398 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 34, NO. 2, MARCH 1996 emote Sensing of Forest Change sing Artificial Neural Networks Sucharita Gopal and Curtis Woodcock, Associate Member, IEEE Abstract-A prolonged drought in the Lake Tahoe Basin in California has resulted in extensive conifer mortality. This phe- nomenon can be analyzed using (multitemporal) remote sensing data. Prior research in the same region used more tradition& methods of change detection [SI, [30]. This paper introduces a third approach to change detection in remote sensing based on artificial neural networks. The neural network architecture used is a Multilayer Feedforward Network. The results of the study indicate that the artificialneural network (ANN) estimates conifer mortality more accurately than the other approaches.Further, an analysis of its architecture reveals that it uses identifiable scene characteristics-the same as those used by a GrammSchmidt transformation.ANN models offer a viable alternative for change detection in remote sensing. I. INTRODUCTION EURAL networks hold the potential for improving a variety of tasks in remote sensing and image processing. They represent a fundamentally different approach to problems like pattern recognition, as they do not rely on statistical relationships. Instead, neural networks adaptively estimate continuous functions from data without specifying mathemat- ically how outputs depend on inputs (Le., adaptive model-free function estimation using a nonalgorithmic strategy). To date, most of the efforts to use neural networks in remote sensing have involved image classification with considerable success (e.g., [2], [12], [14], [18], [22]). Neural network classifiers outperform conventional classifiers mainly due to their lack of assumptions about normality in datasets, considerable ease in using multidomain datasets, and perhaps, in capturing some of the inherent nonlinearity in such data. The purpose of this paper is to test the utility of neural networks in change detection. As far as we know, this is the first paper to use neural networks for change detection in remote sensing. In particular, neural networks are used to estimate the degree of conifer mortality in the Lake Tahoe Basin. This area has been studied in the past [8], [30], and this project uses their datasets. Change detection studies in remote sensing involve the use of multitemporal datasets, Le., sequential images taken of the same area. Techniques to analyze the location, nature, and magnitude of changes serve two distinct purposes [37]: comparative analysis of independently produced classifica- tions, and simultaneous analysis of multitemporal data. Singh E371 provides an excellent review and comparison of various Manuscript received December 16, 1994; revised July 20, 1995. This work was supported by the National Science Foundahon under Grant SBR-9300633 The authors are with the Department of Geography, Boston University, Boston, MA 02215 USA. Publisher Item Identifier S 0196-2892(96)01005-4. change detection methods such as image differencing, princi- pal component analysis, and multitemporal regression. Since that time, much additional literature has been published on the logic of change detection. In the area of timber inventory and forest management, changes in forest cover caused due to defoliation by insects [31], [32], [39], [41] and landuse andor other human-induced factors (e.g., pollution stress) [7], [9], [38] are significant. The perspective on change detection in this paper is a bit different from the normal in remote sensing, which is to detect changes in land use or land cover. In these studies the nature of the change is categorical, or between different land- cover classes 1371. In this study, the intent is to measure the magnitude of change [25], [26], which in this case corresponds to the number of trees in a forest that have died. E. BACKGROUND: THE MAPPING PROJECT AND PRIOR &PROACHES TO DETECTION OF CONIFER MORTALITY For the past several years, Boston University Center for Remote Sensing and the U.S. Forest Service (USFS) Region 5 Remote Sensing Group have been involved in developing a set of new methods for mapping and inventorying forest vegetation using remote sensing and geographical information systems (GIS). An important innovation in this mapping project has been the use of an image segment procedure [42] to define the map units early, in the mapping process, enabling subsequent analysis to be conducted on stands rather than a per-pixel basis. The image segmentation algorithm uses raw TM bands and a texture channel and produces regions that are the polygons used in the final map. The other primary innovation is the use of a forest canopy reflectance model to map forest stand structure. An overview of the mapping procedure and its results are given in Woodcock et al. [43]. During the period of this project, the region experienced prolonged drought that has resulted in considerable mortality of conifer trees. Two prior approaches for detecting conifer mortality in the Lake Tahoe Basin have been tested. Macomber and Woodcock [30] used a method that measures the decrease in crown cover between the two dates of the images. Estimates of crown cover are obtained from the Li-Strahler model [29] for each conifer stand for two dates of imagery. Three levels of change in crown cover were combined with three crown cover classes to stratify the forest areas. Field samples were collected for each stratum and total conifer mortality was estimated at 15% of the total timber volume between 1988 and 1992. The patterns in the means for the strata followed the anticipated patterns with respect to mortality, indicating the viability of ’ 0196-2892/96$05.00 0 1996 IEEE