Avian egg and nestling detection in the wild: should we rely on visual models or behavioural experiments?

Abstract

The fields of avian egg and nestling colour detection have rapidly advanced owing to the application of visual models, which have allowed assessing of evolutionary questions considering receiver perception. Here, I first review the literature aiming to identify patterns of avian visual model usage. Second, I elaborate on limitations in the application of the receptor-noise limited perceptual (RNL hereafter) model. A systematic literature review revealed that the RNL model was the most used approach (81% of studies) in the field, and that most studies (76%) were concerned with classic evolutionary questions in avian brood parasitism. Some known limitations of the RNL model deal with model assumptions and parameterization, or, a poor consideration of post-detection neural processes. Others, however, are specific of the fields of egg and nestling discrimination and deal with the highly variable nature of ambient light at the nests, the complex colour design of eggs and nestlings, the multi-dimensional nature of perception, and the possible implication of learning. I, therefore, conclude that visual models should be used with caution to establish inference about egg and nestling discrimination, and rather be used to provide reasonable hypotheses which need to be validated with behavioural experiments.

This article is part of the theme issue ‘Signal detection theory in recognition systems: from evolving models to experimental tests'.

1. Introduction

One of the most striking features of birds is the extraordinary diversity of colours they exhibit, which has long attracted the attention of evolutionary ecologists aiming to understand the adaptive value of coloration [1–3]. Among birds’ traits, eggs and nestlings show meaningful intra- and interspecific variability in colour designs and patterning. Two main groups of hypotheses have been posited to explain the functional basis of such variation based either in the mechanical and thermal properties of the pigments or nanostructure producing the colours [4–7], or in the perception of egg and nestling colour differences by different visual receivers (i.e. predators, hosts of brood parasites, mates, parents) that may exert a selective pressure on coloration. These hypotheses have been thoroughly discussed in excellent previous review articles [8–11] and, hence readers are referred to these for a more in depth information on the functional basis of egg and nestling coloration.

Hypotheses based on the transfer of visual information from eggs and nestlings to parents, parasites and predators are inherently based in detection and discrimination of colour differences, including chromatic and achromatic aspects, by potential receivers. Receivers use colour differences to choose the most adaptive strategies of parental care, either detecting the salience of parasite eggs or nestlings which could then be discarded, or detecting a fitness-related quality associated with colour which could be favoured through a bias in parental care (see example in figure 1). Hence, a fundamental step to critically study recognition within the frame of these hypotheses is to consider what do potential receivers actually see [16]. During the twentieth century, evolutionary ecologists mostly relied on human-based methods to quantify colour differences, which might be a flawed approach given fundamental differences between the visual system of humans and birds (i.e. receivers) ([17–19], but see [20]). However, during the past decade the field of egg and nestling colour recognition has rapidly advanced owing to the emergence of more objective methods to quantify colour differences [21], to a better knowledge of the physiological properties and composition of avian retina [22], and, notably, to the application of mathematical models of avian colour vision which has allowed approaching the relevant evolutionary questions considering avian perception [16,23,24]. Briefly, avian visual models allow the calculation of perceptual differences in coloration of eggs and nestlings based on their reflectance spectrum, the reflectance spectrum of contrast elements (e.g. nest materials), the ambient light (irradiance) spectrum illuminating the eggs and nestlings, and the physiological properties of the receiver (i.e. the viewer's ocular media transmission spectrum (including oil droplets in birds), and the absorption spectrum of the photoreceptors) [17,25]. Readers interested in the physiological basis of avian vision are referred to several excellent reviews [17,26,27]. Moreover, the foundations of the different visual model approaches and their limitations for the study of animal coloration have already been well established (e.g. [28–30]), and hence are out of the scope of this review.

Figure 1.

Conceptual pathway highlighting the different elements involved in discrimination of egg colour differences by male spotless starling (Sturnus unicolor) in the frame of the sexual selection hypothesis. The questions from a signal detection perspective would be if males can detect eggs against the nest background of if they can differentiate between the colours of eggs of different females to decide food allocation. The two questions can be addressed using visual models (see main text). (a) Spectral reflectance of eggs and of nest material in spotless starling nests [12]. (b) Irradiance spectra of the nest light in nest-box cavities occupied by spotless starlings after Avilés et al. [13]. (c) The detection of egg colour is a consequence of male spotless starling physiology and depends on spectral sensitivities of single (used to model chromatic contrasts) and double (used to model achromatic contrasts) cones [14]. (d) Egg colour signals are integrated in the brain of a male spotless starling through unknown complex neural processes. Egg perception would collectively refer to egg detection and brain processing of egg coloration. (e) Provisioning visits performed by spotless starling males in experimental nests with pale or dark blue-greenish eggs at three different stages of nestling development [15]. The interaction between all the elements involved in this hypothetical communication scenario triggers a male behavioural response which in the long term may affect the expression and evolution of egg coloration. Image credits: (a) spotless starling eggs, C. Navarro, (f) spotless starling adult male, F. Barroqueiro/Macaulay Library at the Cornell Laboratory of Ornithology (ML101681121).

