCT1, TCT2, and TCT3. The TCT bands show good differentiation of between various stages of forest recovery when compared to the RGB image.   Figure 2.4 a comparison of TOA RapidEye (RGB) image and Tasselled Cap image, approximately 55 km southeast of Vanderhoof, BC. Recently cleared areas appear red, recovering regions are orange, stable areas are in blue. Image size is approximately 4 X 4 km.  TCT1, T TCT2, TCT3 are displayed using red, green, and blue colour guns respectively  33  Across all 7 images and both disturbance types, the mean disturbed pixel value was µ = 1.62 (ơ = 1.62) while the unchanged pixel value was µ = -0.03 (ơ = 0.79). A t-test between disturbed and undisturbed classes revealed that differences were statistically significant (p<0.05). For individual disturbances, the pixel values differed with the mean harvest pixels (µ = 2.99) differing significantly from fire (µ = 0.64) (p<0.05) as shown in Figures 2.5 and 2.6. A universal threshold value across all scenes was calculated as 0.79, above which all pixels were classified as changed.  Figure 2.5 A comparison of random stratified disturbance index values for burned and un-burned areas (sampled at a density of 10pts/ha). Unburned areas were sampled at random from a 500 m buffer around the indicated disturbance (‘Fire’ = from within burned area, ‘NC’ = from within no-change area). 34   Figure 2.6 A comparison of random stratified disturbance index values for harvested and non-harvested areas (sampled at a density of 10pts/ha). Non-harvested areas were sampled at random from a 500 m buffer around the indicated disturbance (‘Cut’ = from within har harvested area, ‘NC’ = from within no-change area).   Figure 7 shows the application of the threshold for recently disturbed features using this 0.79 threshold. Areas outlined in red show MoFLNRO harvest polygons.  As indicated by Figure 2.7, several MoFLNRO polygons do not align with the extracted disturbance layer.  35  Figure 2.7 Harvest activities near Burns Lake, BC. Areas indicted as harvest by MoFLNRO data are outlined in red. Left to right: disturbance index for image 1; disturbance index for image 2; difference image; and extracted disturbance using threshold >0.79  Table 2.5 reports the accuracy of the disturbance classification using the methods of Olofsson et al. (2013a) using the VRI polygons as reference data.  The overall accuracy across the landscape is high, ranging between 89% and 96%. Overall, harvest was more successfully detected than fire. The large number of unchanged pixels, however, heavily weights the overall accuracy of the delineation. This weighting of non-disturbed areas results in very high user’s and producer’s accuracy for un-disturbed areas and lower accuracies for the disturbed areas; as a result it is best to compare individual user’s and producer’s accuracies.  Classification accuracies using the VRI polygons as reference data resulted in varying accuracies amongst tiles (Table 2.5). For example tile 1062622, near Kamloops, BC reported respective user’s and producer’s accuracies of 8.71 ± 1.63 and 65.83 ± 13.94. Tile 964123, near Burns Lake, BC demonstrated much better alignment with the VRI polygons and overall higher classification accuracies. Tile 964123 reported user’s and producer’s accuracies of 51.83 ± 5.65 and 56.58 ± 5.37, respectively. Classification results from tile 964123 aligned well with the VRI polygons, with 34 of 46 harvest polygons overlaying consistently. Other individual tile classification results, such as tile 962221 on northern Vancouver Island, did not perform as well and contained substantial mismatches between the VRI disturbance layers and the detected disturbances.    36  Table 2.5 Disturbance accuracies for individual tiles used in disturbance classification, D = Disturbance, N = No Change classes. Columns D and N represent sample proportions of map class (n = 10,000 samples/tile) Tile ID (map)   D (%) N (%) User's Acc. (%) Producer's Acc. (%) Overall Acc. (%) 962221 DD 0.66 2.02 24.56 ± 3.97 58.88 ± 8.07 97.52 ± 0.17   NC 0.46 96.87 99.53 ± 0.14 97.96 ± 2.24   963622 DD 0.58 2.09 21.86 ± 3.77 57.09 ± 8.44 97.47 ± 0.17   NC 0.44 96.89 99.55 ± 0.13 97.89 ± 2.14   964123 DD 1.39 1.29 51.83 ± 5.65 56.58 ± 5.37 97.65 ± 0.25   NC 1.06 96.26 98.91 ± 0.21 98.68 ± 2.65   1062622 DD 0.23 2.44 8.71 ± 1.63 65.83 ± 13.94 97.44 ± 0.08   NC 0.12 97.20 99.88 ± 0.07 97.55 ± 1.45   1063413 DD 0.25 2.42 9.40 ± 1.98 33.03 ± 7.78 97.07 ± 0.15   NC 0.51 96.82 99.48 ± 0.15 97.56 ± 1.5   1063414 DD 1.94 0.73 72.57 ± 3.33 28.36 ± 2.02 94.36 ± 0.44   NC 4.90 92.42 94.96 ± 0.44 99.21 ± 2.49   1063908 DD 2.48 0.19 92.85 ± 1.28 13.32 ± 0.57 83.65 ± 0.77  NC 16.16 81.17 83.40 ± 0.79 99.77 ± 1.66  Total Harvest DD 0.91 2.93 23.75 ± 1.70 57.24 ± 3.44 96.38 ± 0.10  NC 0.68 95.47 99.28 ± 0.08 97.02 ± 3.16  Total Fire  DD 11.54 3.53 76.58 ± 1.80 62.83 ± 1.28  89.64 ± 0.43  NC 6.83 78.11 91.96 ± 0.39 95.67 ± 14.65  Grand Total  DD 3.19 4.43 41.87 ± 1.31 52.22 ± 1.32  92.64 ± 0.15  NC 2.92 89.46 96.83 ± 0.13 95.27 ± 0.10   The manually verified classification reported much higher measures of user’s and producer’s accuracies (Table 2.6). User’s and producer’s accuracies were 75 ± 9% and 87.75 ± 10.56% for 37  the disturbance class and 99 ± 0.98% and 96.64 ± 7.81% for the no-change class, respectively. The Overall classification accuracy was 96.91 ± 1.16%. During the manual classification process it was noted that 11 of the 25 pixels that were misclassified as disturbance occurred within regions of unmasked cloud or snow cover. The misclassification snow and cloud suggested that classification accuracy could be increased by implementing a more effective cloud and snow mask. Removing the 11 misclassified cloud and snow pixels resulted in disturbance classification accuracies of 84.27 ± 7.61%, 88.95 ± 9.64%, 97.72 ± 1.11% for user’s, producer’s and overall accuracies, respectively.  Table 2.6 Manually verified accuracy for overall disturbance classification, D = Disturbance, N = No Change classes. Columns D and N represent sample proportions of map class (n = 100 samples/tile) Tile ID (map)   DD (%) NC (%) User's Acc. (%) Producer's Acc. (%) Overall Acc. (%) Total Area  D 0.07 0.02 75 ± 8.53 87.75 ± 10.56 96.91 ± 1.16  N 0.01 0.90 99 ± 0.98 97.64 ± 7.81  2.4 Discussion  RapidEye and similar ongoing and proposed constellations provide opportunities for combining fine spatial resolution with frequent monitoring. Continuous coverage provided by constellation platforms offers improved data availability over single sensor platforms. Furthermore, we found that cloud free imagery was readily available, even for coastal BC, where cloud cover is extremely high. Our simple disturbance index, derived from selective PCA on RapidEye imagery proved useful in detecting disturbance. Further research should be carried out to assess the relationship between the RapidEye TCT values and the spectral characteristics of real surface features allowing for more detailed interpretation of indices.  38  Although conventional principal components literature suggests that only the first two components from our analysis should be retained (Jackson 1993), a visual assessment of the third derived TCT band (Figure 2.2) clearly distinguishes between recently cleared areas and standing vegetation, providing further support for the inclusion of the third TCT component. Horne (2003) chose to interpret all 4 components derived from 4 IKONOS bands, despite components 3 and 4 explaining only 1.54% and 0.17% of the image variance, respectively. Similar to Horne’s (2003) transformation for IKONOS, our third component can be generalized as the blue and green bands minus red and red-edge bands, and appears useful in the delineation of disturbed areas.  Overall, the locations of data structures within TCT space are consistent with descriptions of TCT from other studies (e.g., Crist and Cicone 2004; Horne, 2003; Yarbrough et al., 2005). Although RapidEye contains a unique red-edge band (690-730 nm), the relative weighting of the red-edge did not provide a distinctive response for any components. A visual assessment of 2D scatter plots (Figure 2.3) shows many similarities to Crist and Cicone’s (1984) original description of Landsat TM TCT. 2.4.1 Accuracy Interpretation It is crucial to use appropriate accuracy estimation metrics for any land cover classification exercise. Best practices for assessing the accuracy of the change should calculate confidence intervals surrounding each respective class (Strahler et al. 2006). Furthermore, commonly used statistical measures, such as Cohen’s Kappa statistic do not provide adequate description of classification accuracy. Cohen’s Kappa is simply a probabilistic measure of random agreement between two maps and does not account for estimates of uncertainty (Stehman 1997). The unbiased estimator approach described by Olofsson et al. (2013a) meets “best practice” 39  requirements for the assessment and classification of land cover classes and was therefore employed in this study.  The threshold approach proved useful in the identification of structural forest disturbances. Unbiased estimates of producer’s accuracy revealed that cumulative producer’s accuracy was greater than cumulative user’s accuracy for all disturbed areas (42 ± 1.3% and 52 ± 1.3%, respectively) using the VRI as reference data . This result implies that approximately half of the disturbance areas provided in VRI data did not align with areas classified using the threshold method, likely due to a variety of sources including error within VRI (Olofsson et al. 2013a).  Large discrepancies between detected disturbance and VRI reference data resulted in a low estimate of cumulative user’s accuracy. The overall 23.8 ± 1.8% estimate of user’s accuracy for harvest can be interpreted as “23.8 ± 1.8% of the area identified as harvest matched the VRI reference data”. The low user’s accuracy estimate should not discredit the utility of the disturbance index, which was supported by the manually verified dataset. Low estimates of user’s accuracy can be caused by (1) overestimating of the extent of disturbances, or (2) errors within reference data. Errors within reference data are common and likely contributed to the low accuracy assessment results (Congalton and Green, 1993; Foody, 2010). In general terms, the user’s accuracy is the number of correctly classified units divided by the total number of units mapped as that particular class, suggesting there may have been many disturbances that were not included in the VRI reference data.  In several instances, the VRI data depicted obvious discrepancies between the location of disturbances and observed changes in the disturbance index, both of which can be seen in Figure 2.7.  40  The estimated producer’s accuracy from the detection disturbance by harvest was much higher with an estimated 57.2 ± 3.4% of the harvest data classified as disturbance. The producer’s accuracy is a measure of omission error and is not related to the overestimation of disturbance. Failing to detect disturbances will produce a low estimate of producer’s accuracy. Similarly, a low producer’s accuracy may be obtained if disturbances did not actually occur between image acquisition times. Producer’s accuracy was likely affected by the reference data containing disturbances that had not occurred between image acquisition times, also apparent in Figure 2.7.  Interpretation of the differences between the user’s and producer’s accuracies provides an indication of the accuracy of the VRI data itself. For example, high producer’s accuracies are indicative of VRI disturbances aligning with detected disturbance layers. This producer’s accuracy is also consistently higher across all harvest sites, indicating that if RapidEye detected a disturbance, it was very often included within the VRI. Lower user’s accuracies potentially indicate that there are disturbances that are not included in the VRI. Alternatively, a low user’s accuracy could also imply a hypersensitive disturbance threshold, resulting in a subsequent overestimation of disturbance. In reality, estimates of classification error derived from confusion matrix data will likely contain both types of error included in the accuracy assessment. However, the stratified approach provides a useful way to estimate the level of accuracy to which we can assess the degree of disturbance (Olofsson et al. 2013a). The manually interpreted reference data supports the contention that many of the VRI polygons did not serve as adequate reference data, as much higher accuracies reported by the 500 manually interpreted pixels.  