Enhanced photothermal absorption in iridescent feathers

Abstract

The diverse colours of bird feathers are produced by both pigments and nanostructures, and can have substantial thermal consequences. This is because reflectance, transmittance and absorption of differently coloured tissues affect the heat loads acquired from solar radiation. Using reflectance measurements and heating experiments on sunbird museum specimens, we tested the hypothesis that colour and their colour producing mechanisms affect feather surface heating and the heat transferred to skin level. As predicted, we found that surface temperatures were strongly correlated with plumage reflectivity when exposed to a radiative heat source and, likewise, temperatures reached at skin level decreased with increasing reflectivity. Indeed, nanostructured melanin-based iridescent feathers (green, purple, blue) reflected less light and heated more than unstructured melanin-based colours (grey, brown, black), as well as olives, carotenoid-based colours (yellow, orange, red) and non-pigmented whites. We used optical and heat modelling to test if differences in nanostructuring of melanin, or the bulk melanin content itself, better explains the differences between melanin-based feathers. These models showed that the greater melanin content and, to a lesser extent, the shape of the melanosomes explain the greater photothermal absorption in iridescent feathers. Our results suggest that iridescence can increase heat loads, and potentially alter birds' thermal balance.

1. Background

Feathers are complex structures with large phenotypic variation and a diversity of functions. Feathers not only aid in flight [1], display [2] and camouflage [3] but also play a substantial role in heat retention [4]. Indeed, the plumage provides a critical thermal buffer between the animal and its environment [4]. Radiation penetration into avian plumage not only varies with feather density and positioning, but also with coloration [4,5]. Feather coloration can have significant thermal properties because coloured tissues differentially reflect or absorb the solar radiation that can strongly affect the acquired heat loads [6,7]. As such, colour can play a significant role in a bird's energy budget, having a direct impact on the metabolic costs of maintaining a constant body temperature [6–10]. In a cool environment, exposure to solar radiation can significantly reduce the energetic costs of thermoregulation, while at high ambient temperatures, high insolation can result in an additional cost [8–13]. There is increasing evidence from comparative studies that plumage brightness correlates positively with temperature in a way that fits expectations for a thermoregulatory function [14,15]. However, the correlation between plumage coloration and solar heat loads is not always straightforward. Even though darker tissues reflect less light than brighter ones [16], most radiation is absorbed by the upper feather layers, while in light plumage the radiation penetrates much deeper into the plumage [4,17]. Thus, the heat transfer to skin level, where it could directly affect a bird's thermal balance [18], strongly depends on the optical and insulating properties of feathers (figure 1).

Figure 1.

Thermal effects of solar radiation illustrated for a highly melanized dark (left) and a bright coloured feather (right). The darker feather absorbs more light at the surface, which leads to a higher increase in surface temperature. Although less light is transmitted through the feather, the skin heats up due to conduction and convection. The brighter feather reflects more and absorbs less sunlight and its surface heats up less. The skin surface beneath bright plumage heats up due to deeper light penetration, higher transmittance and backscattering of the light. Heat transfer mechanisms were illustrated based on the studies of Cena & Monteith [16], Burtt [19], Walsberg [17] and Wolf & Walsberg [4]. Larger font/symbol sizes were used to visualize increasing intensities.

The tremendous diversity in feather coloration is produced by feather microstructure, as well as pigments and nanostructures within the feather barbule [20]. Pigments selectively absorb visible light and produce either brown to black (melanins) or yellow to red colours (carotenoids), while nanostructured tissues produce iridescent or non-iridescent structural colours through coherent light scattering [20]. Nanostructured iridescent colours are produced in feather barbules by combinations of air, keratin and melanin in the form of melanosomes that differ in size, shape and spatial distribution at a nanometer-scale [21]. The combination of these materials, that differ in their complex refractive indices (including both a real component describing the bending of light and an imaginary component describing absorption), strongly affects the scattering and absorption of light [21]. Nanostructural arrangements, with particles that are smaller than the wavelengths of light, can further lead to plasmonic photothermal effects, where absorbed light is dissipated into heat [22].

Despite the effect that diverse pigments and feather structures could have on the heat gain of the integument, most studies have focused on thermal properties by comparing dark with bright colours ([10,13,23], but see [24,25]). However, radiative heating of feathers is a strong function of the type and amount of pigments, and the spectral emission of the light source [26]. For example, melanin has a broadband absorption [27,28] and carotenoids absorb in a narrow range [29]. Here, we first investigated the relationship between plumage reflectivity and colour producing mechanism, i.e. non-pigmented whites, carotenoids, unstructured melanin-based colours, olives (a mixture of melanin and carotenoids [30]) and nanostructured iridescent colours. We then explored the thermal properties of each colour producing mechanism. Due to the highly absorptive properties of melanin [31], we hypothesized that melanin-based feathers would absorb more light, and heat more, than feathers with other pigments (carotenoids) or without any pigment (whites) (figure 1). Within melanin-based feathers, we hypothesized that nanostructured iridescent feathers would heat more due to greater amounts of melanin within the feather barbules [32,33], their lower reflectance in the near-infrared [24] and plasmonic photothermal effects (i.e. highly localized conversion of light to thermal energy) [22]. This study is the first to describe the thermal properties of iridescent feathers and to bring these properties into context with those of other colour producing mechanisms.

