Escape Distance in Ground-Nesting Birds Differs with Individual Level of Camouflage

Camouflage is one of the most widespread antipredator strategies in the animal kingdom, yet no animal can match its background perfectly in a complex environment. Therefore, selection should favor individuals that use information on how effective their camouflage is in their immediate habitat when responding to an approaching threat. In a field study of African ground-nesting birds (plovers, coursers, and nightjars), we tested the hypothesis that individuals adaptively modulate their escape behavior in relation to their degree of background matching. We used digital imaging and models of predator vision to quantify differences in color, luminance, and pattern between eggs and their background, as well as the plumage of incubating adult nightjars. We found that plovers and coursers showed greater escape distances when their eggs were a poorer pattern match to the background. Nightjars sit on their eggs until a potential threat is nearby, and, correspondingly, they showed greater escape distances when the pattern and color match of the incubating adult’s plumage—rather than its eggs—was a poorer match to the background. Finally, escape distances were shorter in the middle of the day, suggesting that escape behavior is mediated by both camouflage and thermoregulation.


Introduction
Camouflage is a classic example of evolution through natural selection, and the selective advantage of cryptic phenotypes in avoiding predation has received considerable attention in recent years (Vignieri et al. 2010;Chiao et al. 2011;Zylinski and Johnsen 2011;Troscianko et al. 2013). A widespread camouflage strategy is background matching, whereby an animal closely resembles its surroundings in color, brightness, and pattern (Stevens and Merilaita 2009a). However, in a heterogeneous habitat, an animal's ability to match the background can vary (Merilaita 1999), and it should be adaptive for individuals to monitor their own degree of camouflage and use this information to adjust protective behavior appropriately. For cryptic animals, the degree of individual background matching should influence the decision of when to sit tight or to flee from an approaching predator. Movement has been shown to break the effects of camouflage (Stevens et al. 2011;Hall et al. 2013), presenting a cost to fleeing by revealing the animal's presence and location and suggesting that camouflage and escape behavior are likely to be linked. In situations where animals are caring for vulnerable offspring, the movement of a fleeing parent may betray the location of eggs or immobile young, such that escape decisions might also be influenced by the camouflage of offspring.
In this study, we test the hypothesis that individual animals vary their escape behavior in relation to their degree of background matching and that of their offspring. If escape reveals the location of adults and young, then we should expect adults to escape at greater distances from an approaching predator when either they or their young are poorly camouflaged, since the nest will be under heightened risk of discovery at close range. We tested this prediction across individuals in the wild, using a number of African groundnesting bird species (plovers, coursers, and nightjars). These are an ideal system to test whether camouflage influences escape decisions because nests are typically in open habitats, where they are susceptible to visually hunting predators (Martin 1993), and because in the absence of any nest structure to conceal the eggs, crypsis is the primary defense against predation (Kilner 2006;Šálek and Cepáková 2006). Further-more, previous work on this study system found that the camouflage of adult birds and their offspring is a strong predictor of nest survival (Troscianko et al. 2016). Our aim is to test not whether escape distance and camouflage strategies differ between species with different ecological traits, but rather whether individual differences in camouflage are associated with behavioral variation across our focal species. We thus control for each species' shared ecological traits, such as life history and morphology, which might also influence escape behavior.
