Articles
High Spatial Resolution Remotely Sensed Data for Ecosystem Characterization MICHAEL A. WULDER, RONALD J. HALL, NICHOLAS C. COOPS, AND STEVEN E. FRANKLIN
Characterization of ecosystem structure, diversity, and function is increasingly desired at finer spatial and temporal scales than have been derived in the past. Many ecological applications require detailed data representing large spatial extents, but these data are often unavailable or are impractical to gather using field-based techniques. Remote sensing offers an option for collecting data that can represent broad spatial extents with detailed attribute characterizations. Remotely sensed data are also appropriate for use in studies across spatial scales, in conjunction with fieldcollected data. This article presents the pertinent technical aspects of remote sensing for images at high spatial resolution (i.e., with a pixel size of 16 square meters or less), existing and future options for the processing and analysis of remotely sensed data, and attributes that can be estimated with these data for forest ecosystems. Keywords: remote sensing, forest ecology, biodiversity, general ecology, landscape ecology
R
emotely sensed data have been employed for the characterization of ecologically important variables from local through global contexts. These data may be used to generate a wide range of estimates that are valuable to ecologists, including information on land cover, vegetation cover, habitat, forest structure, and forest function (Kerr and Ostrovsky 2003), and to track changes in these variables. Recent technological developments in remote sensing have resulted in new capabilities for data capture and data processing, making it possible to generate and analyze digital images at high spatial resolution (fine grain, defined here as a pixel size of 16 square meters [m2] or less). A wide variety of options exists for using data processing and data analysis to estimate a range of ecologically important attributes. For instance, in the characterization of vegetation, applications have been developed for the estimation of stand structural attributes and leaf area. These applications have been developed using airborne sensors, and the lessons learned are currently being applied to spaceborne data at high spatial resolution. For example, it is now possible to identify and map individual trees and groups of trees over large areas, or as part of a strategy for forest sampling. The information that may be generated from remotely sensed data is directly linked to the sensor and to the related characteristics of the images it produces: spatial resolution (pixel size; table 1), spectral resolution (wavelength ranges utilized), temporal resolution (when and how often images are collected), and spatial extent (ground area represented)
(Turner W et al. 2003). Thus, the most appropriate remotely sensed data may be selected for a given application (Lefsky and Cohen 2003). The sensor and image characteristics of the MODIS (Moderate Resolution Imaging Spectroradiometer) satellite, for instance, allow for regular monitoring of global vegetation productivity (Running et al. 2004) but not for spatially detailed characterization of vegetation. Rather, information at the national scale is commonly generated using remotely sensed data at coarse spatial resolution for attributes such as fractional cover, land cover and change, fire, leaf area, and productivity (see Cihlar and colleagues for overviews of methods [2003a] and applications [2003b]). In many cases, information generated to represent ecological phenomena at the global scale is too general to meet regional or local objectives. Sensors such as the Landsat series (see Cohen and
Michael A. Wulder (e-mail:
[email protected]) is a research scientist at the Canadian Forest Service, Pacific Forestry Centre, Natural Resources Canada, Victoria, British Columbia V8Z 1M5, Canada. Ronald J. Hall is a remote sensing inventory specialist at the Canadian Forest Service, Northern Forestry Centre, Natural Resources Canada, Edmonton, Alberta T6H 3S5, Canada. Nicholas C. Coops is an associate professor in the Forest Resources Management Department, University of British Columbia, Vancouver, British Columbia V6T 1Z4, Canada. Steven E. Franklin is vice president of research and a professor in the Department of Geography at the University of Saskatchewan, Saskatoon, Saskatchewan S7N 5A2, Canada. © 2004 American Institute of Biological Sciences.
June 2004 / Vol. 54 No. 6 BioScience 511
Articles Table 1. Ranges of spatial resolution associated with different instruments or photographic scales and corresponding levels of expected plant recognition. Type of instrument or photographic scale
Approximate range of spatial resolution (meters)
General level of plant discrimination
Low-resolution satellite images
1000 (AVHRR) 500 (MODIS)
Broad land-cover patterns (regional to global mapping)
Medium-resolution satellite images
30 (Landsat) 20 (SPOT-4 multispectral) 10 (SPOT-4 panchromatic)
Separation of extensive masses of evergreen versus deciduous forests (stand-level characteristics)
High-resolution satellite images (e.g., IKONOS)
> 1 (panchromatic) > 4 (multispectral)
Recognition of large individual trees and broad vegetative types
Airborne multispectral scanners
> 0.3
Initial identification of large individual trees and stand-level characteristics
Airborne video
> 0.04
Identification of individual trees and large shrubs
Digital frame camera
> 0.04
Identification of individual trees and large shrubs
1:25,000 to 1:100,000
0.31 to 1.24a
Recognition of large individual trees and broad vegetative types
1:10,000 to 1:25,000
0.12 to 0.31
Direct identification of major cover types and species occurring in pure stands
1:2500 to 1:10,000
0.026 to 0.12
Identification of individual trees and large shrubs
1:500 to 1:2500
0.001 to 0.026
Identification of individual range plants and grassland types
AVHRR, Advanced Very High Resolution Radiometer; MODIS, Moderate Resolution Imaging Spectroradiometer; SPOT-4, Systeme Pour l’Observation de la Terre. a. Based on a typical aerial film and camera configuration using a 150-millimeter lens. Source: Wulder (1998).