We chose to study sunbirds (Nectariniidae) because their plumage is extremely colourful and produced by diverse mechanisms (figure 2). Most sunbird species are sexually dimorphic, whereby males have bright iridescent colours and females mostly dull, unstructured melanin-based plumage. For display, sunbirds select highly visible places exposing themselves to direct sunlight [35]. If iridescent feathers have higher photothermal absorption than unstructured melanin-based plumage, male sunbirds might be exposed to greater solar heat loads than females, particularly under high intensities of solar radiation. Nectariniidae consists of 147 species, distributed across Africa, Asia and Australasia. These small (4–22 g) passerines inhabit a variety of ecosystems, occupying tropical to temperate climate zones [36]. Although all birds are endotherms, sunbirds show heterothermic responses (i.e. changes in metabolic rate and body temperature following environmental cues); their body temperatures vary with ambient conditions ranging from 26.8°C (torpor) to 44.2°C [37,38]. Since their body temperature is thermolabile and plumage provides a critical thermal buffer between the animal and its environment [4], sunbird feather coloration could play a significant role in thermoregulation.

Figure 2.

Species that were used for the heating experiment and specifically chosen for their differences in feather coloration at the different body parts: (a) Hedydipna collaris, (b) Anthreptes anchietae, (c) Cinnyris cupreus, (d) Leptocoma aspasia, (e) Deleornis fraseri, (f) Cinnyris bifasciatus, (g) Chalcomitra adelberti, (h) Anthreptes aurantius, (i) Aethopyga siparaja, (j) Aethopyga ignicauda, (k) Cinnyris superbus, (l) Cinnyris chalybeus, (m) Chalcomitra senegalensis, (n) Chalcomitra amethystina, (o) Nectarinia famosa. Species plates of the male sunbirds shown here were reproduced by permission of Lynx Edicions [34].

2. Methods

To test the thermal properties of differently coloured sunbird feathers, we combined an experimental and a modelling approach. We performed heating experiments on museum specimens to assess how differently coloured feather surfaces heat up and how heat is transferred to the skin. To explore the effect of melanin and keratin composition within the barbule, we performed optical and heat modelling based on idealized structures obtained from electron microscopy images of nanostructured iridescent and unstructured melanin-based coloured feather barbules.

2.1. Sample collection and preparation

For reflectance measurements and heating experiments, we studied 84 skins from 15 different sunbird species (three males, two to three females per species) housed at The Field Museum, Chicago (electronic supplementary material, table S1). We chose species that differ in coloration and colour composition (figure 2). Since the curvature of the body of bird specimens could lead to differences in heating rates of the different body parts, we included the uniformly coloured Fraser's sunbird (Deleornis fraseri) as a reference species (n = 10) to test how heat transfer rates differ with body part. We chose this species because all feathers are based on the same colour producing mechanism and it shows the least variation in colour between body parts (figure 2e). For optical and heat modelling we sampled white, grey, black and three iridescent feather colours (copper, green, purple) from the museum's collection.

2.2. Reflectance measurements

We used spectrophotometry to measure reflectance across most of the solar spectrum (380–2000 nm) on different colour patches and body parts (figure 3a) of sunbird museum specimens following Shawkey et al. [24] and Stuart-Fox et al. [39]. We measured reflectance using a dual system of Avantes spectrometer and light sources (AvaSpec-ULS2048 L StarLine Versatile Fibre-optic Spectrometer + AvaSpec-NIR256/512-2.0/2.2TEC NIRLine Near-Infrared Fibre-optic Spectrometer + AvaLight-HAL-(S)-MINI Tungsten-Halogen Light Source + AvaLight-DH-S Deuterium-Halogen Light Source). We connected the fibre-optic cable with a reflection probe holder that was held at an angle of 90° and set on top of the measured colour patch. For the analyses, we averaged three reflectance measurements per colour patch and body part using the R package pavo [40]. To account for the spectral emission of the light source (see Methods, Heating experiment), we calculated the spectral reflectivity (R) as a function of reflectance (S) and irradiance (I) of the heat lamp according to Smith et al. [41] over 380–2000 nm. Hence, reflectivity was calculated as

$$R\hspace{0.17em}(\text{\%})=\frac{{\int }_{{\lambda }_{\min }}^{{\lambda }_{\max }}S(\lambda )I(\lambda )d(\lambda )}{{\int }_{{\lambda }_{\min }}^{{\lambda }_{\max }}I(\lambda )d(\lambda )}\times 100.$$

Figure 3.

Experimental set-up: (a) Division of body parts in sunbird specimens. (b) Temperature at skin level was measured with thermocouples placed beneath the feathers. To control for feather density, we counted the number of feathers above a 25 mm² squared plate and assessed the feather area per unit skin surface $(\mathrm{pl})$ as a function of $(\mathrm{feather}\hspace{0.17em}\mathrm{length}\hspace{0.17em}(\mathrm{mm})\times \mathrm{feather}\hspace{0.17em}\mathrm{width}\hspace{0.17em}(\mathrm{mm})\times \mathrm{number}\hspace{0.17em}\mathrm{of}\hspace{0.17em}\mathrm{feathers})/(25\hspace{0.17em}\mathrm{m}{\mathrm{m}}^2)$ per body part. (c) Feather surface temperature was measured with thermal imaging.

