Comparisons among Vegetation Indices and Bandwise Regression in a Highly Disturbed, Heterogeneous Landscape: Mount St. Helens, Washington

Spectral vegetation indices have been used extensively to predict ecological variables, such as percent vegetation cover, above-ground biomass, and leaf-area index. We examined the use of various vegetation indices and multiple linear regression using raw spectral bands for predicting vegetation cover in a landscape characterized by high variability in vegetation cover and soil properties. We were able to improve the explanatory value of several vegetation indices by using regression fitting techniques including log transformations and polynomial regressions. We expected soil-adjusted indices to perform better than nonadjusted indices. However, soil-adjusted vegetation indices based on a ratio of red and near-infrared bands explained 55–65% of the variability in vegetation cover, while two nonadjusted indices each explained 70%. An index using six spectral bands explained 40%. The best multiple regression model used the red and near-infrared bands and explained 75% of the variability in vegetation cover. Among the soil-adjusted indices, an index which used a computed soil line performed best. Ratio-based vegetation indices were less sensitive to shadow influences, but this influence was outweighed by the advantages of multiple regression against original bands.