We also consider a second cost that parental escape behavior can impose on offspring, whereby reduced attendance to other fitness-related activities can complicate decisions of whether or when to flee (Ydenberg and Dill 1986). In nesting birds, the act of fleeing trades off against the maintenance of optimal developmental temperatures (Webb 1987;Conway and Martin 2000). For the eggs of groundnesting birds, the risk of overheating has been shown to be higher for eggs at a tropical site (where solar radiation is more intense) than at a temperate site and when egg coloration is darker or more maculated (Gómez et al. 2016). Our study species breed in Zambia (167 south of the equator) during the dry season when air temperatures commonly exceed 357C (Harris et al. 2014). Fleeing the nest exposes eggs to potentially harmful ambient temperatures and solar radiation, which are more intense at midday (Mougeot et al. 2014). We therefore predicted that the risk of thermal stress to offspring should exacerbate the costs of fleeing, such that birds should be more reluctant to leave their nests at times of day when thermal costs are highest. In our study system, escape distances corresponded to two different ecological settings: nightjars (Caprimulgiformes) are reported to initiate escape at distances under 10 m (Langley 1984;Jackson 2002), whereas some plover species initiate escape at more than 40 m (Charadriiformes; Blumstein 2006). Although these differences in natural his-tory are likely to be partly phylogenetically determined, they stimulate different predictions about the relationship between camouflage and escape behavior, according to the relative risks to adult and eggs (Lack 1968). For species with long escape distances, such as plovers and coursers, predation risk to adults is low, whereas eggs are exposed to potentially harmful solar radiation as well as visually hunting predators. In this ecological context, we expect strong selection for cryptic egg coloration and for egg rather than adult appearance to be most relevant in modulating escape behavior (Lack 1968). For species that do not readily flee the nest, such as nightjars, predation risk to adults is high if they are detected, since predators reach close proximity before the adult initiates escape. In these species, we expect that the camouflage of the adult's own plumage will be more important than that of their eggs in modulating escape decisions (Lack 1968). These predictions are summarized in figure 1.
Finally, we expect that for all birds, this trade-off between sitting tight and fleeing should be influenced by circadian patterns of solar radiation and ambient temperature. We predict that adults will have shorter escape distances when these environmental risks are most intense (i.e., at midday) and therefore when prolonged exposure is most likely to threaten embryonic viability. The characteristics of the eggs may also influence their thermal properties, with eggs expected to be more sensitive to ambient temperatures when darker (Kilner 2006;Gómez et al. 2016) and smaller (Turner 1985). We therefore expect that adults will have shorter escape distances when their eggs are less bright and are smaller in size, because these qualities should be coupled with a greater risk of overheating. Finally, shorter escape distances may also be expected when backgrounds are darker, since this may cause the surrounding ground to reach greater temperatures faster and present a greater risk that eggs will overheat.  Following previous studies (e.g., Blumstein 2006;Møller 2010), we assessed escape behavior by using an approaching human as a proxy for a predatory threat and measured the distance at which the bird fled its nest. In nesting birds, escape distances are known to be related to degree of concealment through vegetation cover (Klvan et al. 2004;Miller et al. 2013) as well as stage of egg development (Osiejuk and Mickiewicz 2007) and colonial versus solitary breeding (Šálek and Cepáková 2006). However, to our knowledge, this study is the first to investigate escape behavior in relation to directly quantified camouflage, as seen by the visual systems of relevant predators.

Study System
We studied ground-nesting birds within an area of ∼3,100 ha around Musumanene and Semahwa Farms (centered on 16746 0 S, 26754 0 E) and ∼400 ha on Muckleneuk farm (centered on 16739 0 S, 27700 0 E) in the Choma District of southern Zambia. The study was conducted during September-November 2012-2013, corresponding to the late dry season. Monthly averages of daily minimum air temperatures for the Choma region during this period were 13.57-18.07C, and monthly averages of daily maxima were 30.57-32.17C (Harris et al. 2014). The habitat is a mixture of deciduous miombo woodland, grassland, and fallow and active agricultural fields. Nests were principally located by local farm laborers when flushing the birds while walking around farm fields or herding livestock. We studied three plover and two courser species ( (table 1) and lengthy maximum recorded longevities (8-22 years for the two plover and two nightjar species with available data; Hockey et al. 2005). These shared life-history traits suggest that they all are likely to prioritize their own survival over that of their offspring.

