Assessment of anthropocentric representations of color patterns for biological relevance

Studying animal coloration is a notoriously difficult task as animals differ in their visual system properties (number of cone cells, spectral sensitivities of cone cells, filtering properties of ocular media, etc.) and thus may perceive and respond to stimuli differently than we do. These observations have led some to propose that human vision and the anthropocentric expressions of coloration based on our vision can never be used to detect variation in animal coloration. Yet this assertion is at odds with a long history of studying animal coloration in ecological, evolutionary, and behavioral frameworks. Cryptic coloration, male secondary sex traits, and mimicry patterns have been the subject of many fruitful research programs that have, until recently, relied on human perception of animal coloration. This suggests that human vision does well in detecting relevant animal coloration in nature, at least in some cases. Here, we attempt to reconcile these two opposing views. We tackle this issue by intentionally employing methodology that is heavily biased by human subjectivity and compare this to objective measures of coloration. Specifically, we used the reflectance spectra of the field guides and the RGB values from digital photography of birds as our anthropocentric representations of coloration. Images from field guides are potentially haunted by human bias as the bird coloration is (a) perceived by the human visual system, (b) painted to resemble a given bird, and then (c) printed in a guide using black, cyan, yellow, and magenta inks. Similar issues arise with digital photography as the total number of colors that can be recorded (the gamut) does not always account for all the colors a human could see and the resulting data is reduced down to only three data points per color: an R, G, and B value. In our first analysis, we compare the reflectance spectra of plumage elements for birds from museums with the corresponding spectra of plumage elements from a field guide. We also compare the actual bird reflectance spectra with RGB values from digital photographs of actual birds. We measured multiple plumage elements across 14 bird species. For sexually dimorphic species, we discerned between males and females. This resulted in 73 unique combinations of species, sex, and color element. The reflectance spectra from the field guides were drastically different from the actual birds, and this discrepancy was greatest for blue color elements. Yet despite this drastic variation in reflectance spectra, all three data sets (actual reflectance, field guide reflectance, and actual RGB values) captured the major components of variation in animal coloration as indicated by principal components analyses (PCAs). All three PCAs indicated similar data structure (i.e., similar amounts of variation in PC 1-3) with similar PC loadings. From this analysis, we conclude that human vision is capable of detecting the major sources of variation in animal coloration in the visible range (380nm to 700nm). Of course, humans cannot detect relevant variation in UV reflectance (as indicated by our data) nor can they detect polarized visual signals. Still, humans can generally rely on their visual systems to detect variation in color patterns that serve as the fodder for subsequent studies in ecology, evolution, behavior, and visual ecology. In our second analysis, we use avian and human visual detection models to determine how biased field guide images are toward the human visual system. The expectation is that the reflectance spectra from the field guide will differ from the original but will still stimulate human cone cells in a manner that creates the same approximate appearance to the human visual system. It is also expected that the filtering should create a large, more noticeable difference to non-human animals whose visual system properties differ from that of humans. In this analysis, we test these predictions by comparing the above mentioned reflectance spectra of the plumage of actual birds to the corresponding field guide images within visual detection models developed for birds and for humans. For each visual system, we compared the just-noticeable differences (JNDs) between corresponding spectra from the actual birds and field guide representations. To our surprise, we found that the JNDs were larger for the human visual system than for the bird visual systems. We discuss the possible mechanisms creating this pattern