You should really get it into the published literature. The question really needs to be qualified.
You can begin with a simple Dr Google, but I would start looking at referenced papers like...
particular if you are getting into anything more than ice-no ice studies
well, the most obvious is just use the visible bands... ice looks like ice and not the surrounding non-ice covered surfaces.... assuming that the ice isn't surrounded by snow.
If you are interested in differentiating ice versus snow and a whole load of other things, then that is an active field of investigation. Not all things are as they appear... and they all appear differently depending on your bands. So the answer may be really simple or not
Thanks Dan. Surely, but in some case there are more visible bands. I have chosen images from August and September only so that its the end of melt season and there is no snow in the pictures. I am looking at the period between 1984 and 2016 with few years gaps because of clouds and the Landsat 7 error. However, I have some images with minor clouds that I should consider. I am not sure if I should mask them or If I am going to have time for that as my dissertation is quite complex.
I would be useful to mask the cloud years to add into the analysis. If the year before did/didn't have ice and the year after was the same condition, then it would be safe to assume that the intervening year was the same. In the other areas (ie non-cloud obscured) it would be possible to use that data to show that year's advance/retreat. It really depends on how much you need to fill in the time gaps.
I see your point but currently I need to save time because I have 4 months to finish my dissertation. It sounds enough but I work full time and I need to look up many of these baby steps on Esri or other websites. I do study GIS but not remote sensing so that I need more practice in this field.
I will see how difficult it is to cloud mask rasters and try do do so. Even if I don't I will end up with about 25 images that is not that bad. Would you also recommend combining two images in Landsat 7 to fill the error gap? I appreciate your help.