Goward 2004), which gather images at higher spatial resolution, have a long history of use for spatial and temporal scaling in ecology. With a pixel size of 30 by 30 m and an image extent of 185 by 185 kilometers (km), the Landsat systems have generated data that are ideal for characterizing attributes at the landscape to regional level. For greater spatial detail than that offered by Landsat, data at higher spatial resolution are often sought (table 1). While satellite-based data at high spatial resolution are now available, forest inventory data, developed by generalizing the characteristics of vegetation from aerial photographs, have often been the most widely available source of detailed information. The estimation of forest stand structure, including species composition, crown closure, stand height, stem density, age, and volume, has been the focus of forest inventories used to develop resource management plans in the quest to ensure that forests are managed on a sustainable basis. Timber growth and yield have dominated the forest inventory agenda in the past few decades (Lin and Päivinen 1999), with detailed data on the location and extent of forest resources being collected principally to meet timber planning and harvest scheduling needs (Gillis and Leckie 1993). However, management practices are now increasingly designed to maintain and enhance the long-term health of forest ecosystems while providing environmental, social, and cultural opportunities for present and future generations (Kohm and Franklin 1997). For example, conservation of biodiversity and of wildlife habitat has become 512 BioScience June 2004 / Vol. 54 No. 6
a critical goal of forest management as part of an ecosystembased approach to adaptive management of forested landscapes (Kimmins 1999). In response, systems for the classification of forest inventories are changing. New biophysical information is required, perhaps combined with a larger number of more refined classes of traditional inventory attributes (Franklin SE 2001). Forest biophysical data are related to forest structure and function (e.g., leaf area index [LAI], biomass, net primary productivity [NPP]), but these data are often not collected routinely in current programs for forest inventory and monitoring because they are poorly understood by those in operational forestry, and the feasibility of their estimation and mapping has been advanced significantly only with recent developments. This situation has created opportunities to develop complementary methods of data acquisition with remotely sensed aerial or satellite data at high spatial resolution. New technological developments (greater spatial, radiometric, temporal, and spectral resolution), coupled with increasingly sophisticated computer processing techniques, enable biophysical remote sensing and estimation of traditional forest inventory data at reasonable cost (Wulder and Franklin 2003). A fundamental premise in using remotely sensed data to estimate or map biophysical or forest inventory data is that there is a predictable relationship between the spectral response measured by the sensor and the magnitude of the parameter of interest. To interpret such a relationship requires knowledge
Articles of the optical properties of forests (e.g., canopy geometry, leaf spectral properties), the effects caused by the size of the viewed area (e.g., spatial resolution), and environmental and systems factors such as topography, sun elevation, haze, wind speed, and the orientation and inclination of the view axis between the surface and sensor. The spatial resolution of a remotely sensed measurement is determined by the sensor’s instantaneous field of view (IFOV). With any image data, the user must consider the relationship between the ground objects of interest and the IFOV, which strongly influences the information content. When the pixel size exceeds the size of an individual tree crown, decreasing visible and infrared reflectance has been observed in stands as trees age and become larger. As a stand ages, the areas between the tree crowns become smaller, and the effects of crown closure and shadow increase (Franklin J 1986, Gemmell 1995). As trees grow taller, foliage biomass increases, and the spectrally brighter understory is covered with shadows, reducing the overall spectral reflectance of the stand. One potential advantage of using data at medium spatial resolution (e.g., from the Landsat Thematic Mapper and Enhanced Thematic Mapper Plus sensors; see Cohen and Goward 2004) has been the ability to measure the structural properties of forests, because pixel values are an integrated measure of the overstory, understory, soil, and shadow components (Scarth et al. 2001). However, the characteristics of individual trees are difficult to determine using these data. By contrast, remotely sensed images at high spatial resolution commonly have a pixel size smaller than the size of an individual tree crown; in other words, the objects of interest (usually the trees) are larger than the spatial resolution used to measure them. One way to understand these data is to consider that there are many pixels per object rather than many objects in a single pixel (Strahler et al. 1986). Thus, a group of pixels may be combined to characterize individual trees in a way that is impossible with medium-resolution data (Hay et al. 1996, Moskal and Franklin 2002, Culvenor 2003). The optimal approach for processing remotely sensed images depends primarily on the definition of the output products desired (e.g., maps) and is influenced by spatial resolution and interpixel variance. Per-pixel image classifications are commonly applied at low or moderate spatial resolution but sometimes do not operate well at high spatial resolution. The spectral variability that is found for an individual tree (e.g., pixels representing sunlit crown, shaded crown, and the influence of factors such as branches, cones, and tree morphology) is not helpful in developing unique spectral signatures for tree or object classification. A spectral endmember approach may be appropriate (Asner et al. 2003, Ustin et al. 2004). Alternatively, approaches for extracting individual tree information from imagery at high spatial resolution are based on techniques that operate principally in the spatial domain (Culvenor 2003). For example, the texture of an image may be an important characteristic that can be related to forest parameters (Wulder et al. 1998), and an un-
derstanding of the expected spectral response for trees may be used to develop rules for delineating tree crowns before they are classified using a spectral signature (or other) approach (Fournier et al. 1995). Remotely sensed data may be used to represent ecological phenomena across a range of scales (Kerr and Ostrovsky 2003). These data are most commonly applied to characterize global or landscape-level properties. The operational use of satellite and airborne data at high spatial resolution is developing quickly. The ability to use multiple spatial resolutions of imagery, in conjunction with field data, in multiscale analyses is a burgeoning area for research and applications. However, there is often a perceived mismatch between the data wanted by ecologists and the data collected with remote sensing instruments (Kerr and Ostrovsky 2003). This perception is declining, thanks to the availability of high-resolution data that can be directly linked to traditional field-based ecological measurements (Turner W et al. 2003). The nesting of data from differing scales in an information hierarchy provides opportunities for gathering detailed, site-specific information over increasingly large areas. To use this information, researchers must understand the potential and limits of data at high spatial resolution and understand how these newly available data may be integrated with other spatial data sources (including lower-resolution satellite data). The number of options for sampling, scaling, and stratifying data, and the ease with which these options can be selected and applied, are poised to increase, allowing researchers to characterize ecological attributes with data from a range of scales. In this article, we review data at high spatial resolution, and related applications, that offer unique options for generating information to characterize ecosystems (with a focus on forests). We introduce a range of systems—photographic and digital, airborne and satellite optical (approximately 400to 2500-nanometer [nm] wavelength sensitivity)—for gathering high-resolution data (see also Lefsky and Cohen 2003). We also discuss potential and current limitations in the ability to estimate a suite of inventory and biophysical attributes. While much research and development has focused on standlevel attributes (e.g., height, age, density, species composition) gathered through forest inventory, techniques now exist to estimate values for attributes that are process-related (e.g., crown morphology, biomass, LAI) and are based on recognition of individual trees or objects rather than on stand-level or “average-tree” conditions. Further issues include the way in which accuracy is assessed and the level of accuracy with which attributes are estimated.
