Do not be distracted by pretty colors: a comment on Olsson et al.

spatial and temporal integration can counterbalance it. Consider the consequences of spatial summation at the first stage of visual processing. If inputs of multiple receptors are averaged, the signal-to-noise ratio of the receptor channel can be increased at the expense of spatial detail. For example, in honeybees, chromatic contrast sensitivity in behavioral experiments exceeds what we predict from electrophysiological measurements of receptor noise (Dyer et al. 2008); accordingly, the minimal visual angle for color contrast detection (15 degrees) covers 59 ommatidia, equipped with 9 photoreceptors each (Giurfa et al. 1996). It follows that discrepancies between electrophysiological and behavioral estimates of noise from a species (e.g. those reported in Table 1 of Olsson et al. (2018) for rock doves, brown owls, and American kestrels) may actually reflect real differences between the noise of receptors and the noise in more downstream color processing neurons. The differences in achromatic and chromatic discrimination abilities of any given animal, rightly emphasized by the review, also point to the importance of postreceptor processing. Again, in honeybees, achromatic contrast detection is possible at a visual angle of approximately 5 degrees, employing 7 ommatidia, as opposed to the 15 degrees and 59 ommatidia required for chromatic contrast detection (Giurfa et al. 1996). Thus, both achromatic and chromatic signals are summed up, but to a different extent. As a result, the behaviorally estimated noise of the receptor channel is higher for achromatic than chromatic vision (see Table 1 of Olsson et al. 2018). Interestingly, a similar phenomenon appears in human perceptual studies that indicate a much higher sensitivity to achromatic than chromatic blur (Kingdom et al. 2015). The same distinction does not hold true for bumblebees (Dyer et al. 2008). Such differences originate from species-specific postreceptor processing that should not be ignored when dealing with visual perceptual spaces. It is also useful to remember that the RNL model was originally introduced for determining color thresholds and not for calculating perceptual differences between easily distinguishable colors. Perceptual differences may or may not scale linearly with differences in opponent receptor responses, and we caution against making such an assumption before this issue has been convincingly settled. In any case, current evidence indicates nonlinearity of visual spaces. For example, a set of behavioral experiments, using several species of bees, showed that the success of discrimination from a gray background scales nonlinearly with color difference (Dyer et al. 2008; Dyer and Neumeyer 2005; Garcia et al. 2017; Spaethe et al. 2014). Finally, it is important that color vision has many cognitive elements (Skorupski and Chittka 2011). In humans for example, even language constrains color discrimination (Winawer et al. 2007). It is thus impossible to predict receptor noise from behavioral data; instead, noise must be measured with appropriate electrophysiological procedures (e.g. Skorupski and Chittka 2010). Color spaces, including those based on receptor noise, are useful in the same way as a Metro map is: they provide a rough guidance as to what is where, and how far A is from B, but one should be cautious in making overly precise predictions based on them.