Predicting Ecologically Important Vegetation Variables from Remotely Sensed Optical/Radar Data Using Neural Networks

The large data sets engendered during the EOS era will enhance the temporal, spatial, and spectral coverage of the earth (Asrar and Greenstone 1995; Wharton and Myers 1997). The satellite digital data sets and ancillary data products will require the development of efficient algorithms that can incorporate and functionally utilize disparate data types. Numerous vegetation variables, e.g. leaf area, height, canopy roughness, land cover, stomatal resistance, latent and sensible heat flux, radiative properties, and many others, are required for global and regional studies of ecosystem processes, biosphere/atmosphere interactions, and carbon dynamics (Asrar and Dozier 1994; Hall et al. 1995). The success of efforts to extract vegetation variables such as these from remotely sensed data and available ancillary data will determine the degree and scope of vegetation-related science performed using EOS data.