yes, this is where it becomes tricky. If there are signatures in the pixels which are not covered by the endmembers, you will get really weird results (because fully constrained unmixing will always try to sum all three classes to 1 [=100%] per pixel)
Therefore you can also try unconstrained unmixing. The pixel values of the three classes will no longer sum up to 1 but you will get more realistic relative distributions.
I recommend this paper, it helped me a lot. https://archive.ll.mit.edu/publications/journal/pdf/vol14_no1/14_1survey.pdf or also this one: https://ieeexplore.ieee.org/abstract/document/97472
Both are well illustrated.
You can use the mask manager and use expressions to locate pixels which fulfill certain criteria. Examples are given on page 23 of this tutorial: Synergetic use of S1 (SAR) and S2 (optical) data and use of analysis tools
If you want to overlay information from the unmixing with the imported map, you will have to collocate both products (page 15, same tutorial)