In this review, I specifically focus on the use of avian visual models to study evolutionary questions dealing with recognition of eggs and nestlings in the nests (i.e. the natural environment where visual receivers actually face detection challenges). I begin by reviewing the literature on avian visual models aiming (i) to identify patterns of use of the different visual model approaches in relation to relevant evolutionary hypotheses in egg and nestling detection, and to (ii) assess the use of visual models regarding their known limitations. Secondly, I elaborate on possible limitations in the application of the receptor-noise limited Vorobyev-Osorio [24] perceptual (RNL hereafter) model, and based on these, provide a list of new directions for research worth exploring in egg and nestling discrimination. The aim of the review is to contribute to a better understanding of how variation in ambient light, the complexity of colour designs, the simultaneous implication of different sensory channels and stimulus in egg and nestling detection and learning may affect visual model expectations in the wild.

2. The use of avian visual models in egg and nestling detection

To identify studies dealing with egg and nestling detection and using avian visual models I performed a systematic literature review by searching the Web of Knowledge (search conducted on 1 October 2019) combining the following keywords (search 1: ‘brood parasitism’ and ‘avian visual model’; search 2: ‘egg discrimination’ and ‘avian visual model’; search 3: ‘egg recognition’ and ‘avian visual model’; search 4: ‘nestling’ and ‘avian visual perception’; and search 5: ‘nestling’ and ‘visual perception’). The literature search returned 48 publications, 42 of which were retained after skimming the abstracts and methods and verifying they were using a visual model approach to test an evolutionary hypothesis dealing with egg or nestling detection (table 1). From these 42 publications, I obtained information about (i) whether they targeted egg or nestling detection; (ii) the studied evolutionary framework; (iii) the conceptual use of models (i.e. if they were used to assess how a visual receiver perceive differences between two or more colours, to predict thresholds for colour detection, or both simultaneously); and (iv) the used visual model approach (distinguishing among the tetrachromatic colour space approach sensu Endler & Mielke [25], the receptor-noise discrimination model sensu Vorobyev & Osorio [24]; and pattern analysis: granularity analyses sensu Stoddard & Stevens [31]). I also retrieved information on (v) whether the studies modelled chromatic (Chr), achromatic (Achr) or both (Chr + Achr) kinds of perceptual differences; (vi) whether the studied sensorial challenge involved contrasting patterns from the human point of view (i.e. spotted eggs, or nestlings with evident contrast between the target trait (gape or plumage patch) and other traits, figure 2) or not; (vii) whether the studies applied (yes or not) some analyses of patterning; and (viii) whether detection was evaluated or not in dark luminal conditions.

Table 1.

Studies using avian visual models for investigating egg and nestling detection. (Aiming to discern the pattern of use studies are classified based on whether they targeted on egg or nestling detection, the studied evolutionary framework, the conceptual use of models (i.e. if they were used to assess how a visual receiver perceive differences between two or more colours, to predict the limits for the detection of colour salience or both), the used visual model approach (tetrahedral: tetrachromatic colour space approach sensu Endler & Mielke [25]; RNL, receptor-noise discrimination model sensu Vorobyev & Osorio [24]; pattern analysis: granularity analyses sensu Stoddard & Stevens [31]) and whether the studies modelled chromatic (Chr), achromatic (Achr) or both (Chr + Achr) kinds of perceptual differences. In addition, aiming to qualify the use of visual models regarding their possible limitations, studies are classified based on the application (yes or no) of some analyses of patterning, the presence of complex visual designs (yes: if eggs were spotted or nestlings had contrasting begging designs), or if detection was evaluated or not in dark conditions.)

