Countershading enhances camouflage by reducing prey contrast

Abstract

A three-dimensional body shape is problematic for camouflage because overhead lighting produces a luminance gradient across the body's surface. Countershading, a form of patterning where animals are darkest on their uppermost surface, is thought to counteract this luminance gradient and enhance concealment, but the mechanisms of protection remain unclear. Surprisingly, no study has examined how countershading alters prey contrast, or investigated how the presence of a dorsoventral luminance gradient affects detection under controlled viewing conditions. It has also been suggested that the direction of the dorsoventral luminance gradient (darkest or lightest on top) may interfere with predators' abilities to resolve prey's three-dimensional shape, yet this intriguing idea has never been tested. We used live fish predators (western rainbowfish, Melanotaenia australis) and computer-generated prey images to compare the detectability of uniformly pigmented (i.e. non-countershaded) prey with that of optimally countershaded prey of varying contrasts against the background. Optimally countershaded prey were difficult for predators to detect, and the probability and speed of detection depended on prey luminance contrast with the background. In comparison, non-countershaded prey were always highly detectable, even though their average luminance closely matched the luminance of the background. Our findings suggest that uniformly pigmented three-dimensional prey are highly conspicuous to predators because overhead lighting increases luminance contrast between different body parts or between the body and the background. We found no evidence for the notion that countershading interferes with predator perception of three-dimensional form.

1. Introduction

One of the most ubiquitous colour patterns associated with animal camouflage is countershading, where the dorsal surface of the body is more darkly pigmented than the ventral side [1,2]. Countershading is common in both terrestrial [3–6] and aquatic animals [7–9], and is considered to reduce the probability of detection and/or recognition by prey and by predators. For example, a number of experiments with artificial prey and avian predators have shown an average survival benefit to countershaded prey compared with uniformly pigmented prey that match background reflectance [10–14]. However, the visual and perceptual mechanisms that make countershading effective in reducing predation are less clear.

The hypothesis originally proposed to explain the cryptic effect of countershading is self-shadow concealment (SSC), where an appropriately countershaded animal can negate the dorsoventral luminance gradient across its body that results from overhead illumination [2,15,16]. Recent theoretical models [17], experimental studies [18–20] and phylogenetic approaches [5] have provided convincing evidence that SSC is the predominant mechanism of protection for countershaded prey. Importantly, these studies have revealed that the optimum patterning for SSC depends on the illumination conditions, such as the elevation of the sun [17], the weather [18,20] and the habitat type [5,18]. Prey that are optimally countershaded for the illumination conditions have higher survival rates than those with non-optimal countershading, suggesting that the dorsoventral luminance gradient has an important effect on detectability in a given light environment [18].

It has long been suggested that optimal patterning for SSC might enhance camouflage by removing the cues associated with three-dimensional form [2,15,21]. This is because an object's shading can provide useful depth information to visual systems [22], and animals such as pigeons [23], cuttlefish [24] and humans [25,26] use these cues to resolve three-dimensional form. Animal brains have evolved under the assumption that light generally comes from above, which is the reason why objects that are shaded as if lit from above appear convex, while those that are shaded as if lit from below appear concave [25]. Theoretically, a prey's three-dimensional shape could provide an important cue to predator presence, even for prey that have countershading coloration. This is because the optimum countershading pattern is constrained by season, weather, time of day and prey orientation [17,19,20] and is therefore typically sub-optimal. This means that some level of shading is usually present and selection for predators to hunt for three-dimensional prey may be expected. Despite this, computer vision models have revealed that countershading is an effective means of camouflage when visual detection is based on convexity [27], yet the role of shape perception in camouflage using animal vision is not known.