2.3. Feather area per unit skin surface

To account for plumage thermal resistance that increases with feather density, we simplified the approach by Walsberg et al. [42] and estimated feather area per unit skin surface $(\mathrm{pl})$ for each body part of each specimen as

$$\mathrm{pl}=\hspace{0.17em}\frac{\mathrm{feather}\hspace{0.17em}\mathrm{length}\hspace{0.17em}\hspace{0.17em}(\mathrm{mm})\times \mathrm{feather}\hspace{0.17em}\mathrm{width}\hspace{0.17em}\hspace{0.17em}(\mathrm{mm})\times \mathrm{number}\hspace{0.17em}\mathrm{of}\hspace{0.17em}\mathrm{feathers}}{25\hspace{0.17em}\mathrm{m}{\mathrm{m}}^2},$$

where length was measured from feather pin to feather tip, the width was measured at the feather's widest part and the number of feathers was counted above a 5 × 5 mm² squared plate (figure 3b). Details are described in electronic supplementary material, appendix.

2.4. Heating experiment

To assess how feather coloration affects solar heat gain, both on feather surfaces and at skin level, we performed heating experiments on sunbird museum specimens. The specimen was laid horizontally under a light (Philips IR 250 W) that mimics natural solar radiation (electronic supplementary material, figure S1 and table S2). The same lamp has been used for similar experiments [43–45], and was installed 50 cm above the specimen. Irradiance measured at the height of the specimen with a handhold pyranometer (Dr Meter SM206 Digital Solar BTU Power Meter) was 1000 W m⁻², an intensity reached during a sunny day [46]. However, larger specimens would be slightly (a few millimetres) closer to the heat source, an effect we controlled for by including wing chord length as a proxy for body size (see Methods, Statistical analysis). Following Marder [47] and Rogalla et al. [45], specimens were heated for 10 min, which insured that the surface temperature had enough time to stabilize and reached the asymptotic maximum temperature. Specimens were first heated ventrally and then, after cooling down to ambient temperature, were heated dorsally. Temperatures were measured for different colour patches and body parts (figure 3). Feather surface temperature was assessed with a thermal camera (FLIR T530 24°). Thermal videos were analysed with the ResearchIR software (FLIR Systems Inc., Oregon, USA) with feather emissivity set at 0.95 [48]. Skin surface temperature was measured with thermocouples (Type 5SC-TT-T-36-36, Omega, Connecticut) attached to the skin beneath the feathers. To test whether thermocouple and thermal camera measurements were comparable, we included a temperature reference, which consisted of a thermocouple placed beneath a piece of electric tape with a known emissivity of 0.95. We calculated the rate of heating $k_2$ of feather surfaces, skin temperature and references according to Voss & Hainsworth [49] as the negative derivative of the slope of the heating curve:

$$\ln \left[\frac{(T_t-T_{\mathrm{\infty }})}{(T_0-T_{\mathrm{\infty }})}\right].$$

$T_t$ is the temperature at a given time t, $T_0$ is the initial temperature and $T_{\mathrm{\infty }}$ is the asymptotic maximum temperature. For $T_{\mathrm{\infty }}$, we averaged the temperature values of the last 100 s of the heating curve. To quantify the heat transfer between plumage surface and skin, we used the differentials between $T_{\mathrm{\infty }}$ measured at feather surface and at skin level. Air temperature during the experiments was 22.8°C ± 0.4 and humidity was 49.7% ± 3.8.

2.5. Feather nanostructure

To explore the effect of feather nanostructure, more precisely the effect of melanin and keratin composition within the feather barbule, on photothermal properties, we used Ansys Lumerical's finite-difference time-domain (FDTD) solver and analytical models. These models allowed us to predict the photothermal properties, i.e. reflectance, transmittance and absorption profile as well as temperatures of feather barbules based on barbule microstructures. First, we modelled cross sections of six feather barbules based on transmission electron microscopy (TEM) and scanning electron microscopy (SEM) images. Second, we explored the effect of bulk melanin concentration and distribution within the keratin. Finally, we explored the effects of structural arrangement and shape of melanosomes on the photothermal properties of our models. Models were based on mean values (barbule height, widths of outer keratin layers, melanosome shapes, sizes and arrangements) from microscopy images of white, grey, black and iridescent barbules that were measured in ImageJ [50]. For all model sets we obtained reflectance, transmittance and absorption profile per wavelength (380–2000 nm) at normal incidence. For the refractive indices, we used the Cauchy dispersion for keratin and melanin in Stavenga et al. [28]. We calculated the steady state temperature attained due to photothermal heating, based on the optical absorption profile. Temperature increase compared to the ambient temperature (20°C) was $\mathrm{\Delta }T=\hspace{0.17em}q/h$, where q is the cumulative energy absorbed at a light intensity of 1000 W m⁻² and h is the convection coefficient (10 Wm⁻² K⁻¹). Details on image collection and modelling are described in electronic supplementary material, appendix (Materials and methods, electronic supplementary material, figures S2, S3 and table S3). To confirm that barbule models corresponded to the actual structure, we compared the modelled reflectance curves with the reflectance curve measured on the feather barbule using microspectrophotometry (CRAIC AX10 UV–visible micro-spectrophotometer, CRAIC Technologies, Inc., USA). We compared modelled barbule temperatures with experimentally measured plumage surface temperatures after 1-min heating, this to minimize the effects of specimen size on measured surface temperature (see Results, Heating of feather surfaces).

