Spectral Discrimination of Plant Functional Types and Species across diverse North American Ecosystems
Dar
A
Roberts, Department of Geography, U.C. Santa Barbara, dar@geog.ucsb.edu
(Presenting)
Keely
Roth, Department of Geography, U.C. Santa Barbara, klroth@geog.ucsb.edu
Philip
E
Dennison, Department of Geography, University of Utah, dennison@geog.utah.edu
Abigail
Guess, Department of Geography, University of Utah, abigail.guess@geog.utah.edu
Imaging spectrometry has considerable potential for providing improved capabilities to discriminate plant species and plant functional types globally. In this poster, we report upon a generalized approach designed to build statistically robust spectral libraries for species discrimination and for map validation. We use multiple endmember spectral mixture analysis (MESMA) with two endmember models to map plant species. We introduce an interative endmember selection approach to MESMA, in which endmembers for each class of interest are selected from a spectral library in such a way to produce maximum accuracies as expressed by a kappa statistic. We apply this generalized protocol to four AVIRIS data sets representing distinct ecosystems including a mid-latitude broadleaf forest (SERC, MD), a continental great basin sagebrush to mixed forest (Wasatch Range, UT), a semi-arid coastal shrubland (Santa Barbara, CA) and a pacific northwest, coniferous forest (Wind River, WA). We also apply this approach to evaluate spectral separability at species and plant functional type levels. We conclude with an analysis of the impact of coarsening spatial resolution on the ability to discriminate PFT at Wasatch at 20, 40 and 60 m spatial resolutions.
Using iterative endmember selection, map accuracy varied between a low of 59% at SERC, mapping eight tree genera to a high of 73% at Wind River for 11 species dominants. An intermediate accuracy was observed at Santa Barbara, of 65% for 15 dominant species. Map accuracy tended to improve at the life form or PFT level, improving to 73% for Santa Barbara (10 PFTs plus soil) and 96.8% for Wind River. PFT map accuracies at Wasatch were also high at 85.6%. Spatial degradation of 20 m Wasatch data to 60 m, resulted in only a minor decline in accuracy if finer spatial resolution endmembers were used for modeling (83.7%) but a larger drop in accuracy if the original endmembers were derived from 60 m data (75.9%).
Presentation Type: Poster
Poster Session: Ecosystems Science
NASA TE Funded Awards Represented:
Roberts, Dar
Spatial, Spectral and Temporal Requirements for Improved Hyperspectral Mapping of Plant Functional Type, Plant Species, Canopy Biophysics, and Canopy Biochemistry