Landsat-derived serpentine barren classification for locating McDonald's rock cress (Arabis macdonaldiana)

Spectral signatures were developed using Landsat Thematic Mapper imagery to find serpentine barren habitat occupied by the U.S. federally listed endan gered plant species Arabis macdonaldiana Eastwood. Focussing on mapping bar rens in the Klamath Mountains of Del Norte County, California, within the N orth Fork Smith River Basin, we developed four spectral signatures. We samp led spectral classes using 60 randomly located accuracy assessment points. Results were analyzed using chi-square contingency tables (alpha = 0.20) an d forward variable-selection logistic regression. The dependent variable wa s presence/absence of A. macdonaldiana. Independent variables were spectral signature, percent slope, transformed aspect, and elevation. Using geograp hic information systems, we were able to determine the amount of barren hab itat in each spectral class, by watershed. The null hypothesis that all fou r spectral classes had the same frequency of A. macdonaldiana was rejected (X-2, 6.07, 0.10 <p < 0.20), since class 4 was significantly different from the others (X-2 = 5.49, 0.05 <p < 0.025). The logistic regression model sh owed an 85% probability of detecting A. macdonaldiana using spectral classe s, percent slope, and elevation (R-2 = 0.39, df = 3, X-2 = 36.20). Probabil ity of detection increased as percent slope increased at all elevations. Th e Upper North Fork Smith River had the highest amount of barren area. Spect ral class 3 described 47.2% of the total amount of barren area, 1668.5 ha, in the basin.