Visual habitat geometry predicts relative morph abundance in the colour-polymorphic ornate rainbowfish

Abstract

During colour signalling in aquatic environments, the colour of the ambient light, the background against which signals are viewed and signal transmission through the environment can all have profound impacts on the efficacy of a given signal. In colour-polymorphic species, where alternative morphs persist owing to a balance in the natural and sexual selection for each, changes to the visual context can have large effects on the local success and relative abundance of competing phenotypes. The ornate rainbowfish, Rhadinocentrus ornatus, is composed of populations that vary in the relative frequency of red and blue individuals, and inhabit sites that vary in water transmittance from clear (white) to heavily tannin-stained (red-shifted). Using spectroradiometry, we measured the downwelling and sidewelling irradiance, bank radiance and water transmittance of 10 R. ornatus habitats. We found that the relative local abundance of each morph was predicted not by water transmittance but by chromatic differences between the vertical (downwelling light) and horizontal (bank colour) components of the habitat. This visual habitat geometry should increase contrast between the colour signal and background, with large potential to influence the strength of natural and sexual selection in this system.

1. Introduction

Effective intraspecific communication has a large effect on individual fitness in most animal species, both in terms of attracting and securing potential mates, and in agonistic encounters. Animal communication systems are often heavily reliant on colour signals to confer information between individuals, placing great importance on their efficacy. Individuals that are better able to communicate with conspecifics, either through increased signal efficiency, accuracy or visibility, are likely to gain a fitness advantage over those whose signals are less conspicuous or contain content that is less readily deciphered.

The efficacy of colour signals depends not only on the characteristics of the signal—such as its hue, chroma (saturation) and brightness—and the visual properties of the conspecific receiver, but also on the entire visual context [1–3]. The visual habitat is made up of the ambient light (reflected by the colour signal), the visual background against which the signal is viewed and the medium (air or water) through which the signal is transmitted [1,3]. Individuals will achieve greatest signal efficacy when their signal matches the ambient spectrum (e.g. blue light and blue signal), contrasts against the visual background and is subject to little degradation in quality or intensity as it passes from signaller to receiver. As a consequence of the restrictions imposed by the signalling environment and the need for effective communication, directional selection generally acts to optimize colour signals to the conditions under which they are used.

The persistence of alternative colour signals in colour-polymorphic (CP) species has therefore attracted considerable attention, as any fitness advantage of one morph over another should result in its inevitable fixation. The maintenance of competing colour morphs within a population requires a fitness equilibrium between morphs, generated through multiple agents of sexual and natural selection that are unique to every system examined to date (see [4,5] for a review). Currently, our knowledge of how interactions between colour signals and the visual environment affect the fitness equilibrium between competing morphs is largely limited to aquatic systems. Most well-studied terrestrial CP species occupy habitats where the visual context does not favour one morph over another, with maintenance processes generally taking the form of behavioural or physiological traits that are correlated with the external phenotype [4,6]. Through correlated selection, these traits act to offset the low performance of a morph in one area of their ecology with high performance in another [7,8]. Where habitat variation does favour one morph over another in terrestrial environments this variation alters natural but not sexual selection, as in light and dark substrates that create disruptive selection for anti-predator crypsis ([4] and references therein).

A better indication of how the visual environment influences the sexual selection exerted on competing colour morphs comes from our growing knowledge of CP maintenance in aquatic systems, where there is large potential for variation in the ambient light spectrum, visual background and the signal transmission medium. Microhabitat variation, such as the presence of multiple visual backgrounds, can directly affect the visibility of colour signals to balance the selection exerted on competing morphs. For example, the fitness equilibrium between competing morphs of the sharpfin silverside, Telmatherina sarasinorum, is generated by chromatic contrast of yellow males against the blue-shifted background of the open-water ‘roots’ spawning site, and contrast of blue males against the mid-wavelength-dominated (yellow) rocky shore site [9,10]. Each male morph concentrates its courting efforts to the habitat in which they are most visible to conspecifics and receive the greatest female attention and spawning success [9].

