I have to carry out a Principal Component Analysis (sklearn.decomposition.PCA) on a dataset composed of several images. Some of these images have very large gaps due to the removal of clouds.
Is there the possibility to apply a mask to all the NoData values for each layer so as the algorithm does not take them into consideration, and consider only the valid values of other underlying/overlying layers (which do not have NoData in that position)? So to have final PCs without gaps.
I'm not sure if my request is clear, I try to give an example: I have a dataset consisting of 4 layers. In the pixel in position (x = 27; y = 32) the different 4 layers have: Layer-1 = value; Layer-2 = value; Layer-3 = NoData; Layer-4 = value. I need that the alghoritm takes into consideration the Layers-1,2,4 and discard the Layer-3 to build the PCs values of only that pixel.