study	egg/nestling	evolutionary framework	conceptual use of visual model	visual model approach	perceptual differences	patterning	complex visual design	dark conditions
[32]	eggs	mimicry hypothesis	estimating differences in colour	tetrahedral	Chr + Achr	yes	yes	no
[33]	eggs	colour trait evolution	estimating differences in colour	tetrahedral	Chr	no	yes	no
[34]	eggs	evolution of defences	estimating differences in colour	tetrahedral	Chr	no	yes	no
[35]	eggs	mechanisms of recognition	estimating differences in colour	tetrahedral	Chr	no	yes	no
[36]	eggs	mimicry hypothesis	estimating differences in colour	tetrahedral and RNL	Chr + Achr	no	yes	yes
[37]	eggs	crypsis hypothesis	estimating differences in colour	RNL	Chr + Achr	no	yes	yes
[38]	eggs	crypsis hypothesis	estimating differences in colour	RNL	Chr + Achr	no	yes	yes
[39]	eggs	luminosity hypothesis	estimating differences in colour	RNL	Chr + Achr		no	yes
[40]	eggs	luminosity hypothesis	estimating differences in colour	RNL	Chr + Achr	no	yes	yes
[41]	eggs	mechanisms of recognition	estimating differences in colour	RNL	Chr + Achr		no	no
[42]	eggs	mechanisms of recognition	estimating differences in colour	RNL	Chr + Achr		no	medium
[43]	eggs	mechanisms of recognition	estimating differences in colour	RNL	Chr + Achr	yes	yes	no
[44]	eggs	mechanisms of recognition	estimating differences in colour	RNL	Chr	no	yes	no
[45]	eggs	mechanisms of recognition	estimating differences in colour	RNL	Chr	no	no	no
[46]	eggs	mechanisms of recognition	estimating differences in colour	RNL	Chr + Achr	no	yes	no
[47]	eggs	mechanisms of recognition	estimating differences in colour	RNL	Chr + Achr	no	yes	no
[48]	eggs	mechanisms of recognition	estimating differences in colour	RNL	Chr + Achr	yes	yes	no
[49]	eggs	mechanisms of recognition	estimating differences in colour	RNL	Chr + Achr	yes	yes	no
[50]	eggs	mimicry hypothesis	estimating differences in colour	RNL	Chr + Achr	no	yes	no
[51]	eggs	mimicry hypothesis	estimating differences in colour	RNL	Chr + Achr	no	yes	no
[52]	eggs	mimicry hypothesis	estimating differences in colour	RNL	Chr + Achr		yes	no
[53]	eggs	sexual selection hypothesis	estimating differences in colour	RNL	Chr	no	yes	no
[54]	eggs	camouflage hypothesis	estimating differences in colour and thresholds for detection	RNL	Chr + Achr	no	yes	no
[55]	eggs	mimicry hypothesis	estimating differences in pattern	tetrahedral and RNL	Chr + Achr		yes	no
[56]	eggs	mechanisms of recognition	estimating differences in pattern	pattern analysis		yes	yes	no
[31]	eggs	mimicry hypothesis	estimating differences in pattern	pattern analysis		yes	yes	no
[57]	eggs	mimicry hypothesis	estimating differences in colour	tetrahedral	Chr + Achr	no	no	no
[58]	eggs	mimicry hypothesis	estimating thresholds	RNL	Chr + Achr	no	some species	some species
[59]	eggs	mimicry hypothesis	estimating thresholds	RNL	Chr + Achr	no	yes	no
[60]	eggs	sexual selection hypothesis	estimating thresholds	RNL	Chr	no	no	no
[61]	eggs	sexual selection hypothesis	estimating thresholds	RNL	Chr + Achr	no	yes	yes
[12]	eggs	sexual selection hypothesis	estimating thresholds	RNL	Chr + Achr		no	yes
[62]	eggs	colour trait evolution	estimating variability	tetrahedral	Chr + Achr		no	no
[63]	eggs	colour trait evolution	estimating variability	RNL	Chr	no	no	no
[64]	eggs	colour trait evolution	estimating variability	RNL	Chr + Achr	yes	yes	no
[65]	nestlings	nestling colour evolution	comparing perception by different receivers	RNL	Chr + Achr		yes	some species
[13]	nestlings	nestling colour evolution	estimating differences in colour	RNL	Chr + Achr		yes	some species
[66]	nestlings	nestling colour evolution	estimating differences in colour	RNL	Chr + Achr	no	yes	yes
[67]	nestlings	nestling colour evolution	estimating differences in colour	RNL	Chr		yes	yes
[68]	nestlings	nestling colour evolution	estimating differences in colour	RNL	Chr		yes	yes
[69]	nestlings	mimicry hypothesis	estimating differences in colour and thresholds for detection	RNL	Chr + Achr	no	yes	yes
[70]	nestlings	mimicry hypothesis	estimating thresholds	RNL	Chr	no	yes	no

Figure 2.