Data Collection
Methods followed Troscianko et al. (2016); in brief, once a nest was shown to us by its finder, it was photographed and its coordinates were recorded with a Garmin eTrex 20 global positioning system (GPS). Camouflage was quantified from digital photographs taken with a Nikon D7000 camera converted to full spectrum sensitivity (Advanced Camera Services Limited, Norfolk, United Kingdom) and fitted with a 105-mm Micro-Nikkor lens. Human-visible spectrum photographs were taken through a Baader ultraviolet-infrared (UV-IR) blocking filter (Baader Planetarium, Mammendorf ), and UV photographs were taken with a Baader UV pass filter. Consistency between UV and visible images was maintained by using a custom-made filter holder that facilitated a smooth transition between filters without the need to move the camera. Ambient lighting conditions were controlled for by photographing a 40% Spectralon (Labsphere) gray standard (Stevens et al. 2007;Troscianko et al. 2016) beside nests from 2 m, using identical camera settings (a sequential calibration method; Stevens et al. 2009). Photographs were not taken within 2 h of sunrise or 2 h before sunset and were taken only in direct sunlight with a fixed aperture of f8 (ISO 400) in raw image format. These lighting conditions are representative of the dry season's weather and ensured consistency in lighting between photographing the gray standard and target (adult or eggs). Consequently, photographs were not always taken on the first visit to the nest. Photographs of adult nightjars were taken from 5 m, with the camera angled toward the most visible flank of the adult. A few nightjar individuals and all adult plovers and coursers fled at distances greater than 5 m, such that it was not possible to photograph them. Eggs were photographed under natural lighting conditions from 1.25 m directly above and again in controlled lighting conditions: shaded on a uniform white background, alongside a scale bar and gray standard.
Nests were revisited every other day until hatching or depredation, using binoculars and a GPS to relocate the nest from a distance. On every visit, time of day was recorded in addition to escape distance (when possible). An approaching human (either J. T. Troscianko or J. K. Wilson-Aggarwal) was used as a model predator, a method widely used in studies of escape behavior (e.g., Frid and Dill 2002;Stankowich and Blumstein 2005). At some nests, camera traps were set up to identify the main nest predators (for details, see below and Troscianko et al. 2016). One such predation event involved human predators, further supporting the use of humans as a model predator. Nest visits were not constrained by lighting conditions and so were conducted throughout the day. Nests were checked by one observer, except for the first visit when the nest finder was present and when the nest was photographed. Escape distance was measured from when the incubating adult was seen fleeing the nest; for plovers and coursers, escape distance was measured using GPS, and for nightjars, escape distance was paced out, with distances !1.5 paces measured in foot lengths to the nearest 10 cm (approximating distance in meters, since GPS was not reliable within ∼5 m). Nests were always approached from the same direction at normal walking pace. We did not directly measure egg temperature, since accurately doing so would involve inserting a temperature probe directly into the egg, preventing its development. Other ecological variables affecting egg temperature (such as air temperature, radiance, wind, convection, and conduction) have complex interactions and are difficult to measure in situ without disturbing the nest. Instead, we used time of day as a proxy for thermal risk, since at midday solar radiation is at its most intense because of the sun's elevation.

Image Processing
Camera traps revealed a broad range of diurnal predators, including dichromats (banded mongoose Mungos mungo), trichromats (vervet monkey Chlorocebus pygerythrus and human) and tetrachromats (gray-headed bushshrike Malaconotus blanchoti), all of which consumed the entire clutch (Troscianko et al. 2016). In the absence of data on the visual systems of these particular species, we modeled images to the most phylogenetically relevant predator visual systems; ferret Mustela putorius furo vision (representing banded mongoose), human vision (representing primates), and common peafowl Pavo cristatus vision (representing the violetsensitive gray-headed bushshrike). A companion article found that this approach to quantifying camouflage is biologically realistic, since clutch survival was predicted in this suite of ground-nesting birds (Troscianko et al. 2016).
