I am classifying glaciers using the iso cluster unsupervised classification tool. However, parts of the glaciers are missing due to debris and shadow. Since my composite images(Landsat TM bands 2,4,5) clearly distinguish ground from ice I use them as backrgound layer to correct the partial misclassification using on-screen digitising (I found the 'reshape tool' the most useful editing function for this task). In order to be scientifically correct I should use ground truth for each image for each year in the same month when the image was acquired. Since this is clearly impossible my question is how do spatial analyst experts solve this problem. This is the only way to classify ice extent for my images, therefore I feel like it is a Catch 22 of remote sensing analysis.