Animal visual systems have evolved to detect contrast rather than absolute luminance [28], thus crypsis should be maximized when prey have patterning that minimizes all aspects of contrast between the body and the visual background (with disruptive coloration being a clear exception [29]). Under directional, overhead lighting, a luminance profile is created across the body of a uniformly pigmented (3D) animal that enhances (i) the internal contrast between the brighter dorsal surface and the darker ventral surface and (ii) the contrast at the dorsal and ventral boundaries (edges) of the body when prey are viewed against a uniform background. These sources of contrast can potentially increase prey detectability, yet no study has investigated how predator detection behaviours are affected by changes in prey contrast due to countershading patterning. In this context, testing the assumptions of SSC requires a fine level of control of predator viewing angle [15]. This is because a prey's dorsoventral luminance gradient will generally be strongest when lighting is overhead and when prey are viewed from the side. For example, a prey animal being viewed by a predator from above will not receive any benefits from concealing a ventral shadow. Yet, many studies that have measured behavioural responses to countershaded prey have failed to control for viewing angle and thus have struggled to disentangle the underlying mechanisms of concealment.

Our experimental design used computer-generated prey and live fish predators allowing for manipulation of prey contrast while controlling for the predator viewing angle and the visual background. We first examined the detectability of prey with optimal countershading for different levels of average prey contrast with the background, allowing us to model the basic relationship between prey contrast and detection probability. We then compared the detectability of optimally countershaded prey with that of uniformly pigmented (i.e. non-countershaded) prey to determine the equivalent increase in contrast caused by the prey's dorsoventral luminance gradient. To test for the role of shape recognition in detectability, we exposed uniformly pigmented prey to overhead lighting and then reversed the luminance gradient, which would cause them to appear concave to most visual systems [25]. If the luminance profile is used as a cue to three-dimensional form, and/or if predators hunt for three-dimensional prey, then convex prey should be more readily detected or recognized than concave prey.

2. Materials and methods

(a). Study species and maintenance

We measured the predator detection/recognition behaviours of western rainbowfish, Melanotaenia australis [30] that had been trained to approach and ‘attack’ computer-generated prey. The western rainbowfish is a small (total length less than 10 cm) freshwater fish endemic to the north of Western Australia. Rainbowfish were chosen for this study because they have a well-developed cognitive ability and can learn spatial association tasks [31]. Wild-caught adults were maintained at 26°C (±1°C) on a 12 L : 12 D cycle. A total of seven males and nine females were used in this study.

(b). Computer-generated prey

The prey were modelled as grey cylinders (11.25 mm width × 15.75 mm length) that represented potential food items to the rainbowfish. We generated optimally countershaded prey (CS) which were designed to counteract the overhead illumination and thus had uniform radiance distribution across the surface of the body (figure 1). We also generated two types of uniformly pigmented prey: a non-countershaded prey (NCS) that would appear three-dimensional (convex) under directional, overhead illumination and a reverse non-countershaded prey (RNCS) that should appear concave under the same conditions (figure 1). The NCS prey was generated from a side-viewing direction using the open source, computer animation software Blender (http://www.blender.org) where an idealized three-dimensional cylinder was exposed to an overhead hemispherical light source and the imaged rendered from a side-viewing angle. The NCS image (.TIFF) was then imported into MATLAB where the background and ends of the cylinder were cropped to remove these shape cues. The RNCS was generated by rotating the NCS image through 180°. The mean luminance of the NCS and RNCS prey was 119.4, designed to match closely with the mean luminance of the background (details below).

Figure 1.

Eleven optimally countershaded prey were presented during the predator detection trials, with contrast ranging from −0.6 (60% darker than the background) to 0.6 (60% brighter than the background). Two prey were uniformly pigmented: one had a luminance gradient as if lit from overhead (i.e. non-countershaded prey, NCS), which would cause it to appear convex to most visual systems, while the other was inverted (reverse non-countershaded prey, RNCS), which would allow it to appear concave. Average prey contrast with the background (%) is shown above each image.