Before converting images to predator vision, both visible and UV images were linearized, scaled, and aligned (Stevens et al. 2007;Troscianko and Stevens 2015). Predicted cone catch values for each predator visual system were modeled from digital images after a transformation from camera to animal color space, following a widely used mapping technique (Párraga et al. 2002;Stevens et al. 2007;Pike 2010;Troscianko and Stevens 2015). Cone catch images (32 bits/channel) were used for all image processing for each predatory visual system. Target selections were made for adult nightjars, cutting them out using a freehand selection tool. Target metrics were then compared with the surrounds (excluding the in situ eggs) in the same photograph. Egg targets in photographs with controlled lighting conditions were cut out using an egg selection tool (Troscianko 2014) and downsized using bilinear interpolation to match the pixels/millimeter of the nest surrounds.
Luminance (lightness as perceived by a visual system [Osorio and Vorobyev 2005]) distribution differences (luminance diff ) were calculated by comparing absolute differences in counts of the numbers of pixels in each target (egg or adult nightjar plumage) to its background (Troscianko and Stevens 2015). Luminance diff values describe the difference between the target's and background's overall reflectance values in terms of predator vision. In addition, we measured the intrinsic mean luminance of both the target and the background, as well as their intrinsic contrast (by calculating the standard deviation of luminance pixel values following a square root transformation to generate a normal distribution of luminance values). Similar to luminance diff , spatial frequency differences (pattern diff ) were calculated by summing the absolute differences in energy between target and background at different spatial scales (Troscianko and Stevens 2015). Fast Fourier transform bandpass was used with filters at 17 levels, and the energy was determined by the standard deviation of luminance values at each spatial scale. This allowed us to calculate how similar birds/eggs were in terms of marking sizes to those of the substrate, providing a measure of background matching camouflage. Color analysis was based on a widely used model of noise-determined color discrimination (Vorobyev and Osorio 1998), using visual system-specific cone ratios (supplementary materials) and a Weber fraction of 0.05 for generating just noticeable differences (JNDs), whereby a JND of !1 means that two colors should be indiscriminable to an observer. A script was used to determine the dominant colors in an image (up to 32 different colors). Color difference (color diff ) for targets was the mean difference (in JNDs) between the most abundant color of the target and all the colors found in its surrounds, weighted by coverage (Troscianko and Stevens 2015); for more information on how camouflage metrics were calculated, see the supplementary material (appendix, available online) and Troscianko et al. (2016).

Statistical Methods
R (ver. 3.1.0; R Development Core Team 2013) was used to conduct all statistical tests. Potential predictors of var-iation in escape distance were simultaneously tested using linear mixed-effect models implemented using the lme4 package v1-6 (Baayen et al. 2008), fitted with restricted maximum likelihood and a Gaussian error structure. Model simplification was done through backward fitting the fixed effects with Akaike information criterion and log likelihood, facilitated by the function fitLMER from LMERConve-nienceFunctions (ver. 2.5). Data on ferret vision and peafowl vision were not put through the model simplification process and instead were interrogated using the simplified model for human vision, since humans were the relevant approaching threat. The pamer function was used to obtain P values, and the most conservative values were reported. Before model simplification, variables were transformed to meet assumptions of homogeneity of variance and a normal error structure. Time of day was converted to decimal hour and treated as a polynomial since it was used as a proxy for temperature. We analyzed the two orders of birds (Charadriiformes and Caprimulgiformes) in separate models because of different predictions based on their different ecology (Lack 1968). To confirm the differences in escape distances between the different species groups (nightjars, plovers, and coursers), we ran an additional model with data on escape distances from all species; the mcposthoc function was used for testing planned contrasts between variables. For all models, nest identification and visit number were included as random effects; the latter controlled for any habituation effect across multiple visits to the same nest. Species was retained in all models, meaning any remaining effects found were detected across all species. Covariance between model variables was tested for, using a combination of the vcov and cor2cov functions.

