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

Spectral vegetation indices have been used extensively INTRODUCTION to predict ecological variables, such as percent vegetation On 18 May 1980, Mount St. Helens in the State of cover, above-ground biomass, and leaf-area index. We exWashington erupted with catastrophic, landscape-scale amined the use of various vegetation indices and multiple effects (Lipman and Mullineaux, 1981). In a mere 10 linear regression using raw spectral bands for predicting min, an area of approximately 500 sq km was devastated, vegetation cover in a landscape characterized by high with destruction of substantially all above-ground vegetavariability in vegetation cover and soil properties. We tion (Frenzen, 1992). The ensuing debris, mud, and pywere able to improve the explanatory value of several roclastic flows further transformed the landscape from vegetation indices by using regression fitting techniques one of lush Pacific Northwest forests into a seemingly including log transformations and polynomial regresbarren expanse. sions. We expected soil-adjusted indices to perform better The Mount St. Helens eruptions and subsequent rethan nonadjusted indices. However, soil-adjusted vegetacovery have gathered world-wide public attention. tion indices based on a ratio of red and near-infrared Mount St. Helens National Volcanic Monument remains one of the most visited natural wonders in the Pacific bands explained 55–65% of the variability in vegetation Northwest. The attention from scientists has been at cover, while two nonadjusted indices each explained least as great. A compendium published in 1994 listed 70%. An index using six spectral bands explained 40%. 637 publications and 69 research abstracts related to the The best multiple regression model used the red and eruptions and their aftermath (Frenzen et al., 1994). near-infrared bands and explained 75% of the variability However, the research opportunities afforded by Mount in vegetation cover. Among the soil-adjusted indices, an St. Helens have gone essentially overlooked by the reindex which used a computed soil line performed best. mote sensing community. The research reported in this Ratio-based vegetation indices were less sensitive to article is part of a program designed to exploit some of shadow influences, but this influence was outweighed by the remote sensing opportunities afforded by this site. the advantages of multiple regression against original Mount St. Helens provides a nearly unique opportubands. Elsevier Science Inc., 1998 nity to address some of the more perplexing problems in remote sensing. Since the early days of satellite remote sensing, scientists have sought to use multispectral imagery to measure assorted ecological variables related to vegetation amount. Much of this research has attempted