We used Weber's contrast ((I_f − I_b)/I_b, where I_f is intensity of the foreground and I_b is intensity of the background) to generate 11 variants of the CS prey with luminance contrasts between −0.6 and 0.6. Five were brighter than the background, five were darker than the background and one had the same average luminance as the background average (figure 1). Prey were presented on a heterogeneous background with a uniform random distribution (each screen pixel was assigned a value of 0 (black) or 255 (white)), mean luminance of 127.5 and brightness of 63.8 cd m⁻² (figure 1; see electronic supplementary material for details).

(c). Design and experimental procedure

Training and experiments took place in a 108 l (L 60.5 cm × W 25.5 cm × H 45.5 cm) aquarium with opaque walls (figure 2). One end panel was removed from the aquarium and replaced with an LCD computer monitor (GL2230-8 Benq). Attaching the computer monitor directly to the aquarium ensured that there was no refraction at the glass–air interface, thus reducing any optical distortion of the computer-generated prey. The computer monitor was calibrated for linearity of grey values using a spectrometer ILT1700 (International Light Technologies). Transparent silicon tubing (diameter = 5 mm) was attached to the monitor directly above the prey stimuli to provide a remotely delivered food reward during the training and subsequent experimental trials. The presentation of the stimulus was controlled through Psychtoolbox-3 (www.psychtoolbox.org) for MATLAB.

Figure 2.

Schematic showing the experimental set-up. The presentation of prey was controlled from a laptop computer (not shown) and allowed prey to appear on the computer screen at a pre-determined time and location. The companion fish was separated from the focal fish by a one-way mirror attached to the Perspex barrier.

Each focal fish was given a 5 min acclimation period before the first trial of each training or test session. A ‘companion fish’ was present in a separate compartment to the focal fish, providing a social stimulus to limit the positional variability from which the focal fish viewed the prey. The Perspex divider separating the focal fish from the companion fish was covered with a one-way mirror window film (KNG) preventing the companion fish from seeing the prey (figure 2). A new trial was only started once the focal fish was positioned at the far end of the experimental compartment, approximately parallel to the dividing wall with the companion compartment (electronic supplementary material, figures S1 and S2 for starting position distribution). When the fish responded by rapidly approaching the prey and moving within one body length of the monitor, they were rewarded with food to maintain motivation during both training and testing. Both the timing (between 130 and 160 s) and the position (left, right or centre) were randomized to prevent the fish from anticipating presentations.

(d). Training and testing

Each fish engaged in three consecutive days of training followed by the experimental procedure. Fish were trained to associate the appearance of prey on the computer screen with a food reward (bloodworm). MATLAB Psychtoolbox-3 was used to manipulate stimulus size (range: 950 × 860 pixels (236 × 213 mm) to 8 × 17 pixels (2 × 4 mm)), position (left, right and centre) and contrast with the background (range: −1 to 1). Fish (n = 16) were individually trained in either the morning or the afternoon, and this was kept consistent during the 3-day training period and for the experimental test to avoid any time of day effects on behaviour. Fish began training with large, high-contrast stimuli (white stimuli on a black background), which were made progressively smaller over the course of the training procedure until the final prey size was reached. Starting with large stimuli made it easier for the fish to develop a relationship between the computer monitor and a food reward. The task was also made increasingly difficult by introducing a heterogeneous background and reducing the contrast of the stimuli with the background. During the second and third training sessions, the stimuli were composed of a random combination of prey with an even luminance distribution as well as those with a strong luminance gradient, to prevent fish from developing a bias for a certain level of shading during training. We took the conservative approach of ensuring that the final training stimulus was smaller than the test stimulus to filter out the non-responsive fish before they proceeded to the experiment. The rainbowfish very quickly learnt to associate images appearing on the computer monitor with a food reward, and only fish who were showing a rapid and consistent response took part in the experiment (n = 16/18) (see electronic supplementary material for further details).