Conversely, among-population variation in the visual habitat can drastically alter the relative fitness of alternative colour phenotypes. The greatest potential for variation among aquatic visual habitat is in the transmission medium, which can additionally affect both the ambient light and visual background. Turbid and eutrophic water degrade signals to such an extent that sympatric sister species of cichlids are unable to distinguish between red and blue signals, and are punished through increased hybridization [11,12]. Unlike air, water can also become ‘stained’ with organic pigments, altering its transmission properties and causing it to act as a colour filter. In tannin-stained environments, attenuation of blue wavelengths results in a ubiquitous shift in the visual habitat as it becomes red-wavelength-dominated. This is most strongly evident in the visual background. In open-water habitats such as lakes, the sidewelling light is subject to strong filtration over a large distance, attenuating short wavelengths and causing it to become heavily red-shifted [13–15]. In more enclosed environments such as streams, where the bank forms the visual background, ambient light (downwelling and sidewelling) illuminating the background will have been filtered by the tannin content. Moreover, light radiating off the bank will be further subjected to filtration before reaching the signal receiver. Against the red visual background, blue signals should have much greater chromatic contrast than red (and other ‘warm’-coloured) signals. The strongest evidence to date comes from the blue-fin killifish, Lucania goodei, where blue-finned males are more abundant in tannin-stained habitats than red- or yellow-finned alternatives [13]. While this aspect of the signalling system (chromatic contrast against the visual background) is intuitive, to be successful under these conditions blue signals must overcome the constraints of having reduced blue wavelengths in the ambient light to reflect and the attenuation of their signal before it reaches the receiver. The solution to this conundrum may lie in the fundamental difference between clear aquatic and terrestrial habitats: illumination geometry. All light entering the aquatic environment comes from above (due to Snell's window)—evidenced by the counter-shading on most fish, and the maximal reflection of downwelling light by fish colour signals at a 45° angle. This means that the primary source of signal illumination, the downwelling light, should be subject to less tannin filtration than the sidewelling light or visual background (bank), and may provide enough dissimilarity in chromaticity between the ambient light and visual background to allow blue signals to maximize both strength and contrast. We propose that this visual habitat geometry may provide adequate chromatic contrast for blue individuals of polymorphic species to succeed in tannin-stained habitats.

The ornate rainbowfish, Rhadinocentrus ornatus, provides an ideal opportunity to test this hypothesis. Populations from Fraser Island, Australia and the adjacent mainland (Tin Can Bay basin) are comprised of red and blue individuals that vary in their relative local abundance [16]. Unlike many other colour-polymorphic systems, both morphs occur in males and females, and we have previously demonstrated strong, non-random mate choice with the ability to contribute to colour polymorphism maintenance in this species, a process that could be significantly altered by changes in the signalling context [16]. Furthermore, the visual habitats of these populations vary from clear to heavily tannin-stained, and our experience with this system suggests that the extent of tannin stain does not directly predict relative local abundance of red and blue individuals. Investigating how different visual habitat components vary among populations along a red-shifted scale could help in understanding the trade-off between red and blue colour signals in these environments. In the present study, we conducted a correlative field experiment to examine among-population variation in key aspects of the visual habitat to assess whether chromatic differences between key visual habitat components predict differential morph frequencies among populations of R. ornatus.

2. Material and methods

(a). Study system

The ornate rainbowfish is a small (standard length < 8 cm) obligate freshwater fish species restricted to the coastal ‘wallum’ habitat of eastern Australia [17], typified by soft, acidic and often tannin-stained water [18]. Populations from Fraser Island and the Tin Can Bay basin, Queensland exhibit a within-population colour polymorphism absent from populations to the south of this region [18]. This polymorphism is comprised of blue individuals whose colouration is generated through iridescent structures on the fins, and red individuals who display varying amounts of carotenoid pigment on the fins and body in addition to iridescent blue colouration, but primarily on the caudal fin [16]. Whether this represents a truly discrete polymorphism with a continuous distribution of carotenoid pigment in the red morph is yet to be established. However, as the amount of red colouration on individuals scales with standard length (SL) [16] but SL does not predict whether an individual is red or blue, we believe that classifying individuals to ‘red’ and ‘blue’ morphs is valid in this system (this is further supported by mate choice outcomes as described in [16]).