2.6. Spectral effects

To explore which part of the spectrum best explains feather surface heating, we calculated reflectivity over the visible spectrum (380–700 nm) and the near-infrared (701–2000 nm) and tested the correlation with the experimentally measured $T_{\mathrm{\infty }}$ of the feather surfaces. In feather barbule models, we modelled the temperature increase as a function of the cumulative energy absorbed over the visible spectrum (380–700 nm) and over the near-infrared spectrum (701–2000 nm) separately.

2.7. Statistical analysis

To explore the experimentally measured heat loads on feather surfaces and skin of museum specimens, we first controlled for effects, such as variation in $T_0$, different methods for temperature acquisition (thermocouples and thermal imaging), grouping of body parts and phylogenetic signal. Details are described in electronic supplementary material, appendix. Repeatability, i.e. the proportion of variance among and within groups [51], of the surface temperatures from 180 colour patch measurements (six body parts in 15 species of both sexes), was high for $T_{\mathrm{\infty }}$ (r = 0.798), but relatively low for the heating rate (r = 0.142), which could be explained by differences in the specimens, such as specimen size. Indeed, larger specimens would be slightly closer to the heat source, which we controlled for by measuring tarsus, head size and wing length of each specimen. We ran a linear mixed effects analysis (LMM) with surface $T_{\mathrm{\infty }}$ as response variable, and tarsus, head size, wing chord length, body region and sex as fixed effects, and species and individual as random effects. Tarsus and head size did not show any significant effect on $T_{\mathrm{\infty }}$ (tarsus: β = −0.48, CI: −1.23, 0.29; head: β = 0.43, CI: −0.17, 1.04). Hence, we only included wing chord length as a control variable for body size in our analyses. Due to the low repeatability of the heating rate, we focused on $T_{\mathrm{\infty }}$ in our analyses.

To test how the heating of feather surfaces and at skin level changed with colour producing mechanism, we divided the colours into five categories: structural iridescent colours, non-structural melanin-based colours, olives, carotenoid-based colours and whites. Colours were categorized both visually and based on the shape of the plumage reflectance spectra [52]. Sample sizes per colour category differed for each body part. Since the greatest diversity in colour was found in the breast feathers, we put a specific focus on these results.

We first used an analysis of variance (ANOVA) model to test whether colour producing mechanisms differed in reflectivity (380–2000 nm). We ran Tukey HSD tests to perform multiple pairwise-comparisons between the means of the groups. We then used an LMM to examine the relationship between feather surface heating, plumage reflectivity and colour producing mechanism. We set $T_{\mathrm{\infty }}$ of the feather surfaces as response variable, and reflectivity, colour category, body region, sex and wing chord length as fixed effects, and species and individual as random effects. Similarly, we examined this relationship for the temperatures reached at skin level. In a second LMM, we set $T_{\mathrm{\infty }}$ reached at skin level as response variable, and plumage reflectivity, colour category, feather area per unit skin surface, body region and sex as fixed effects and species and individual as random effects. We further tested whether feather area per unit skin surface $(\mathrm{pl})$ differed between species, sex and body parts in an ANOVA.

To quantify the heat transfer between plumage surface and skin, we used the differentials between $T_{\mathrm{\infty }}$ measured at feather surface and at skin level. To test whether temperature differentials differed with feather area per unit skin surface $(\mathrm{pl})$ and plumage reflectivity, we used an LMM. We set surface-to-skin temperature differentials as response variable, plumage reflectivity, colour category, feather area per unit skin surface, body region and sex as fixed effects and species and individual as random effects. Additionally, we used ANOVA models and Tukey HSD tests to compare the mean temperatures (surface $T_{\mathrm{\infty }}$, skin $T_{\mathrm{\infty }}$, differential temperature) of each colour producing mechanism. These models were run separately for each body part.

To test whether surface $T_{\mathrm{\infty }}$ could be explained better by reflectivity over the visible spectrum (380–700 nm) or by reflectivity over the near-infrared (701–2000 nm), we used linear models with $T_{\mathrm{\infty }}$ as response variable and reflectivity as predictor variable. We further compared mean reflectivity over each spectrum using an ANOVA model.

We ran all our statistical analyses in R [53] using the following packages: dplyr [54], lme4 [55], phytools [56] and pavo [40]. For data visualization, we used ggplot2 [57].