Testing took place on the day following the last training session. We performed two replicates of the experimental trials for each fish to evaluate the repeatability of object detection as well as gain additional data without needing to train more fish. Therefore, each fish took part in two experimental blocks (over 2 days), with each of the 14 artificial prey stimuli presented to the fish once within each block. The order of treatments was randomized according to 14 × 14 Latin squares where each fish was randomly allocated to a column and each treatment to a row. The behaviour of the fish was recorded using a camcorder (HDR-CX550, Sony) positioned above the aquarium (figure 2), and video data were imported to MATLAB for analysis.

(e). Video analysis and response variables

Video images were captured at 25 frames s⁻¹ and analysed using a custom tracking program in MATLAB [32] to measure four predator detection/recognition behaviours:

• (1)Probability of response was defined as the probability that a rainbowfish approached to within one body length of the monitor within 5 s of the prey appearing.
• (2)Delay to respond was determined as the time taken from the appearance of the prey to the start of the rapid ‘reorientation’ towards the screen; a rapid body rotation which was typically observed before the fish approached the prey.
• (3)Time to orientate: the time taken to orient the body towards the prey following the initial delay to respond.
• (4)Time to approach was measured as the time it took for the fish to swim within one body length of the computer monitor following the time to orientate. Rainbowfish that failed to respond when a prey item was presented were not included in this analysis.

The electronic supplementary material contains video footage of a fish response with these response behaviours labelled (electronic supplementary material, videos S1 and S2).

(f). Data analysis

The responses of the fish to the CS prey were used to generate a contrast-response function, which allowed us to compare the detectability of optimally countershaded prey at various brightness levels with the detectability of uniformly pigmented prey. Data were imported into the open-source software R (R Core Team v. 0.98.1103) for statistical analysis. Generalized linear mixed models (GLMMs) were used to test for an effect of CS prey contrast on the probability of response. Linear mixed effects models (LMEs) were used to test for an effect of CS prey contrast on the time to approach, fitted with maximum likelihood using the lme4 package [33]. Time to approach was transformed (Box–Cox, λ = −1) to improve the distribution of the residuals. Since transformation did not improve the distribution of residuals for the delay to respond and time to orientate, we used a non-parametric permutation approach. We entered absolute contrast as a fixed effect in all models. To test whether fish responded differently to negative or positive values of contrast we included an interaction between absolute prey contrast and whether the stimulus was darker or brighter than the background. The term ‘individual’ was included as a random effect in all cases.

We considered the potential effect of additional variables, such as sex, body length, experimental block, predator viewing eye (left or right visual hemisphere facing the monitor), prey position (left or right) and predator starting angle (in degrees), but these were only kept in the final model if they had a significant effect (p < 0.05). We initially included all interactions involving contrast that were determined a priori to be biologically meaningful; these included all interactions between contrast and prey position, contrast and starting angle, and contrast and viewing eye. All p-values presented were estimated by comparing the fit of each alternative model against the final model (using likelihood-ratio tests) which only contained significant terms. The assumptions of the models were checked by exploring the distribution of the residual values (using Q-Q plots) and examining plots of the standardized residuals against the fitted values for each model.

Post hoc tests were conducted to test for a difference in detectability between the NCS, the RNCS and the CS prey at equivalent levels of contrast (CS: −0.1; NCS and RNCS: 0.06). For the probability of approach, we used Fisher's exact test, and for the time to approach, post hoc tests were performed using Tukey's HSD with the Holm–Bonferroni adjustment using the R package ‘multcomp’ [34]. For delay to respond and time to orientate, post hoc pairwise testing was done in MATLAB by permuting a subset of the data. The full dataset used for analysis is available from Dryad Digital Repository [35].

(g). Contrast regions

To determine the sources of contrast that might account for the increased detectability of non-countershaded prey (NCS and RNCS) compared with CS prey of equivalent average contrast, we measured all sources of contrast using the image pixel values (internal, average and edge contrast against the background) and plotted them for each stimulus type. Using Weber's contrast, we generated four potential sources of contrast. (i) Edge contrast (top and bottom edge) was taken as the luminance contrast between the top and bottom pixel row of the prey and the background. As a result of light reflecting off the substrate, the bottom row did not represent the darkest row, and therefore, we measured (ii) the contrast between the brightest row and the darkest row of the prey with the background. (iii) The average contrast was a measure of the average luminance of the prey and the background, and finally, (iv) the internal contrast was calculated as the difference between the darkest and brightest rows compared with the background luminance ((brightest − darkest)/background).