We have identified a number of aspects of the R. ornatus mating system through observations in the field. During courtship, male R. ornatus defend small, temporary territories by chasing away rival males. These potential spawning sites are generally part of or close to the stream bank and composed of aquatic vegetation or submerged roots near the top of the water column. Males display to females in front of the spawning site by extending their dorsal and anal fins, and tilting their heads down to reveal a red nuptial blaze, while attempting to lure the targeted female into the spawning area. Evenly sized males also engage in sparring bouts consisting of fin-flaring and circling behaviour, which can escalate into biting and chasing. As in other fish species, small males may elect not to compete, but adopt a ‘sneaker’ mating strategy and attempt to intercept females at the point of gametic release.

(b). Habitat measurements

We used an Ocean Optics USB2000+ spectrometer and UV-VIS fibre-optic cable to measure a selection of key visual habitat components at each site. This involves two types of spectroradiometric measurements. Irradiance—the quantum flux (energy) per unity area coming from all or many angles—is a measure of incident light [3]. Radiance, also measured as quantum flux per unit area but at a narrow acceptance angle, is used to quantify the light being reflected by a surface (e.g. visual background, colour signal) [3]. At each site, we measured the downwelling and sidewelling space light (irradiance), as well as the colour (radiance) of the submerged bank. The selection of these visual habitat components was based on a female or male viewing another individual against the background of a potential spawning site. From our own observations, colour signalling in R. ornatus is generally initiated by a male seeking to attract females into the aquatic vegetation or submerged roots that comprise the species's spawning substrate, or to exclude males from this temporary territory. This scenario is where the majority of sexual or social selection is likely to take place. We also obtained bank irradiance (as opposed to radiance) and sidewelling radiance (as opposed to irradiance) measurements for some but not all populations, preventing us from including these in our analyses to represent an individual viewed against an open-water background. While incorporating this information would be ideal and allow us to consider a reversal of the position of signaller and receiver, it is unlikely that the results of our study would be dramatically altered. All measurements were taken at both 10 and 20 cm below the surface of the water column and at a fixed distance of 30 cm from the bank. Unlike male Trinidadian guppies, which are known to display at times and locations that simultaneously maximize their conspicuousness to females and crypsis to predators [1,19,20], the R. ornatus populations in the Fraser–Cooloola region lack natural piscivorous predators and males are free to display under conditions that maximize their conspicuousness to conspecifics. We observed male displays throughout the day and most commonly in sunlit patches. Our habitat measurements were therefore taken after fish collection and processing (12–3 pm), in a sunlit patch encompassed by the area of highest density of individuals. Time of day (see the electronic supplementary material, table S1 for sampling times for each site) had no effect on the quality of spectral curves or the outcome of our statistical analyses (see §3 for further details). Water samples were collected for transmittance measurement and riparian vegetation scored as a discrete proportion of canopy cover (see the electronic supplementary material for additional information).

(c). Fish collection and digital photography

We sampled 611 adult R. ornatus from 10 populations (four on Fraser Island and six in the Tin Can Bay region) from November 2009 to December 2010 (see the electronic supplementary material, including figure S1 for additional site information). Fish were caught using 40 × 20 × 20 cm box traps and wide-mouthed dip nets, and immediately transferred to shaded 70 l holding bins that received constant aeration throughout the processing period. Fish were immobilized by immersion in chilled (7°C) creek water, and lateral images of immobilized fish were taken using a Canon Powershot A540 or Casio Exilim EX-FH25 digital camera. We have successfully implemented this technique on our study species under both field and laboratory conditions, and have observed no adverse effects of this treatment [16]. Upon recovery, fish were released immediately downstream of the sampling site.

(d). Analysis of spectral curves

Spectral curves were truncated to 325–685 nm for analysis (to avoid secondary fluorescence by aquatic vegetation), representing the majority of the probable visible range of R. ornatus based on related [21,22] and ecologically similar species [23,24]. The area under the resultant curves was standardized to a nominal value of 1 [3,25], and the spectral index (λp₅₀, median wavelength) was calculated for each curve as an estimate of the dominant wavelength [26]. We also conducted a separate principal components analysis (PCA) for each habitat measurement to assess the primary axes of variation in our dataset (see the electronic supplementary material for further details). A PCA was also conducted on the spectral indices for all habitat measurements to test for a simultaneous red shift in the visual environment.

