# Analysing surprising clustering results in spectral clustering

I have two-dimensional data that I am trying to cluster. The data looks like this:

I have been trying to cluster this using spectral clustering, as in sklearn.cluster.SpectralClustering and am somewhat surprised by the results I get. For example, for a number of clusters of 4, I get the following clusters:

All looks fine, except for the light blue point at the top left very close to the green cluster. How do I figure out why this point was clustered with the light-blue cluster (as opposed to the green one), i.e. why the algorithm thinks this point is more similar to the other blue points rather than the green ones?