3. Results

(a). Detectability of optimally countershaded prey

We found that for optimally countershaded prey (CS), contrast had a significant effect on the probability of predator approach, delay to respond, time to orientate and time to approach (black lines and black dots in figure 3a–d and table 1). Specifically, fish were more likely to approach and approached more rapidly when stimuli were high contrast than when they were low contrast. We found that this relationship did not depend on whether the CS prey were darker or brighter than the background for the probability of approach (figure 3a; χ² = 0.18, p = 0.67), the delay to respond (figure 3b; F_1,254 = 4.8, p = 0.93) and the time to orientate (figure 3c; F_1,254 = 1.96, p = 0.16). However, we found a significant interaction between prey contrast and whether the prey was darker or brighter than the background for the time taken to approach variable (figure 3d; χ² = 11.85, p = 0.001; table 1). In this case, the time to approach the prey was dependent on contrast for prey that were darker than the background but not for prey that were brighter than the background.

Figure 3.

The relationship between average prey contrast and the probability of approach by fish predators (a), the delay to respond (s) to the prey (b), the time taken (s) to orientate towards prey (c), and the time taken (s) to approach the prey (d). The solid lines represent the fit of the models for optimally countershaded prey (CS: black circles) that are 60% darker (negative contrast values) or 60% brighter (positive contrast values) than the background. Predator responses towards non-countershaded prey (NCS: light grey circle) and reverse non-countershaded prey (RNCS: dark grey circle) are also shown, and the dashed lines indicate their predicted equivalent contrast given the predators' responses. The shaded region indicates pairwise tests conducted for the NCS and/or RNCS stimuli and the CS prey (−0.1) with equivalent average contrast with the background.

Table 1.

Binomial GLMMs, LMEs and permutation models used to test for an effect of prey contrast (for optimally countershaded prey) on the probability of approach, the delay to respond (s), the time to orientate towards prey (s) and the time taken to approach (s). The contrast × bright/dark interaction determines whether the time to approach depends on whether prey were brighter (bright contrast) or darker (dark contrast) than the background. Starting angle (angle) was included as a covariate in the delay to respond and time to orientate models. The test statistic for the probability of approach and the time to approach is χ², and for the delay to respond and the time to orientate, it is the F statistic. Table shows only final models with significant results. See electronic supplementary material, table S1 for results of all fixed effects and their interactions.

response variable	fixed effect	slope (±s.e.)	intercept (±s.e.)	d.f.	test statistic	p-value
probability of approach	contrast	4.75 (0.72)	1.17 (0.29)	1	52.54	<0.001
delay to respond	contrast + angle	−0.5 (0.11)	0.38 (0.06)	1, 254	10.15	0.002
time to orientate	contrast + angle	−0.13 (0.02)	0.12 (0.02)	1, 254	33.99	<0.001
time to approach	contrast × bright/dark	1.06 (0.21)	0.62 (0.09)	1	11.85	0.001

(b). Detectability of optimally countershaded prey compared with non-countershaded prey

Fish were significantly more likely to approach the NCS prey than the CS prey of similar average contrast to the background (odds ratio = 8.65, CI = 2.42–36.87, p < 0.001; grey bar in figure 3a). In fact, the probability of approaching the NCS and RNCS prey was comparable to the probability of approaching the CS stimuli that were 60% and 55% darker or brighter than the background, respectively (dashed lines in figure 3a). Mirroring this finding, fish predators approached the NCS and RNCS prey more rapidly (z = 3.50, p = 0.01; grey bar in figure 3d) and orientated faster (F_1,78 = 8.35, p = 0.02; grey bar in figure 3c) than countershaded prey of comparable contrast. There was no significant difference in the delay to respond (figure 3b; F_1,58 = 0.052, p = 0.84) between the uniform prey (RNCS and NCS) and between the CS prey stimuli with equivalent average contrast.

