CinSWUS: Mapping C Pools in the SW US with Multiangle EOS Data

Late 2005

• We have shown that community type (NB not land cover type) maps with an overall accuracy of ~77% can be produced using multi-angle imagery and support vector machine algorithms, an important improvement over the use of nadir-spectral approaches using maximum likelihood algorithms (~45%). See papers on the publications page that have Su as the first author.
  
  
• woody shrub fractional cover can be mapped at landscape scales by exploiting the directional signal in multi-angle data from MISR, when the background contribution is estimated using the weights of isotropic, geometric, and volume-scattering kernels (this approach has been further validated with a similar study using data from the European CHRIS/Proba mission).
  
  
• Serendipitous discovery: the GO modeling approach with injected background response provides a means to map estimates of crown cover, mean canopy height, woody biomass (and potentially understory density) for all woody plants, including forest. See Spring 2006 results and the improved Spring 2007 results.

| |