41  2.4.2 Assessment of Burned Areas  Classifying the full extent of burned areas presented several challenges to the threshold-based approach. Although the general extent of the detected burned area appears to have aligned well with the VRI data, the inherent patchy nature of forest fires appears to have caused several discrepancies between reference and mapped areas (Rowe and Scotter 1973; Kachmar and Arturo Sánchez-Azofeifa 2006). Reference polygons represented a simplification of fire extent and did not incorporate patchiness or residual unburned areas within the fire bounds. It was also apparent that many regions within burned areas exhibited a negative disturbance index, indicative of increased vegetation. This was likely due, in part, to the large degree of heterogeneity within burned areas. As a result, the single universal threshold produced maps depicting extremely patchy burn scars.  2.4.3 Impact of Atmospheric Distortion The disturbance index threshold appears to overestimate disturbed pixels in the presence of smoke or haze. Clouds present a variety of challenges in land cover monitoring studies (Sano et al. 2007). These challenges appeared to have been exacerbated by RapidEye’s lack of a thermal band, which is often used to detect cloud cover in many screening algorithms (Hollingsworth et al. 1996; Choi and Bindschadler 2004; Goodwin et al. 2013). Although RapidEye does provide an unusable data mask (UDM) targeted at detecting clouds in imagery, smaller clouds, haze, smoke, and cloud edges are often missed. Cloud shadow also provided an additional source of error evident in a number of images. Cloud shadows in the T2 image were often misclassified as disturbance as overall reflectance values were lower in the second image. Presence of haze can also be clearly seen in images 1062622, and to a lesser extent in 1063413, and in many cases this was classified as disturbance. Clouds present challenges to all optical remote sensing exercises.  While cloud contamination may be inevitable given seasonal constraints of image acquisition, 42  the constellation format of RapidEye allows for greater image selection over single-sensor platforms. 2.4.4 Threshold Selection Our universal binary threshold provided a simple, robust method for differentiating between two cover classes, and may not result in optimal extraction of disturbance features across vegetation types and across different seasons. A more accurate disturbance classification could be produced depending on the context of disturbance and the area of interest. Once difference images are produced a variety of more complex methods can be used to differentiate threshold values using the image histogram. Furthermore, the application of a specific thresholding algorithm could also be used to potentially estimate an optimal change threshold. In an assessment by Patra et al. (2011), six image threshold algorithms were evaluated based on their ability to differentiate between changed and unchanged regions. After deriving a difference image, any single band thresholding algorithm could be applied to extract change classes from within imagery.  Our universal threshold likely overestimated the area of stand replacing forest disturbances. This could be attributed, in part, to pixels sampled from large fire polygons which encompassed both severely and less severely burnt vegetation. This overestimation of disturbance was also notable in pixels where there was some evidence of atmospheric haze and unmasked clouds, apparent for tiles 1062622 and 962221. Figure 2.8 demonstrates how changing the threshold value can affect the classification accuracies. In this particular image, increasing the disturbance index threshold appears to increase both user’s and producer’s accuracies for the detection of disturbance, suggesting that our estimated threshold may have been overly sensitive to inherent spectral differences between the bitemporal images. 43   Figure 2.8 User's and Producer's accuracy for a disturbance threshold gradient using RapidEye tile 964123 (Burns Lake, BC) 2.4.5 Future Applications The readily available imagery and fine spatial resolution of RapidEye could allow for the exploration of many additional applications. In Canada, forest inventories typically combine spatial data on forest attributes such as stand age, species composition, as well as various structural and health traits (McRoberts and Tomppo 2007). Canadian forest inventory monitoring approaches combine metrics derived from ground plot data, aerial photography, and satellite imagery to estimate forest cover and change at provincial and national scales (Pitt and Pineau 2009). These inventories can take years to produce (Leckie and Gillis 1995); thus, current information on forest stands is often lacking. In addition, newer inventories are often conducted using inconsistent measurement techniques as novel technologies are adopted (Gillis 2001). Comprehensive inventories, such as those produced by Light Detection and Ranging (LiDAR), provide opportunities for estimating detailed structural stand attributes.  However, wall-to-wall 44  LiDAR coverage is not yet operationally feasible (Woods et al. 2011). In order to routinely update these inventories, maps are often updated every one to five years using disturbance information. Other nations routinely use satellite imagery to detect disturbance on the forested landbase. The Brazilian PRODES programme is a successful forest monitoring project aimed at tracking annual deforestation throughout the Amazon basin (INPE, 2013). The PRODES dataset is updated using 117 Landsat Thematic Mapper (TM) scenes to derive a comprehensive spatial product depicting deforestation. The annual product provides a snap-shot of deforestation throughout the Amazon, yet it lacks frequent temporal updates required to address illegal harvest activities as they occur (Shimabukuro et al., 2006). A complementary project, the Deforestation Detection in Real Time (DETER), is focused on the enforcement of regulations regarding logging activity using MODIS. The DETER programme is capable of achieving bi-weekly updates, however the spatial resolution is limited to a minimum mapping unit of 25 ha (Assunção et al. 2013). Both the Canadian and Brazilian examples highlight the need for more frequent and finer scale monitoring to meet national forest monitoring needs. The spatial and temporal resolutions provided by RapidEye would complement both forest inventory programmes and improve their ability to effectively monitor forest cover at scales relevant to the countries’ respective needs.   In addition to forest inventory applications, monitoring disturbance using RapidEye could provide substantial improvements for monitoring resource extraction activities. For example, concerns over the integrity of pipelines in BC and abroad could potentially be addressed through frequent satellite monitoring. Pipeline corridors often occur in remote regions and are affected by geophysical changes to the landscape. Forest disturbance events can affect slope stability and should be considered within a pipeline monitoring context (Pradan et al. 2008). To ensure 45  pipeline safety, remote sensing of pipeline corridors could become an integral component of the monitoring requirements used by the oil and gas industry (Hausamann et al. 2005).   Assessing habitat fragmentation metrics is another potential application that could be exploited using the RapidEye disturbance index approach. Land cover maps provide vital information on a suite of ecological characteristics used to quantify important variables associated with habitat and biodiversity (Boykin et al., 2013). Assessing trends, changes, and patterns in vegetation cover are common goals shared amongst global and national monitoring administrations (Ståhl et al. 2011).  The ability of the RapidEye disturbance index to detect and quantify spatial patterns relevant to habitat fragmentation scales presents opportunities for fine scale habitat assessment.  The following suggestions are provided to improve the quality and accuracy of change detection using imagery from the RapidEye constellation:   Our approach to change detection applied a basic form of radiometric correction (conversion to TOA reflectance). More rigorous radiometric corrections could be applied prior to implementing our change detection technique   Threshold optimization could potentially serve as a method of feature extraction, improving our ability to extract disturbed areas from satellite imagery (see Patra et al. 2011)  More robust snow, cloud, and cloud shadow detection techniques should be employed  2.5 Conclusion  Our British Columbian example highlighted the ability of RapidEye to detect disturbances over large areas, showing the potential for an automated threshold index type approach to be applied over the entire province. Our findings suggest that RapidEye could be employed to monitor 46  change at a regional- or national-level and could aid in wall-to-wall forest monitoring programmes. The use of the RapidEye constellation to achieve continuous monitoring on a yearly basis, or near real-time updates of specific areas, could be beneficial to administrative or government commitments to assessing forest cover. In addition, RapidEye’s ability to provide imagery at much finer spatial and temporal resolutions compared to conventional single-sensor platforms makes it an attractive alternative to monitoring forest change over other conventional approaches. Deriving a simple disturbance index for RapidEye imagery successfully delineated disturbances throughout the 7 image pairs tested. In many instances the VRI disturbance layer did not align with obvious disturbances detected by RapidEye, and this likely deflated our estimates of accuracy when using the VRI as reference data. Our results suggest that these methods could be implemented in an automated or semi-automated fashion to detect disturbance covering large areas.   47  3 2DETECTING FOREST DAMAGE AFTER A LOW-SEVERITY FIRE USING REMOTE SENSING AT MULTIPLE SCALES 3.1 Introduction Fires play an important role in determining the structure and composition of forest ecosystems (Franklin et al. 2002, Turner 2010). The legacy of fire on landscapes can have major impacts on the composition and ecological function of forests in western North America (Palmer et al. 1997). Fire suppression during the 20th century has resulted reduced forest heterogeneity, potential reductions in biodiversity, and the accumulation of fuels, particularly in dry forest types (Schoennagel et al. 2004, Adams 2013). The lack of surface fires in some dry forests has led to the fuel accumulations woody surface fuels and has increased chance of severe wildland and interface fires (Miller et al. 2009). Yet management protocols regarding the normative state of fire on the landscape remains a contentious issue, even amongst researchers (Fule et al. 2013, Odion et al. 2013).  Fire is a dominant agent of forest disturbance in many North American forests (Bond-Lamberty et al. 2007, Perry et al. 2011). The effects of fire may exhibit a range of spatial and temporal changes to forested ecosystems (Landres et al. 1999). Post fire effects may be immediate, through the direct consumption of living and dead vegetation (Simard et al. 2001) or as lasting legacies, such as changes in forest structure (Turner et al. 1994). Fire can also change the geophysical and geochemical properties of soil, including the soil’s availability to store nutrients and retain water (Certini 2005). In western North America many forest types have evolved to                                                  2 A version of this chapter has been submitted for publication. Arnett, J.T.T.R., Coops, N.C. Daniels, L.D., Gergel, S.E., Falls R.W. 2014. Detecting forest damage after a low-severity fire using remote sensing at multiple scales.  48  rely on fires as an important process for maintaining the ecological integrity of the system (Habeck and Mutch 1973, Allen et al. 2002).  Emissions from wildfires contribute substantially to global carbon emissions (Schimel and Baker 2002), thus global, national, and regional carbon accounting frameworks are tuned to address concerns over emissions from wildfires (Hurteau et al. 2008). Currently research is underway to monitor carbon emissions from fires in both the United States and Canada (Masek et al. 2013). The Canadian National Carbon Forest Monitoring programme (NCFM) as well as the American Monitoring Trends in Burn Severity project (MTBS) are both mandated to estimate emissions produced by forest fires at a national scale (Eidenshink et al. 2007).  Wildfires present a paradox with respect to sustainable forest stewardship and carbon emission targets. Prescribed burning of forest fuels is often seen as a means of managing ecosystems by reducing fuel loading and increasing forest resistance to severe fires; however, these practices also contribute considerably to atmospheric carbon through fuel combustion and smoke production (Weidinmyer et al. 2010, Pan et al. 2011). Improved monitoring of fires will allow for a better understanding of the trade-offs between ecological and climatological objectives.  