(c). Shape from shading cues and object detectability

We found no significant difference between the detection of the NCS prey compared with the RNCS prey for the probability of approach (figure 3a; odd ratio = 0.80, CI = 0.17–3.60, p = 0.99), the delay to respond (figure 3b; F_1,59 = 1.92, p = 0.22), the time to orientate (figure 3c; F_1,59 = 0.02, p = 0.91) and the time to approach (figure 3d; z = 0.39, p = 0.99) response variables.

(d). Contrast regions

The contrast between the brightest row of the prey, the top edge of the NCS prey and, therefore, the bottom edge of the RNCS fell outside the range of the CS stimuli (figure 4). However, the internal contrast generated by the difference between the brightest and darkest rows represents, by far, the highest region of contrast on the NCS and RNCS prey (figure 4). The internal contrast was 140% brighter than the background, compared with 80% brighter for the next largest source of contrast.

Figure 4.

Replicate of figure 3a that includes potential sources of contrast for each prey stimulus type (NCS: black symbols; RNCS: grey symbols). Weber's contrast was used to calculate average contrast (circles), contrast of the bottom edge (squares) and the top edge (diamonds) against the background, the internal contrast (difference between the brightest and darkest pixels: triangles) and the maximum brightest (inverted triangles) and darkest (plus sign) pixels relative to the background. The red circle highlights internal contrast as the largest source of contrast on the NCS prey stimulus. (Online version in colour.)

4. Discussion

Our results provide evidence that directional light falling on three-dimensional prey generates a luminance gradient that increases the probability of prey detection by predators. Our experimental design controlled for viewing angle and average internal contrast, thus the increased detectability of prey is due to other sources of contrast, such as the edges between prey and the background, and/or the internal contrast generated by the dorsoventral luminance gradient. Our findings also reveal that prey detection/recognition was not dependent on the direction of the dorsoventral luminance gradient (i.e. prey that appear convex or concave), suggesting that the perception of shape from shading does not influence the predators' behaviour. Overall, these results show that uniformly pigmented three-dimensional prey will never be able to perfectly blend into a relatively homogeneous background. For example, the edges of the body, which are known to play an important role in prey detection and recognition [29,36], will become brighter on the dorsal side and darker on the ventral side compared with the background, resulting in increased conspicuousness. Our findings support the notion that countershading patterning can reduce the probability of predator detection/recognition by reducing sources of contrast across the body surface.

(a). Internal contrast and background matching

Given our modelling of the potential sources of contrast in uniformly patterned three-dimensional animals, we suggest that the main driver for the evolution of countershading patterning is the elimination of internal contrast rather than the contrast between the edges and the background (figure 4). However, it is important to note that changes in illumination caused by the elevation of the sun, the weather and the orientation of the prey will have a big effect on these sources of contrast. Indeed, a previous study based on human visual search of virtual three-dimensional objects found that small changes in the orientation (pitch, yaw and roll) of optimally countershaded prey generated large differences in internal contrast, which increased the speed and accuracy of prey detection [19]. These results are consistent with our conclusion that any deviation from optimal conditions will increase internal contrast and therefore detectability. However, we predict that any level of countershading will reduce internal contrast and hence provide a protective benefit. This remains to be tested.

Our results also provide important insights into the interaction between SSC and background matching; countershaded prey with optimal patterning for SSC were still highly detectable when their average contrast did not match the background. Our findings, thus, demonstrate the costs, in terms of the increased probability of detection, of displaying patterning that achieves SSC but does not provide protection through background matching. In theory, animals can achieve both SSC and background matching if the reflectance of the animal's body, combined with the incoming irradiance, matches the radiance of the background when the animal is viewed from the side [2,37]. However, because incoming irradiance is very low for the parts of the body that are not directly exposed to light, these surfaces would need to have reflectance greater than 1 to achieve both SSC and background matching for some lighting conditions (e.g. midday sun), which would require the generation of light [17]. Given these constraints, we predict that countershading patterns primarily try and reduce internal contrast, but we should expect a compromise with the need to also reduce overall contrast against the background.

