While classifying Landsat 8 imagery for change detection purposes,what is the maximum acceptable co -variance level of training sites in ArcMap? I have seen co- variance figures going to 1,000,000 - plus, yet for Landsat 7 Parece and Campbell (2013) talk of figures not higher than 2000 as acceptable.
covariance is covariance... can you explain what you mean by an acceptable level? specifically, why a particular level would be acceptable when using data from one sensor and not another? Do you have some results you are trying to assess?
