Abstract
The potential impact of environmental change on human welfare has renewed interest in understanding the patterns and processes associated with global climate change. Goals of the Committee on Earth Sciences (1989) regarding the U.S. Global Climate Change Program concentrated on the development of sound scientific strategies for monitoring and predicting environmental change. The scaling of ecological characteristics from local to regional and global scales were identified by the Committee as key priorities. The scaling of ecological information is not simply done by integrating or aggregating information from local scale investigations to regional and global scales (Caldwell et al., 1993). The complexity of the effects of scale variations rules out the use of simple generalizations (Foody and Curran, 1994). Information that is significant at local scales may be trivial when evaluated at regional or global scales. Biological interactions with the environment occur over many scales, suggesting a role for multiscale analysis in the description of these interactions (Sclmeider, 1994). Methods must be developed to better understand and evaluate ecological processes operating at multiple scales. Forest structure attributes have been measured using remotely sensed data. Leaf area index (LAI), for example, has been related to the infrared/red ratio (Running et al., 1986 Peterson et al., 1987), the normalized difference vegetation index (NDVI) (Leblon et al., 1993), and gap fractions (Nel and Wessman, 1993). These methods generate values for each pixel in a satellite scene based on the relationship between one or more spectral and/or ancillary data channels and the attribute of interest. The spatial autocorrelation or spatial dependence present in surface phenomena and satellite data are usually not ex-ploited during attribute assignment because of difficulty in quantifying the spatial patterns present (Woodcock et al., 1988). Geostatistics provides a statistically based technique to quantify spatial pattern. Geostatistical techniques, in particular cokriging, can serve as an efficient means of modeling forest canopy structure at a variety of spatial scales to serve as inputs to global change models. The key issue will be to determine the factors that influence remotely sensed spectral reflectance and relating them to the ecological model across scales (Ustin et al., 1993). The geostatistical techniques considered in this research include the following: the semivariograrn, which allows the user to compare values of a random variable at two points separated by a given lag distance (Milne, 1991); kriging which uses the information on spatial dependence present in the semivariogram to estimate values at unsampled locations based on scattered sample data (lsaaks and Srivastava, 1989); and cokriging, the multivariate extension of kriging, which is appropriate when two or more variables are spatially interdependent and the variable of interest is undersampled (McBratney and Webster, 1983; Leenaers et al., 1989). Geostatistical techniques have been successfully applied to remotely sensed data. Variograms have been used to determine components of coniferous canopy structure (Cohen et al., 1990), and to determine the spatial autocorrelation structure of Landsat Thematic Mapper (IM) imagery and intercepted photosynthetically active radiation (IPAR) (Lathrop and Pierce, 1991). Atkinson et al. (1992) used cokriging of ground-based radiometer data to estimate LAI, dry biomass and percent cover. Satellite imagery is an excellent candidate for inclusion as an explanatory variable in the cokriging process because it is an exhaustive sample of a given area.