High spatial resolution data Analog (hard-copy) aerial photographs have been used extensively and continue to be used in vegetation-mapping and forestry applications (Hall and Fent 1996), although digital aerial photographs are expected to become more prominent in the future (Caylor 2000). The components of aerial photography include the camera, lens, aerial film, processing, and printing, which together influence the quality of June 2004 / Vol. 54 No. 6 BioScience 513
Articles the resultant photograph. Recently introduced improvements in aerial photography acquisition include forward motion compensation, high-performance optical systems with lenses that achieve 100 line pairs per millimeter, GPS (global positioning system) interfaces, film data recording systems, and gyrostabilized mounts (Hall and Fent 1996, Read and Graham 2002). Characteristics such as camera type, aerial film type, speed, processing, and resolution influence the sharpness and quality of images (Eastman Kodak Company 2002, Hall 2003). The interpretation of analog aerial photographs relies heavily on the skills of a trained analyst, and the process can be both time-consuming and subjective (Holopainen and Wang 1998). Scanning hard-copy photographs is increasingly common. It is seldom necessary or even desirable to scan at a level equivalent to the resolving power of the original aerial photograph (Nelson et al. 2001). Experience suggests that it is best to scan for a pixel size that is approximately 20% of the output size of the object of interest (Hall 2003). The pixel value of a scanned aerial photograph depends on the film type (e.g., spectral sensitivity, resolution), lens resolution, film processing, scale, elevation, and illumination at the time the photo was acquired, as well as, in the case of vegetation, the stage of phenology. Scanning is best undertaken using the original film or diapositive, since the resolving power of film is greater than that of paper. Atmospheric effects are also inherent in photographic images, and there may be bidirectional reflectance effects caused by differential shading of exposed and shaded sides of objects (e.g., tree crowns), especially at larger photographic scales (Fournier et al. 1995). Scanned aerial photo images can be converted to orthophotos through a differential rectification process that includes a digital elevation model, known exterior orientation parameters, and collinearity equations that relate object and image coordinates (Baltsavias 1999, Wolf and Dewitt 2000). Digital aerial photographs can be obtained directly from small-format (e.g., Kodak DCS 420 and DCS 460) and largeformat (e.g., Leica Helava ADS40, Zeiss/Intergraph DMC 2001) digital cameras (Graham and Koh 2002). Advances in video and digital camera technologies and in computer processing speed have enabled the development and testing of a number of airborne multispectral and hyperspectral imaging systems to complement existing aerial photographic programs (Wright et al. 2003). A single CCD (charge-coupled device) sensor with mosaic optical filtering and near-infrared airborne camera systems can produce imagery at very high spatial resolutions (pixel size < 25 centimeters [cm]). The use of definable and changeable spectral filters or spectrographs allows spectral resolution to be defined to within 10 nm. Driving the development of airborne remote sensing instruments, such as the Compact Airborne Spectrographic Imager (CASI; Anger et al. 1994), is the capacity to mobilize quickly at opportunistic times and at user-specified locations. Speed and flexibility of response are important considerations in data acquisition. For example, in monitoring forest health, infestation and infection are often linked to variables such as dis514 BioScience June 2004 / Vol. 54 No. 6
turbance, infecting agents, and the drought phenology of a particular forest type (Stone 1996). The seasonal date of image acquisition is an important factor in maximizing the discriminating potential of algorithms for image interpretation (Treitz and Howarth 1996; see Leckie and colleagues [1995] for a summary of image processing issues). High spatial resolution data have recently become available from commercially operated satellite systems. Satellite-borne high-resolution sensors allow for the collection of data from a stable platform, at regular time intervals, with a relatively large footprint size. For example, the IKONOS system, launched in September 1999, collects panchromatic data (with a spectral range of 450 to 900 nm) with approximately 1-m spatial resolution and four channels of multispectral data (blue, green, red, and near-infrared) with approximately 4-m spatial resolution over a swath 11 km wide. The QuickBird satellite, launched in October 2001, collects panchromatic data (also 450 to 900 nm) with approximately 0.61-m spatial resolution and multispectral data (blue, green, red, and nearinfrared) with approximately 2.44-m spatial resolution over a swath 16.5 km wide. Listings of current and proposed highresolution satellite sensor systems may be found in Glackin and Peltzer (1999), Franklin SE (2001), Donoghue (2000), and Aplin (2003).
Applications A wide range of applications for high spatial resolution data, including the remote measurement of structural attributes, may be applicable to ecological issues. In some instances, remotely sensed data is combined with other commonly used data sources to better portray ecological attributes. Forest structure. The extraction of information on forest structure from airborne or satellite data at high spatial resolution has been the subject of a large number of experimental and empirical studies. One approach has been based on per-pixel classification or regression analyses in which the spectral data are used to predict classes or continuous variables of structure. Some examples include the following: · Conifer species composition and forest cover classifica-
tion (Franklin SE et al. 2000, Key et al. 2001). · Estimation of individual tree species and crown closure
(Cohen et al. 1995, 2001, Gerylo et al. 1998). · Prediction of stem density and stand height (Franklin
SE and McDermid 1993) and of stand volume and basal area (Wulder et al. 2002). · Classification of relative stand age or stage of develop-
ment (Cohen et al. 1995, 2001, Franklin SE et al. 2001, Scarth et al. 2001, Nelson et al. 2003).
The utility of these studies has been in developing an understanding of which forest attributes and which sensor systems might be integrated into an operational system for the collection of forest inventory data. Because of the high in-
Articles vestment that the forest industry and government agencies make in producing vegetation inventories, it is unlikely that any single method of digital remote sensing could be used to replace operational inventories entirely. Instead, remotely sensed data must be considered at multiple scales together with, or as a complement to, the existing system for forest data collection. The estimation of crown closure on a continuous scale, rather than from the ordinal class data typically obtained by interpreting aerial photographs, is one advantage of a digital remote sensing approach (St-Onge and Cavayas 1997); the integration of remotely sensed data with multitemporal data or with data from other sources, such as lidar (light detecting and ranging), is another (Lefsky et al. 2001, Lim et al. 2003). A third possibility is the hierarchical nesting of image data to represent different scales of forest information, from individual trees to stand-level conditions, forest ecosystems, and larger areal units (Marceau et al. 1994). A promising approach to image processing or information extraction has focused on the use of texture variables and vegetation indexes computed using spatial domain and different band combinations of image data at high levels of spatial detail. Applications have included assessment of forest productivity and crown gap (Blackburn and Milton 1997), monitoring of forest health (Lévesque and King 2003), and
