Abstract:
Analyses of ecosystem processes are advanced through remote sensing and geostatistical modeling methods capable of capturing landscape pattern over broad spatial and temporal scales. Many ecological studies rely on land cover data classified from satellite imagery. In this, changes in land cover are often presumed to correlate with changes in ecosystem processes or services provided by ecosystems (e.g., watershed protection). Documenting changes in land cover requires that images be classified over time, often using historical images to document landscape change. But this is difficult to do for historical images because we cannot ground-truth old images, lacking actual land cover data from the past. I developed a land cover classification scheme using a classification and regression tree (CART) model generated from 2001 National Land Cover Dataset (NLCD) and Summer, Fall, and Winter triplets of Landsat 5 Thematic Mapper (TM) imagery. The model is robust to inter-annual variability in surface reflectance, and thus can be extended in time to classify land cover from images from any time, past or future. The model was used to predict land cover from 1985 to 2005, for a study region in the Piedmont of North Carolina. Temporal and spatial analyses focused on ecosystem services of carbon sequestration and biodiversity support as affected by forest fragmentation. This study offers a landscape-level identification of the relationships between spatial and temporal development patterns and the provision of ecosystem services. The project also represents the creation of a multi-annual land cover classification dataset of which few exist, thus providing a framework for further studies of landscape pattern and ecological processes.