Remote sensing provides the only feasible solution to operationalize both local and national monitoring of fires (Kerr and Ostrovsky 2003).  Conventional assessment of fire metrics often relies on the collection of field data after the fire has occurred, these data may include estimates of canopy consumption, surface fuels consumption (Key 2005), plant mortality (Larson and Franklin 2005), and changes to soil properties (Ice et al. 2004). Since fire is arguably unpredictable across the landscape, pre-fire conditions are often inferred or estimated after the fire has occurred (Lentile et al. 2006). Several 49  standardized surveys have been adapted to assess the severity and intensity of fire on forested communities. The composite burn index (CBI) has been adopted as a means of assessing fire severity due to the indices ability of correlate with remotely sensed imagery and combines measures of fire severity from below and above the canopy (Kasischke 2008)   Using optical remote sensing technologies to assess fire damage relies on the attribution of damage to vegetation to spectral changes recorded by the sensor. Mapping the extent and impacts of fire using remote sensing can provide information on the presence/absence of fires or discrete severity classes along a gradient of spectral changes caused by the loss or damage of vegetation (Lentile et al. 2006). Fire severity classes derived from remote sensing are often somewhat arbitrary and do not necessarily reflect empirical changes in forest cover or consumed biomass (Lentile et al. 2006). Furthermore, the mapping of specific burn attributes using remote sensing has been most frequently applied to large, severe and stand replacing disturbances (Epting et al. 2005). As fires encompass a vast range of severity, intensity, and scalar gradients across a range of ecosystems it is necessary to develop fine scale methods to accurately monitor post-fire effects on the vegetation (Eidenshink et al 2007). New remote sensing platforms offering higher spatial resolution, more frequent image acquisition times, and specific band configurations offer many opportunities to study the effects and impacts of fire on vegetation (Arroyo et al. 2008).  The Landsat series of satellites has been successfully used to map and monitor large fires (Masek et al. 2013). However, finer spatial detail and more frequent acquisition of imagery may be more desirable to a variety of disciplines including ecology, risk assessment, and carbon accounting (Whitman et al. 2013, de Groot et al. 2007).  50  The RapidEye satellite constellation includes 5 identical satellite sensors, which acquire high-spatial resolution (5 X 5 m) imagery at daily intervals. In addition to the sensors’ spatial resolution and temporal frequency, RapidEye has a unique red-edge band, as well as bands strategically placed in the visible and near-infrared (NIR) portions of the electromagnetic spectrum, making it an ideal sensor for monitoring vegetation (Bindel et al. 2011). Conventional Earth observing satellites, such as Landsat, provide imagery at a coarser spatial resolution but may offer several advantages over RapidEye, including wider swath coverage, more bands including in the shortwave infrared region of the spectrum, a lengthy historical data archive, as well as radiometrically-corrected imagery provided at no cost. Furthermore the longstanding availability of Landsat data has allowed for the much research on the sensor’s abilities to detect and monitor change on the Earth’s surface (Wulder et al. 2008).   A variety of change detection techniques can be used to identify fire and other disturbances using satellite imagery. Tracking of ground surface change using satellite imagery is generally conducted by the use of multitemporal, co-registered imagery (Coppin et al. 2004). Differences are then mapped by observing relatively large magnitude changes in surface reflectance between individual pixels (Coppin et al. 2004). The simplest change detection technique involves the subtraction of pixel values from one image to another co-registered image (Singh et al. 1989). Mapping change, however, is generally not undertaken on individual spectral bands. Vegetation indices, or band ratios, are usually derived prior to any image differencing (Singh et al. 1989; Lu et al. 2004). Indices derived from algebraic operations using band reflectance values are preferable because (i) indices are able to normalize the magnitude of change between images that may exhibit slightly different spectral levels (Huete et al. 2002) and (ii) because vegetation indices have been correlated with specific structural, physiological, and chemical responses of 51  vegetation due to stress or disturbance (Gao, 1996; Carlson and Ripley, 1997; Sims and Gamon 2003). Most vegetation indices are used to create a single grey-scale univariate difference image however some indices, such as the Normalized Burn Ratio (NBR), can rely on a single image to detect fire (Epting et al. 2005). To map the extent of subtle disturbances, such as non-stand-replacing fires it is necessary to incorporate detailed spatial data, beyond the 30 X 30 m resolution provided by Landsat. In this paper we explore the potential of a RapidEye image time series to detect changes in vegetation following a low-intensity prescribed fire that aimed to reduce surface fuels in a dry forest in western Canada. Spatial semi-variance statistics were calculated to compare the level of variability between imagery from the RapidEye and Landsat sensors. Several commonly used spectral vegetation indices were then used to assess spectral differences before and after fire. These differences were then correlated with aggregate measures of damage to individual trees.  Estimating burn severity using optical remote sensing tends to emphasize changes associated with the forest canopy (Falkowski et al. 2004, Key and Benson 2006). Estimates of consumed sub-canopy fuels are largely unavailable when using optical remote sensing (Miller et al. 2003); however, these fuels may contribute substantially to forest biomass stocks. It is therefore necessary to include sub-canopy measures of burn severity along with overall estimates of canopy scorch to calculate consumed biomass due to fires. To account for damage to the sub-canopy, metrics of individual tree damage were combined into a Simple Burn Index (SBI), which was then compared to remotely sensed imagery.   Finally, we derived and mapped a set of burn severity classes and estimated consumption of woody surface fuels and biomass throughout the study area. In this study we hypothesize that 52  estimates of burn impacts derived from RapidEye satellite imagery will be able to resolve stand damage at the individual tree level. It is expected that finer-scale disturbance attributes derived from RapidEye will be able to provide a more detailed assessment of the burn damage than conventional analysis of Landsat-imagery.  3.2 Site Description  This study was conducted in the West Vaseux Lake unit of the Vaseux-Bighorn National Wildlife Area (VBNWA), located in the Southern Okanagan Region of British Columbia, Canada (Figure 3.1). The VBNW presents an ideal opportunity to assess the capacity of remote sensing due to the availability of pre- and post-fire forest and fuels data, an attribute often lacking in disturbance remote sensing studies. The forest includes a mix of Douglas-fir (Pseudotsuga menziesii) and Ponderosa pine (Pinus ponderosa), with an approximate density of 250 trees ha-1. The VBNWA is a protected area managed by the Canadian Wildlife Service to sustain habitat and populations of more than 30 threatened and endangered species (Cooper et al. 2011). Conservation interest for the area has led to the reintroduction of fire as an ecological process that has largely been suppressed since European settlement. In March 2013, the Canadian Wildlife Service (CWS), in collaboration with the BC Wildfire Management Branch, conducted a prescribed fire in the VBNWA to re-introduce fire, restore critical wildlife habitat, mitigate fuel build-up to decrease the risk of a severe wildfire and increase ecosystem resilience to climate change (Mottishaw 2013).   53   Figure 3.1 A map depicting the location of the Vasuex lake Study Area 3.3 Methods 3.3.1 Field Data Trees (n =116, diameter at breast height (DBH) > 5cm) were sampled at 15-m intervals along 4 established transects using a point-center-quarter method (Cottam and Curtis 1953) and measured before and after the burn. Indicators of burn damage included (1) burn circumference; a measure of the scorched area around the base of the tree, (2) char height; the height of charring up the tree divided by the extrapolated tree height (derived from DBH), and (3) estimated crown scorch expressed as a percent of the total unburned crown. Summaries of the fire damage estimates are provided in Figure 3.2. 54  Woody surface fuels (diameter ≥ 2.6cm) were sampled along each transect using the line intercept method (van Wagner 1968).  Before the burn, the locations of each piece of wood along transects were recorded, species was determined and diameter measured with calipers.  Steel wire was tied at the intersection point so that the wood could be re-measured after the burn.  Species-specific wood density (Gonzalez 1990) was used to calculate the biomass of each piece of wood before and after the burn.  Measures of burn circumference, char height, and crown scorch were combined to derive a simple burn index (SBI). The SBI provided a meaningful estimate of burn severity for individual trees. Based on the relative stand closure we estimated weightings for the sub-canopy attributes to contribute to 20% (evenly split between the burn circumference and char height measures) and canopy reflectance to contribute to 80% of the SBI. Figure 3.2 provides a visual depiction of the variability of burn attributes recorded for individual trees in this study.    Figure 3.2: Field data for individual tree damage estimates for the 115 trees sampled. The combined burn percentage ranges from 0-300%, as three variables were summed.  55  3.3.2 Satellite Imagery Two satellite image pairs were used in the analysis. Image pairs from before and after the fire were acquired from both RapidEye and Landsat (TM+ and OLI) sensors. A summary of the images used is provided in Table 3.1. Image pairs were selected to be cloud free and within 20 days of each other’s respective anniversary date to mitigate any seasonal radiometric differences. Table 3.1 A summary of Landsat and RapidEye images selected in this study Image  Date Recorded  Specific Sensor  Spatial Resolution (m) RapidEye T1  June 11, 2012 RapidEye 5 5 RapidEye T2  June 30, 2013 RapidEye 3 5 Landsat T2  June 21, 2011 Landsat 5 (TM) 30 Landsat T2 June 10, 2013 Landsat 8 (OLI) 30 3.3.3 Pre Processing Image to image registration was conducted using ENVI 5’s automated tie point analysis workflow (Exelis Visual Information Solutions 2013). Mean pixel error was verified to be within less than half a pixel for both RapidEye and Landsat image pairs. Simple atmospheric correction was conducted by subtracting the darkest pixel value from deep water bodies in all four images (Chavez 1996). Lastly, image to image radiometric normalization was conducted using a multivariate alteration detection algorithm (MAD) described by Canty et al. (2004).  3.3.4 Derivation of Vegetation Indices  A survey of the literature was conducted to determine an appropriate selection of common vegetation indices used to detect fires. The Enhanced Vegetation Index (EVI), the Soil Adjusted Vegetation Index (SAVI), and the Normalized Difference Vegetation Index (NDVI) were selected because they were applicable to both RapidEye and Landsat sensors. In addition, the NBR (only applicable to Landsat) was also used due to its proven ability to detect fires and 56  predestinated burn severity classes (Key and Benson 2006). All of the indices, except for the NBR, take advantage of changes in reflectance exhibited in the red and NIR portions of the electromagnetic spectrum following fire.  The NBR emphasizes the change in reflectance caused by combustion of forest fuels from two of the least correlated bands, the NIR and the short-wave infrared (SWIR) (López García and Caselles, 1991). Spectral bands in the SWIR and the MIR regions of the electromagnetic spectrum have been shown to exhibit opposing responses following fire. In general reflectance in the NIR regions tends to decrease (Sunar and Özkan, 2001) while MIR reflectance increases (van Watendonk et al. 2004). Although the RapidEye sensor does not provide a MIR band to derive and NBR or similar index, Smith et al. (2005) suggested NBR may not be appropriate for finer spatial resolution imagery due to smaller pixels being influenced by the presence of ash, resulting in higher reflectance values. Table 3.2 provides a summary of the vegetation indices used in this study and their respective formulas.   