Results
Here we report only the results for data from the trichromatic primate vision model. All results for data from ferret and peafowl vision models did not alter the conclusions and are reported in the supplementary material (appendix). All data, including that for the different predator visual systems, are deposited in the Dryad Digital Repository: http://dx.doi .org/10.5061/dryad.3h6r1 (Wilson-Aggarwal et al. 2016). We found that escape distances did not differ between plover and courser species (t 385 p 1:256, P p :210), but nightjar escape distances were significantly shorter than both plover (t 385 p 19:590, P ! :001) and courser (t 385 p 16:663, P ! :001) escape distances.

Plovers and Coursers
The distance at which incubating adults initiated escape was shorter when egg background matching was better with respect to pattern (i.e., pattern diff was lower), and this effect differed with time of day: it declined as midday approached and increased thereafter, with this effect more pronounced when egg pattern diff values were lower ( fig. 2; pattern diff # time of day: F 2, 61 p 4:509, P p :014). Color diff was not retained in the model after simplification, indicating that escape distance was not influenced by degree of color match. Luminance diff was retained in the simplified model but was not found to be significant (F 1, 61 p 3:594, P p :063). Mean egg luminance was not retained in the simplified model; however, it was found to positively cocorrelate with egg contrast (R 2 p 0:760). Escape distances were greater for higher-contrast eggs, and this effect was stronger when eggs were larger (egg contrast # logged mean egg volume: F 1, 61 p 16:634, P ! :001). Because of the collinearity between mean egg luminance and egg contrast, it is not clear which variable is driving this relationship. Last, birds initiated escape at greater distances when background contrast was higher (i.e., when backgrounds had a greater variance in luminance; F 1, 61 p 14:551, P ! :001).

Nightjars
No aspect of egg camouflage predicted escape distance, with no model better than the null. However, there were complex relationships between adult camouflage and escape distance. When adult pattern diff values were low (good pattern match), escape distances did not differ with varying degrees of color diff . However, when pattern diff was high (poor pattern match), escape distances were greater when adult color diff values were higher (color diff # pattern diff : F 1, 270 p 14:441, P ! :001). After simplification, the model did not retain luminance diff , adult mean luminance, adult contrast, or background mean luminance. The distance at which escape was initiated varied according to time of day, depending on background contrast: escape distances were shortest at midday and when background contrasts were high (background contrast # time of day: F 2, 270 p 5:682, P p :004).

Discussion
We investigated whether components of background matching camouflage predicted escape behavior in ground-nesting birds. We found that plovers and coursers initiated escape at greater distances when their eggs were less well camouflaged in terms of pattern, as expected if escape behavior at close quarters would exacerbate the costs of poor camouflage by revealing nest location. This implies that groundnesting birds are able to assess the camouflage of their eggs against their nesting background and respond appropriately. We found that this effect was most pronounced at midday (fig. 2). The strong relationship between escape behavior and time of day is consistent with previous studies demonstrating that birds adjust their incubation behavior according to seasonal and daily variations in environmental temperature (Brown and Downs 2003;Yasué and Dearden 2006;Tieleman et al. 2014). While the circadian correlation we observed could potentially be explained by another factor that covaries with time of day (e.g., predator activity), temperature seems the most likely, given its strong circadian pattern and the extremes it reaches at ground level in our study area during our focal species' breeding season. Conditions at another tropical site have recently been shown to impose greater thermal stress on experimentally placed eggs in ground-nesting species' natural nest sites than those at a temperate site (Gómez et al. 2016). Some of our study species in particular are known to engage in thermoregulatory behavior when incubating (including gular fluttering and wetting their eggs using soaked belly feathers [Hockey et al. 2005]). Taken together, our results suggest that camouflage is able to mitigate not only predation risk but also thermal risks from predator-induced disturbance by permitting adults to shade their eggs for longer when the risk of overheating is highest.
