Calculating Change Curves for Multitemporal Satellite Imagery: Mount St. Helens 1980–1995

We developed and tested a method for analyzing multiINTRODUCTION temporal satellite imagery using change curves. The The recovery of vegetation following the 1980 eruption method is flexible and allows an analyst to extract specific of Mount St. Helens in southwest Washington has been change parameters from the curves depending on the reone of the major ecological stories of the Pacific Northsearch question of interest. Eight Landsat TM images of west during the past two decades. From a 550 sq km the Mount St. Helens, Washington, blast zone from 1984 homogeneous “moonscape” following the eruption on 18 to 1995 were geometrically and radiometrically corrected. May 1980, to a heterogeneous landscape that includes They were then transformed to estimates of green vegetaareas of lush vegetation today, the story of Mount St. tion cover. Unsupervised clustering was performed on the Helens’ recovery is unfolding in many ways. The study set of eight transformed images and polynomial curves reported in this article presents one way of reading this were fit to the cluster means. From these fitted curves, story, using a series of satellite images to characterize parameters of interest were extracted and returned to vegetation change since 1980. GIS layers, including number of years to reach 10% cover, Analysis of vegetation is one of the most important the greatest rate of cover increase during the study peuses of satellite remote sensing data (Lillesand and Kiefer, riod, and time-integrated cover. Statistical analysis indi1994; Coppin and Bauer, 1996). Although a variety of cated that the curves did a good job of representing the remote sensing techniques has been used to analyze vegchange trajectories of unclustered pixels. We demonetation change (e.g., Muchoney and Haack, 1994), much strated the use of change curve analysis by analyzing the of the research has focused on year-to-year changes, importance in the revegetation of Mount St. Helens of the rather than trends. Coppin and Bauer (1996) listed 11 different types of disturbance resulting from the volcanic general approaches for remote sensing change detection. eruption. The change curve analysis is useful in a variety Of these, one is based on a single date of imagery, eight of applications where the data are continuous, more than on year-to-year changes, and two (composite analysis and two dates of data are available, and the underlying quesmultitemporal linear data transformation) can analyze tion of interest relates to trends in the data. Elsevier two or more dates. Even when long time series have been Science Inc., 1999 used, the ecological analysis often focuses on year-to-year