Table 3.2: Common vegetation indices and their formulas, * L refers to a constant used to shift the relationship between NIR and red wavelengths, in this study we used the conventional value of ‘1’ (see Huete 1988) Index Name  Formula  Applicable to Examples Enhanced Vegetation Index (EVI) (NIR - Red) / (NIR + Red) Landsat and RapidEye Huete et al. 1994 Normalized Burn Ratio (NBR) (NIR - SWIR) / (NIR + SWIR) Landsat  Miller and Thode 2007  Normalized Difference Vegetation Index (NDVI) (NIR - Red) / (NIR + Red) Landsat and RapidEye  Colwell 1974 Soil Adjusted Vegetation Index (SAVI) (NIR – Red) / (NIR + Red + L) *(1+L) Landsat and RapidEye  Huete 1988  57  For each calculated index and difference image was created by subtracting the T2 image (post-fire) from the T1 (pre-fire) images. The detection of change due to fire was then conducted on four difference images from RapidEye and four difference images from Landsat.  3.3.5 Exploring Spatial Variability and Image Semi-variance  An analysis of semi-variance across the study area was undertaken to analyze the spatial and spectral variability for both the RapidEye and Landsat sensors (Johansen et al., 2007). Metrics estimated from semi-variograms can aid in the determination of a suite of spatial and ecological mapping concerns, including image element size and image minimum mapping unit (MMU) (Johansen et al. 2007), while the shape of the semi-variogram can be used to determine optimal distances between sample locations, removing the effects of spatial autocorrelation (Curran 1988¸Trietz and Howarth 2000). For a detailed review of applications of semi-variance analyses using remote sensing see Naimi et al. (2011). In order to compute semi-variograms from the imagery, we used R package USDM (Version 1.1-12) to calculate semi-variance on the entire suite of VBNWA images (Naimi 2014). Lag distances were specified as the respective Landsat and RapidEye cell lengths (30m and 5m respectively). 3.3.6 Relating field data to Vegetation Indices  The mapped tree data was overlaid with difference image raster data using a GIS and regression analysis was conducted between the SBI and spatially corresponding pixel values. For the Landsat imagery it was necessary to average SBI measures for Landsat pixels containing multiple mapped trees; this was not necessary for RapidEye imagery since no trees occupied the same pixel. Because of potential geo-locational error associated with the GPS ground location data and the remotely sensed images a low pass, smoothing, filter with a 3 by 3 kernel was applied prior to the derivation of the vegetation indices. Pixels from filtered vegetation index 58  difference images were then extracted to shapefile point features and subjected to regression analysis.  To provide an indication of the efficacy of each vegetation indices’ ability to differentiate fire severity, ground estimates of crown scorch were divided into three arbitrary classes (0-33%, 33-66%, and 66-100% scorch). Analysis of variance was used to test for differences between these classes.  Burn severity classes were adapted from Key and Benson’s (2005) description to map burn severity throughout the study area. The dNBR severity classes from Landsat imagery were sampled using the Zonal Statistics tool provided in ArcMAP’s Spatial Analyst toolbox and applied to overlying RapidEye pixels (ESRI 2014).  Fuel consumption data gathered along transects was used to estimate woody surface fuel biomass consumed for each severity class. Field estimates of consumed woody surface fuels were stratified by RapidEye severity classes to provide an average measure of fuel consumption by class. Estimated biomass consumed was extrapolated to estimate surface fuels consumed throughout the entire burn area.  3.4 Results: Semi-variograms comparing spectral variability between RapidEye and Landsat imagery revealed that Landsat had less overall variability compared to RapidEye (Figure 3.3). The RapidEye band semi-variograms revealed a definitive sill around 400m, while the Landsat band semi-variograms did not have a definitive shape or sill. The abrupt curve depicted by the RapidEye semi-variogram at between 0 and 30 meters provides an indicator of the spatial 59  variability of the stand due to its open structural composition. The lack of a sill on the Landsat imagery indicates at 30m spatial resolution this finer spatial detail is not captured.    Figure 3.3 Semi-variograms comparing RapidEye and Landsat imagery for visible and near infrared bands   Stratified estimates of crown scorch demonstrated that both sensors were able to differentiate crown scorch levels; however some indices performed better than others (Figure 3.4). The mean SAVI values provided the best differentiation of crown scorch classes for both the RapidEye and Landsat sensors, suggesting that SAVI would be the best vegetation index for mapping fire severity. Analysis of variance among crown scorch classes revealed that all stratified canopy estimates exhibited significant differences (p < 0.05). A post-hoc Tukey HSD test revealed significant differences between crown all scorch classes except RapidEye dNDVI and Landsat 60  dNBR which resulted in no significant differences between the low and medium crown scorch categories (p > 0.05).  Figure 3.4 Comparison of crown scorch severity estimates for Landsat and RapidEye Sensors. * indicates indices are not comparable across sensors.   The resulting coefficients of determination (R2) for regression analysis between SBI and vegetation indices are provided in Figure 3.5. For both Landsat and RapidEye the SAVI index provided the best fit with the SBI.  61   Figure 3.5:Commparisons of Coefficients of Determination (R2) for correlations between SBI and vegetation indices derived from Landsat and RapidEye Imagery  Figure 3.6 provides scatter plots of the best fit between the RapidEye and Landsat SAVI indices. In the case of Landsat imagery the 30m data resulted in field sample locations averaged by pixel, thereby reducing point samples from 116 to 26 for the Landsat regression.  62   Figure 3.6 Best fit between Landsat Soil Adjusted Vegetation Index and Simple Burn Index, p<0.05, R2 = 0.56, RE SAVI p<0.05, R2 = 0.51  A map of fire severity is provided in Figure 3.7. Due to the relatively low severity of the fire, only ‘low’ and ‘moderate’ severity classes were interpreted using the dNBR classes described by Key and Benson (2006). Furthermore, since much of the fire appears to have burned under the canopy; only 2 of the 40 ha mapped were classified by moderate severity burn.  In addition to the ‘low severity’ class derived from the dNBR values, it was necessary to add an additional ‘grass burn’ severity class, with a threshold calculated by the median value between ‘no-change’ and ‘low severity’ burn classes.  63   Figure 3.7 Map of dSAVI fire severity classes using transferred dNBR classes described by Key and Benson (2006)  64  Estimated fire severity classes from SAVI are provided in Table 3.3. The relative percentages of woody surface fuels consumed increased with severity class; however several consumed pieces of wood (n = 21) occurred in areas where no change was detected. This resulted in an estimated 19.3 ± 5.9% of fuels consumed within the no change class.  Table 3.3 Estimate of woody surface fuels consumed grouped by burn severity class, error is estimated as standard error of the mean. Class RapidEye dSAVI Values % LWD Fuels Consumed Area Mapped (ha) Biomass Consumed by class (Mg/ha) No Change  0.000 to 0.022 19.3 ± 5.9 12.6 3.38 ± 1.03 Grass Burn  0.022 to 0.027  33.3 ± 8.6 16.5 5.84 ± 1.51  Low 0.027 to 0.034 52.0 ± 6.2 9.3 9.12 ± 1.09 Moderate 0.034 to 0.060 66.9 ± 10.5 2.1 11.72 ± 1.84  3.5 Discussion Improving accuracy of carbon budget models through remote sensing is an ever-increasing focus of research. With forests presenting a potential sink and source of atmospheric carbon, detailed forest monitoring will become increasingly important for scientists and policy makers. Although forests are considered a net sink at the global scale, carbon fluxes have been shown to vary dramatically by region (Pan et al. 2011). Estimating the contribution of individual forest attributes, such as the combustion of live crowns and woody fuels, to specific carbon pools is also a challenge for carbon monitoring (Vaillant et al. 2013).  3.5.1 Assessment of Spatial Variability  As expected, semi-variogram analysis of both post-burn images revealed a greater degree of spatial detail within the RapidEye imagery. The ability to resolve finer scale spatial features 65  using RapidEye imagery, coupled with the potential ability to detect damage to individual tree crowns, could provide benefits to fine scale disturbance mapping projects. Although the effective range for both sensors was achieved at a distance of approximately 400 meters (Figure 3.3), the abrupt partial sill in the RapidEye semi-variograms indicates fine-scale stand attributes such as gaps between individual trees can be detected and offers advantages for fine-scale ecological applications (Johansen and Phinn, 2006). This large partial sill was depicted in the RapidEye imagery between approximately 5 and 30 meters lag, a level of spatial detail not available in the Landsat imagery.  This also indicates potential for RapidEye imagery to be used to map fires, which leave behind a complex burn patterns on the landscape (Kachmar and Sanchez-Azofeifa 2006), particularly in relatively low-severity fires.  3.5.2 Relating Field Data to Vegetation Indices  Boxplots differentiating crown scorch classes yielded similar results for comparable indices across sensors (EVI, NDVI, and SAVI). Although ANOVA tests conducted on crown scorch classes revealed significant differences between classes for all indices from RapidEye and Landsat, post-hoc Tukey’s HSD revealed that RapidEye dNDVI and Landsat dNBR were unable to differentiate between low and medium crown scorch classes; a possible indication of these indices’ insensitivity to minor changes in vegetation damage and condition (Epting et al. 2005).   RapidEye’s red-edge band produced unexpected results compared to the other indices with change vectors decreasing along the crown scorch severity gradient. Interestingly Ramoelo et al. (2012) found that RapidEye’s red-edge band could be used to improve foliar nitrogen estimates when used in tandem with other vegetation indices, suggesting that the red-edge band may not be particularly useful as a single index. It should also be noted that the red-edge was the only single band index used in our analysis, (i.e. not a ratio of two or more bands).  66  Regression analysis revealed that the dSAVI index provided the highest R2 value of all indices for both RapidEye and Landsat. This relationship is not surprising, as the SAVI index was developed to account for inherent soil reflectance from below the canopy (Huete 1988). The ability of the SAVI to shift the relationship between reflectance emitted in the NIR and red regions of the electromagnetic spectrum using the constant ‘L’ appears to have accounted for proportion of the increased reflectance caused by soil and relatively open stand structure, characteristic of the study area (FLNRO 2014).  Results may have varied if the study area exhibited a higher level of crown closure.  Apart from the single red-edge band, dNBR provided a relatively poor correlation between the SBI, likely due to low severity of the fire and little overall change to stand structure. Although Epting et al. (2005) found a higher correlation was achieved in a relatively closed stand with the dNBR index, the forest stands in their study experienced more severe stand damage and displayed a relatively higher crown closure, suggesting that scalar, structural, and severity attributes may influence stand correlation between SBI and indices. Based on the results of our analysis, we do not recommend using dNBR to map low-severity fires.  3.5.3 Estimation of Sub-canopy Fuel Consumption Adapting dNBR severity classes to dSAVI produced four classes ranging from ‘grass-burn’ (no apparent canopy damage, but evidence of spectral changes to the sub-canopy) to ‘moderate severity’ (Figure 3.7). Estimates of fuels consumed by severity class revealed that 19.3 ± 5.9% of fuels were consumed in the ‘no-change’ class, a potential challenge when mapping low severity fires. This estimate of fuels consumed by the no-change class was likely attributed to four potential factors as follows: (1) the inherent low-severity nature of the fire; (2) the T2 image being acquired 19 days after the later T1 anniversary date; (3) substantially different amounts of 67  precipitation leading prior to the acquisition of the imagery with T2 being markedly drier than T1 or (4) issues associated with transferring dNBR derived severity classes to other types of vegetation indices. Meteorological data from the nearby Penticton weather station revealed that spring 2013 was also relatively dry (Climate Canada); potentially affecting the estimated severity of the mapped burn classes (van Wagtendonk et al. 2004). In the grass, low, and medium burn severity classes the fire consumed 33.3 ± 8.6%, 52.0 ± 6.2%, and 66.9 ± 10.5% of woody fuels, respectively.  