That color chart is just one of those that are used normally in forensics. It's based on probabilities. Most "very fair" skinned people in modern populations are blue eyed. However, not all. I'm in category one and have brown eyes. It's not common, but it happens.
If you go to the text of the paper and scroll down you'll see the color chart the authors of that paper were using. The actual test based on that probably has a shortened chart. Regardless, the strong probability based on 36 snps was dark brown-black. Based on current knowledge of pigmentation that's as much as they can say.
It would be interesting to see the prediction for samples like Loschbour, LaBrana, Bichon, Villabruna, etc., to see if there was some variation among the WHG. This test wasn't available when those studies and predictions were published.