Examples illustrating the complexity of eggshell and gape colour designs in terms of visual perception. (a) Eggs laid by different tawny-flanked prinia Prinia subflava females (outside circle) around eggs from different females of the parasitic cuckoo finch Anomalospiza imberbis. (b) Eggs laid by one magpie Pica pica female (six eggs in top row), two parasitic great spotted cuckoo females (female a: 1–2 eggs in bottom row; female b: 3–5 eggs in bottom row), and a model great spotted cuckoo egg (egg six in bottom row) used to study egg rejection [71]. (c) Nestling begging display of the brood parasitic pin-tailed whydah Vidua macroura (left) and common waxbills Estrilda astrild (right) from the same brood. (d) Nestling begging display of the Gouldian finch Chloebia gouldiae. (e) Adult cedar waxwing Bombycilla cedrorum feeding its begging nestlings. Potential receivers (hosts of avian brood parasites and feeding parents) should discriminate between complex visual egg and nestling designs involving colour, pattern, shape, size, movement and other sensory channels. Image credits: (a) tawny-flanked prinia and cuckoo finch eggs, C. Spottiswoode, (b) magpie, great spotted cuckoo and model eggs, M. Molina-Morales, (c) pin-tailed whydah and common waxbill nestlings, J.G. Schuetz, (d) Gouldian finch nestlings, G. Hofmann & F. Scheffer, (e) cedar waxwing, S. Kelling/Macaulay Library at the Cornell Laboratory of Ornithology (ML176630681).

Of the 42 retrieved studies, 35 (83%) dealt with egg detection and 7 (17%) with nestling detection (table 1). The most widely used visual model was by far the RNL (34 (81%) out of 42 studies), followed by the tetrachromatic colour space model (8 (19%) out of 42 studies) and pattern analyses (2 (4.8%) out of 42 studies) (table 1). Most of the studies targeted the detection of both Chr and Achr aspects of coloration of eggs and nestlings (28 (66.7%) out of 42 studies). However, still a large proportion of studies ignored the modelling of the Achr dimension (12 (28.6%) out of 42 studies) (table 1), or did not differentiate between the Chr and Achr aspects when using granularity analyses (table 1).

Regarding the evolutionary framework, the largest proportion of studies (32 (76%) out of 42) were concerned with classic questions in the field of avian brood parasitism, followed by studies of nestling colour evolution (five studies), the sexual selection hypothesis for egg coloration (four studies), and the camouflage hypothesis (one study) (table 1), which are reviewed in relation to use of visual models in the following subsections.

(a). Avian brood parasitism

Obligate avian brood parasites lay their eggs in the nest of another species, the host, which carries out the parental duties, from the incubation of eggs to chick feeding [72–75]. In some brood, parasite-host systems brood parasites impose large fitness costs to hosts [72], which have favoured the evolution of certain behaviours preventing parasitism such as egg discrimination and rejection of parasite eggs (e.g. [72,73,76]), or discrimination of parasitic nestlings (e.g. [77–79]). On the other hand, parasite discrimination selects further counter-defences on the brood parasités side, such as more accurate mimicry between parasite and host eggs [31,58,80], or between parasite and host nestlings [70,81] hindering parasite discrimination by hosts, and resulting, in many instances, in a coevolutionary arms race [73].

The study of the mechanisms leading to parasite recognition and of mimicry is central to identify signs of coevolution and has long been based on assessment of colour differences. Indeed, among the studies dealing with eggs, I report that visual models were primarily used to study egg recognition (11 (31%) out of 35 studies). In these studies, the most widely used visual model approach was the RNL (9 (81.8%) out of 11 studies), which was used to estimate the relative importance of perceptual differences in colour between host and parasite eggs in eliciting egg rejection. These studies used conditions of bright light (table 1), when the performance of the RNL model is not impaired [24,29]. However, in many instances (8 (72.7%) out of 11 studies, table 1), visual models were used to study discrimination of spotted eggs. Only half of these eight studies (table 1) included patterns of spots as additional predictors likely to influence parasite egg discrimination. Notably, some of these studies relied on painted model eggs, which are very different in colour from host eggs (e.g. [41,42,44,45]), and which may suppose supra-threshold stimulus in terms of the performance of the RNL model [28,29].

The second main evolutionary hypothesis studied with visual models was the mimicry hypothesis (10 out of 35 (28.6%)) studies on eggs, and two further studies dealing with mimicry of host fledglings by different brood parasites (table 1)). Four (33.3%) out of these 12 studies were conceptually based on the estimation of thresholds for discrimination to explain the observed absence of either egg or nestling recognition by hosts [58,59,69,70]. All these studied relied on the RNL model, and the predictions about discrimination were supported by behavioural observations, suggesting model validity in predicting perceptual differences. The remaining eight studies dealing with mimicry calculated perceived differences based on an RNL model (3 (37.5%) out of 8 studies), the tetrachromatic colour space approach (2 (25%) out of 8 studies), these two approaches simultaneously (2 (25%) out of 8 studies), or based on granularity analyses of patterning (1 study (12.5%) out of 8; table 1). Notably, in some of these studies, mimicry was estimated on spotted eggs (table 1), and only two (20.0%) of these explicitly considered differences in spottiness [32,55].

