Using artificial neural networks to explain the attraction of jewel beetles (Coleoptera: Buprestidae) to colored traps.

Jewel beetles pose significant threats to forestry, and effective traps are needed to monitor and manage them. Green traps often catch more beetles, but purple traps catch a greater proportion of females. Understanding the function and mechanism of this behavior can provide a rationale for trap optimization. Jewel beetles possess UV-, blue-, green-, and red-sensitive photoreceptors, and perceive color differently from humans. Jewel beetle photoreceptor signals were calculated for tree leaf and tree bark stimuli, representing feeding and oviposition sites of adult jewel beetles respectively. Artificial neural networks (ANNs) were trained to discriminate those stimuli using beetle photoreceptor signals, providing in silico models of the neural processing that might have evolved to drive behavior. ANNs using blue-, green-, and red-sensitive photoreceptor inputs could classify these stimuli with very high accuracy (>99%). ANNs processed photoreceptor signals in an opponent fashion: increasing green-sensitive photoreceptor signals promoted leaf classifications, while increasing blue- and red-sensitive photoreceptor signals promoted bark classifications. Trained ANNs were fed photoreceptor signals calculated for traps, wherein they always classified green traps as leaves, but often classified purple traps as bark, indicating that these traps share salient features with different classes of tree stimuli from a beetle's eye view. A metric representing the photoreceptor opponent mechanism implicated by ANNs then explained catches of emerald ash borer, Agrilus planipennis, at differently colored traps from a previous field study. This analysis provides a hypothesized behavioral mechanism that can now guide the rational selection and improvement of jewel beetle traps.