Escape behavior of plovers and coursers also differed according to egg contrast (intrinsic variation in egg lumi-nance) and egg size. However, egg contrast was positively correlated with mean egg luminance, with darker eggs having lower variation in luminance values (i.e., less pronounced patterns), and it is therefore unclear which variable is the driver of this relationship. As expected if darker (Kilner 2006;Gómez et al. 2016) and smaller (Turner 1985) eggs absorb more solar radiation, parents escaped at greater distances when egg contrast/luminance was higher, and this effect was greater for larger eggs. More research is needed to test whether egg contrast/luminance and egg size reflect risk of overheating, since the evidence for the relationship between egg color and heat transfer has been disputed and still requires appropriate quantitative evidence (Ruxton 2012).
For birds such as nightjars that initiate escape only at very close range, we expected to find an association between escape distance and the camouflage of adult plumage rather than that of the eggs (Lack 1968). As predicted, nightjar escape distance showed no association with egg camouflage. Instead, escape distance was associated with the degree of both color and pattern matching between adult plumage and the background. Irrespective of their color match, adults initiated escape at shorter distances when their pattern was a good match to the background. However, when their pattern match was poor, adults with less effective color match initiated escape at greater distances. This suggests that pattern may be the more important cue in determining escape behavior for nightjars but that these birds may also be sensitive to color when making the decision to flee. Background contrast was found to be an important predictor of escape decisions of all species, with escape distances shorter when background contrast was high, and for nightjars this effect was most prominent at midday. It is possible that high-contrast backgrounds reduce detection risk. This is plausible, given that higher-contrast backgrounds are typically more heterogeneous, and predator search times can be longer in these complex habitats (Merilaita 2003;Dimitrova and Merilaita 2009). Longer search times would reflect a lower detection risk from the prey's perspective, and background contrast may therefore be a reliable indicator of when best to flee. These results imply that nesting birds may use absolute properties of their surrounding habitat to modulate escape behavior, in addition to using their degree of background match.
A strength of this study is that it was conducted on a community of wild, free-ranging animals under natural conditions with clear fitness consequences. However, our results are inevitably correlative, and experimental manipulations are needed to confirm the mechanisms underlying the patterns we have uncovered. Ideally, background matching and thermal costs should be experimentally manipulated, but doing so in a biologically realistic way is very challenging and potentially destructive. Such a manipulation would also shed light on how adult birds assess their degree of camouflage. We might speculate that egg camouflage could be directly assessed visually, predicting that escape behavior should respond to experimentally manipulated background matching. Alternatively, camouflage might be indirectly assessed through experience: individuals with poor camouflage may experience more predation attempts and therefore associate those circumstances with the need to initiate escape at greater distances when subsequent breeding attempts are in similar habitats. Such self-assessment of camouflage may also be relevant to other behavioral decisions, such as when to initiate an attack from an ambush predator's perspective.
To our knowledge, this study is the first to show empirically that animals modulate risk-taking behavior according to their direct degree of camouflage, as perceived by relevant visual systems. We also found strong circadian patterns in escape distance, consistent with the hypothesis that ambient temperatures and solar radiation influence escape decisions and suggesting for the first time that this tradeoff is modulated by camouflage. Similarly, we found that escape distances were correlated with habitat and egg char-acteristics that could influence risk of overheating. Future studies should monitor egg temperatures in addition to quantifying camouflage, in order to directly measure the thermal costs of escape behavior and how they vary in a circadian manner. Although our work has focused on background matching camouflage with regards to color, luminance, and pattern, it would be valuable in future studies to consider other potential types of camouflage that are important to concealment, such as disruptive coloration (Cuthill et al. 2005). However, this will be challenging because defining and quantifying disruptive markings on threedimensional objects is problematic (Stevens and Merilaita 2009b). Overall, our results suggest that animals are able to assess their degree of camouflage against the background and use this information to fine-tune behavioral decisions in response to predation risk. Zylinski, S., and S. Johnsen. 2011. Mesopelagic cephalopods switch between transparency and pigmentation to optimize camouflage in the deep. Current Biology 21:1937-1941.