Using Delaunay triangulation to sample whole-specimen color from digital images

Color variation is one of the most obvious examples of variation in
nature, but biologically meaningful quantification and interpretation of
variation in color and complex patterns is challenging. Many current
methods for assessing variation in color patterns classify color patterns
using categorical measures, provide aggregate measures that ignore spatial
pattern, or both, losing potentially important aspects of color pattern.
Here, we present Colormesh, a novel method for analyzing complex color
patterns that offers unique capabilities. Our approach is based on
unsupervised color quantification combined with geometric morphometrics to
identify regions of putative spatial homology across samples, from
histology sections to whole organisms. Colormesh quantifies color at
individual sampling points across the whole sample. We demonstrate the
utility of Colormesh using digital images of Trinidadian guppies (Poecilia
reticulata), for which the evolution of color has been frequently studied.
Guppies have repeatedly evolved in response to ecological differences
between up- and downstream locations in Trinidadian rivers, resulting in
extensive parallel evolution of many phenotypes. Previous studies have,
for example, compared the area and quantity of discrete color (e.g., area
of orange, number of black spots) between these up- and downstream
locations neglecting spatial placement of these areas. Using the Colormesh
pipeline, we show that patterns of whole-animal color variation do not
match expectations suggested by previous work. Colormesh can be deployed
to address a much wider range of questions about color pattern variation
than previous approaches. Colormesh is thus especially suited for analyses
that seek to identify the biologically important aspects of color pattern
when there are multiple competing hypotheses, or even no a priori
hypotheses at all.