LAI estimation (Wulder et al. 1998). Combining the perpixel approach with spatial- or texture-processing approaches has been shown to provide additional accuracy in the estimation of forest structure variables such as crown closure, LAI, and stand age (e.g., Franklin and McDermid 1993, Wulder et al. 1996a). A relatively new approach—commonly considered part of the emerging model of “image understanding” (Guindon 1997)—has been developed on the basis of algorithms for the recognition of individual trees (Gougeon 1995a, Meyer et al. 1996, Culvenor 2003). The incentives for developing such an approach are obvious. The isolation of individual tree crowns on digital images creates the potential for more accurate classification of tree species, allowing researchers to construct area compilations of crown closure and stem density (as illustrated in figure 1). In turn, these estimates can be used in modeling of stand volume (Hall et al. 1998). Several methods have been developed for isolating individual tree crowns (Culvenor 2003), and their relative merits continue to be tested in a variety of forest environments. Such individual tree information is best considered as a sampling tool within the spatially explicit operational forest inventory. Images at high spatial resolution are data-storage and data-processing intensive when considered as a tool for managing forests at
Figure 1. Example of species classification and regrouping of individual tree crowns and tree clusters into structurally homogenous units, akin to forest inventory polygons. Individual tree isolation (Gougeon 1995a) and classification (Gougeon 1995b) were applied to IKONOS imagery at 1-meter spatial resolution. The study was located in the Lac à l’Ours region of Quebec, Canada. June 2004 / Vol. 54 No. 6 BioScience 515
Articles the landscape level. Sampling of high-resolution images to develop continuous estimates of stand attributes at the single-tree level may be a viable approach for short-term implementation of this technology. Forest biophysical data. Biophysical data include measures of forest ecosystem productivity, structure, and amount of vegetation. Typically, the measures of interest include LAI, aboveground forest biomass, and NPP (Wulder 1998). The first of these measures, LAI, has been defined as one-half of the total intercepting area per unit ground surface area (Chen and Black 1992). An important measure of vegetation structure, LAI is thought to govern many physical and biological processes (e.g., photosynthesis, transpiration, evapotranspiration) related to vegetation dynamics (Chen et al. 2002). Forest biomass, or the dry mass of live plant material, is a measure used in studies of forest ecosystem processes and in models that calculate or forecast carbon budgets (Price et al. 1997, Kurz and Apps 1999). Finally, NPP is the change or accumulation of foliage, branch, stems, and root biomass over time. Estimates of NPP are often based on ecological models that require detailed inputs, many of which are feasible only when acquired by remote sensing. Detailed maps of the magnitude and spatial distribution of biomass are also required for these estimates (Bernier et al. 1999, Fournier et al. 2000) but are often impossible to acquire using coarse spatial resolution data, because of mixed pixel effects and scaling relationships from the field to large-area image mosaics (Chen et al. 2002). Regional maps of LAI are routinely produced using satellite data at medium or coarse spatial resolution (Spanner et al. 1994, Fassnacht et al. 1997, Chen et al. 1999), but LAI maps for a local area, individual stand, or tree are possible only with high-resolution data (Gong et al. 1995, Hu et al. 2000, Pellikka et al. 2000). Uncertainties and errors in LAI estimation may occur because of the influence of understory vegetation, background reflectance,image registration, differences in time between field measurements and image acquisition, and insufficient spatial resolution to resolve land surface variability. These uncertainties are resolved more readily with high-resolution airborne or spaceborne data because of the greater flexibility in timing of data acquisition and the increased pixel size. The process of LAI mapping usually starts with the measurement of LAI in the field. Destructive sampling for direct LAI estimation is time-consuming and costly (Cutini et al. 1998). Indirect, optical methods are often employed, as they can offer more rapid and consistent estimates of leaf area at scales larger than the individual tree (Fournier et al. 2003). From these field-based estimates, statistical models are developed to estimate LAI as a function of a spectral measure (e.g., vegetation index) (Chen et al. 2002). Image preprocessing is necessary when using any image to estimate LAI. Geometric registration, conversion of image radiance to reflectance, and correction for bidirectional reflectance help improve the relationship between the image and LAI (Chen et al. 1999). Stratification of images by landcover class has also been recommended as an approach to 516 BioScience June 2004 / Vol. 54 No. 6
improve LAI estimation over the landscape (Wulder et al. 1998, Turner DP et al. 1999). High spatial resolution data have also been used in scaling LAI to coarser-resolution data for mapping at regional and large-area levels (Fernandes et al. 2002, Hall et al. 2003). Interest in biomass estimation and mapping has grown for a variety of reasons, including the national and international commitments to reporting indicators of forest sustainability (e.g., Sips 1996, CCFM 1997). Previous mapping efforts based on inventory data or on satellite data at coarse spatial resolution (Penner et al. 1997, Brown et al. 1999) are useful for meeting national reporting needs, but these methods are not suited to detecting change and are too broad for many of the forest management applications that are of increasing interest (Franklin SE 2001). One approach has been to develop models that estimate aboveground forest biomass as a function of inventory data; these models can be scaled to regional and national levels (Bonner 1982). Because of the effect of image calibration on model parameters, empirical models are based on image-by-image calibration requirements. There are limitations to the extent that this can be achieved at the levels of accuracy required for satellite data. However, satellite data at coarse spatial resolution can be used for the initial classification and stratification of the landscape, from which higher-resolution airborne or satellite data might logically be used to derive more detailed structural elements of a stand. Recent studies with scanning lidar sensors show promise for aboveground biomass estimation (Lefsky et al. 1999). A combination of lidar data with optical data at fine spatial resolution may be used to derive the elements of stand composition and structure at the level of detail required to permit accurate biomass estimation and mapping. Some airborne sensors can be configured to collect hyperspectral imagery at high spatial resolution. The CASI, for example, can obtain imagery in 10 to 12 spectral channels at 60-cm spatial resolution, or in up to 196 channels at 20-m spatial resolution from the same flying height. Hyperspectral data can be collected at higher spatial resolution if the sensor is configured to collect fewer channels (the trade-offs are presented by Wulder and colleagues [1996b]). This combination of high spectral and spatial resolution allows for detailed physiological information on the status of the forest canopy or individual trees. Recent work has suggested that hyperspectral data at high spatial resolution can be correlated empirically with the following: · Chlorophyll pigments (Curran et al. 1997, Coops et al.
2002). · Band ratios, including wavelengths representing
carotenoids and chlorophyll (Peñuelas et al. 1995). · Anthophyll cycle pigments and related photosynthetic
performance (Gamon et al. 1992), and other measures
of integrated leaf stress (Carter 1994).
Articles Advances in the understanding of leaf physiology and spectral reflectance suggest that foliage health indicators may be developed on the basis of observations from hyperspectral data and from remotely sensed data at high spatial resolution (Zarco-Tejada et al. 2000). Reviews of imaging spectroscopy for ecological applications can be found in Curran (2001) and Ustin and colleagues (2004). Emerging methods and applications. In addition to individual tree–based and inventory-based assessment of forest and vegetation attributes, a wide range of plot-based assessments of vegetation structure, including structural diversity, land cover, classification of habitat diversity, and within-stand variability, can be determined from imagery at high spatial resolution. Of the wide variety of methods and successful applications, five are cited here: 1. Standard image-based, per-pixel or texture classifica-
tions used to extract land-cover classes in mixed-wood and boreal forests (Franklin SE et al. 2000).