Estimates of woody surface fuels’ contribution to forest carbon pools can range from 5-26% (Muller and Lui 1991), and, therefore, should be considered when assessing forest management objectives. In a more complex analysis, Smith and Hudak (2005) used aerial photography and Landsat imagery to estimate the woody surface consumed by fire by estimating the area of white ash within imagery. However, white ash is only formed from the full combustion of woody fuels under very high heat and was present in negligible quantities in our study area. In our medium severity class, we estimated 66.9 ± 10.5% of woody surface fuels were consumed, which is comparable to consumption estimates in similar ecosystems (Harington 1987, Hille and Stephens 2005).  The extrapolated estimates of total wood fuel biomass consumed were comparable to estimates presented in other studies (e.g. Hille and Stephens 2005). In total, an estimated 247.7 ± 51.8 Mg of woody surface fuels were consumed. Comparing our estimates to other studies we find that our consumed fuels in the VBNWA to be similar; for example Vaillant et al. (2013) found prescribed burn to have reduced woody surface fuels by 10 Mg ha-1 in a managed forest in California, our highest severity class (moderate) predicted a surface woody fuel loss of 11.7 Mg ha-1.  68  3.5.4 Monitoring Implications   Although our study demonstrated a relatively simply approach to estimate the amount of consumed large woody surface fuels, the ability to achieve fine level of spatial detail offers valuable advantages to assess changes in ecosystems at a scale relevant to management.  Criticism of severity classes applied to vegetation indices are generally founded on the basis that metrics of severity attributed to remotely sensed indices do not directly correlate to empirical changes in vegetation and biomass consumption (Lentile et al. 2005). In addition, transferability of severity classes across ecosystems presents several challenges to designating universal severity metrics (Jain and Graham 2004). While optical remote sensing can provide good estimation of spatial distribution of vegetation types and disturbances, optical imagery still must be supplemented by vertical structural attributes in order to achieve accurate estimates of carbon stocks (Wulder et al. 2009). Metrics derived from other sensors capable of resolving structural attributes such as Light Detection and Ranging (LiDAR) may be fused with optical indices to provide detailed inventories of change due to disturbances (Bolton et al. 2013), however routine monitoring using LiDAR is generally considered expensive and beyond the scope of most management budgets.  As strategies for carbon monitoring become increasingly applied to achieve forest management objectives, frameworks will inevitably develop to assess and actively monitor wildfires. Currently active monitoring of fires relies on very coarse resolution satellites to detect changes (resolution of 1 km2) (Miettinen et al. 2013); however, RapidEye could augment active fire monitoring given its high spatial and temporal resolution. Sensors such as RapidEye and other fine-scale remotely sensed imagery could provide advantages to estimating forest carbon in reserves or to provide detailed mapping of specific burn areas. Furthermore, since above ground 69  forest carbon can change quickly as the result of different management regimes (Hines et al. 2010), the fine temporal resolution of RapidEye could be used to assess the immediate effects of management treatments and monitor their outcomes over time.  3.6 Conclusion  The greater spatial variability provided by the RapidEye data demonstrated potential for fine scale mapping of burn damage throughout the study area. Overall RapidEye did not appear to correlate any better with the ground data than the averaged SBI estimates correlated with the Landsat data. Inherent seasonal differences may have influenced our ‘grass burn’ severity class, potentially exaggerating the extent of that category and thus over-representing LWD consumed for the grass-burn class. The temporal availability of RapidEye provided an advantage over Landsat, with several cloud free images readily available covering the research area. Image analysis suggests that RapidEye was able to provide a greater detail of spatial patterns and objects compared to Landsat. The estimation of burn severity classes in this research simply provided an empirical methodology for estimating consumption of large wood fuels weightings as well as burn severity classes should be implemented on a site-by-site basis, but should have some grounding in empirical vegetation damage. Cost considerations regarding the specific level of detail should therefore be assessed prior to considering the acquisition of RapidEye imagery over the free, but less detailed Landsat imagery. Practical applications could include park and reserve management designs and planning, assessment of forest disturbances, and monitoring sensitive areas.  70  4 CONCLUSION   The overarching goal of this study was to assess the ability of the RapidEye constellation to detect forest disturbances in British Columbia. Optical remote sensing has been consistently proven in its ability to assess stand replacing disturbances (Achard et al. 2002; Kasischke and Turetsky 2006; Healey et al. 2005; Maskek et al. 2008), however assessment of non-stand replacing and structural aspects of disturbance using optical imagery remain a challenge (Kimes et al. 2006; Lefsky et al. 2002; Frokling et al. 2009). Medium and low-spatial resolution sensors such as Landsat and MODIS have demonstrated accurate identification of stand replacing disturbances for decades; yet achieving metrics other than presence/absence of disturbance has proven difficult using optical imagery alone (Wulder et al. 2009). Although technologies such as Light Detection and Ranging (LiDAR) can provide precise estimates of structural forest attributes, these technologies remain limited due to high operational costs (Ma et al. 2014) and are not considered feasible for routine forest monitoring.     The results of this study demonstrate that RapidEye offers potential advantages to assess both stand and non-stand replacing disturbance over other sensors. In addition, the relatively fine spatial resolution and frequent image acquisition may offer many advantages to improve routine monitoring of forest disturbances.  The first research question examined the potential of RapidEye to assess the presences/absence of stand-replacing disturbance, while the second study addressed finer-scale disturbance metrics after a single, non-stand-replacing fire. Both studies highlight the sensors’ ability to assess stand replacing and non-stand-replacing disturbances at yearly intervals, a timeframe relevant to meet the needs of most forest management initiatives. Although the research was conducted in a British Columbian context, the overarching research is intended to aid in forest inventory 71  globally and support a variety of applications including improving forest inventories, prioritizing conservation efforts, ecological mapping and monitoring, and carbon accounting. We hope that results of this study will aid in the ongoing objectives associated with broad-scale forest management and will be adapted to accommodate a variety of research and monitoring initiatives. 4.1 Key Findings In Chapter 2 a method was developed for mapping the extent of stand replacing disturbances over a set of diverse forest types in British Columbia. By adapting the methods of Healey et al. (2005) a modified disturbance index was developed specifically for RapidEye imagery. A simple threshold was applied to the modified disturbance index to differentiate stand replacing disturbances and unchanged land. The modified disturbance index and simple threshold technique was able to identify stand replacing disturbances at an accuracy of 92.64 ± 0.15% using forest inventory data and 96.91 ± 1.16% using an independently verified dataset. It was also noted that classification of disturbance could be improved if a more sensitive cloud identification technique had been employed. Furthermore, the seven 25 X 25 km RapidEye tiles selected for the study were all available as image pairs with <10% cloud cover and within ±20 days of each other’s respective anniversary date, a level of temporal detail that is largely unavailable from other high spatial resolution sensors. The high classification accuracy coupled with the extensive image archive, and potential to capture daily imagery over a given region suggested the RapidEye constellation could aid in the ongoing monitoring of disturbances such as wildfire, or illegal logging activity, especially in regions where access is difficult or unfeasible (Grainger and Obersteiner, 2011).  72  In Chapter 3 the potential of RapidEye to assess the condition of vegetation following a low-severity, non-stand-replacing burn in a dry western Canadian forest was explored. In this study metrics derived from both Landsat (TM and OLI) and RapidEye imagery were compared on their ability to detect fine-scale disturbance attributes. Estimates of tree damage and consumption of woody surface fuels were used to compare effectiveness of vegetation indices derived from RapidEye and Landsat.  Analysis of the spatial variability provided by the post-fire imagery revealed that RapidEye was able to achieve spatial detail capable of resolving damage on an individual tree basis. This level of detail occurred between 5 and 30 meters, and was therefore unattainable using Landsat with its 30 meter resolution. Regression analysis revealed significant relationships between all vegetation indices and the derived Simple Burn Index (SBI) -a weighted measure of collated burn damage. The SBI and Soil Adjusted Vegetation Index (SAVI) provided the best fit of regression for both Landsat and RapidEye. Although Landsat was ultimately able to provide a better fit of regression (SAVI, R2 = 0.56) from mean aggregated SBI scores, RapidEye imagery produced a comparable coefficient of determination (SAVI, R2 = 0.51) using SBI observations for individual trees. The estimation of consumed woody surface fuels classes from derived spectral classes also exposed the additional detail provided by RapidEye and demonstrated the sensors’ potential to quantify specific ground-level attributes, a detail often overlooked in burn severity studies (Lentile et al. 2006). In addition estimates of consumed woody surface fuels were comparable to other studies using more detailed and extensive field measures (Harington 1987; Hille and Stephens 2005; Vaillant et al. 2013). The results of this study demonstrate the ability of RapidEye to map fine-scale attributes associated with ecological change caused by fire.  73  4.2 Future Work  Remote sensing technologies will become increasingly used to address multitude of land use concerns. To meet the demands of a growing industry and research community it is necessary to develop remote sensing-derived products and applications to support a range of disciplines. RapidEye’s ability to capture many images per year over a given area makes it an ideal candidate for strategic monitoring of forest resources. This research has clearly demonstrated that potential of RapideEye to augment forest inventories throughout a variety of forest types. Based on the findings of this research the we recommend: (1) exploring of sensor automation to monitor forest disturbances using satellite constellations should  be explored (2); derived disturbance indices from RapidEye across various biomes be assessed, and finally (3) the integration of RapidEye with active sensors, particularly LiDAR, should be investigated.  Satellite constellations provide opportunities for near real-time mapping of disturbance and other abrupt land change events. Full or partial automation of land cover change is an emerging field of study that will arguably lead to enhanced decision making, policy, and disaster response (Verbesselt et al. 2012). Real-time and near real-time frameworks are already in place to detect large disturbances such as wildfires (Ruminski et al. 2008). However, the sensors used to detect these events are typically mounted on high orbital, geostationary satellites which provide relatively coarse spatial resolution imagery (Zhang et al. 2012). The integration of finer spatial resolution sensors, capable of frequent image acquisition could aid in near-real time mapping of disturbance or natural disasters. Augmenting coarse real-time monitoring frameworks with finer spatial resolution data could potentially aid in natural disaster response and relief efforts.  