(e). Analysis of digital photographs

(i). Camera calibration and colour measurement

Digital camera sensors often exhibit a nonlinear response to linear increases in colour saturation [27]. We assessed the response of our digital cameras prior to use in data collection following the procedure outlined by Stevens et al. [27] (see the electronic supplementary material for additional information). The caudal fin was chosen for colour analysis as it is the primary area of red expression (almost all red individuals have red caudal fins) and is free from shadows cast by the body of the fish. Red, green and blue (RGB) values were measured in Adobe Photoshop CS3 and transformed using the linearization equation for the appropriate camera, from which hue (colour), saturation (chroma) and brightness (value) of the red pigment were then calculated. We conducted a PCA to account for covariance in these three parameters.

(ii). Estimating proportion of red

We estimated the proportion of the total lateral area of red fish that displayed carotenoid pigmentation using the morphometric analysis software ImageJ v. 1.43u [28]. Distance calibration was conducted for each image using the 5 mm grid, and the lateral area of each individual's body and caudal, dorsal and anal fins was calculated, as well as proportion of each body segment displaying red pigmentation. As the proportion of red displayed by an individual was found to scale with their SL, we used the residual values (arcsine-transformed) from a linear regression (LR) of these two parameters in our analyses.

(f). Data availability

The final dataset used in our statistical analyses is available as electronic supplementary material, table S2.

3. Results

(a). Habitat variation

We obtained similar results for spectroradiometry measurements taken at 10 cm below the water surface and those taken at 20 cm, and so limit our presentation and discussion to those taken at 10 cm for reasons of simplicity and clarity.

Water transmittance had a large impact on spectral indices of habitat measurements, explaining a large proportion of observed between-population variation in downwelling irradiance (LR: R² = 0.5741, F_{1, 8} = 10.78, *p = 0.01), sidewelling irradiance (LR: R² = 0.6234, F_{1, 8} = 13.24, *p = 0.006) and bank radiance (LR: R² = 0.647, F_{1, 8} = 14.66, *p = 0.005). This is observed as a simultaneous red shift in all habitat components, with PC1 from a PCA on the spectral index being positive and approximately equal for downwelling irradiance (0.49), sidewelling irradiance (0.5), bank radiance (0.482) and water transmittance (0.527), and accounting for 79.9 per cent of the observed variation in visual habitat.

Despite this relationship, the locations sampled exhibited step-wise differences in the spectral index for downwelling irradiance, sidewelling irradiance and bank radiance (figure 1). Downwelling irradiance (mean = 575.3 ± 6.52 nm, range = 59.9 nm) was significantly ‘bluer’ than sidewelling (587.3 ± 6.14 nm, 55.7 nm), with a mean difference of 11.96 nm (paired t-test: t =−2.63, d.f. = 9, *p = 0.027), and sidewelling irradiance significantly bluer than bank radiance (600.3 ± 5.59 nm, 44.7 nm), with a mean difference of 13.04 nm (t =−2.64, d.f. = 9, *p = 0.027). This resulted in a mean difference of 25 nm between downwelling irradiance and bank radiance (t =−4.7, d.f. = 9, *p = 0.001).

Figure 1.

Mean downwelling irradiance (short-dashed line), sidewelling irradiance (long-dashed line) and bank radiance (solid line) spectral curves for the visual habitat of 10 populations of Rhadinocentrus ornatus. Shaded areas represent ± s.e. and vertical lines indicate the mean spectral index for each component.

This blue shift in downwelling irradiance was partially explained by an increased riparian vegetation (standardized coefficient β =−0.68, t =−4.77, *p = 0.002) and, in conjunction with water transmittance (β = 1.13, t = 7.92, *p < 0.0001), accounted for the majority of variation in this component (GLM: R² = 0.9, F_{2, 7} = 31.5, *p < 0.001).

