The Evolution of Visual Patterning in North American Freshwater Fishes

Abstract

Visual patterns make up some of nature's most remarkable traits, which have puzzled biologists for centuries. The evolution of visual patterns can occur through multiple mechanisms, such as natural selection for cryptic camouflage, or sexual selection. In particular, sexual selection has led to some exceptionally flamboyant displays. However, our understanding of how the pressures of sexual selection shape visual patterns lacks a sufficient theoretical and empirical basis. Recent extensions of the sensory drive and sensory bias models of sexual selection have hypothesized that pattern evolution may be related to the visual environments that animals inhabit. According to the processing bias hypothesis, an extension of the efficient coding hypothesis, sexual signals will mimic the visual structure of their habitats so that these visual signals can be efficiently encoded by the receiver's visual systems. Here, I test the predictions of the processing bias hypothesis through both computational and behavioral trials. I use fish called darters (Percidae: Etheostoma) as a model system, given the diversity of male visual patterns and habitat preferences in this genus. First, I collected images of eleven species of darters, as well as underwater photographs of their habitats. To examine the relationship between the spatial statistics of fish and their habitats, I used Fourier analysis and convolutional neural networks. I found that the Fourier slope of male darter nuptial patterns significantly correlates with the Fourier slope of their habitats, while no such correlation exists for females. Using convolutional neural networks, I did not find evidence to reinforce the results of the Fourier analysis, but did find evidence that female darters are more camouflaged with their environments than their male counterparts. Finally, I used behavioral assays to explicitly test preferences in one species, Etheostoma caeruleum, for visual stimuli matching the visual statistics of their preferred habitat. Although my results do not demonstrate clear support for the processing bias hypothesis, this work advances a new theoretical framework for understanding complex visual traits. Furthermore, my results show how convolutional neural networks and Fourier analysis provide tools that can advance our understanding of the evolution of visual patterns.