2. An image classification approach used to classify 1-m
multispectral imagery into a number of land-cover types associated with conifer habitat in an alpine environment (McGregor 1998). 3. Habitat classification (coastal, marine, and wetland)
using hyperspectral imagery at high spatial resolution (Marcus 2002), 4-m multispectral IKONOS imagery (Mumby and Edwards 2002), and multitemporal IKONOS imagery (Dechka et al. 2002). 4. Derivation of image variance and semivariance to pro-
vide measures of stand structure (St-Onge and Cavayas 1997, Wulder et al. 1998, Lévesque and King 2003). 5. Fitting of spatial statistical models such as semivari-
ograms to represent forest structure (Lévesque and King 2003). In this last approach, the semivariogram range, sill, and nugget are fitted to image objects and interpreted (Treitz and Howarth 2000). In seven mixed-wood stands in Alberta, Canada, Zhang and colleagues (2003) improved the accuracy a
b
c
Figure 2. Schematic representation of three forest plots, each with varying complexity in terms of vertical distribution of live and dead biomass: (a) a young (regrowth) forest, (b) a mature forest, and (c) an oldgrowth forest. To the left are typical remotely sensed images at high spatial resolution over similar forest types; to the right are photographs representing each general category. The ability to resolve individual features in an image, combined with the image’s texture, provides an indication of the forest’s structural characteristics and of the vertical distribution of biomass. June 2004 / Vol. 54 No. 6 BioScience 517
Articles of forest species classification by more than 10% using texture obtained from kriging surfaces that they derived from imagery at high spatial resolution. Earlier, Bowers and colleagues (1994) used spatially detailed airborne sensor data to model balsam fir damage caused by insect infestation. Coops and Catling (1996) used a modified local variance method that assessed changes in the standard deviation of a moving 3 ´ 3 window on successively spatially degraded images. They developed a relationship that related the maximum level of variance observed to the vertical distribution of biomass within a forested canopy (see figure 2 for an example). The assessments were applied to estimate mammalian distribution and abundance over large areas (Coops and Catling 2002). These studies demonstrate that, by capturing the inherent image variance on a pixel-by-pixel basis, data from images at high spatial resolution can significantly improve the description, classification, and interpretation of patterns of vegetation. Data integration. The integration of data from differing sources enables the generation of complex or derived ecological attributes. For instance, a third dimension can be incorporated into traditional two-dimensional remotely sensed data by using lidar (Lefsky et al. 2002) or radar data (Treuhaft et al. 2004). The integration of remotely sensed data into models can provide for model parameter initialization, incorporated as condition or state variables, or for validation of model outcomes (Coops and White 2003, Turner DP et al. 2004). Remotely sensed data can convey information regarding the amount, distribution, structure, and condition of vegetation. The integration of remotely sensed data with other spatial data sources and models provides a powerful approach for generating ecological information. One example of such an approach is the US Geological Survey’s National Gap Analysis Program, or GAP (Jennings 2000), in which remotely sensed data with differing spatial resolution (i.e., Landsat, aerial videography, and photographs), spatial data (e.g., soils, elevation), and modeling are combined to produce detailed information on vegetation cover for maps of the United States.
Conclusions Choosing which remotely sensed data to collect over forests for ecological studies and management is a function of what is needed and what is possible within the economic models that currently dominate forest assessment and planning. These needs are changing, and information resources are expanding. Remotely sensed data at high spatial resolution are available from a wide range of sensor and platform configurations, including analog aerial photography, digital aerial photography, video and digital cameras, and multispectral airborne and spaceborne sensor systems. The multitude of possible sources for high-resolution images is increasing the possibilities for extracting detailed information about forest structure, function, and ecosystem processes. 518 BioScience June 2004 / Vol. 54 No. 6
The extraction of structural and biophysical information about forests from imagery at high spatial resolution requires nontraditional approaches to digital analysis and, often, the careful use of complementary data such as lidar. Lidar provides unprecedented accuracy in estimates of forest biomass, height, and the vertical distribution of forest structure. To enable characterization at the landscape level, samples of remotely sensed data at high spatial resolution may be combined with more spatially extensive, yet less detailed, information sources such as Landsat or existing GIS (geographic information system) inventory databases. One emerging role of data at high spatial resolution is the precise estimation of forest biophysical data within units of land cover. Combining estimates of remotely sensed attributes with the analytical utility of GIS and advanced forest process models is a powerful means of generating information that describes forests. This combination of methods contributes to better understanding of the influence of disturbances, management practices, and changing climate on the sustainability of forest ecosystems. The applications reviewed here suggest that timely spatial characterization of forest condition, extent, and structure is possible when using high-resolution data from airborne or spaceborne platforms as an information source. We expect continuing refinements to the need for information to characterize forest ecosystems, a need that may be addressed, in part, by the increasing availability of remotely sensed data at high spatial resolution. In attempts to understand ecological processes more fully, from the leaf to the globe, the sampling, scaling, and stratification of remotely sensed data at high spatial resolution are likely to play a key role.
Acknowledgm ents We acknowledge financial support from the Natural Sciences and Engineering Research Council of Canada and the Canadian Forest Service. We also thank François Gougeon (Canadian Forest Service); CLC-Camint, Inc.; Industries Davidson, Inc.; and the Quebec Department of Natural Resources for providing the graphic on individual tree isolation used in figure 1. Two anonymous reviewers are heartily thanked for their insights and suggestions.
References cited Anger CD, Mah S, Babey SK. 1994. Technological enhancements to the compact airborne spectrographic imager (casi). Pages 205–213 in Proceedings of the First International Airborne Remote Sensing Conference and Exhibition, Strasbourg, France, 11–15 September 1994, vol. 2. Ann Arbor (MI): Environmental Research Institute of Michigan. Aplin P. 2003. Remote sensing: Base mapping. Progress in Physical Geography 27: 275–283. Asner G, Hicke J, Lobell D. 2003. Per-pixel analysis of forest structure: Vegetation indices, spectral mixture analysis and canopy reflectance modeling. Pages 229–254 in Wulder M, Franklin S, eds. Remote Sensing of Forest Environments: Concepts and Case Studies. Boston: Kluwer Academic. Baltsavias EP. 1999. On the performance of photogrammetric scanners. Pages 155–173 in Fritsch D, Spiller R, eds. Photogrammetric Week 1999.