3. Results

3.1. Reflectance measurements

Averaged reflectance curves of melanin-based feathers (nanostructured iridescent and unstructured) differed little in the visible spectrum (figure 4a). Only beyond 600 nm and within the near-infrared spectrum did iridescent feathers reflect less radiation than unstructured melanin-based colours (figure 4a). Plumage reflectivity varied with colour producing mechanism (F_4,497 = 78.22, p < 0.001). White and carotenoid-based feathers were brightest and did not differ significantly in their reflectivity (figure 4b). All other colour producing mechanisms differed significantly, whereby iridescent feathers were least reflective, followed by non-iridescent melanin-based colours and olives (figure 4b).

Figure 4.

Colour producing mechanisms differ in reflectance and surface heating. (a) Averaged reflectance curves over the solar spectrum with the shaded area as standard deviation and (b) reflectivity as a function of reflectance and irradiance of the heat lamp (380–2000 nm) are shown for each colour category. For (a) and (b) measurements for all colour patches and body parts are provided and corresponding sample sizes are given in (b). For the reflectivity (b) we plotted the mean, standard deviation, minimum and maximum value. Significant differences between the means of the groups were assessed with a Tukey HSD test. For colour categories that do not differ significantly (p > 0.05) the same letters are provided. (c) Feather surface heating is shown as a function of time for male breast feathers. Sample sizes are provided in the plot. We used LOESS curve fitting (local polynomial regression) for fitting the smoothed curves. The shaded area is the standard error of the heating curves.

3.2. Heating of feather surfaces

Heating curves of the feather surfaces were very steep in the beginning and reached $T_{\mathrm{\infty }}$ within the first 200 s of the experiment (figure 5, electronic supplementary material, figure S4). In this short time the feather surfaces, starting at room temperature, reached maximum values of 83°C (mean $T_{\mathrm{\infty }}=64.6\pm {7.7}^{\circ }\mathrm{C}$). Heating curves varied with colour (figure 5) and colour producing mechanism (figure 4c). In breast feather surfaces (n = 252), nanostructured melanin-based iridescent feathers reached higher $T_{\mathrm{\infty }}$ than unstructured melanin-based feathers, olives, carotenoid-based colours and whites (F_4,247 = 29.1, p < 0.001). Unstructured melanin-based feathers became hotter than olives, carotenoids and whites (figure 6a). In the throat (n = 84), nanostructured melanin-based iridescent feather surfaces heated up more than unstructured melanin-based colours and olives (F_3,80 = 7.78, p < 0.001). In the mantle (n = 84), iridescent feather surfaces heated up more than olives and carotenoids, and unstructured melanin-based feathers warmed up more than carotenoids (F_3,80 = 6.01, p < 0.001; electronic supplementary material, figure S5). Maximum surface temperature decreased with plumage reflectivity (figure 7a): $T_{\mathrm{\infty }}$ declined by about −0.35 ± 0.03°C (t-value = −10.34, CI: −0.41, −0.28) per 1% increase in reflectivity. By contrast, $T_{\mathrm{\infty }}$ increased by about 0.20 ± 0.06°C (t-value = 3.125, CI: 0.07, 0.33) per 1 mm increase in wing chord, indicating that larger specimens that are slightly closer to the heat source warm up more.

Figure 5.

(a) Reflectance curves for a selection of colour patches with the shaded area as standard deviation and (b) the corresponding heating curves of the feather surfaces as a function of time. We used LOESS curve fitting (local polynomial regression) for fitting the smoothed heating curves. The shaded area is the standard error. For the raw data of the single heating curves please refer to electronic supplementary material, figure S4. In both (a) and (b) iridescent plumage is shown with dashed, and non-iridescent plumage with continues lines. In general, feather surfaces that reflect less light heat up more under exposure to the heat lamp.

Figure 6.

Variation in (a) maximum surface temperature, (b) maximum skin temperature and (c) surface-to-skin temperature differentials (heat transfer) for all colour producing mechanisms. We plotted the mean, standard deviation, minimum and maximum value for temperatures measured at the breast (n = 252) of sunbird museum specimens. Results for further body parts can be found in electronic supplementary material, figure S5. We ran a Tukey HSD test to perform a multiple pairwise comparison between the means of the groups. For categories that do not differ significantly (p > 0.05) the same letters are provided above the corresponding boxes.

Figure 7.

(a) Maximum surface temperature, (b) maximum skin temperature and (c) surface-to-skin temperature differential decrease with plumage reflectivity (380–2000 nm). Data are shown for breast feathers only (n = 252). The shaded band is a pointwise 95% confidence interval on the fitted values.

3.3. Feather area per unit skin surface

Feather area per unit skin surface $(\mathrm{pl})$ differed among species (F_14,483 = 8.83, p < 0.001) and body part (F_5,483 = 510.09, p < 0.001) but did not differ between sexes (F_1,483 = 1.83, p = 0.177). We found the highest feather density at the belly, followed equally by mantle and lower breast band. Feathers were less dense in the upper breast band and least in both throat and crown (figure 8).

Figure 8.