Once brood parasites mimic the eggs of their hosts, hosts are expected to produce more homogeneous clutches in appearance [82,83], which, as a by-product, will induce higher levels of egg polymorphism in parasitized host populations [82,84,85]. A low intra-clutch variation in host egg appearance together with a higher inter-clutch variation in egg appearance will hinder brood parasites in matching their eggs to the proper host egg-type facilitating parasite egg discrimination [86]. Visual models have been used to estimate perceptual variability in host eggs in relation to parasitism levels or rejection abilities in three studies (table 1). One of these dealt with variability of spotted eggs [64], but accounted for spottiness variability apart from perceptual coloration.

Finally, parasite egg or nestling discrimination often occur in a vast range of environmental conditions (e.g. open versus holes nests, cloudy versus clear days) that may influence the discriminatory tasks [87]. For instance, given the known effects of ambient light on perception of bird colour patterns [17,88], low nest light environments could constrain Chr-based parasite egg discrimination and, thus, under some circumstances, lead to the acceptance of parasitism. Two studies have tested predictions of the luminosity hypothesis using the RNL model to estimate how variation in ambient light at the nest (i.e. spectral irradiance) affected perceptual differences in matching between parasite and host eggs (table 1). In addition, visual models have been used to study the egg crypsis hypothesis (two studies; table 1), which aims to explain why Chalcites spp. hosts accept Australian and New Zealand cuckoo eggs which are so different in appearance from their ones [38,89]. These studies used an RNL model to calculate contrasts between parasite eggs and the nest background and between parasite eggs and host eggs. Both the luminosity and the crypsis hypotheses implicitly deal with colour detection under low ambient light levels, and therefore perceptual differences were estimated using spectral irradiance values typical of dark luminal conditions, when model prediction reliability might be impaired [90] (see discussion below).

(b). Nestling colour evolution

Nestlings of altricial birds commonly beg to obtain food from their parents. The coloration of gapes is a key component of such a begging display (reviewed in [11]), which can inform parents on nestling level of satiation or health (e.g. [91–93]), or provide a conspicuous target towards which they direct their feeds [13,94,95], influencing parental allocation strategies [96–98]. The colour of the skin [99,100] and feathers of nestlings (e.g. [101,102]) may also indicate offspring’ quality and mediate parental favouristim [103]. Visual models have been used to estimate the efficacy of begging displays in terms of conspicuousness toward their parents. All five studies dealing with the evolution of nestling coloration relied on an RNL model to calculate contrasts between a given begging feature and other body regions of nestlings or the nest background (table 1). The five studies targeted cavity-nesting species and hence reliability of colour contrast calculations should be considered with caution. In addition, as Achr perceptual differences vary much more than Chr ones with changes in ambient lighting (irradiance) [17], discrimination in low ambient light conditions might be based on perception of Achr differences rather than on Chr ones [17]. Indeed, comparative work has shown that nestlings of hole nesters birds showed a more evident Achr contrast between flanges and other traits than did nestlings of open nesters [13]. Also, hole-nesting species have more Achr eggs than open- and semi-hole-nesting species [104], suggesting that the Achr dimensions may play a key role in recognition in low light conditions.

(c). The sexual selection hypothesis

Another evolutionary scenario where colour-based egg discrimination may play an adaptive role is sexual selection. Moreno & Osorno [105] proposed that avian blue-green egg colorations may have evolved as signals of female quality that males can use to modulate their parental investment (sexual selection hypothesis). The reasoning behind this appealing hypothesis is that biliverdin, which is the pigment responsible for blue-green egg coloration [106,107], possesses antioxidant activities [108,109], and, hence, only those females with a high antioxidant capacity could use biliverdin as an eggshell pigment ([105], but see [110]). Accordingly male birds are expected to modulate their allocation of parental effort by using eggshell coloration as it would inform on female phenotypic quality or condition (i.e. attractiveness) [105]. A key premise of the sexual selection hypothesis is that males can discriminate between the eggs of different females in the light conditions of their nests (figure 1), which has been tested in three different studies using the RNL model (table 1).