Articles Heidelberg (Germany): Wichmann Verlag. (13 April 2004; www.ifp. uni-stuttgart.de/publications/phowo99/baltsavias99.pdf) Bernier PY, et al. 1999. Linking ecophysiology and forest productivity: An overview of the ECOLEAP project. Forestry Chronicle 75: 417–421. Blackburn GA, Milton EJ. 1997. An ecological survey of deciduous woodlands using airborne remote sensing and geographic information systems (GIS). International Journal of Remote Sensing 18: 1919–1935. Bonner GM. 1982. Canada’s Forest Inventory 1981. Chalk River (Canada): Forestry Statistics and Systems Branch, Canadian Forest Service, Department of the Environment. Information Report PI-X-45. Bowers WW, Franklin SE, Hudak J, McDermid GJ. 1994. Forest structural damage analysis using image semivariance. Canadian Journal of Remote Sensing 20: 28–36. Brown SL, Schroeder P, Kern JS. 1999. Spatial distribution of biomass in forests of the eastern USA. Forest Ecology and Management 123: 81–90. Carter GA. 1994. Ratios of leaf reflectances in narrow wavebands as indicators of plant stress. International Journal of Remote Sensing 15: 697–703. Caylor J. 2000. Aerial photography in the next decade. Journal of Forestry 98: 17–19. [CCFM] Canadian Council of Forest Ministers. 1997. Criteria and Indicators of Sustainable Forest Management in Canada. Ottawa (Canada): CCFM. Chen JM, Black TA. 1992. Defining leaf area index for non-flat leaves. Plant, Cell and Environment 15: 421–429. Chen JM, Leblanc SG, Miller JR, Freemantle J, Loechel SE, Walthall CL, Innanen KA, White HP. 1999. Compact Airborne Spectrographic Imager (CASI) used for mapping biophysical parameters of boreal forests. Journal of Geophysical Research–Atmospheres 104: 27945–27948. Chen JM, et al. 2002. Derivation and validation of Canada-wide coarseresolution leaf area index maps using high-resolution satellite imagery and ground measurements. Remote Sensing of Environment 80: 165–184. Cihlar J, Latifovic R, Beaubien J, Trishchenko A, Chen J, Fedosejevs J. 2003a. National scale forest information extraction from coarse resolution satellite data,pt. 1: Data processing and mapping land cover types. Pages 337–357 in Wulder M, Franklin S, eds. Remote Sensing of Forest Environments: Concepts and Case Studies. Boston: Kluwer Academic. Cihlar J, Fernandes R, Fraser R, Chen J, Chen W, Latifovic R, Liu J, Li Z. 2003b. National scale forest information extraction from coarse resolution satellite data, pt. 2: Forest biophysical parameters and carbon. Pages 359–387 in Wulder M, Franklin S, eds. Remote Sensing of Forest Environments: Concepts and Case Studies. Boston: Kluwer Academic. Cohen WB, Goward SN. 2004. Landsat’s role in ecological applications of remote sensing. BioScience 54: 535–545. Cohen WB, Spies TA, Fiorella M. 1995. Estimating the age and structure of forests in a multiownership landscape of western Oregon, USA. International Journal of Remote Sensing 16: 721–746. Cohen WB, Maiersperger TK, Spies TA, Oetter DR. 2001. Modelling forest cover attributes as continuous variables in a regional context with Thematic Mapper data. International Journal of Remote Sensing 22: 2279–2310. Coops NC, Catling PC. 1996. Utilising airborne multispectral videography to predict forest structure in eucalypt forests for wildlife management. Wildlife Research 24: 691–703. ———. 2002. Prediction of the spatial distribution and relative abundance of ground-dwelling mammals using remote sensing imagery and simulation models. Landscape Ecology 17: 173–188. Coops NC, White JD. 2003. Modeling forest productivity using data acquired through remote sensing. Pages 411–431 in Wulder M, Franklin S, eds. Remote Sensing of Forest Environments: Concepts and Case Studies. Boston: Kluwer Academic. Coops NC, Stone C, Culvenor DS, Chisholm L, Merton R. 2002. Predicting chlorophyll content in eucalypt vegetation at the leaf and canopy level using high spectral resolution imagery. Tree Physiology 23: 23–31. Culvenor D. 2003. Extracting individual tree information: A survey of techniques for high spatial resolution imagery. Pages 255–277 in Wulder M, Franklin S, eds. Remote Sensing of Forest Environments: Concepts and Case Studies. Boston: Kluwer Academic.
Curran PJ. 2001. Imaging spectrometry for ecological applications. International Journal of Applied Earth Observation and Geoinformation 3: 305–312. Curran PJ, Kupiec IA, Smith GM. 1997. Remote sensing the biochemical composition of a slash pine canopy. IEEE Transactions on Geoscience and Remote Sensing 35: 415–420. Cutini A, Matteussi G, Mugnozza GS. 1998. Estimation of leaf area index with the Li-Cor LAI 2000 in deciduous forests. Forest Ecology and Management 105: 55–65. Dechka J, Franklin S, Watmough M, Bennett R, Ingstrup D. 2002. Classification of wetland habitat and vegetation communities using multitemporal IKONOS imagery in southern Saskatchewan. Canadian Journal of Remote Sensing 28: 679–685. Donoghue D. 2000. Remote sensing: Sensors and applications. Progress in Physical Geography 24: 407–414. Eastman Kodak Company. 2002. Aerial imaging. (13 April 2004; www. kodak.com/go/aerial) Fassnacht KS, Gower ST, MacKenzie MD, Nordheim EV, Lillesand TM. 1997. Estimating the leaf area index of north central Wisconsin forests using Landsat Thematic Mapper. Remote Sensing of Environment 61: 229–245. Fernandes RA, Miller JR, Hu B, Rubinstein IG. 2002. A multi-scale approach to mapping effective leaf area index in boreal Picea mariana stands using high spatial resolution CASI imagery. International Journal of Remote Sensing 23: 3547–3568. Fournier RA, Edwards GE, Eldridge NR. 1995. A catalogue of potential spatial discriminators for high spatial resolution digital images of individual tree crowns. Canadian Journal of Remote Sensing 21: 285–298. Fournier RA, Guindon L, Bernier PY, Ung C-H, Raulier F. 2000. Spatial implementation of models in forestry. Forestry Chronicle 76: 929–940. Fournier R, Mailly D, Walter J-M, Soudani K. 2003. Indirect measurement of forest canopy structure from in situ optical sensors. Pages 77–113 in Wulder M, Franklin S, eds. Remote Sensing of Forest Environments: Concepts and Case Studies. Boston: Kluwer Academic. Franklin J. 1986. Thematic Mapper analysis of coniferous forest structure and composition. International Journal of Remote Sensing 7: 1287–1301. Franklin SE. 2001. Remote Sensing for Sustainable Forest Management. New York: Lewis. Franklin SE, McDermid GJ. 1993. Empirical relations between digital SPOT HRV and CASI imagery and lodgepole pine (Pinus contorta) forest stand parameters.International Journal of Remote Sensing 14: 2331–2348. Franklin SE, Hall RJ, Moskal LM, Maudie AJ, Lavigne MB. 2000. Incorporating texture into classification of forest species composition from airborne multispectral images. International Journal of Remote Sensing 21: 61–79. Franklin SE, Wulder MA, Gerylo GR. 2001. Texture analysis of IKONOS panchromatic imagery for Douglas-fir forest age class separability in British Columbia. International Journal of Remote Sensing 22: 2627–2632. Gamon JA, Peñuelas J, Field CB. 1992. A narrow-waveband spectral index that tracks diurnal changes in photosynthetic efficiency. Remote Sensing of Environment 41: 35–44. Gemmell FM. 1995. Effects of forest cover, terrain, and scale on timber volume estimation with Thematic Mapper data in a Rocky Mountain site. Remote Sensing of Environment 51: 291–305. Gerylo GR, Hall RJ, Franklin SE, Roberts A, Milton EJ. 1998. Hierarchical image classification and extraction of forest species composition and crown closure from airborne multispectral images. Canadian Journal of Remote Sensing 24: 219–232. Gillis MD, Leckie DG. 1993. Forest Inventory Mapping Procedures across Canada. Chalk River (Canada): Forestry Canada, Petawawa National Forestry Institute. Information Report PI-X-114. Glackin D, Peltzer G. 1999. Civil, Commercial, and International Remote Sensing Systems and Geoprocessing. El Segundo (CA): Aerospace Press. Gong P, Pu R, Miller JR. 1995. Coniferous forest leaf area index estimation along the Oregon transect using Compact Airborne Spectrographic Imager data. Photogrammetric Engineering and Remote Sensing 61: 1107–1117.