Feather area per unit skin surface $(\mathrm{pl})$ as a function of $(\mathrm{feather}\hspace{0.17em}\mathrm{length}\hspace{0.17em}(\mathrm{mm})\times \mathrm{feather}\hspace{0.17em}\mathrm{width}\hspace{0.17em}(\mathrm{mm})\times \mathrm{number}\hspace{0.17em}\mathrm{of}\hspace{0.17em}\mathrm{feathers})/(25\hspace{0.17em}{\mathrm{mm}}^2)$ per body part. We plotted the median as the line that divides the box and the upper and lower quartiles as the end of the boxes. The extreme lines each show the highest and the lowest value.

3.4. Heating at skin level

At skin level, $T_{\mathrm{\infty }}$ differed with colour producing mechanism in the breast (F_4,247 = 11.15, p < 0.001; figure 6b) and to a small extent in the mantle (F_3,80 = 3.91, p < 0.012; electronic supplementary material, figure S5). Other body regions showed no differences in skin temperature between colour production mechanisms. The skin beneath nanostructured iridescent breast feathers became hotter than the skin beneath olive feathers, and both got hotter than the skin beneath orange, yellow or red feathers (figure 6b). Beneath non-iridescent melanin-based feathers, such as grey or black, the skin warmed up more than under carotenoid-based feathers, while temperatures beneath white feathers did not differ significantly from the other colour categories (figure 6b). $\hspace{0.17em}{T}_{\mathrm{\infty }}$ at skin level did not differ between nanostructured iridescent and non-iridescent melanin-based feathers. A more reflective feather surface led to a temperature decrease at skin level (figure 7b). $T_{\mathrm{\infty }}$ at skin level decreased −0.21 ± 0.04°C (t-value = −5.61, CI: −0.29, −0.14) for each increase in 1% plumage reflectivity. Feather area per unit skin surface did not affect skin $T_{\mathrm{\infty }}$ (β = 0.006 ± 0.005, CI: −0.003, 0.016).

3.5. Heat transfer

In olive, carotenoid-based and white feathers we observed a negative mean surface-to-skin temperature differential, indicating that the temperature at skin level became hotter than at the feather surface (figure 6c). Temperature differentials in the breast were lowest in white feathers and highest in nanostructured melanin-based iridescent feathers and unstructured melanin-based feathers (F_4,247 = 5.27, p < 0.001; figure 6c). Per 1% increase in plumage reflectivity differential, temperatures decreased by −0.15 ± 0.04°C (t-value = −4.04, CI: −0.23, −0.08) (figure 7c). We found no effect of feather area per unit skin surface on surface-to-skin temperature differentials (β = 0.002 ± 0.004, CI: −0.01, 0.01).

3.6. Feather nanostructure

Reflectance peaks of idealized models of feather barbules based on TEM and SEM images corresponded with microspectrophotometer measurements for nanostructured iridescent colours indicating a good match between models (keratin blocks with integrated melanosomes) and measurements (figure 9a,b). The white, grey and black barbule model matched less well, most likely due to the difficulties in replicating unstructured barbules in the optical models (figure 9b).

Figure 9.

(a) In the idealized models of feather barbules based on TEM images melanin is shown in brown and keratin in grey. (b) Normalized modelled reflectance peaks (dashed line) correspond with microspectrophotometer measurements (continuous line). Insets in each plot are close-up pictures of the plumage. (c) Brighter feather barbules transmit more light. (d) Absorption varies with colour and is low in the near-infrared. (e) Feather surface temperatures of museum specimens after 1-min heating show a similar trend to modelled temperatures of feather barbules (f): iridescent green heated up most followed by iridescent purple, copper, black, grey and white. (g) Modelled temperatures increase with greater melanin content in the feather barbules, here expressed as the total width of melanin layers. Black dots represent temperature values modelled for a 3 µm keratin block with an incorporated melanin layer of widths ranging from 50 to 2000 nm, while squares represent the temperatures modelled for feather barbules in their representative colour.

Brighter feather barbules, such as white and grey, transmitted more light and most light was transmitted in the near-infrared spectrum (figure 9c). Modelled photothermal absorption was greatest in iridescent and black feather barbules, such that most light was absorbed in the visible spectrum (figure 9d). We found a similar trend in both modelled barbule temperatures and experimentally measured plumage surface temperatures after 1-min heating: iridescent green heated up most followed by iridescent purple, iridescent copper, non-iridescent black, grey and white (figure 9e,f). Temperatures clearly increased with increasing melanin content in the feather barbule (expressed as the total width of melanin layers within the model (figure 9g)). In fact, modelled transmittance decreased, and absorption increased, with increasing melanin content leading to a rise in temperature for both models in which melanin was considered as an increasing single layer and with increasing numbers of melanin layers, similar to the structures found in iridescent feather barbules (electronic supplementary material, figure S6).

We observed an effect of melanosome shape on the photothermal properties of our models (figure 10). In both organized and random arrangements of melanosomes with a constant melanin content (ca 7.3%) within the keratin block, photothermal absorption was higher for shapes with greater aspect ratios (figure. 10a,b). As such, the model that contained melanin platelets heated up more than the models that contained rods, ellipsoids or spheres (figure 10c). We observed a slight effect of structural arrangement on temperature, which was not consistent in all shapes and considerably smaller than the effect induced by melanosome shape. However, temperature increase based on shape-effects was substantially smaller than the temperature increase associated with increasing melanin contents as described for the feather barbule models in figure 9.

