A Quantitative Comparison of Change-Detection Algorithms for Monitoring Eelgrass from Remotely Sensed Data Robb D. Macleod and Russell G. Congalton Abstract nent of coastal and estuarine ecosystems (Milne and Milne, The eelgrass (Zostera marina L.) population in Great Bay, 1951; Ackleson and Klemas, 1987; Short, 1989; Ferguson et New Hampshire has recently undergone dramatic changes. A 1993). It grows in bays, and coastal Oceans reoccurrence of the 1930s wasting disease and decreasing throughout the northern temperate regi0ns the and water quality due to pollution led to a reduction in the eel- Can rival the productivity of agricultural crops ( ~ h a ~ e r et al., grass population during the late 1980s. Currently, the eel- 1984). In addition, eelgrass meadows ~rovide habitat for nu- grass populations in ~~~~t B~~ have a remark- merous organisms, including coastal fish, lobsters, crabs and able recovery from the decline in the late 1980~. Eelgrass is scallops, and a food source for waterfowl. Eelgrass meadows important in our estuarine ecosystems because it is utilized also increase water quality by filtering sediments and nutri- as habitat by many and non-commercial ents within the water (Short, 1989). It is therefore important isms and is a food source for wateqfowl. In order to monitor to maintain healthy ~ o ~ u l a t i o n s of eelgrass in to en- the eelgrass populations in Great Bay, a change detection Sure the continuing ~ r o s p e r i t ~ of coastal and estuarine eco- analysis was performed to determine the fluctuation in eel- systems. grass meadows over time. Currently, the two major problems that are severely im- change detection is a technique used to determine the pacting eelgrass meadows throughout the world are the wast- change between two or more time periods of a ob- ing disease (Labyrinthula zosterae) and pollution (short et ject of study. Change detection is an important process in ~1.' 1991). As a result, the eelgrass population in e re at B~Y, monitoring and managing natural resources and urban de- New Hampshire has gone through dramatic changes in the velopment because it quantitative analysis of the last decade (Short et al., 1993). Monitoring the spatial distri- spatial distribution in the population of interest. A large bution in eelgrass habitat is an important part of understand- number of change-detection techniques have been developed, ing the changes in eelgrass which in turn but little has been done to assess the accura- ensure the viability of coastal and estuarine ecosystems (Fer- cies of these techniques. guson et al., 1993). Historically, eelgrass and other sub- In this study, post-classification, image differencing, and merged aquatic vegetation (SAV) have been monitored in the principal components change-detection techniques were used field with either permanent transects or stations (~ckleson to determine the change in eelgrass meadows with Landsat and Klemas, 1987). However, the cost of field sampling has Thematic Mapper (TM) data. Low altitude (1,000 m), oblique become expensive for large areas and is now used primarily aerial photography combined with boat surveys were used as to assess the accuracy of more efficient techniques such as reference data. A proposed change-detection error matrix aerial photography (Short et al., 1986; ~ckleson and ~lemas, was used to quantitatively assess the accuracy of each 1987; Ferguson et al., 1993). More recently, satellite imagery change-detection technique. The three different techniques has been used to detect eelgrass and sAV (AcklesOn were then compared using standard accuracy assessment and Klemas, 1987; Jensen et al., 1993; Luczkovich et al., procedures. The image differencing change-detection tech- lgg3; Zainal et ~l.9 lgg3). nique performed significantly better than the post-classifica- An increasingly popular application of remotely sensed tion and principal components analysis. The overall accuracy data is for change detection. Change detection is the process of the image differencing change detection was 66 percent of identifying differences in the state of an object or phenom- with a Khat value of 0.43. enon by observing it at different times (Singh, 1989). Four ~ h j ~ study provided an of ~~~d~~~ ~ h ~ ~ ~ ~ i ~ aspects of change detection are important when monitoring Mapper to detect submerged aquatic vegetation and the natural resources: (1) detecting that changes have occurred, methodology for comparing change detection techniques us- (2) identifying the nature of the change, (3) measuring the ing a proposed change detection error matrix and standard areal extent of the change, and (4) assessing the spatial pat- accuracy assessment procedures. In addition, this study tern of the change (Brothers and Fish, 1978; Malila, 1980; showed that image differencing was better than the post-clas- Singh, 1986). Techniques to ~erform change detection with sjfication or components techniques for detecting satellite imagery have become numerous because of increas- changes in submerged aquatic vegetation. ing versatility in manipulating digital data and increasing Introduction Photogrammetric Engineering & Remote Sensing, Eelgrass (Zostera marina L.), a true flowering plant that com- Vol. 64, No. 3, March 1998, pp. 207-216. pletes its life cycle in shallow sea water, is a critical compo- 0099-1112/98/6403-207$3.00/0 Department of Natural Resources, University of New Hamp- O 1998 American Society for Photogrammetry shire, Durham, NH 03824 (russ.congalton@unh.edu). and Remote Sensing PE&RS March 1998 207