Do crown shape retrievals match vegetation type distributions?
Yes, they do (remembering that both radius (r) and crown aspect ratio (b/r) are effective parameters). The crown shape map below was made by inverting the SGM model against MISR red band data (9 looks), with the soil-understory (non-shrub/tree) background estimated a priori using the isotropic, geometric and volume scattering kernel weights of a LiSparse-RossThin BRDF model. The details are in various papers on the GO model results that we published in 2006-2008. The vegetation type map is a generalized version originally from the Jornada LTER. Apart from the intrinsic usefulness of mapping crown shape, this highlights one of the advantages of using model inversion over purely empirical methods (e.g., regression techniques, artificial neural networks, k-NN, and learning machine algorithms): model internal parameters can be checked for the reasonableness of their ranges and spatial distributions. When implementing these GO model retrievals, we were able to ensure that the retrievals made sense in terms of the ranges and distributions of both crown shape (mean aspect ratio) and mean crown radius; maps of these can be seen to be very reasonable (even though the radius retrievals are effective rather than true). Note the more oblate crown shape values for mesquite-dominated areas, the more spherical crown shape values for the creosotebush-dominated areas, and the more prolate crown shape values for the woodland areas on the San Andres Mountains to the east. The boundaries between creosotebush (cyan), tarbush (blue/purple), and the town of Organ (purple) can also be seen in the SE quadrant of the b/r ratio map.
| 
| 