Figure 10.

Effect of shape and organization of nanostructures on photothermal properties. We modelled shapes that are common in feather barbules, i.e. spheres, ellipsoids, rods and platelets both in an organized and random arrangement. Melanin content was constant for all models (ca 7.3% of the total system). Each shape is shown in a different colour, whereby random arrangements compared to the corresponding organized structure are shown in a darker brightness level of the same hue. (a,b) Smoothed absorption spectra for organized and random arrangements of differently shaped melanosomes. (c) Modelled temperatures for different shapes and arrangements of nanostructures.

3.7. Spectral effects

In experimentally heated feather surfaces, $T_{\mathrm{\infty }}$ was influenced by both reflectivity over the visible spectrum (β = −0.37 ± 0.02, p < 0.001, R² = 0.53; electronic supplementary material, figure S7a) and reflectivity over the near-infrared (β = −0.68 ± 0.05, p < 0.001, R² = 0.47; electronic supplementary material, figure S7b). However, reflectivity of nanostructured melanin-based iridescent feathers and unstructured melanin-based feathers differed slightly more in the near-infrared (visible: F_1,91 = 4.79, p < 0.032; near-infrared: F_1,91 = 6.61, p < 0.012; electronic supplementary material, figure S7c). Interestingly, the temperature increase in modelled feather barbules was influenced to a greater extent by light absorption across the visible spectrum and only to a small extent by near-infrared absorption (electronic supplementary material, figure S7d).

A summary of our main findings is presented in table 1.

Table 1.

Summary of main findings.

section	summary of results
reflectivity	iridescent feathers were least reflective, followed by non-iridescent melanin-based colours and olives. White and carotenoid-based feathers were most reflective
heating of feather surfaces	maximum surface temperature decreased with increasing plumage reflectivity
	iridescent breast feathers reached higher maximum temperatures than unstructured melanin-based breast feathers, olives, carotenoid-based colours and whites
heating at skin level	maximum temperature at skin level decreased with increasing plumage reflectivity
	the skin beneath iridescent breast feathers became hotter than the skin beneath olive breast feathers, and both got hotter than the skin beneath carotenoid-based colours
heat transfer	surface-to-skin temperature differentials decreased with increasing plumage reflectivity
	mean temperature at skin level became hotter than at the feather surface in olive, carotenoid-based and white breast plumage patches
models of feather barbules	photothermal absorption was highest in iridescent feather barbule models
	temperatures increased with increasing melanin content in the model
models of different melanosome shapes	photothermal absorption increased with increasing aspect ratio of melanosome shapes
	shape-induced temperature increases were substantially smaller than increases linked to greater melanin contents

4. Discussion

As predicted, radiative heating of feather surfaces was strongly correlated with plumage reflectivity. Reflectivity, however, varied with the underlying colour producing mechanism. We showed that iridescent feathers, which are less reflective than carotenoid-pigmented and non-pigmented white feathers, heated up most when exposed to a radiative heat source. Iridescent feathers contain highly absorbent melanin pigment [21,31]. Increasing the amount of melanin in our models led to greater photothermal absorption and higher temperatures, while modifying the shape and arrangement of the melanosomes had little effect. This is surprising because smaller melanosomes with lower aspect ratios should reflect more light [22,58]. Melanosomes in iridescent feathers generally have higher aspect ratios [59] and in iridescent sunbird feathers were mostly platelet-shaped. In our models, these platelets heated up more than the spherical melanosomes typically found in grey feather barbules or the rod-shaped melanosomes of black barbules. However, temperature increased considerably less with shape than with melanin content. Therefore, the large number of melanosomes in iridescent feathers [32,33] likely explains their higher surface temperatures after radiative heating, with shape and structural effects playing only a minor role.

Consistent with previous research, we found clear differences between experimentally measured iridescent and unstructured melanin-based feather reflectance spectra in the near-infrared range [24]. Interestingly, experimentally measured surface temperatures were explained by both parts of the spectrum (visible and near-infrared), while the temperature increase in our models was explained more by differences in the visible spectrum. The absorption spectrum of melanin, which is high in the visible range but low in the near-infrared [27,60], could explain this difference for the models of single barbules. When considering the entire feather surface, however, differences in the near-infrared can be caused by variation in feather microstructure, such as arrangement or density of barbules [61]. Based on our experimental results, both parts of the spectrum likely explain the differences in asymptotic maximum temperatures of feather surfaces.

Feathers play a significant role in heat transfer processes [4,62]. We found a strong correlation between plumage reflectivity, particularly with feather surface temperatures and, to a smaller extent, with temperatures measured at skin level. However, for brighter feathers maximum skin temperatures were higher than maximum surface temperatures, likely due to deeper light penetration into the plumage [4], backscattering effects [63] and greater transmittance [19]. Interestingly, some sunbird species have black skin, which not only protects them against high ultraviolet irradiation [64] but can further lead to higher heat loads at skin level and more so under higher light transmitting and heat insulating tissues (e.g. as in polar bears [65]). Here, skin temperatures beneath highly melanized feathers reached values similar to surface temperatures, likely as a result of heat transfer by thermal conduction and convection. Overall, the skin reached higher maximum temperatures under less reflective and hotter surfaces, and became hottest beneath highly melanized feathers.

