Classification of Forest and Crop Types From Hyperspectral and Multi-Angular Data
Mitchell
Andrew
Schull, Boston University, schull@bu.edu
(Presenting)
Liang
Xu, Boston University, bireme@gmail.com
Pedro
Carmona, Univeristat Juame I, latorre@lsi.uji.es
Arindam
Samanta, Boston University, arindam.sam@gmail.com
Yuri
Knyazikhin, Boston University, jknjazi@bu.edu
Ranga
Myneni, Boston University, ranga.myneni@gmail.com
Numerous studies have demonstrated the ability of hyperspectral data to discriminate vegetation types, however most methods rely on empirical data and are site specific and therefore may not be extendable to operational use. In this poster we provide a physically based approach for separation of dominant forest and crop types using hyperspectral and multi-angular data. The radiative transfer theory of canopy spectral invariants facilitates the parameterization of the canopy reflectance in terms of the leaf spectral scattering and two spectrally invariant and structurally varying variables – recollision and directional escape probabilities. The methodology is based on the idea of retrieving spectrally invariant parameters from hyperspectral multi angular data first, and then relating their values to structural characteristics of three-dimensional canopy structure. Theoretical and empirical analyses of ground, airborne data acquired by Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) over two sites in New England and satellite suggest that the canopy spectral invariants convey information about canopy structure at both the macro and micro scales. The total escape probability (one minus recollision probability) varies as a power function with the exponent related to the number of nested hierarchical levels present in the pixel. Its base is a geometrical mean of the local total escape probabilities and accounts for the cumulative effect of canopy structure over a wide range of scales. The ratio of the directional to the total escape probability becomes independent of the number of hierarchical levels and is a function of the canopy structure at the macro scale such as crown shape and size, ground cover, within-crown foliage density and portion of sunlit and shadow leaves. These properties allow for the natural separation of dominant forest and crop classes based on the location of points on the total escape probability vs the ratio log-log plane.
Presentation Type: Poster
Poster Session: Carbon Cycle Science
NASA TE Funded Awards Represented:
Knyazikhin, Yuri
Vegetation Biophysical Parameter Suite from MISR for Ecological Applications
Myneni, Ranga
A SYNERGISTIC STUDY FOR LIDAR AND PASSIVE OPTICAL REMOTE SENSING OF FOREST HORIZONTAL STRUCTURE IN SUPPORT OF DESDYNI MISSION