Cassey et al. [60] calculated thresholds for detection to assess the potential for the sexual selection hypothesis in 46 species of the superfamily Muscicapoidea. Based on these calculations they argued that in the majority of species, birds would not be able to discriminate between eggs in different conspecific clutches. Also, questioning the functionality of this hypothesis, it has been suggested that low luminosity levels in cavities might constraint the use of visual signals given the low performance of colour vision in dim light levels [111–113]. However, two studies in which an RNL model was used to estimate thresholds for detection in cavity-nesting species showed that estimated egg detectability depended entirely on model assumptions about visual limitations [12,61]. Moreover, experiments in which egg colour was successfully manipulated showed that birds are potentially capable of responding to changes in egg colour within their holes [12,15,104], suggesting that behavioural experiments, rather than visual model calculations, are needed to test the sexual selection hypothesis.

(d). The camouflage hypothesis

Eggs of species that lay directly on the ground are more cryptic than others because they are under a higher risk of predation [9,114,115]. However, egg colour may also serve a communication function to conspecifics, aiding to locate or identify own eggs by parents in colonial ground-nesting birds (e.g. [116]). Therefore, egg colour in these species may have evolved under the antagonistic selective pressures of signalling to intended receivers (e.g. conspecifics) while minimizing the risk of being detected by unintended receivers (e.g. predators). So far this hypothesis has been tested using the RNL model to calculate brightness and colour contrasts of black-legged kittiwake (Rissa tridactyla) speckled and soiled eggs as seen from a distance by conspecifics and their main avian predator, the common raven (Corvus corax).

3. Limitations in the application of the receptor-noise limited perceptual model for studying egg and nestling discrimination in nature

The most widely used visual model to test evolutionary hypotheses related with egg and nestling discrimination in nature is the RNL model developed by Vorobyev & Osorio [24] (table 1), which assumes that the ability to discriminate between coloured objects is limited by the noise at the photoreceptors [24,117]. Based on a reduced number of physiological parameters of the visual system of the receiver, and on colour (i.e. spectral reflectance) and ambient light (i.e. spectral irradiance) measures, the model gives back values of Chr contrasts between two colour patches in perceptual units of discrimination call ‘just noticeable differences' (jnds) [17]. The model was originally developed for Chr-based discrimination, and, therefore disregarded luminance (i.e. intensity) based discrimination. However, a subsequent extension allowed calculating luminance based on double cone calculations [117].

Previous works have thoroughly discussed limitations of this model, focusing on frequent violation of assumptions and poor parameterization, and, a general poor understanding of post-detection neural processes [28,29]. In this section, I briefly review this, but preferentially extend on specific limitations in the context of egg and nestling discrimination that were not covered in previous work.

(a). Reliability of model predictions depends on noise at the photoreceptors and luminosity levels

The reliability of RNL model predictions greatly depends on parameter choice and particularly of noise detected at photoreceptors [118], which have been only estimated for a few bird species based on careful behavioural and electrophysiological analyses in the laboratory [29]. Moreover, the RNL model is based on the assumption that noise at the photoreceptors is independent of light intensity [24,90], a condition which is violated at low light levels where other source of noise can affect discrimination [119]. These limitations have been extensively discussed in previous theoretical work [17,28,120], but not have been fully considered in egg and nestling detection so far, as many of the retrieved studies dealt with detection challenges in cavities (table 1). Indeed the luminosity and the crypsis hypotheses implicitly focus on colour detection under low light levels. Also, many of the species in which the sexual selection hypothesis has been tested are cavity-nester species (table 1), and therefore perceptual differences were estimated under dark luminal conditions, when reliability of model prediction might be impaired [90].

(b). Reliable sensory-based analyses require knowledge of photoreceptor spectral sensitivity of signal receivers which is lacking for most species

Visual models need to be parameterized with data for the visual receiver species. The visual perception of a bird depends ultimately on the type, number and distribution of different photoreceptors in its retina, the absorption properties of the pigments included in the outer segment of those photoreceptors, and the filtering properties of oil droplets located in the cone inner segments of the photoreceptors [27,121]. Some of these features might be evolutionary conserved in birds [22,122]. For instance, visual pigments are highly conserved other than the UVS/versus shortwave opsin (SWS1) [27]. However, oil droplet absorbance can vary among species, and the relative densities of each cone photoreceptor type may substantially vary even within the same species [27]. Moreover, a large body of growing evidence is showing that combinations of these features would be species–specific [27,123]. Therefore, model calculations that do not consider the particularities of each species might potentially lead to misleading inference. Illustrating this issue, a recent study has shown that the relative abundance of different photoreceptor types may have a large influence on estimated colour dichromatism [124].