Surprisingly, we found no effect of feather density on maximum temperature at skin level nor on surface-to-skin temperature differentials. This is likely because feather density differed strongly with body region. Interestingly, feather density was lowest in the crown and throat. These body parts are most exposed to solar radiation and could thus heat up quickly under the sun, but also release heat faster. The belly, which is not directly exposed to solar radiation, had high feather density and thus low heat transfer, indicating good insulative properties. However, birds can behaviourally adjust their plumage thickness, e.g. by fluffing up in the cold, to maintain their body temperature [62]. Erected feathers could reduce heat gain at the skin, and this effect is stronger in darker than brighter feathers [4].

Many sunbird species occur in high altitudes in tropical climate zones [36], where they are exposed to very hot days, but freezingly cold nights with daily temperature fluctuations of up to 20°C [38]. Since basal metabolic and thermoregulatory costs in small birds account for 40–60% of their total daily energy expenditure [66], living under these extreme conditions can be energetically costly. Sunbirds can lower their body temperature at low air temperatures, thereby decreasing heat loss and reducing their metabolic rate to save energy [37,67]. But overheating might result in higher acute lethality than cooling [68] and little is known about how sunbirds prevent it. Since the thermal dynamics of plumage depends on feather microstructure and microoptics [4], these features could provide thermal protection and directly affect a birds' energy balance. Here, we showed that plumage reflectivity affects the temperature reached at skin level in sunbird museum specimens. In sunbirds, females often display dull, olive or grey colours, while more light-absorbing highly melanized iridescent colours are typically found in males. In male sunbirds, iridescence is particularly prominent in crown and mantle, i.e. the body parts that are most exposed to solar radiation, implying that under high solar irradiance males could be exposed to higher heat loads than females. Since plumage coloration can affect the metabolic costs of maintaining a constant body temperature [6–10] male sunbirds could experience a stronger reduction of energetic costs than females in early morning hours when ambient temperatures are low. When ambient temperatures are high, greater solar heat loads would add an additional cost to thermoregulation. Furthermore, although developing structurally arranged melanosomes within the feather barbule might not be energetically demanding [33], there might be additional costs for iridescent birds associated with greater melanin production [69,70].

Our experimental conditions were artificial and aimed at uncovering physical data, but still have implications for wild animals. First, it could be argued that feather surfaces of museum specimens reached unrealistically high temperatures (up to 83°C) when exposed to radiative heating (1000 W m⁻² under windless conditions). However, these surface temperatures are similar to those measured in live ravens under direct exposure to solar radiation [47]. Additionally, live birds have higher skin temperatures (near body temperature) that can affect the heat transfer through the plumage. Furthermore, birds in the wild may select cooler, shaded microclimates during the hottest time of the day [71] and already low wind speeds can have strong cooling effects [4] reducing the solar heat loads. Second, while using natural sunlight may have made our experiments more realistic, it would also have been extremely challenging to maintain consistent experimental conditions (e.g. cloud cover, temperature, wind speed). The spectral power in the 800–1300 nm range was indeed lower for the light source than for sunlight. However, the average spectral power distribution between 800 and 2000 nm is similar, and the absorption of melanin is not very high between the 800 and 1300 nm region [27] and is, therefore, unlikely to affect the results. But differences in the spectral power distribution may explain why iridescent and unstructured melanin-based plumage did not differ in surface temperature across all body regions despite lower near-infrared reflectance in iridescent feathers. We predict greater surface temperature differences between iridescent and unstructured melanin-based plumage under exposure to natural sunlight, and this can be tested in future work. Third, our experiments were run on museum specimens that were placed horizontally under a heat lamp. This does not fully represent the positioning of live birds towards the sun and limits the comparison of different body parts. Temperature differences between body parts are probably linked to differences in the curvature of the body of the specimens, an effect we controlled for in our analyses by (i) grouping the body parts based on the uniformly coloured reference species, D. fraseri and (ii) by including body region as a fixed factor in our analyses. Other possible explanations for the results on different body parts could be potential differences in feather microstructure, which can be explored in future research.

Here, we tested hypotheses on the effect of radiative heating on differently coloured feather surfaces using an experimental and a modelling approach. This study may lay the foundation for future research that should not only explore the effects of feather surface heating on body temperature in heterothermic sunbirds thereby testing for temperature differences between males and females but should further address the effect of feather positioning on surface heating with a specific focus on angle-dependency in iridescent feathers.

Data accessibility

The data are provided in electronic supplementary material [72].

Authors' contributions

S.R. collected and processed the experimental data, performed optical and heat modelling and wrote the manuscript. A.P. and A.D. performed optical and heat modelling and critically revised the manuscript. M.D.S. critically revised the manuscript and supervised the project. L.D.A. assisted with the data analysis and optical modelling and supervised the project.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by the Research Foundation Flanders (FWO) (grant no. GOG2217N) and the US Air Force Office of Scientific Research (MURI grant no. FA9550-18-1-0477 and MURI grant no. FA9550-18-1-0142) as well as by HFSP (grant no. RGP0047).