Sampling thematic mapper imagery for land use data

Sampling Landsat TM imagery for land use is a very efficient method of data acquisition for many environmental models. These grid-based models are often used in the modeling of the long range transport and deposition of air pollutants. In this article, a simple procedure for estimating proportions of different land use in grid cells with sample unit Landsat TM picture element is presented. In the procedure, the bias of the maximum likelihood classification of the sample is corrected using double sampling. For a small number of random pixels, the true classification has to be provided additionally. From the comparison of the true classification and the maximum likelihood classification, the amount of bias in each estimate from misclassification is estimated. In this article, the case with several land-use classes is discussed.