The first three principal components explained between 95.8 per cent (bank radiance) and 99.9 per cent (water transmittance) of variation in habitat measurements. The majority of the variance (59.2–90.3%) was explained by PC1, which in each case was positive for long wavelengths (greater than 550 nm) and negative for all shorter wavelengths, with the exception of water transmittance, which was also positive for medium wavelengths (see the electronic supplementary material, figure S2). PC1 for each habitat variable was highly correlated with the corresponding λp₅₀ (see the electronic supplementary material, table S2), indicating that red–blue shift is the primary axis of variation in our dataset.

The time at which spectroradiometry measurements were taken had no direct effect on downwelling irradiance (LR: R² = 0.035, t_{1, 8} =−0.54, p = 0.60), sidewelling irradiance (R² = 0.128, t_{1, 8} =−1.08, p = 0.31) or bank radiance (R² = 0.007, t_{1, 8} =−0.25, p = 0.81), nor influence on any of our models.

(b). Phenotypic variation

Blue morph frequency varied among populations from 0 (all red) to 1 (all blue), with a mean (0.56 ± 0.13) not significantly different from 0.5 (one-tailed t-test: t = 0.45, p = 0.67). Three of our 10 populations consisted of only one colour morph, with two being all red (Teewah Creek and Lake Coomboo) and one all blue (Woralie). Morph frequency differed significantly among populations (two-way ANOVA, F_9,19 = 43.6, p < 0.0001) but not between sexes (F_1,19 = 2.90, p = 0.12). The proportion of red on individuals ranged from 0.004 to 0.813, with a mean of 0.179 ± 0.009. Population means for the proportion of red on individuals was not significantly correlated with the frequency of red individuals for arcsine-transformed raw proportion values (Pearson r = 0.05, t = 0.14, p = 0.9) or arcsine-transformed residual proportion values (r = 0.38, t = 1.10, p = 0.31). PC1 from a PCA on tail pigment measurements consisted of redder (hue =−7.01), more saturated (saturation = 7.06) and slightly darker (value =−1.0) pigment, and explained 44.2 per cent of the variation in this trait.

Water transmittance did not predict relative morph abundance (LR: R² = 0.023, F_{1, 8} = 0.18, p = 0.6758). The downwelling spectrum had a significant effect on relative morph abundance, with blue individuals increasing in frequency where the downwelling spectrum was less red-shifted (LR λp₅₀: R² = 0.4, F_{1, 8} = 5.325, *p = 0.05; PC1: R² = 0.426, F_{1, 8} = 5.938, *p = 0.04). In combination with this effect, blue morph relative abundance increased as bank radiance became red-shifted, with the final model composed of a blue shift in downwelling irradiance (λp₅₀: β =−0.04, *p = 0.008; PC1: β =−0.98, *p = 0.01) and a red shift in bank radiance (λp₅₀: β = 0.65, p = 0.053; PC1: β = 0.56, p = 0.088; GLM λp₅₀: R² = 0.6606, F_{2, 7} = 6.812, *p = 0.028; PC1: R² = 0.6319, F_{2, 7} = 6.008, *p = 0.03; figure 2).

Figure 2.

Scatter plot comparing the observed blue morph relative abundance from 10 populations of R. ornatus with that predicted by a linear model containing main effects of downwelling irradiance (negative) and bank radiance (positive). Closed circles represent spectral index values and open circles represent PCA values.

We found a marginally significant model based on spectral indices where mean proportion of red coloration on individuals increased as water transmittance became red-shifted (λp₅₀: β = 1.31, *p = 0.02) and bank radiance blue-shifted (λp₅₀: β = −0.98, p = 0.057; GLM λp₅₀: R² = 0.6262, F_{2, 6} = 5.026, p = 0.052). However, this relationship was less evident when using principal components (R² = 0.5451, F_{2, 7} = 3.594, p = 0.094).

Tail pigment PC1 (redder hue, more saturated, slightly darker) was associated with a general red shift in the visual environment, increasing significantly with increasing values for habitat λp₅₀ PC1 (LR: R² = 0.56, F_{1, 7} = 8.928, *p = 0.02). However, bank radiance alone was a slightly better predictor of tail pigment (GLM λp₅₀: R² = 0.6424, F_1,7 = 12.57, *p = 0.0094; PC1: R² = 0.5374, F_{1, 7} = 8.133, *p = 0.025).