June 2004 / Vol. 54 No. 6 BioScience 519
Articles Gougeon FA. 1995a.A crown-following approach to the automatic delineation of individual tree crowns in high spatial resolution aerial images. Canadian Journal of Remote Sensing 21: 274–284. ———. 1995b. Comparison of possible multispectral classification schemes for tree crowns individually delineated on high spatial resolution MEIS images. Canadian Journal of Remote Sensing 21: 1–9. Graham RW, Koh A. 2002. Digital Aerial Survey: Theory and Practice. Boca Raton (FL): CRC Press. Guindon B. 1997. Computer-based aerial image understanding:A review and assessment of its application to planimetric information extraction from very high resolution satellite images. Canadian Journal of Remote Sensing 23: 38–47. Hall RJ. 2003. The roles of aerial photographs in forestry remote sensing image analysis. Pages 47–75 in Wulder M, Franklin S, eds. Methods and Applications for Remote Sensing of Forests: Concepts and Case Studies. Boston: Kluwer Academic. Hall RJ, Fent L. 1996. Influence of aerial film spectral sensitivity and texture on interpreting images of forest species composition. Canadian Journal of Remote Sensing 22: 350–359. Hall RJ, Gerylo GR, Franklin SE. 1998. Estimation of stand volume from high resolution multispectral images. Paper presented at the 20th Canadian Symposium on Remote Sensing; 10–13 May 1998, Calgary, Alberta, Canada. Hall RJ, Davidson DP, Peddle DR. 2003. Ground and remote estimation of leaf area index in Rocky Mountain forest stands, Kananaskis, Alberta. Canadian Journal of Remote Sensing 29: 411–427. Hay G, Niemann K, McLean G. 1996. An object-specific image texture analysis of H-resolution forest imagery. Remote Sensing of Environment 55: 108–122. Holopainen M, Wang G. 1998. The calibration of digital aerial photographs for forest stratification. International Journal of Remote Sensing 19: 677–696. Hu B, Inannen K, Miller JR. 2000. Retrieval of leaf area index and canopy closure from CASI data over the BOREAS flux tower sites. Remote Sensing of Environment 74: 255–274. Jennings MD. 2000. GAP analysis: Concepts, methods, and recent results. Landscape Ecology 15: 5–20. Kerr J, Ostrovsky M. 2003. From space to species: Ecological applications for remote sensing. Trends in Ecology and Evolution 18: 299–314. Key T, Warner TA, McGraw JB, Fajvan MA. 2001. A comparison of multispectral and multitemporal information in high spatial resolution imagery for classification of individual tree species in a temperate hardwood forest. Remote Sensing of Environment 75: 100–112. Kimmins JP. 1999. Biodiversity, beauty and the “beast”: Are beautiful forests sustainable, are sustainable forests beautiful, and is “small” always ecologically desirable? Forestry Chronicle 75: 955–960. Kohm KA, Franklin JF, eds. 1997. Creating a Forestry for the 21st Century: The Science of Ecosystem Management.Washington (DC): Island Press. Kurz WA, Apps MJ. 1999. A 70-year retrospective analysis of carbon fluxes in the Canadian forest sector. Ecological Applications 9: 526–547. Leckie DG, Beaubien J, Gibson JR, O’Neill NT, Piekutowski T, Joyce SP. 1995. Data processing and analysis for MIFUCAM: A trial of MEIS imagery for forest inventory mapping. Canadian Journal of Remote Sensing 21: 337–356. Lefsky MA, Cohen WB. 2003. Selection of remotely sensed data. Pages 13–46 in Wulder M, Franklin S, eds. Remote Sensing of Forest Environments: Concepts and Case Studies. Boston: Kluwer Academic. Lefsky MA, Harding D, Cohen WB, Parker GG. 1999. Surface lidar remote sensing of basal area and biomass in deciduous forests of eastern Maryland, USA. Remote Sensing of Environment 67: 83–98. Lefsky MA, Cohen WB, Spies TA. 2001. An evaluation of alternate remote sensing products for forest inventory, monitoring, and mapping of Douglas-fir forests in western Oregon. Canadian Journal of Forest Research 31: 78–87. Lefsky MA, Cohen WB, Parker GG, Harding DJ. 2002. Lidar remote sensing for ecosystem studies. BioScience 52: 19–30.