Assessing RapidEye’s ability to derive universal metrics of disturbance across biomes would inevitably increase the applicability of RapidEye to detect disturbance. Although our initial 74  detection algorithm was developed using a diverse set of forest types in British Columbia, it would be beneficial to assess the modified disturbance index in other biomes, particularly tropical and broadleaf-dominated forests, which may exhibit different spectral responses than the primarily conifer dominated forests used in this study. Furthermore, standard vegetation indices derived from satellite imagery may vary significantly by biome (Huete et al. 1997). It is therefore recommended that RapidEye’s disturbance detection capabilities be assessed across a diverse range of ecological gradients, particularly broadleaf and tropical forest types.   Finally the integration of RapidEye disturbance detection with non-spectral sensors, capable resolving structural attributes associated with vegetation could be advantageous for quantifying disturbance events. Two dimensional data is inherently limited with respect to resolving structural attributes of vegetation (Wulder et al. 2009). As seen in chapter 3, the variability associated with fine-scale attributes could potentially be improved by adding metrics of structural attributes associated with tree damage. LiDAR, in tandem with satellite imagery has been shown to enhance ecological mapping and modeling application using optical imagery at a range of spatial resolutions (Bolton et al. 2013; Hlakdik et al. 2013). Integrating RapidEye with LiDAR could provide powerful analytical tools aimed at forest inventories, ecological mapping and monitoring, and carbon accounting and should be explored.    75  REFERENCES Achard, F., Eva, H. D., Stibig, H.-J., Mayaux, P., Gallego, J., Richards, T., & Malingreau, J.-P. (2002). Determination of deforestation rates of the world’s humid tropical forests. Science (New York, N.Y.), 297, 999–1002. doi:10.1126/science.1070656 Adams, M. A. (2013). Mega-fires, tipping points and ecosystem services: Managing forests and woodlands in an uncertain future. Forest Ecology and Management, 294, 250–261. doi:10.1016/j.foreco.2012.11.039 Allen, C. D., Savage, M., Falk, D. A., Suckling, K. F., Swetnam, T. W., Schulke, T., … Klingel, J. T. (2002). Ecological Restoration of Southwestern Ponderaos Pine Ecosytems: a Broad Persepective. Ecological Applications, 12, 1418–1433. doi:10.1890/1051-0761(2002)012[1418:EROSPP]2.0.CO;2 Anderson, J. (2007). Global Cloud Cover Maps. The University of Manitoba. Retrieved from http://home.cc.umanitoba.ca/~jander/clouds/globalclouds.html Arroyo, L. A., Pascual, C., & Manzanera, J. A. (2008). Fire models and methods to map fuel types: The role of remote sensing. Forest Ecology and Management, 256, 1239–1252. doi:10.1016/j.foreco.2008.06.048 Asner, G. P. (2001). Cloud cover in Landsat observations of the Brazilian Amazon. International Journal of Remote Sensing, 22(18), 3855–3862. doi:10.1080/01431160010006926 Barford, C. C., Wofsy, S. C., Goulden, M. L., Munger, J. W., Pyle, E. H., Urbanski, S. P., … Moore, K. (2001). Factors controlling long- and short-term sequestration of atmospheric CO2 in a mid-latitude forest. Science (New York, N.Y.), 294(5547), 1688–1691. doi:10.1126/science.1062962 Bindel, M, S Hese, C. B., Bindel, M., Hese, S., Berger, C., & Schmullius, C. (2011). Evaluation of red-edge spectral information for biotope mapping using RapidEye. In Proceedings of SPIE (Vol. 8174, p. 8174). Retrieved from http://www.scopus.com/inward/record.url?eid=2-s2.0-80455135101&partnerID=40&md5=336642b3576f16eb59ff2c7a139df275 Bolton, D. K., Coops, N. C., & Wulder, M. A. (2013). Measuring forest structure along productivity gradients in the Canadian boreal with small-footprint Lidar. Environmental Monitoring and Assessment, 185, 6617–6634. doi:10.1007/s10661-012-3051-9 Bond-Lamberty, B., Peckham, S. D., Ahl, D. E., & Gower, S. T. (2007). Fire as the dominant driver of central Canadian boreal forest carbon balance. Nature, 450, 89–92. doi:10.1038/nature06272 Bouchard, M., Pothier, D., & Gauthier, S. (2008). Fire return intervals and tree species succession in the North Shore region of eastern Quebec. Canadian Journal of Forest Research, 38(6), 1621–1633. doi:10.1139/X07-201 76  Boykin, K. G., Kepner, W. G., Bradford, D. F., Guy, R. K., Kopp, D. A., Leimer, A. K., … Gergely, K. J. (2013). A national approach for mapping and quantifying habitat-based biodiversity metrics across multiple spatial scales. Ecological Indicators, 33, 139–147. doi:10.1016/j.ecolind.2012.11.005 Canty, M. J., & Nielsen, A. A. (2008). Automatic radiometric normalization of multitemporal satellite imagery with the iteratively re-weighted MAD transformation. Remote Sensing of Environment, 112(3), 1025–1036. Canty, M. J., Nielsen, A. A., & Schmidt, M. (2004). Automatic radiometric normalization of multitemporal satellite imagery. Remote Sensing of Environment, 91, 441–451. doi:10.1016/j.rse.2003.10.024 Carlson, T. N., & Ripley, D. A. (1997). On the relation between NDVI, fractional vegetation cover, and leaf area index. Remote Sensing of Environment, 62, 241–252. doi:10.1016/S0034-4257(97)00104-1 Certini, G. (2005). Effects of fire on properties of forest soils: A review. Oecologia, 143, 1–10. doi:10.1007/s00442-004-1788-8 Chambers, J. Q., Asner, G. P., Morton, D. C., Anderson, L. O., Saatchi, S. S., Esprito-Santo, F. D. B., … Souza, C. (2007). Regional ecosystem structure and function: ecological insights from remote sensing of tropical forests. Trends in Ecology and Evolution, 22, 414–423. doi:10.1016/j.tree.2007.05.001 Chavez, P. S. (1996). Image-based atmospheric corrections- revisited and improved. Photogrammetric Engineering and Remote Sensing, 62, 1025–1035. doi:0099-1112/96/6209-1025 Choi, H., & Bindschadler, R. (2004). Cloud detection in Landsat imagery of ice sheets using shadow matching technique and automatic normalized difference snow index threshold value decision. Remote Sensing of Environment, 91, 237–242. doi:10.1016/j.rse.2004.03.007 Chuvieco, E., Martín, M. P., & Palacios, A. (2002). Assessment of different spectral indices in the red-near-infrared spectral domain for burned land discrimination. International Journal of Remote Sensing, 23, 5103–5110. doi:10.1080/01431160210153129 Cilmate Canada. (2014). Daily Data. Retrieved May 01, 2014, from http://climate.weather.gc.ca/climateData/dailydata_e.html?timeframe=2&Prov=&StationID=50269&hlyRange=2012-05-07|2014-06-04&cmdB1=Go&Year=2012&Month=6&cmdB2=Go# Cohen, W. B., Kushla, J. D., Ripple, W. J., & Garman, S. L. (1996). An introduction to digital methods in remote sensing of forested ecosystems: Focus on the Pacific Northwest, USA. Environmental Management, 20(3), 421–435. doi:10.1007/BF01203849 77  Cohen, W. B., Spies, T. A., Alig, R. J., Oetter, D. R., Maiersperger, T. K., & Fiorella, M. (2002). Characterizing 23 Years (1972-95) of Stand Replacement Disturbance in Western Oregon Forests with Landsat Imagery. Ecosystems, 5(2), 122–137. doi:10.1007/s10021-001-0060-X Congalton, R. G., & Green, K. (1993). A practical look at the sources of confusion in error matrix generation. Photogrammetric Engineering & Remote Sensing, 59, 641–644. Cooper, M., Sarell, M., Bryan, A., & Chapman, K. (2011). No TitleVaseux – Bighorn National Wildlife Area Management Plan. Delta, BC. 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), 67–80. doi:10.1016/j.rse.2006.03.012 Coops, N. C., Wulder, M. A., & Iwanicka, D. (2009). Large area monitoring with a MODIS-based Disturbance Index (DI) sensitive to annual and seasonal variations. Remote Sensing of Environment, 113(6), 1250–1261. Coppin, P., Jonckheere, I., Nackaerts, K., Muys, B., & Lambin, E. (2004). Review ArticleDigital change detection methods in ecosystem monitoring: a review. International Journal of Remote Sensing, 25(9), 1565–1596. doi:10.1080/0143116031000101675 Cottam, G., & Curtis, J. T. (1956). The use of distance measures in phytosociological sampling. Ecology, 37, 451–460. doi:10.2307/1930167 Crist, E. P., & Cicone, R. C. (1984). A Physically-Based Transformation of Thematic Mapper Data---The TM Tasseled Cap. IEEE Transactions on Geoscience and Remote Sensing, GE-22. doi:10.1109/TGRS.1984.350619 Crist, E. P. and Kauth, R. J. (1986). The Tasseled Cap de-mystified [transformations of MSS and TM data]. Photogrammetric Engineering and Remote Sensing, 52, 81–86. Curran, P. J. (1988). The Semivariogram in Remote-Sensing - an Introduction. Remote Sensing of Environment, 24, 493–507. Da Silva Curiel, A., Boland, L., Cooksley, J., Bekhti, M., Stephens, P., Sun, W., & Sweeting, M. (2005). First results from the disaster monitoring constellation (DMC). In Acta Astronautica (Vol. 56, pp. 261–271). doi:10.1016/j.actaastro.2004.09.026 Davin, E. L., & de Noblet-Ducoudré, N. (2010). Climatic Impact of Global-Scale Deforestation: Radiative versus Nonradiative Processes. Journal of Climate, 23(1), 97–112. doi:10.1175/2009JCLI3102.1 De Groot, W. J., Landry, R., Kurz, W. A., Anderson, K. R., Englefield, P., Fraser, R. H., … Pritchard, J. M. (2007). Estimating direct carbon emissions from Canadian wildland fires. International Journal of Wildland Fire, 16(5), 593. doi:10.1071/WF06150 78  DeFries, R. S., Field, C. B., Fung, I., Collatz, G. J., & Bounoua, L. (1999). Combining satellite data and biogeochemical models to estimate global effects of human-induced land cover change on carbon emissions and primary productivity. Global Biogeochemical Cycles, 13(3), 803. doi:10.1029/1999GB900037 Douglass, L. L., Possingham, H. P., Carwardine, J., Klein, C. J., Roxburgh, S. H., Russell-Smith, J., & Wilson, K. A. (2011). The effect of carbon credits on savanna land management and priorities for biodiversity conservation. PLoS ONE, 6(9). Eidenshink, J., Schwind, B., Brewer, K., Zhu, Z.-L., Quayle, B., & Howard, S. (2007). A Project for Monitoring Trends in Burn Severity. Fire Ecology, 3, 3–21. doi:10.4996/fireecology.0301003 Epting, J., Verbyla, D., & Sorbel, B. (2005). Evaluation of remotely sensed indices for assessing burn severity in interior Alaska using Landsat TM and ETM+. Remote Sensing of Environment, 96, 328–339. doi:10.1016/j.rse.2005.03.002 ESRI. (2013). ArcGIS Desktop: Release 10.2. Redlands CA. Falkowski, M. J., Gessler, P. E., Morgan, P., Hudak, A. T., & Smith, A. M. S. (2005). Characterizing and mapping forest fire fuels using ASTER imagery and gradient modeling. Forest Ecology and Management, 217(2-3), 129–146. FAO. (2009). State of the world’s forests 2009. SciencesNew York (p. 152). New York. doi:10.1209/epl/i2005-10515-2 Fensholt, R., Anyamba, A., Huber, S., Proud, S. R., Tucker, C. J., Small, J., … Shisanya, C. (2011). Analysing the advantages of high temporal resolution geostationary msg seviri data compared to polar operational environmental satellite data for land surface monitoring in Africa. International Journal of Applied Earth Observation and Geoinformation, 13(5), 721–729. Filella, I., & Penuelas, J. (1994). The red edge position and shape as indicators of plant chlorophyll content, biomass and hydric status. International Journal of Remote Sensing, 15, 1459–1470. doi:10.1080/01431169408954177 Fleming, M. D. (1988). An integrated approach for automated cover-type mapping of large inaccessible areas in Alaska. Photogrammetric Engineering and Remote Sensing, 55, 357–362. FLNRO. (2014). BECdb: Biogeoclimatic Ecosystem Classification Codes and Names, Version 8, 2012. Victoria, BC. Retrieved from http://www.for.gov.bc.ca/hre/becweb/resources/codes-standards/standards-becdb.html   Foody, G. M. (2010). Assessing the accuracy of land cover change with imperfect ground reference data. Remote Sensing of Environment, 114, 2271–2285. doi:10.1016/j.rse.2010.05.003 79  Foster, D. R. (1992). Land-use history (1730-1990) and vegetation in central New England , USA. Journal of Ecology, 80, 753–771. doi:10.2307/2260864 Franke, J., Navratil, P., Keuck, V., Peterson, K., & Siegert, F. (2012). Monitoring Fire and Selective Logging Activities in Tropical Peat Swamp Forests. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 1–10. doi:10.1109/JSTARS.2012.2202638 Franklin, J. F., Lindenmayer, D. B., Thornburgh, D. A., Pelt, R. Van, Chen, J., Spies, T. A., … Bible, K. (2002). Disturbances and structural development of natural forest ecosystems with silvicultural implications, using Douglas-fir forests as an example. Forest Ecology and Management, 155(1-3), 399–423. doi:10.1016/S0378-1127(01)00575-8 Frolking, S., Palace, M. W., Clark, D. B., Chambers, J. Q., Shugart, H. H., & Hurtt, G. C. (2009). Forest disturbance and recovery: A general review in the context of spaceborne remote sensing of impacts on aboveground biomass and canopy structure. Journal of Geophysical Research, 114. doi:10.1029/2008JG000911 Fule, P. Z., Swetnam, T. W., Brown, P. M., Falk, D. A., Peterson, D. L., Allen, C. D., … Taylor, A. (2013). Unsupported inferences of high-severity fire in historical dry forests of the western United States: Response to Williams and Baker. Global Ecology and Biogeography. doi:10.1111/geb.12136 Gao, B. C. (1996). NDWI - A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment, 58, 257–266. doi:10.1016/S0034-4257(96)00067-3 Gao, X., Huete, A. R., Ni, W., & Miura, T. (2000). Optical-biophysical relationships of vegetation spectra without background contamination. Remote Sensing of Environment, 74(3), 609–620. García, M. J. L., & Caselles, V. (1991). Mapping burns and natural reforestation using thematic Mapper data. Geocarto International, 6, 31–37. doi:10.1080/10106049109354290 Gil Docampo, M. L., Armesto Gonz lez, J.,  Ca as Guerrero, I. (2005). 