(c). Models do not incorporate the complexity of higher-order brain processing

Colour perception results from the combination of physiological detection and neural processing mechanisms [17] (figure 1). Previous theoretical reviews have extended on the poor understanding of the mechanisms driving higher-order processing of colour and patterns in animals and on how this may limit model-based inference [28,29,90,125]. Emerging evidence in the context of egg discrimination would provide support for a role of post-detection mechanisms in processing colour stimulus and suggest that there could be a plastic component of Chr-based recognition in birds. A recent study measured the behavioural response of two host species, the American robin Turdus migratorius and the European blackbird Turdus merula, to gradients of natural (blue-green to brown) and artificial (green to purple) egg colour [126]. Interestingly, it was found that host discrimination decisions were not based on perceived visual differences between host and experimental eggs as estimated with a visual model. Indeed, hosts were more prone to reject brown eggs and to accept blue ones regardless of the fact that they showed similar absolute perceived differences from their own eggs [126]. Similar findings have been shown in magpies, Pica pica hosts of great spotted cuckoos Clamator glandarius [40], and in the chalk-browed mockingbird, Mimus saturninus host parasitized by the shiny cowbird Molothrus bonariensis [127], as well as in great reed warblers, Acrocephalus arundinaceus [128]. Also, one comparative work has shown that common cuckoo Cuculus canorus hosts parasitized with more blue, green and ultraviolet cuckoo eggs, or those laying more blue-green eggs, were more prone to accept experimental parasitism with artificial cuckoo eggs [129]. These findings would suggest that at least some birds do not base recognition decisions on pure colour distances as delivered to the visual system and that post-detection perceptual biases could be widespread in birds, meriting further investigation.

(d). The ambient light is not stable and may affect receptor-noise limited perceptual model calculations

Ambient light for detection of colour patches is contingent on time of day, weather, season or microhabitat complexity [18,88]. For instance, a study in European magpies has shown that the architecture of nests may affect the quality of light for egg discrimination, and that late breeders with experimentally increased luminosity in their nests were more prone to reject model eggs [40], suggesting a potential role for ambient light in egg detection. Moreover, it has been shown how visual thresholds for cuckoo egg discrimination estimated with an RNL model may change when using spectral irradiance measures collected in two extreme nest light environments, although this only minimally affected the proportion of cuckoo eggs that could be potentially discriminated based on Chr signals [58].

(e). Perception is multifaceted and information gathered through different sensory channels may modify behavioural responses to visual cues

In nature, it is difficult to isolate visual information from that a potential receiver (i.e. host or feeding parent) may gain through other sensory channels, and that could contribute to egg and/or nestling discrimination [27]. For instance, experimental work on egg discrimination has shown that exposure to non-natural tobacco olfactory cues [130], or the sight of the brood parasite at the nest may affect host proneness to reject parasite eggs based on stimulus summation mechanisms [76,131]. Also, it is very well established that coloration is just one of the features involved in the multicomponent and multimodal signalling begging strategy of nestlings [132,133], and that parents may gain information to decide their food allocation strategy based on several independent coloured traits [11,100,103] or acoustic signals [132]. The possibility that parents, parasites, and predators receivers may incorporate multiple distinct components within and across sensory modalities is not considered in the RNL model.

(f). Eggs and nestlings often have complex visual designs

Egg and nestling colour designs can be very complex as illustrated in figure 2. Many birds have highly polymorphic eggs regarding coloration and spottiness; such has been shown in the tawny-flanked prinia, Prinia subflava parasitized by the cuckoo finch Anomalospiza imberbis, or in the European magpie parasitized by the great spotted cuckoo (figure 2). Also, the colour designs of nestlings of many altricial birds can be very complex, and mimicry may for instance involve coloration, but also the size and distribution of colour patches, as in the interaction between the brood parasitic pin-tailed whydah, Vidua macroura and the common waxbill, Estrilda astrild ([134]; figure 2). Beyond spottiness, discrimination might also be influenced by differences between eggs or nestlings in shape [135] or in size [136]. The RNL model was explicitly designed to deal with differences in blocks of colour, and not with differences in complex visual designs implying variation in spottiness, shape or size. Although in many cases researchers have added information on aspects other than coloration to refine their analyses (e.g. patterning in table 1), we still know very little about how that variation integrates with colour differences in the brain of visual receivers [125]. Interpretation of model outcomes could be further complicated by scramble competition among siblings which often result in fast movement of signalling colour patches, which may affect noise at the photoreceptors and reliability of model predictions [29].

(g). Visual receivers can learn in nature

Discriminatory tasks have an important learning component, which has been the focus of extensive work in avian brood parasitism. In particular, it has been suggested that egg rejection of parasite eggs should be less frequent in young individuals, because a learning period is needed to express that capacity [137,138]. Some source of evidence indicates that this may happen. For instance, in great reed warblers, Acrocephalus arundinaceus, naive females are less likely to reject foreign eggs in their nests [139], also in magpies it was found that some individual hosts shift from accepting to rejecting foreign eggs when exposed to parasitic eggs several times during their lives [71]. Learning may lead to potential predictive errors where behavioural thresholds for discrimination estimated based on trained individuals in the laboratory were used to predict recognition of individuals in nature.

