Spectral Mixture Analysis of Montane Forest Biophysical Parameters: I a Comparison of Endmembers from Airborne Imagery and a Field Spectroradiometer

Sub-pixel scale fractions computed from spectral mixture analysis (SMA) provide improvements over vegetation indices for extracting forest biophysical information such as LAI, biomass and NPP for use in forest inventories and regional scale carbon budget models. The acquisition of endmember spectra of forest canopy, ground vegetation and shadow is a critical input to spectral mixture analysis. In this paper, we compare the use of three sets of endmembers: image endmembers extracted directly from airborne imagery, reference endmembers measured in the field using a portable spectroradiometer, and an integrated set which combined both image and reference endmembers. Each set of endmembers was used in spectral mixture analyses of multiscale airborne CASI imagery at 60cm, 1m and 2m resolutions acquired July 1998 over mountainous terrain in Kananaskis Provincial Park, Alberta. The scene fractions were validated using a sub-pixel multi-resolution classification strategy. To test their ability to predict biophysical variables, independent LAI measurements were first collected with LAI-2000 and TRAC instruments. Separate linear regressions were performed for each of the SMA fractions as well as for NDVI. The reference endmember set was the best predictor ofTRAC based measurements of LA I, with r = 0.69 and 0.67 for 1m and 2m imagery respectively. The integrated and reference endmember sets predicted effective LAI from the LAI-2000 with r = 0.62 and 0.72 for the 1m and 2m data. NDVI results were r = 0.33 and 0.34 for the TRAC and 0.45 and 0.44 for LAI-2000 measurements at 1m and 2m resolutions, respectively. These results suggest that the acquisition of reference endmembers is needed to achieve the best overall predictive ability using spectral mixture analysis, and that fractions from image endmembers also show significant improvements over NDVI without the need of reference spectra.