4. Discussion

(a). Habitat

As expected, water transmittance had a large effect over the other three visual habitat measurements in this study, resulting in a ubiquitous red shift in the visual habitat as tannin concentration increased. Furthermore, the primary axis of variation (depicted by PC1) in each habitat variable was the relative red–blue shift in the spectral curve. However, we found significant differences between mean spectral indices of each visual habitat measurement. As predicted, the downwelling spectrum was significantly less red-shifted than the sidewelling or bank spectrum, indicating that chromatic geometry is present in the visual habitat of R. ornatus. This is likely to arise because the sidewelling light and bank radiance are subject to greater filtration by dissolved tannin than the downwelling light, and because the sidewelling light and bank radiance are both affected by the organic content of the visual background (e.g. submerged roots and leaves). A significant portion of the among-site variation in the downwelling spectrum is attributable to the amount of riparian vegetation, with greater values for riparian vegetation leading to less red-shifted downwelling spectra. This relationship can intuitively be interpreted as a shading effect [29,30] shifting the ambient spectrum towards shorter wavelengths, which combined with tannin content results in a broader spectrum. However, our measurements were taken in small or large sunlit patches (as defined by Endler [29]), which in terrestrial systems have a broad or red-shifted spectrum compared with open environments [29]. To our knowledge, no detailed investigation has been made into how these spectral differences are conserved across the air–water interface. It may be that all of our measurement sites have a slightly red-shifted ambient light source (thus minimizing its effect on our dataset) or that any variation because of patch size is masked by the filtering effect of tannin. However, neither explains how an effect of increased riparian cover is translated into broader downwelling spectrum in the aquatic environment. Future studies would benefit from direct measurements of the projecting foliage cover (using a spherical densiometer or hemispherical photography) across multiple microhabitats within creeks, with corresponding irradiance data for both aerial and downwelling light.

(b). Relative morph abundance

As predicted, we found no direct effect of water transmittance on the relative abundance of red and blue individuals. There was, however, a weak but significant effect of downwelling irradiance on relative morph abundance, with the number of blue individuals decreasing as the downwelling spectrum became red-shifted. This indicates that matching the illumination spectrum is important for the success of colour signals in this system, which is unsurprising given the well-established role of ambient light in shaping the evolution of both terrestrial [31] and aquatic [32] colour signals. Ultimately, the relative abundance of red and blue individuals depends on the magnitude of the chromatic difference between downwelling irradiance and bank radiance, which maximizes signal efficacy by allowing signals to match the ambient light while achieving strong chromatic contrast against the visual background. The greater effect of downwelling irradiance than bank radiance on relative abundance is most likely to be due to its greater range and variance, providing greater scope for interactions between sexual selection and the visual environment. This result provides indirect evidence that the chromatic habitat geometry in this system is altering the direction of sexual selection in R. ornatus.

Our reasons for basing our interpretation of this relationship on the blue morph are threefold. Lower spectral index values are based on the downwelling spectrum being broader rather than blue-shifted, such that long-wavelength content should always be adequate for red signals. This also means that chromatic contrast is positive for blue signals and approaches parity as geometry decreases; the downwelling spectrum never becomes more red-shifted than the background. Finally, blue signals are iridescent and are highly efficient at reflecting target wavelengths from the downwelling irradiance at 45° [23,33]. As such, they contain a far greater geometric component than pigment-based signals, which reflect light in all directions (isotropic) [34].

The greater intensity of blue iridescent signals may also contribute to their success, as they should be able to make more efficient use of the ambient light and travel through the tannin-stained water column. However, the issue of blue signal attenuation may be less problematic than it would intuitively seem. Small, colourful freshwater fish species (including poecilids [35,36], fundulids [13] and sticklebacks [37,38]) often communicate over very short distances, reducing the likelihood of significant signal degradation. In fact, the signals of such species may be designed specifically to degrade quickly to reduce detectability by predators, as evidenced by the incorporation of UV wavelengths, which are the first to be attenuated, into both blue and red signals.