3-D expression of relief in Landsat Thematic Mapper and SPOT P images. Optical Engineering, 44, 047003. doi:10.1117/1.1883699 Gong, P., & Xu, B. (2003). Remote Sensing of Forests Over Time. In M. A. Wulder & S. E. Franklin (Eds.), Remote sensing of forest environments Concepts and case studies (pp. 301–333). Kluwer Academic Pub. Retrieved from http://books.google.com/books?hl=en&lr=&id=1JVdXQU9qmYC&pgis=1 Gonzalez, J. S. (1990). Wood density of Canadian tree species. Forestry Canada, Northwest Region, Information Report NOR-X-315. Northern Forestry Centre, Edmonton, Alberta. Retrieved from http://cfs.nrcan.ca/pubwarehouse/pdfs/11744.pdf; retrieved February 3, 2014 80  Goodwin, N. R., Collett, L. J., Denham, R. J., Flood, N., & Tindall, D. (2013). Cloud and cloud shadow screening across Queensland, Australia: An automated method for Landsat TM/ETM+ time series. Remote Sensing of Environment, 134, 50–65. doi:10.1016/j.rse.2013.02.019 Grainger, A., & Obersteiner, M. (2011). A framework for structuring the global forest monitoring landscape in the REDD+ era. Environmental Science and Policy, 14, 127–139. doi:10.1016/j.envsci.2010.10.006 Gupta, A., L??vbrand, E., Turnhout, E., & Vijge, M. J. (2012). In pursuit of carbon accountability: The politics of REDD+ measuring, reporting and verification systems. Current Opinion in Environmental Sustainability, 4(6), 726–731. Gutschick, V. P., & BassiriRad, H. (2003). Extreme events as shaping physiology, ecology, and evolution of plants: toward a unified definition and evaluation of their consequences. New Phytologist, 160(1), 21–42. doi:10.1046/j.1469-8137.2003.00866.x Habeck, J. R., & Mutch, R. W. (1973). Fire dependent forests in the northern Rocky Mountains. Quaternary Research (Orlando), 3, 408–424. Retrieved from <Go to ISI>://BIOSIS:PREV197410019700 Hantson, S., Padilla, M., Corti, D., & Chuvieco, E. (2013). Strengths and weaknesses of MODIS hotspots to characterize global fire occurrence. Remote Sensing of Environment, 131, 152–159. Hardy, C. C. (2005). Wildland fire hazard and risk: Problems, definitions, and context. In Forest Ecology and Management (Vol. 211, pp. 73–82). Harrington, M. G. (1987). Predicting Reduction of Natural Fuels by Prescribed Burning Under Ponderosa Pine in Southt!astern Arizona (pp. 1–4). Rocky Mountain Forest and Range Experiment Station. Retrieved from http://www.fs.fed.us/rm/pubs/rmrs_gtr292/rm_rn472.pdf Hausamann, D., Zirnig, W., Schreier, G., & Strobl, P. (2005). Monitoring of gas pipelines – a civil UAV application. Aircraft Engineering and Aerospace Technology, 77, 352–360. doi:10.1108/00022660510617077 Healey, S. P., Cohen, W. B., Zhiqiang, Y., & Krankina, O. N. (2005). Comparison of Tasseled Cap-based Landsat data structures for use in forest disturbance detection. Remote Sensing of Environment, 97(3), 301–310. doi:10.1016/j.rse.2005.05.009 Hines, S. J., Heath, L. S., & Birdsey, R. A. (2010). An annotated bibliography of scientific literature on managing forests for carbon benefits (p. 49). Hladik, C., Schalles, J., & Alber, M. (2013). Salt marsh elevation and habitat mapping using hyperspectral and LIDAR data. Remote Sensing of Environment, 139, 318–330. doi:10.1016/j.rse.2013.08.003 81  Horne, J. H. (2003). A tasseled cap transformation for IKONOS-2 images. In ASPRS 2003 Annual Conference (pp. 60–70). Anchorage, Alaska. Huete, A. (1997). A comparison of vegetation indices over a global set of TM images for EOS-MODIS. Remote Sensing of Environment, 59(3), 440–451. doi:10.1016/S0034-4257(96)00112-5 Huete, A. . (1988). A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment, 25(3), 295–309. doi:10.1016/0034-4257(88)90106-X Huete, A., Didan, K., Miura, T., Rodriguez, E. P., Gao, X., & Ferreira, L. G. (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83, 195–213. doi:10.1016/S0034-4257(02)00096-2 Hurteau, M. D., Koch, G. W., & Hungate, B. A. (2008). Carbon protection and fire risk reduction: toward a full accounting of forest carbon offsets. Frontiers in Ecology and the Environment, 6, 493–498. doi:10.1890/070187 Ice, G. G., Neary, D. G., & Adams, P. W. (2004). Effects of wildfire on soils and watershed processes. Journal of Forestry, 102, 16–20. Jackson, R. D., & Huete, A. R. (1991). Interpreting vegetation indices. Preventive Veterinary Medicine, 11(3-4), 185–200. doi:10.1016/S0167-5877(05)80004-2 Jain, T. B., & Graham, R. T. (2004). Is forest structure related to fire severity? Yes, no, and maybe: Methods and insights in quantifying the answer. In Silviculture in special places: Proceedings of the National Silviculture Workshop (pp. 217–234). Johansen, K., Coops, N. C., Gergel, S. E., & Stange, Y. (2007). Application of high spatial resolution satellite imagery for riparian and forest ecosystem classification. Remote Sensing of Environment, 110(1), 29–44. Johnson, R. D., & Kasischke, E. S. (1998). Change vector analysis: A technique for the multispectral monitoring of land cover and condition. International Journal of Remote Sensing, 19(3), 411–426. doi:10.1080/014311698216062 Joseph, S., Murthy, M. S. R., & Thomas, a. P. (2010). The progress on remote sensing technology in identifying tropical forest degradation: a synthesis of the present knowledge and future perspectives. Environmental Earth Sciences, 64(3), 731–741. doi:10.1007/s12665-010-0893-8 Kachmar, M., & Sanchez-Azofeifa, G. A. (2006). Detection of post-fire residuals using high- and medium-resolution satellite imagery. Forestry Chronicle, 82, 177–186. Karabashev, G. S., & Evdoshenko, M. A. (2013). On the informativeness of surveys of marine coastal areas using the ocean color scanners. Oceanology, 53(6), 668–675. 82  Kasischke, E. S.,  Turetsky, M. R. (2006). Correction to “Recent changes in the fire regime across the North American boreal region-Spatial and temporal patterns of burning across Canada and Alaska.” Geophysical Research Letters, 33. doi:10.1029/2006GL026946 Kasischke, E. S., Turetsky, M. R., Ottmar, R. D., French, N. H. F., Hoy, E. E., & Kane, E. S. (2008). Evaluation of the composite burn index for assessing fire severity in Alaskan black spruce forests. International Journal of Wildland Fire, 17, 515–526. doi:10.1071/WF08002 Kauth, R., & G Thomas. (1976). The tasseled cap - a graphical description of spectral-temporal development of agricultural crops as seen in Landsat. In LARS Symposia (p. 159). Keane, R. E., Burgan, R., & van Wagtendonk, J. (2001). Mapping wildland fuels for fire management across multiple scales: Integrating remote sensing, GIS, and biophysical modeling. International Journal of Wildland Fire, 10(4), 301–319. doi:doi:10.1071/WF01028 Kerr, J. T., & Ostrovsky, M. (2003). From space to species: ecological applications for remote sensing. Trends in Ecology & Evolution, 18(6), 299–305. doi:10.1016/S0169-5347(03)00071-5 Key, C. H. (2006). Ecological and Sampling Constraints on Defining Landscape Fire Severity. Fire Ecology, 2, 34–59. doi:10.4996/fireecology.0202034 Key, C. H., & Benson, N. C. (2006). Landscape Assessment ( LA ) Sampling and Analysis Methods. In D. Lutes, R. Keane, J. Carati, C. Key, N. Benson, & L. Gangi (Eds.), FIREMON: Fire Effects Monitoring and Inventory System (pp. LA–1–55). USDA Forest Service Rocky Mountain Research Station General Technical Report RMRS-GTR-164-CD. Kimes, D. S., Ranson, K. J., Sun, G., & Blair, J. B. (2006). Predicting lidar measured forest vertical structure from multi-angle spectral data. Remote Sensing of Environment, 100, 503–511. doi:10.1016/j.rse.2005.11.004 Kovalskyy, V., & Roy, D. P. (2013). The global availability of Landsat 5 TM and Landsat 7 ETM+ land surface observations and implications for global 30m Landsat data product generation. Remote Sensing of Environment, 130, 280–293. Kurz, W. A., & Apps, M. J. (1999). A 70-year retrospective analysis of carbon fluxes in the Canadian forest sector. Ecological Applications, 9(2), 526–547. doi:10.1890/1051-0761(1999)009[0526:AYRAOC]2.0.CO;2 Landres, P. B., Morgan, P., & Swanson, F. J. (1999). Overview of the use of natural variability concepts in managing ecological systems. Ecological Applications, 9(4), 1179–1188. doi:10.1890/1051-0761(1999)009[1179:OOTUON]2.0.CO;2 83  Larson, A. J., & Franklin, J. F. (2005). Patterns of conifer tree regeneration following an autumn wildfire event in the western Oregon Cascade Range, USA. Forest Ecology and Management, 218, 25–36. doi:10.1016/j.foreco.2005.07.015 Lefsky, M. A., Cohen, W. B., Parker, G. G., & HardingG, D. J. (2002). Lidar Remote Sensing for Ecosystem Studies. BioScience, 52, 19. doi:10.1641/0006-3568(2002)052[0019:LRSFES]2.0.CO;2 Lentile, L. B., Holden, Z. A., Smith, A. M. S., Falkowski, M. J., Hudak, A. T., Morgan, P., … Benson, N. C. (2006). Remote sensing techniques to assess active fire characteristics and post-fire effects. International Journal of Wildland Fire, 15(3), 319. Loboda, T. V., Zhang, Z., O’Neal, K. J., Sun, G., Csiszar, I. A., Shugart, H. H., & Sherman, N. J. (2012). Reconstructing disturbance history using satellite-based assessment of the distribution of land cover in the Russian Far East. Remote Sensing of Environment, 118, 241–248. Lu, D., Mausel, P., Brondízio, E., & Moran, E. (2004). Change detection techniques. International Journal of Remote Sensing, 25(12), 2365–2401. doi:10.1080/0143116031000139863 Lyon, J. G., Yuan, D., Lunetta, R. S., & Elvidge, C. D. (1998). A Change Detection Experiment Using Vegetation Indices. Photogrammetric Engineering & Remote Sensing, 64(2), 143–150. Retrieved from http://www.asprs.org/publications/pers/98journal/february/1998_feb_143-150.pdf Ma, H., Song, J., Wang, J., Xiao, Z., & Fu, Z. (2014). Improvement of spatially continuous forest LAI retrieval by integration of discrete airborne LiDAR and remote sensing multi-angle optical data. Agricultural and Forest Meteorology, 189-190, 60–70. doi:10.1016/j.agrformet.2014.01.009 Mahanty, S., Suich, H., & Tacconi, L. (2013). Access and benefits in payments for environmental services and implications for REDD+: Lessons from seven PES schemes. Land Use Policy, 31, 38–47. Maness, H., Kushner, P. J., & Fung, I. (2013). Summertime climate response to mountain pine beetle disturbance in British Columbia. Nature Geoscience, 6(1), 65–70. doi:10.1038/ngeo1642 Marsh, W. M., & Grossa, J. J. (2002). Environmental geography: science, land use and earth systems (p. 426). New York: John Wiley and Sons. Maruyama, M., & Morioka, N. (1998). The Impact of Deforestation in the Brazillian Amazonia - The Indigenous People of Rondonia State. Japan Society of Forest Planning, 4, 71–75. Masek, J. G., Goward, S. N., Kennedy, R. E., Cohen, W. B., Moisen, G. G., Schleeweis, K., & Huang, C. (2013). United States Forest Disturbance Trends Observed Using Landsat 84  Time Series. Ecosystems, 16, 1087–1104. Retrieved from http://link.springer.com/article/10.1007/s10021-013-9669-9/fulltext.html Masek, J. G., Huang, C., Wolfe, R., Cohen, W., Hall, F., Kutler, J., & Nelson, P. (2008). North American forest disturbance mapped from a decadal Landsat record. Remote Sensing of Environment, 112, 2914–2926. doi:10.1016/j.rse.2008.02.010 Meslem, A., Yamakazi, F., & Maruyama, Y. (2011). Accurate evaluation of building damage in the 2003 Boumerdes, Algeria earthquake from Quickbird satellite images. Journal of Earthquake and Tsunami, 05(01), 1–18. doi:10.1142/S1793431111001029 Miettinen, J., Hyer, E., Chia, A. S., Kwoh, L. K., & Liew, S. C. (2013). Detection of vegetation fires and burnt areas by remote sensing in insular Southeast Asian conditions: current status of knowledge and future challenges. International Journal of Remote Sensing, 34(12), 4344–4366. Retrieved from http://www.tandfonline.com/doi/pdf/10.1080/01431161.2013.777489 Miettinen, J., & Liew, S. C. (2008). Comparison of multitemporal compositing methods for burnt area detection in Southeast Asian conditions. International Journal of Remote Sensing, 29, 1075–1092. doi:10.1080/01431160701281031 Miles, L., & Kapos, V. (2008). Reducing greenhouse gas emissions from deforestation and forest degradation: global land-use implications. Science, 320(5882), 1454–1455. doi:10.1126/science.1155358 Miller, J. D., Danzer, S. R., Watts, J. M., Stone, S., & Yool, S. R. (2003). Cluster analysis of structural stage classes to map wildland fuels in a Madrean