A test site network for EOS-MODIS global land cover classification based on representativeness criteria

The land remote sensing communities require a database of global training, testing and validation sites to generate reliable maps of land cover based on the new generation of Earth Observing System (EOS) sensors. The MODIS (Moderate Resolution Imaging Spectroradiometer) land cover products to be produced by Boston University include global maps produced on a quarterly basis using the International Geosphere Biosphere Programme (IGBP) land cover classification system, and quarterly change detection products, all at 1 km resolution. Given the large data volume, area and high temporal frequency, it is critical that the authors develop a database of global sites that can be used to train classifiers, test algorithms and validate products. Global site selection is driven by data and analytical costs, the need to use sites where there is a wealth of existing data, and the diverse needs of global research. Thus, the site network must be representative of a number of meteorological, climatological, ecological, geographical and biological criteria. A primary objective of the research presented is to evaluate the effectiveness and representativeness of the existing site network based on physical and biological gradients