Regarding prediction of eye color categories:
The accuracy in predicting 3-category eye color was 0.92 for blue, 0.74 for intermediate, and 0.93 for brown,...There is some discussion about the limited quality of the photos due in part to the un-standardized lighting conditions.
I wonder how much better prediction would have been in a more diverse sample.
Digital quantification of human eye color highlights genetic association of three new loci.
Liu F, Wollstein A, Hysi PG, Ankra-Badu GA, Spector TD, Park D, Zhu G, Larsson M, Duffy DL, Montgomery GW, Mackey DA, Walsh S, Lao O, Hofman A, Rivadeneira F, Vingerling JR, Uitterlinden AG, Martin NG, Hammond CJ, Kayser M.
PLoS Genetics 2010 May 6;6:e1000934.
Abstract: Previous studies have successfully identified genetic variants in several genes associated with human iris (eye) color; however, they all used simplified categorical trait information. Here, we quantified continuous eye color variation into hue and saturation values using high-resolution digital full-eye photographs and conducted a genome-wide association study on 5,951 Dutch Europeans from the Rotterdam Study. Three new regions, 1q42.3, 17q25.3, and 21q22.13, were highlighted meeting the criterion for genome-wide statistically significant association. The latter two loci were replicated in 2,261 individuals from the UK and in 1,282 from Australia. The LYST gene at 1q42.3 and the DSCR9 gene at 21q22.13 serve as promising functional candidates. A model for predicting quantitative eye colors explained over 50% of trait variance in the Rotterdam Study. Over all our data exemplify that fine phenotyping is a useful strategy for finding genes involved in human complex traits.