ecosystem. Journal of Environmental Management, 68(3), 239–252. Miller, J. D., Safford, H. D., Crimmins, M., & Thode, A. E. (2009). Quantitative evidence for increasing forest fire severity in the Sierra Nevada and southern Cascade Mountains, California and Nevada, USA. Ecosystems, 12, 16–32. doi:10.1007/s10021-008-9201-9 Moore, T. S., Dowell, M. D., & Franz, B. A. (2012). Detection of coccolithophore blooms in ocean color satellite imagery: A generalized approach for use with multiple sensors. Remote Sensing of Environment, 117, 249–263. Mottishaw, J. (2013). Prescribed fire burn plan. Penticton, British Columbia. Muller, R. N., & Liu, Y. (1991). Coarse woody debris in an old-growth deciduous forest on the Cumberland Plateau, southeastern Kentucky. Canadian Journal of Forest Research, 21, 1567–1572. doi:10.1139/x91-218 Mumby, P. J., Green, E. P., Edwards, A. J., & Clark, C. D. (1999). The cost-effectiveness of remote sensing for tropical coastal resources assessment and management. Journal of Environmental Management, 55(3), 157–166. doi:10.1006/jema.1998.0255 85  Naimi, B. (2013). Package “usdm.” CRAN. Retrieved from http://cran.r-project.org/web/packages/usdm/usdm.pdf Naimi, B., Skidmore, A. K., Groen, T. A., & Hamm, N. A. S. (2011). Spatial autocorrelation in predictors reduces the impact of positional uncertainty in occurrence data on species distribution modelling. Journal of Biogeography, 38(8), 1497–1509. Retrieved from http://dx.doi.org/10.1111/j.1365-2699.2011.02523.x Naughton, D., Brunn, A., Czapla-Myers, J., Douglass, S., Thiele, M., Weichelt, H., & Oxfort, M. (2011). Absolute radiometric calibration of the RapidEye multispectral imager using the reflectance-based vicarious calibration method. Journal of Applied Remote Sensing, 5(1), 053544. doi:10.1117/1.3613950 Nelson, R. F. (1983). Detecting forest canopy change due to insect activity using Landsat MSS. Photogrammetric Engineering & Remote Sensing, 49, 1303–1314. Odion, D. C., Hanson, C. T., Arsenault, A., Baker, W. L., DellaSala, D. A., Hutto, R. L., … Williams, M. A. (2014). Examining historical and current mixed-severity fire regimes in ponderosa pine and mixed-conifer forests of western North America. PLoS ONE, 9. doi:10.1371/journal.pone.0087852 Oliver, C. D. (1980). Forest development in North America following major disturbances. Forest Ecology and Management, 3, 153–168. doi:10.1016/0378-1127(80)90013-4 Palmer, M. a., Ambrose, R. F., & Poff, N. L. R. (1997). Ecological theory and community restoration ecology. Restoration Ecology, 5, 291–300. doi:10.1046/j.1526-100X.1997.00543.x Pan, Y., Birdsey, R. A., Fang, J., Houghton, R., Kauppi, P. E., Kurz, W. A., … Hayes, D. (2011). A large and persistent carbon sink in the world’s forests. Science (New York, N.Y.), 333, 988–993. doi:10.1126/science.1201609 Pearson, A. F. (2010). Natural and logging disturbances in the temperate rain forests of the Central Coast, British Columbia. Canadian Journal of Forest Research, 40(10), 1970–1984. doi:10.1139/X10-137 Penman, J., Gytarsky, M., Hiraishi, T., Krug, T., Kruger, D., Pipatti, R., … Ngara, T. (2003). Intergovernmental Panel on Climate Change Good Practice Guidance for Land Use , Land-Use Change and Forestry Edited by. (Array, Ed.)Practice (Vol. 177, p. 576). Institute for Global Environmental Strategies (IGES) for the IPCC. doi:citeulike-article-id:1260638 Perry, D. A., Hessburg, P. F., Skinner, C. N., Spies, T. A., Stephens, S. L., Taylor, A. H., … Riegel, G. (2011). The ecology of mixed severity fire regimes in Washington, Oregon, and Northern California. Forest Ecology and Management, 262, 703–717. doi:10.1016/j.foreco.2011.05.004 86  Pickett, S. T. A., & White, P. S. (1985). The ecology of natural disturbance and patch dynamics. (S. T. A. Pickett & P. S. White, Eds.)The Ecology of Natural Disturbance and Patch DynamicsAcademic Press (Vol. 1, p. 472). Academic Press. doi:10.2307/2403105 Pistorius, T. (2012). From RED to REDD+: The evolution of a forest-based mitigation approach for developing countries. Current Opinion in Environmental Sustainability, 4(6), 638–645. Pohl, M., Graf, F., Buttler, A., & Rixen, C. (2012). The relationship between plant species richness and soil aggregate stability can depend on disturbance. Plant and Soil, 355(1-2), 87–102. doi:10.1007/s11104-011-1083-5 Potapov, P., Turubanova, S., & Hansen, M. C. (2011). Regional-scale boreal forest cover and change mapping using Landsat data composites for European Russia. Remote Sensing of Environment, 115(2), 548–561. Power, K.,  Gillis, M. (2006). Canada’s Forest Inventory 2001. Information Report. Ramoelo, A., Skidmore, A. K., Cho, M. A., Schlerf, M., Mathieu, R., Heitk, & Nig, I. M. A. (2012). Regional estimation of savanna grass nitrogen using the red-edge band of the spaceborne rapideye sensor. International Journal of Applied Earth Observation and Geoinformation, 19, 151–162. doi:10.1016/j.jag.2012.05.009 Rogan, J., & Miller, J. (2006). Integrating GIS and Remotely Sensed Data for Mapping Forest Disturbance and Change. Understanding Forest Disturbance and Spatial Pattern: Remote Sensing and GIS Approaches, 133–170. Samonil, P., Kral, K., & Hort, L. (2010). The role of tree uprooting in soil formation: A critical literature review. Geoderma, 157(3-4), 65–79. Sano, E. E., Ferreira, L. G., Asner, G. P., & Steinke, E. T. (2007). Spatial and temporal probabilities of obtaining cloud‐free Landsat images over the Brazilian tropical savanna. International Journal of Remote Sensing, 28(12), 2739–2752. doi:10.1080/01431160600981517 Scheffer, M., Carpenter, S., Foley, J. A., Folke, C., & Walker, B. (2001). Catastrophic shifts in ecosystems. Nature, 413(6856), 591–596. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/11595939 Schimel, D., & Baker, D. (2002). Carbon cycle: the wildfire factor. Nature, 420, 29–30. doi:10.1038/420029a Schoennagel, T., Veblem, T., & Romme, W. (2004). The Interaction of Fire, Fuels, and Climate across Rocky Mountain Forests. BioScience, 54, 661. doi:10.1641/0006-3568(2004)054[0661:TIOFFA]2.0.CO;2 Shimabukuro, Y. E., Duarte, V., Anderson, L. O., Valeriano, D. M., Arai, E., Freitas, R. M. De, … Moreira, M. A. (2006). Near real time detection of deforestation in the Brazilian 87  Amazon using MODIS imagery Detecção em tempo quase real de desflorestamento na Amazônia brasileira utilizando imagens MODIS. Revista Ambi-Agua, 1(12), 37–47. doi:10.4136/1980-993X Sidle, R. C. (1992). A theoretical model of the effects of timber harvesting on slope stability. Water Resources Research, 28(7), 1897–1910. Simard, D. G., Fyles, J. W., Paré, D., & Nguyen, T. (2001). Impacts of clearcut harvesting and wildfire on soil nutrient status in the Quebec boreal forest. Canadian Journal of Soil Science, 81, 229–237. doi:10.4141/S00-028 Sims, D. A., & Gamon, J. A. (2003). Estimation of vegetation water content and photosynthetic tissue area from spectral reflectance: A comparison of indices based on liquid water and chlorophyll absorption features. Remote Sensing of Environment, 84, 526–537. doi:10.1016/S0034-4257(02)00151-7 Singh, A. (1989). Review Article Digital change detection techniques using remotely-sensed data. International Journal of Remote Sensing, 10(6), 989–1003. doi:10.1080/01431168908903939 Singh, A. (1989). Review article-digital change detection techniques using remotely-sensed data. International Journal of Remote Sensing, 10, 989–1003. Smith, A. M. S., & Hudak, A. T. (2005). Estimating combustion of large downed woody debris from residual white ash. International Journal of Wildland Fire, 14, 245–248. doi:10.1071/WF05011 Sunar, F., & Özkan, C. (2001). Forest fire analysis with remote sensing data. International Journal of Remote Sensing, 22, 2265–2277. doi:10.1080/014311601300229818 Sunderlin, W. D., Belcher, B., Santoso, L., Angelsen, A., Burgers, P., Nasi, R., & Wunder, S. (2005). Livelihoods, forests, and conservation in developing countries: An overview. World Development, 33(9 SPEC. ISS.), 1383–1402. Tack, F., Buyuksalih, G., & Goossens, R. (2012). 3D building reconstruction based on given ground plan information and surface models extracted from spaceborne imagery. ISPRS Journal of Photogrammetry and Remote Sensing, 67(1), 52–64. Tapsall, B., Milenov, P., & Tasdemir, K. (2010). ANALYSIS OF RAPIDEYE IMAGERY FOR ANNUAL LANDCOVER MAPPING AS AN AID TO EUROPEAN UNION ( EU ) COMMON AGRICULTURAL POLICY. In 100 Years ISPRS - Advancing Remote Sensing Science (Vol. XXXVIII, pp. 568–573). Retrieved from http://www.isprs.org/proceedings/xxxviii/part7/b/pdf/568_XXXVIII-part7B.pdf\nhttp://www.isprs.org/documents/archive/isprs100vienna/tc-vii-symposium/index.html 88  Townsend, J. R. G., & Justice, C. O. (1988). Selecting the spatial resolution of satellite sensors required for global monitoring of land transformations. International Journal of Remote Sensing, 9(2), 187–236. Tucker, C. J. (1979). Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment, 8(2), 127–150. Turner, B. L., Lambin, E. F., & Reenberg, A. (2007). The emergence of land change science for global environmental change and sustainability. Proceedings of the National Academy of Sciences of the United States of America, 104(52), 20666–20671. Turner, M. G. (2010). Disturbance and landscape dynamics in a changing world. Ecology, 91, 2833–2849. doi:10.1890/10-0097.1 Turner, M. G., Hargrove, W. W., Gardner, R. H., & Romme, W. H. (1994). Effects of fire on landscape heterogeneity in Yellowstone National Park, Wyoming. Journal of Vegetation Science, 5(5), 731–742. doi:10.2307/3235886 Tyc, G., Tulip, J., Schulten, D., Krischke, M., & Oxfort, M. (2005). The RapidEye mission design. Acta Astronautica, 56(1-2), 213–219. Vaillant, N. M., Reiner, A. L., & Noonan-Wright, E. K. (2013). Prescribed fire effects on field-derived and simulated forest carbon stocks over time. Forest Ecology and Management, 310, 711–719. doi:10.1016/j.foreco.2013.09.016 Van Wagner, C. E. (1968). The line intersect method in forest fuel sampling. Forest Science, 14, 20–26. Van Wagtendonk, J. W., Root, R. R., & Key, C. H. (2004). Comparison of AVIRIS and Landsat ETM+ detection capabilities for burn severity. Remote Sensing of Environment, 92, 397–408. doi:10.1016/j.rse.2003.12.015 Vedeld, P., Angelsen, A., Bojo, J., Sjaastad, E., & Kobugabe Berg, G. (2007). Forest environmental incomes and the rural poor. Forest Policy and Economics, 9(7), 869–879. Verbesselt, J., Zeileis, A., & Herold, M. (2012). Near real-time disturbance detection using satellite image time series. Remote Sensing of Environment, 123, 98–108. doi:10.1016/j.rse.2012.02.022 Vitousek, P. M., Mooney, H. A., Lubchenco, J., & Melillo, J. M. (1997). Human Domination of Earth’s Ecosystems. Science, 277(5325), 494–499. doi:10.1126/science.277.5325.494 Wallerman, J., Fransson, J. E. S., Bohlin, J., Reese, H., & Olsson, H. (2010). Forest mapping using 3D data from SPOT-5 HRS and Z/I DMC. Geoscience and Remote Sensing Symposium (IGARSS), 2010 IEEE International. doi:10.1109/IGARSS.2010.5653818 89  Whitman, E., Rapaport, E., & Sherren, K. (2013). Modeling fire susceptibility to delineate wildland-urban interface for municipal-scale fire risk management. Environmental Management, 52(6), 1427–39. doi:10.1007/s00267-013-0159-9 Wiedinmyer, C., & Hurteau, M. D. (2010). Prescribed fire as a means of reducing forest carbon emissions in the western United States. Environmental Science and Technology, 44, 1926–1932. doi:10.1021/es902455e Wilson, E. H., & Sader, S. A. (2002). Detection of forest harvest type using multiple dates of Landsat TM imagery. Remote Sensing of Environment, 80(3), 385–396. Wulder, M. A., & Franklin, S. E. (2007). Understanding Forest Disturbance And Spatial Pattern: Remote Sensing And GIS Approaches. Program, 1–29. Retrieved from http://www.loc.gov/catdir/toc/ecip0610/2006007999.html Wulder, M. A., White, J. C., Alvarez, F., Han, T., Rogan, J., & Hawkes, B. (2009). Characterizing boreal forest wildfire with multi-temporal Landsat and LIDAR data. Remote Sensing of Environment, 113(7), 1540–1555. doi:10.1016/j.rse.2009.03.004 Wulder, M. A., White, J. C., Goward, S. N., Masek, J. G., Irons, J. R., Herold, M., … Woodcock, C. E. (2008). Landsat continuity: Issues and opportunities for land cover monitoring. Remote Sensing of Environment, 112(3), 955–969. doi:10.1016/j.rse.2007.07.004 Xie, Y., Sha, Z., & Yu, M. (2008). Remote sensing imagery in vegetation mapping: a review. Journal of Plant Ecology, 1(1), 9–23. doi:10.1093/jpe/rtm005 Yarbrough, L. D., Easson, G., & Kuszmaul, J. S. (2005). QUICKBIRD 2 Tasseled Cap Transform Coefficients: A Comparison of Derivation Methods. In Pecora 16: Global Priorities in Land Remote Sensing”. Zhang, X., Kondragunta, S., Ram, J., Schmidt, C., & Huang, H.-C. (2012). Near-real-time global biomass burning emissions product from geostationary satellite constellation. Journal of Geophysical Research, 117. doi:10.1029/2012JD017459  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0167248/manifest

Comment

Related Items