520 BioScience June 2004 / Vol. 54 No. 6
Lévesque J, King DJ. 2003. Spatial analysis of radiometric fractions from high-resolution multispectral imagery for modelling forest structure and health. Remote Sensing of Environment 84: 589–602. Lim K, Treitz P, Wulder M, St-Onge B, Flood M. 2003. LiDAR remote sensing of forest structure. Progress in Physical Geography 27: 88–106. Lin C, Päivinen R. 1999. Remote sensing and forestry information needs in the 21st century. Paper presented at the Remote Sensing and Forest Monitoring Conference; 1–3 June 1999, Rogow, Poland. Marceau DJ, Gratton DJ, Fournier RA, Fortin JP. 1994. Remote sensing and the measurement of geographical entities in a forested environment, pt. 2: The optimal spatial resolution. Remote Sensing of Environment 49: 105–117. Marcus W. 2002. Mapping of stream microhabitats with high spatial resolution hyperspectral imagery. Journal of Geographical Systems 4: 113–126. McGregor S. 1998. An integrated geographic information system approach for modeling the suitability of conifer habitat in an alpine environment. Geomorphology 2: 265–280. Meyer P, Staenz K, Itten KI. 1996. Semi-automated procedures for tree species identification in high resolution data from digitized color infrared aerial photography. ISPRS Journal of Photogrammetry and Remote Sensing 51: 5–16. Moskal LM, Franklin SE. 2002. Multi-layer forest stand discrimination with spatial co-occurrence texture analysis of high spatial detail airborne imagery. Geocarto International 17 (4): 53–65. Mumby PJ, Edwards AJ. 2002. Mapping marine environments with IKONOS imagery: Enhanced spatial resolution can deliver greater thematic accuracy. Remote Sensing of Environment 82: 248–257. Nelson T, Wulder MA, Niemann KO. 2001. Spatial resolution implications of digitising aerial photography for environmental applications.Journal of Imaging Science 49: 223–232. Nelson T, Niemann KO, Wulder M. 2003. Spatial statistical techniques for aggregating point objects extracted from high spatial resolution remotely sensed imagery. Journal of Geographical Systems 4: 423–433. Pellikka P, Seed ED, King DJ. 2000. Modelling deciduous forest ice storm damage using aerial CIR imagery and hemispheric photography. Canadian Journal of Remote Sensing 26: 394–405. Penner M, Power K, Muhairwe C, Tellier R, Wang Y. 1997. Canada’s Forest Biomass Resources: Deriving Estimates from Canada’s Forest Inventory. Victoria (Canada): Canadian Forest Service, Pacific Forestry Centre. Information Report BC-X-370. Peñuelas J, Filella I, Lloret P, Muñoz F, Vilajeliu M. 1995. Reflectance assessment of mite effects on apple trees. International Journal of Remote Sensing 16: 2727–2733. Price DT, Halliwell DH, Apps MJ, Kurz WA, Curry SR. 1997. Comprehensive assessment of carbon stocks and fluxes in a boreal forest management unit. Canadian Journal of Forest Research 27: 2005–2016. Read RE, Graham RW. 2002. Manual of Aerial Survey: Primary Data Acquisition. Boca Raton (FL): CRC Press. Running SW, Nemani RR, Heinsch FA, Zhao M, Reeves M, Hashimoto H. 2004. A continuous satellite-derived measure of global terrestrial primary production. BioScience 54: 547–560. Scarth P, Phinn SR, McAlpine C. 2001. Integrating high and moderate spatial resolution image data to estimate forest age structure. Canadian Journal of Remote Sensing 27: 129–142. Sips P. 1996. Sustainable forest management, sustainably produced timber, and certification: A state-of-the-art international initiative. BOS Newsletter 15 (1): 24–36. Spanner MA, Johnson L, Miller J, McCreight R, Freemantle J, Runyon J, Gong P. 1994. Remote sensing of seasonal leaf area index across the Oregon transect. Ecological Applications 4: 258–271. Stone C. 1996. The role of psyllids (Hemiptera: Psyllidae) and bell miners (Manorina melanophrys) in canopy dieback of Sydney blue gum (Eucalyptus saligna Sm.). Austral Ecology 21: 450–458. St-Onge B, Cavayas F. 1997. Automated forest structure mapping from high resolution imagery based on directional semivariogram estimates. Remote Sensing of Environment 61: 82–95.
Articles Strahler A, Woodcock C, Smith J. 1986. On the nature of models in remote sensing. Remote Sensing of Environment 20: 121–139. Treitz P, Howarth P. 1996. Remote Sensing for Forest Ecosystem Characterization: A Review. Sault Ste. Marie (Canada): Natural Resources Canada, Canadian Foreign Service. NODA/NFP Technical Report TR-12. ———. 2000. High spatial resolution remote sensing data for forest ecosystem classification: An examination of spatial scale. Remote Sensing of Environment 72: 268–289. Treuhaft RN, Law BE, Asner GP. 2004. Forest attributes from radar interferometric structure and its fusion with optical remote sensing. BioScience 54: 561–571. Turner DP, Cohen WB, Kennedy RE, Fassnacht KS, Briggs JM. 1999. Relationships between leaf area index and Landsat TM spectral vegetation indices across three temperate zone sites. Remote Sensing of Environment 70: 52–68. Turner DP, Ollinger SV, Kimball JS. 2004. Integrating remote sensing and ecosystem process models for landscape- to regional-scale analysis of the carbon cycle. BioScience 54: 573–584. Turner W, Spector S, Gardiner N, Fladeland M, Sterling E, Steininger M. 2003. Remote sensing for biodiversity science and conservation. Trends in Ecology and Evolution 16: 306–314. Ustin SL, Roberts DA, Gamon JA, Asner GP, Green RO. 2004. Using imaging spectroscopy to study ecosystem processes and properties. BioScience 54: 523–534. Wolf PR, Dewitt BA. 2000. Elements of Photogrammetry with Applications in GIS. 3rd ed. New York: McGraw-Hill. Wright GG, Matthews KB, Cadell WM, Milne R. 2003. Reducing the cost of multi-spectral remote sensing: Combining near-infrared video imagery
with colour aerial photography. Computers and Electronics in Agriculture 38: 175–198. Wulder M. 1998. Optical remote sensing techniques for the assessment of forest inventory and biophysical parameters. Progress in Physical Geography 22: 449–476. Wulder M, Franklin S, eds. 2003. Remote Sensing of Forest Environments: Concepts and Case Studies. Boston: Kluwer Academic. Wulder M, Lavigne M, Franklin S. 1996a. High spatial resolution optical image texture for improved estimation of forest stand leaf area index. Canadian Journal of Remote Sensing 22: 441–449. Wulder M, Mah S, Trudeau D. 1996b. Mission planning for operational data acquisition campaigns with the CASI. Pages 53–62 in Proceedings of the Second International Airborne Remote Sensing Conference and Exhibition, San Francisco, 24–27 June, vol. 3. Ann Arbor (MI): Environmental Research Institute of Michigan. Wulder M, LeDrew E, Franklin S, Lavigne M. 1998. Aerial image texture information in the estimation of northern deciduous and mixed wood forest leaf area index (LAI). Remote Sensing of Environment 64: 64–76. Wulder M, Niemann O, Goodenough D. 2002. Error reduction methods for local maximum filtering of high spatial resolution imagery for locating trees. Canadian Journal of Remote Sensing 285: 621–628. Zarco-Tejada PJ, Miller JR, Mohammed GH, Noland TL. 2000. Chlorophyll fluorescence effects on vegetation apparent reflectance, pt. I: Leaf-level measurements and model simulation. Remote Sensing of Environment 74: 582–595. Zhang C, Franklin S, Wulder, M. 2003. Geostatistical and texture analysis of airborne imagery for forest classification. International Journal of Remote Sensing 25: 859–865.
June 2004 / Vol. 54 No. 6 BioScience 521
