Skip to main content
Dryad logo

Cracks in the mirror hypothesis: high specularity does not reduce detection or predation risk


Franklin, Amanda et al. (2021), Cracks in the mirror hypothesis: high specularity does not reduce detection or predation risk, Dryad, Dataset,


Some animals, including certain fish, beetles, spiders and Lepidoptera chrysalises, have such shiny or glossy surfaces that they appear almost mirror-like. A compelling but unsubstantiated hypothesis is that a highly specular or mirror-like appearance enhances survival by reflecting the surrounding environment and reducing detectability.

We tested this hypothesis by asking human participants to wear a mobile eye-tracking device and locate highly realistic mirror-green and diffuse-green replica beetles against a variety of backgrounds in a natural forest environment. We also tested whether a mirror-like appearance enhances survival to wild predators by monitoring survival of mirror-green and diffuse-green replica beetles in a forested habitat and an open habitat.

Human participants showed no difference in the detection probability or detection latency of mirror versus diffuse replica beetles, indicating that mirror-like appearance does not impair prey capture. The field predation experiment found no difference in survival between the mirror and diffuse replica beetles in forested environments. Similarly, there was no difference in survival when beetles were deployed in open habitat where there is no background to reflect, indicating that predators detect and do not actively avoid mirror-like beetles.

Our results suggest that a mirror-like appearance does not reduce attack by predators. Instead, highly specular, mirror-like surfaces may have evolved for an alternate visual function or as a secondary consequence of selection for a non-visual function, such as thermoregulation.


Reflectance measurements

We measured the total reflectance over all angles using an integrating sphere (400 – 700 nm). We used an integrating sphere with an inbuilt tungsten-halogen light source (ISP-REF; Ocean Optics Inc., Dunedin, FL, USA.) and USB2000+ spectrometer (Ocean Optics Inc.) to measure a 4 mm spot. To extend the spectrum into the UV (300 – 400 nm), we used the same spectrometer coupled to a goniometer with a PX2 pulsed xenon light source. The light source and detector probes were positioned +/- 10° from the normal (i.e. 20° between the probes) and the spectra were then scaled and stitched to the visible spectra. A 99% Spectralon diffuse white reflectance standard (Labsphere, NH, USA) was used as the reference.  (Datafile: Franklinetal2021_MirrorBeetles_Spectra_TotalReflectance.csv)

We also measured the directional reflectance of all samples following the method described by Gruson et al. (2019). We used a Flame UV/VIS spectrometer (Ocean Optics Inc.) and PX2 pulsed xenon light source coupled to a goniometer to control the angle of incident light and detector probe. The angle between the incident light and detector probe was kept at 20° while varying the position of the bisector of this angle from -30°, -20°, -10°, 0°, 10°, 20°, and 30° around the normal. This set of measurements captures the light reflected at the specular angle (i.e. bisector between the light source and collection probe is 0° from the normal) and in incremental angular changes away from the specular angle. We used a 99% Spectralon diffuse white reflectance standard (Labsphere, NH, USA) as the reference for 100% reflectance, and we re-calibrated each time we modified the measuring geometry. (Datafile: Franklinetal2021_MirrorBeetles_Spectra_DirectionalReflectance.csv)

Human detection experiment

We tested whether human participants differed in detection success or detection latency for the diffuse and mirror replica beetles. Four different locations in the rainforest were selected, to account for natural variation in light environments and vegetation. At each location, 12 replica beetles (six diffuse, six mirror) were placed within a 2 m x 2 m x 2 m area of forest. To ensure that the placement of the replica beetles was random and did not bias our results, the placement of treatment types was reversed after four participants (i.e. diffuse beetle replaced with mirror beetle and vice versa). To enable quantification of success rate, participants were fitted with a portable eye tracking unit (Tobii Pro Glasses 2, Tobii Sweden). During each trial, the participant was asked to point to the location of the beetle detected and the experimenter, positioned behind the participant, noted the order, position, and treatment type of successful model detections by the participant. Trials were concluded either after the participant found all 12 replica beetles or the allocated 5 minutes was lapsed. From the eye-tracking data, we recorded whether each beetle replica was detected and the latency until detection for each beetle replica. Eight participants completed trials at each location for a total of 32 experimental subjects. For subsequent analyses, data from two individuals were removed because the quality of the eye-tracking data was unreliable. (Datafile: Franklinetal2021_MirrorBeetles_HumanDetection.csv)

Bird predation experiment

We set up linear transects with ten replica beetles per treatment group, each beetle spaced 2 m apart. In the open habitat replica beetles were tethered to a nail in the ground. In the forest habitat, replica beetles were tethered to a branch at approximately chest height and placed on backgrounds that ranged from chromatically similar (e.g. light green leaves) and highly contrasted (e.g. brown branches) with the replica beetles. Transects were checked for predation daily. Any beetle showing signs of predation was removed and placed in a ziplock plastic bag for scoring. Replica beetles that were attacked or removed were replaced with a new beetle of the same treatment in a different position. At the Daintree site, 11 transects were set up for four days in forested habitat. At the Grampians site, we initially set up 15 transects in forest and 15 transects in the open for six days to increase predation events. We added an additional five forested transects for three days to further increase sample size.  All replica beetles with signs of predation were assessed independently by two people. Replica beetles were scored as either: bird attack, mammal or unknown attack (small ambiguous markings), or completely removed (Supp. Fig. S9). Any beetles scored as mammal or unknown attack were removed from further analysis. Beetles where the target had disappeared were considered a bird attack because the small marsupial mice could not completely remove tethered replica beetles whereas comparatively large birds such as kookaburras and magpies could potentially do so. Survival was number of days until the replica was removed and beetle replicas that were not attacked were marked as censored data. (Datafile: Franklinetal2021_MirrorBeetles_BirdPredation.csv)


Australian Research Council, Award: DP190102203

Australian Research Council, Award: FT180100216