(c). Variation in the red morph

In the current study, the relative abundance of the red morph is not correlated with the amount of red displayed by individuals, which supports our interpretation of the two as independent traits. However, the ecological predictors of each can be reconciled when considered in relation to their competition with blue signals. In the populations sampled, the relative abundance of the red morph decreases when the downwelling light spectrum is less red-shifted and the background more red-shifted. The amount of red pigmentation also decreases with more red-shifted backgrounds in combination with less tannin-stained water. In the former, being out-competed in conditions that favour the blue morph will lead to reduced abundance of the red morph. Within the red morph, fin areas not displaying red pigmentation also display blue iridescence. Selection could favour individuals displaying less red pigmentation in less tannin-stained water due to increased transmission of blue signals if red-shifted visual backgrounds also increase their chromatic contrast. This ultimately means that competition between red and blue signals may be not only limited to between-individual processes but also manifested as a trade-off within individuals.

Surprisingly, the properties of the red pigment itself were also not correlated with the amount of red displayed by individuals. Furthermore, while the other traits examined in this study appear to be under selection for contrast, pigment appears to match the dominant wavelength of the environment; tail pigment becomes redder and more saturated when the habitat is red-shifted. The closest comparison with the patterns found within the red morph can be drawn with another Australian freshwater fish, the southern pygmy perch, Nannoperca australis, which also inhabits locations that vary in their tannin content. As in the current study, Morrongiello et al. [39] found that N. australis males displayed more red on their fins and redder pigmentation at sites with greater tannin content. The authors of that study propose that viewing these red signals against the blue–green water column should result in maximum contrast (as in Telmatherina sarasinorum), neglecting that the water column is likely to become red-shifted in concert with water transmittance. Their alternative suggestion that green aquatic vegetation would provide significant chromatic contrast with red signals is a more probable scenario, as may also be the case in many populations of R. ornatus. Of the sites used in the current study, Carland, Pipeclay, Teewah and Gerowweea contain various amounts of living aquatic vegetation. In sites where aquatic vegetation is sparse, the ramp function of the bank radiance spectral curves (the exponential relationship between wavelength and relative intensity visible in figure 1) may act to maintain chromatic contrast.

Ultimately, it is impossible to know how red or blue signals appear to conspecifics without applying visual modelling techniques, which require accurate quantification of each signal and the visual sensitivity of R. ornatus. However, we do know that the visual sensitivity of species inhabiting environments that contain a dominant wavelength should be tuned to maximize their use of the available light. Rapid colour adaptation means that deviations from the wavelength to which the eyes are adjusted should be highly visible. In the blue-fin killifish, the upregulation of red opsin expression in tannin-stained environments, through developmental [40–42] or reversible plasticity [43], should have a similar effect [41,44].

Our understanding of the efficacy of competing signals in aquatic CP species would be further improved by a consideration of fine-grained heterogeneity in the visual background, which has not been addressed in the current study or similar studies. While the R. ornatus populations sampled inhabit sites that appear homogeneous and measurements were taken in the highest density of individuals, the potential for heterogeneity would merit its own examination. In addition, it would be interesting to look at genuine open-water sites, where the water column should be blue-shifted relative to the downwelling spectrum [13], and to see whether these also result in reduced selection for blue individuals. While these exist for R. ornatus, our attempts to sample these sites were unsuccessful due to inaccessibility, low population density or dry phase of an ephemeral population during the course of the study. Conversely, as our study found that the bank was important in predicting relative morph abundance and not the sidewelling spectrum, the chromaticity of the water column may not be important. However, the scenario depicted in our study is focused on that of a female viewing a courting male against the visual background of a potential spawning site. Future studies may benefit by also considering the reverse—a male viewing a courting target (female) against the water column, or two males sparring, both of which would require radiance measurements of the water column and irradiance measurements of the bank.

5. Conclusions

The results of the current study provide the first correlative evidence that the effects of variation in illumination spectrum and variation in background spectrum combine to drive local selection for colour signals. These results have important implications for the maintenance of colour polymorphism and local shifts to directional selection, as well as informing wider processes of colour phenotype evolution.