High-contrast boundaries, such as two-tone patterns, can increase prey detectability, particularly under conditions of diffuse illumination [18]. However, this effect is highly dependent on viewing distance relative to an animal's spatial resolution thresholds. At larger viewing distances, the fish will not be able to resolve differences in luminance across the body, and there is a critical distance after which the predator will only perceive the average contrast between the patterning and the background. This further highlights the benefit for prey to match the average luminance against the background. However, our results strongly suggest that at distances shorter than this critical distance, there is also a strong selection pressure to reduce internal contrast or edge contrast. Under most natural conditions, only countershaded animals can optimize both sources of contrast.

(b). Shape from shading

Our finding that predators responded similarly to uniformly pigmented prey, irrespective of the direction of the dorsoventral luminance gradient, suggests that countershading does not interfere with predators' abilities to resolve shape, or at least does not override the stronger effects of contrast. These results are unlikely to be due to the limited cognitive ability of fishes because fishes are known to be able to detect and discriminate among three-dimensional shapes, as well as exhibit considerable cognitive skills such as numeracy and contour completion (review in [38]). For example, Abon damselfish (Pomacentrus amboinensis) can quickly learn to distinguish two-dimensional stimuli from three-dimensional shapes [39], and Malawi cichlids (Pseudotropheus sp.) exhibit shape recognition, even when objects are rotated in several planes [40]. Despite this, the role of depth cues in three-dimensional shape perception remains poorly understood outside of human vision.

Our experiment shows that the fish do not show an inherent bias towards using shape from shading cues to distinguish non-countershaded stimuli from reverse non-countershaded stimuli. However, optimal foraging theory predicts that predators should selectively target prey depending on their relative abundance and profitability [41]. Our experimental protocol ensured that all stimuli were encountered at the same frequency and were equally rewarding. Hence, we cannot rule out that in nature, where non-countershaded prey and reverse non-countershaded prey may be rarer (and hence less profitable), predators may learn to use shape from shading cues as part of their search image.

5. Conclusion

In summary, this study provides evidence that small three-dimensional prey are highly conspicuous to predators, even if on average, their coloration matches that of the background. Under our modelling conditions, this effect is explained by the increase in dorsoventral contrast that results from overhead illumination. Sensitivity to contrast is fundamental for the visual detection of objects in animals [28], thus our findings provide a simple explanation for the evolution of countershading that does not invoke complex cognitive processes such as shape recognition. While our modelling presents an important first step in determining how countershading patterning influences predator detection/recognition behaviours, this approach needs expanding to include more realistic lighting scenes and natural prey body shapes. Such an approach has great potential to further our understanding of this intriguing form of protective coloration.

Ethics

This project was approved by the University of Western Australia, Animal Ethics Committee, under ethics protocol RA/3/100/1176.

Data accessibility

The full data used for analysis is available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.cvdncjt0m [35].

Authors' contributions

C.G.D., J.M.H. and J.L.K. conceived and designed the experiment; C.G.D. conducted the experiments and digitized video data; C.G.D., J.M.H. and J.L.K. performed statistical analysis and generated graphics; C.G.D. and J.L.K. wrote original draft; J.M.H. and J.L.K. reviewed and edited all versions. All authors reviewed the final draft and gave approval for publication.

Competing interests

The authors have no competing interests to declare.

Funding

This research was funded by the School of Biological Sciences at The University of Western Australia and an ARC Future Fellowship to J.M.H. (FT110100528). J.L.K. was supported by an ARC Linkage Grant (LP120200002) with industry partners Rio Tinto and BHP Billiton, and is now funded by an ARC Future Fellowship (FT180100491).