I did an unsupervised classifications of depth-invariant processed images of different periods. How to know that one cluster from an image is the same cluster from a different image? As comparing different periods it is important to compare the same cluster from different images to detect coverage changes.
This is not possible. The unsupervised clustering approaches (ED and k-means) in SNAP are based on the random initializaiton of cluster centers. Even if you apply the same method on the same data you will get slightly different results. So you cannot assume that cluster1 of image A is related to cluster1 of image B.
If you want to compare classes between different input images you have to perform supervised classification and use the same training areas for all inputs.
Acknowledged, thank you for the great response, is there any alternatives to detect benthic changes without field data?
there is a chlorophyll and suspended matter processor (C2RCC), but it also needs some input variables on temperature and salinity