A bivariate binary model is developed for estimating the change in land cover from satellite images obtained at two different times. The binary classifications of a pixel at the two times are modeled as potentially correlated random variables, conditional on the true states of the pixel. The model can be fit to a "training" set of pixels for which the true states are presumed from a reference dataset, and two methods are proposed for using the results of that fit to predict the true states in a separate set of pixels having only classification information. Applied to two images taken over Mexico by the LANDSAT Multi-Spectral Scanner, this methodology finds statistically significant temporal correlation of pixel classifications and illustrates that adjustment for this correlation is important for obtaining accurate estimates of changes in land cover.

Key Words
Bivariate binary; Correlation; Remote sensing.

Paul A. Murtaugh is Assistant Professor, Department of Statistics, Oregon State University, Corvallis, Oregon 97331. Donald L. Phillips is Research Biologist, U.S. Environmental Protection Agency, National Health and Environmental Effects Research Laboratory, 200 S.W. 35th Street, Corvallis, Oregon 97333.