4. New directions of research in egg and nestling discrimination

The RNL model was specifically developed for the purpose of estimating discrimination thresholds between colours of similar hues [24], and therefore it may be less reliable when used to predict (large) perceptual differences in coloration [90]. One promising approach in this respect, would be combing the use of visual models with behavioural experiments testing receiver responses to a stimulus varying in a continuous range of natural colours (e.g. [126]).

We clearly need to know more about the role of Achr based discrimination and the mechanisms allowing the assembly of Chr and Achr differences in egg and nestling detection, as this may give us completely new insights into the functioning of colour-based recognition in variable nest light environments. Future studies dealing with egg and nestling detection are urged to jointly evaluate the role of Chr and Achr discrimination.

In the same vein, our knowledge about how colour and patterning information is integrated in the receivers' brain is very limited. The most conservative approach to retrieve all possible visual features that may serve in recognition would be to study it by combining the tetrachromatic colour space approach [25], or the RNL model [24]; with pattern analysis [31].

Given that reliability of RNL model predictions in dim light levels is impaired [29], future studies testing predictions of this hypothesis should preferentially rely on behavioural experiments to confirm the use of visual signals in holes [12]. A potential issue with ambient light variation could partly be alleviated by a more detailed analysis of the influence of spectral irradiance on the perception of colour differences and threshold for detection in these studies. In particular, a fruitful approach would be considering the ambient light in which receivers should discriminate eggs or nestlings.

A promising field of study is to deal with the multi-dimensional nature of communication in egg and nestling discrimination by adopting experimental approaches, in which stimulus perceived by different sensory channels, or different stimulus perceived by the visual channel, were simultaneously manipulated in crossed designs.

Finally, although it has been recently acknowledged of the importance of between-individual variation in visual systems for colour perception [140], I do not know of any study analysing how visual thresholds for discrimination may change with age or previous exposure to discrimination challenges in nature. Analysing the role of both between-individual and within-individual variation in perception seem particularly fruitful fields for future research in the evolution of recognition systems and post-detection neural processing.

5. Conclusion

The study of egg and nestling colour discrimination in birds has undergone an unprecedented explosion in the last decades thanks to the emergence of more objective methods to quantify colour differences and the consideration of sensory ecology [23,30]. In this context, the use of visual models have allowed the consideration of receptor physiology and ambient light that can affect the salience of a visual signal, and have been widely used to either assess how a visual receiver perceives differences between two or more colours, or to predict the limits for the detection of colour salience in the frame of major evolutionary hypotheses. Here I have stressed that although sensory-visual models are a powerful tool that has allowed us to accurately predict the role of vision in detection, discrimination and recognition under laboratory controlled experimental conditions, their use in field studies, in which is important to assess egg and nestling discrimination, need to be treated with caution. Some of the limitations of the visual models are concerned with model assumptions and parameterization or a poor consideration of post-detection neural processes, and were already pointed out and thoroughly discussed in excellent review papers in more general behavioural and evolutionary grounds [28,29,125]. Other limitations, however, are specific of the context of egg and nestling discrimination and deal with the highly variable nature of ambient light at the nests (sometimes very low), the often complex colour designs of eggs and nestlings, the simultaneous implication of different sensory channels and stimulus in egg and nestling detection, and the implication of a learning component in discrimination. Given the strong context-dependent reliability of visual model predictions, I argue that visual models should be used with more caution to establish inference about egg and nestling discrimination in nature, and, rather be used to provide reasonable hypotheses which need to be validated with behavioural discriminatory tests in the field (e.g. [12,61]). In this respect, it is worth stressing that in these systems, either hosts or parents are the receivers ultimately selecting egg and nestling colour designs, as their behavioural responses to differences in colour patterns are likely to have fitness effects [125]. Thus, measuring all the variation in egg/nestling colours, and inferring from visual models what the receivers can see, would not be enough to understand the evolution of egg and nestling colour patterns, unless the behaviour of the receivers in response to such variation is assessed.

Data accessibility

This article has no additional data.

Competing interests

There are no competing interests.

Funding

My research on avian coloration that allowed me to think on the ideas that have given rise to this contribution was funded by the Spanish Ministry of Education and Science/FEDER through the projects CGL2008-00718, CGL2011-27561/BOS, CGL2014-56769-P and CGL2017-83503-P.