Landscape classification and parameterization based on multiscale remote sensing and site data (Landscape ecology, Biogeography)

This research develops a new approach for addressing the interrelated problems of describing the land cover of global sites for land cover classification algorithm training and testing, and for validating global land cover maps. The objective is to relate ground- based and satellite remote sensing-based observations more directly to vegetation attributes including physiognomy, structure, morphology and phenology and cover. This approach is based on the development of the System for Terrestrial Ecosystem parameterization (STEP), a formal model and database which relates ecosystem, vegetation and landscape description via functional attributes and parameters from plot- level data to multiple scales of remote sensing data.; The requirements for site-based descriptions of vegetation and land surface attributes needed to train and test classification algorithms and to validate map products are developed. These site descriptions are critical to understanding the relationship of remote sensing observation to the land surface, and will permit global characterization and monitoring using satellite remote sensing data. An important component is development of the methods for extracting and recording site data using field observation, existing maps and fine-resolution satellite data. This research further explains the interrelated development of classification systems from the ecological, land surface model and remote sensing perspectives.; The application of the site data and classification methodologies was tested at the regional scale of Central America using supervised artificial neural network classification algorithms applied to 1-km AVHRR data. This regional site testing was expanded to more-detailed description and mapping of its vegetation, ecosystems and biogeophysical land surface parameters. The development of the site database permitted exploration of the relationship of classification systems to remote sensing observations using quantitative correlation analysis of site and remote sensing variables using parametric and nonparametric statistical techniques. The result was a methodology for using a classification approach to directly map continuous vegetation parameters required by global models. In addition to the methodological development, this research produced heretofore-unachievable descriptions, databases and maps of the physical properties of Central America's terrestrial ecosystems.