Diurnal biting flies are strongly attracted to blue objects. This behaviour is widely exploited for fly control, but its functional significance is debated. It is hypothesised that: blue objects resemble animal hosts; blue surfaces resemble shaded resting places; and blue attraction is a by-product of attraction to polarised light. We computed the fly photoreceptor signals elicited by a large sample of leaf and animal integument reflectance spectra, viewed under open/cloudy illumination and under woodland shade. We then trained artificial neural networks (ANNs) to distinguish animals from leaf backgrounds, and shaded from unshaded surfaces, in order to find the optimal means of doing so based upon the sensory information available to a fly. After training, we challenged ANNs to classify blue objects used in fly control. Trained ANNs could make both discriminations with high accuracy. They discriminated animals from leaves based upon blue-green photoreceptor opponency, and commonly misclassified blue objects as animals. Meanwhile, they discriminated shaded from unshaded stimuli using achromatic cues and never misclassified blue objects as shaded. We conclude that blue-green opponency is the most effective means of discriminating animals from leaf backgrounds using a fly’s sensory information and that blue objects resemble animal hosts through such mechanisms.

Raw reflectance spectra were obtained as follows:

72 animal reflectance spectra were originally collected by Galván I. & Wakamatsu, K. (2016) Color measurement of the animal integument predicts the content of specific melanin forms. RSC Advances 6: 79135-79142. Two additional spectra were not analysed because reflectance was not available for the complete 300-700nm range. Spectra were provided as x,y values at 10nm wavelength resolution and have been linearly interpolated for 2nm resolution. 

72 leaf reflectance spectra were obtained from the Floral Reflectance Database (www.reflectance.co.uk), Arnold S.E.J. et al. (2010) FReD: The Floral reflectance database - A web portal for analyses of flower colour. PLoS ONE 5: e14287. Sampling was random, based on accession numbers. Spectra were obtained as x,y values and subsampled for 2nm resolution.

11 spectra for biting fly control devices were obtained from a variety of sources. These were: manually extracted from figures in Green C.H. & Flint S. (1986) An analysis of colour effects in the performance of the F2 trap against Glossina pallidipes Austen and G. morsitans morsitans Westwood (Diptera: Glossididae). Bull. Ent. Res. 76: 409-418, and Green C.H. (1988) The effect of colour on trap- and screen-oriented responses in Glossina palpalis palpalis (Robineau-Desvoidy) (Diptera: Glossinidae). Bull. Ent. Res. 78: 591-604; obtained as x,y values from the supplementary materials of Lindh J.M. et al. (2012) Optimizing the colour and fabric of targets for the control of the tsetse fly Glossina fuscipes fuscipes. PLoS Negl. Trop. Dis. 6: e1661, and linearly interpolated for 2nm resolution; and measured by the authors in earlier work published in  Santer R.D. et al. (2019) Optimising targets for tsetse control: taking a fly's eye view to improve the colour of synthetic fabrics. PLoS Negl. Trop. Dis. 13: e0007905.

Irradiance spectra were originally collected by Endler J.A. (1993) The color of light in forests and its implications. Ecological Monographs 63: 1-27, and were provided as x,y values at 2nm resolution. 

Fly photoreceptor sensitivities were based on data from Hardie R.C. (1986) The photoreceptor array of the dipteran retina. Trends Neurosci, 9: 419-423, and Hardie R.C. et al. (1989) The compound eye of the tsetse fly (Glossina morsitans morsitans and Glossina palpalis palpalis). J. Insect Physiol. 35: 423-431. These were obtained from the supplementary materials of Santer R.D. (2017) Developing photoreceptor-based models of visual attraction in riverine tsetse, for use in the engineering of more-attractive polyester fabrics for control devices. PLoS Negl. Trop. Dis. 11: e0005448.

Photoreceptor excitation values were calculated based upon the above functions, following the methods in the accompanying manuscript.

R code to run ANN models was developed by the authors.

The datasheet is provided in .xls format and can be opened in a standard spreadsheet program.

R code is provided as a .R file.