Fluffy feathers: how neoptile feathers contribute to camouflage in precocial chicks

Camouflage is a widespread strategy to increase survival. The plumage of precocial chicks often contains elements of disruptive colouration and background matching to enhance concealment. Chick plumage also features fringed feathers as appendages that may contribute to camouflage. Here, we examine whether and how neoptile feathers conceal the outline of chicks. We first conducted a digital experiment to test two potential mechanisms for outline diffusion through appendages: 1) edge intensity reduction and 2) luminance transition. Local Edge Intensity Analysis (LEIA) showed that appendages decreased edge intensity and a mean luminance comparison revealed that the appendages created an intermediate transition zone to conceal the object’s outline. The outline was most diffused through an intermediate number of interspersed thin appendages. Increased appendage thickness resulted in fewer appendages improving camouflage, whereas increased transparency required more appendages for best concealment. For edge intensity, the outline diffusion was strongest for a vision system with low spatial acuity, which is characteristic of many mammalian predators. We then analysed photographs of young snowy plover (Charadrius nivosus) chicks to examine whether neoptile feathers increase outline concealment in a natural setting. Consistent with better camouflage, the outline of digitally cropped chicks with protruding feathers showed lower edge intensities than the outline of chicks cropped without those feathers. However, the observed mean luminance changes were not consistent with better concealment. Taken together, our results suggest that thin skin appendages such as neoptile feathers improve camouflage. As skin appendages are widespread, this mechanism may apply to a large variety of organisms.


33
Avoiding detection either for protection from predators or to go unnoticed by potential prey is 34 essential for individual survival. The threat of predation has led to the evolution of various camouflage 35 mechanisms, which make potential prey more difficult to detect or recognize. The most prominent 36 mechanism is visual camouflage that includes highly adaptive colouration strategies among animals 37 (Stevens and Merilaita 2009). One strategy to achieve visual camouflage is background matching (also 38 termed "crypsis" by Endler (1981)). For background matching, animals try to match colour, luminance 39 and pattern of their background. 40 While background matching is one of the most common and frequently studied strategies of visual 41 camouflage (Cott 1940, Endler 1981 important mechanism is concealing the outline of the body. Thayer (1909) proposed that detecting the 43 outline of their prey is one of the ways predators locate and identify their prey. In general, the 44 detection of edges is an essential task for object recognition (Marr 1976, Tovée 1996. In this regard, 45 disruptive colouration makes animals less detectable. It involves a set of markings that creates false 46 edges within the animal hindering the detection or recognition of its true outline and shape or parts 47 of it (Thayer 1909, Cott 1940, Stevens and Merilaita 2009). Cott (1940) suggested that structural 48 modifications of the organism's outline themselves could contribute to camouflage by creating an 49 'irregular marginal form'. This makes the animal's true body outline effectively diffused and hence 50 makes it harder to detect (Cott 1940). Recently, support for the 'irregular form' hypothesis was found 51 in an experimental study showing that false holes markings reduce avian predations (Costello et al. 52 2020). 53 Birds with their typically advanced vision and high plumage diversity have been featured prominently 54 in camouflage research, either as predators or as prey (Cuthill et  We did not account for differences in colour vision between different predators as the setup mostly 131 consists of greyscale images that predominantly differ in luminance. Note that in many animals, visual 132 acuity is greater for achromatic than chromatic stimuli (Giurfa et al. 1997, Endler et al. 2018. 133  Calibration (MICA) toolbox (Troscianko and Stevens 2015) for ImageJ version 1.52a (Schneider et al. 143 2012). We converted the generated images into multispectral images containing the red, green and 144 blue channel in a stack and transformed them further into 32-bits/channel cone-catch images based 145 on the human visual system, which are required by the framework. To create the luminance channel, 146 6 we averaged the long and medium wave channel, which is thought to be representative of human 147 vision (Livingstone and Hubel 1988) ( Figure S1a). We modelled the spatial acuity with Gaussian Acuity 148 Control at a viewing distance of 1300 mm and a minimum resolvable angle (MRA) of 0.01389 ( Figure  149 S1b). To increase biological accuracy, we applied a Receptor Noise Limited (RNL) filter that reduces 150 noise and reconstructs edges in the image. The RNL filter used the Weber fractions "Human 0.05" 151 provided by the framework (longwave 0.05, mediumwave 0.07071, shortwave 0.1657), luminance 0.1, 152 5 iterations, a radius of 5 pixels and a falloff of 3 pixels ( Figure S1c We ran LEIA on the chosen region of interest (ROI) with the same Weber fractions used for the RNL 165 filter. The ROI was the contour region, a 180 pixel-wide band that included the area of the appendages 166 extended by 30 pixels towards the object inside and towards the outside (Figure 1a). We log-167 transformed the S values as recommended for natural scenes (Troscianko and van den Berg 2020) to 168 make the results comparable to the natural background images used in Experiment 2 (see below). To 169 test whether the size of the ROI affected our results, we ran an additional analysis using a 1500 x 1500 170 pixel-wide rectangle surrounding the object as the ROI, which included a bigger area of the background 171 and the full object inside ( Figure S2a). 172 We extracted the luminance S values from the four slices of the output image stack in ImageJ and 173 stored them in separate matrices for further analysis using R version 3.5.3 (R Core Team 2019). ImageJ 174 generally assigned values outside the chosen ROI to zero. Thus, we first discarded all values of zero. 175 We then set all negative values that arose as artefacts in areas without any edges to zero, in order to 176 make them biologically meaningful. We then identified the parallel maximum (R function pmax ()) of 177 the four interrelated direction matrices and transferred this value to a new matrix.

7
High luminance and colour contrasts imply high conspicuousness (Endler et al. 2018). Consequently, a 179 lower luminance contrast leads to lower conspicuousness and therefore, better camouflage. As the 180 outline is an important cue for predators locating and identifying a prey item (Thayer 1909), we 181 assumed that especially low contrasts in the outline of an object improve camouflage. Thus, a 182 reduction of edge intensity in the object outline by the appendages indicates a camouflage 183 improvement. To test whether the object outline became less detectable we compared the edge 184 intensity of the outline pixels in the basic scenario without appendages (Table S1,  In the luminance channel of each image, we measured the mean luminance in the three regions and 199 compared them subsequently. Luminance values range from 0 to 1. 200 According to background matching, objects that differ more in luminance from the background are 201 more conspicuous and hence less well camouflaged (Endler 1981). We assumed that detectability 202 based on possible luminance differences between object and background are weakened by the 203 appendages as they form a transition zone helping to blend the object better into the background. 204 Accordingly, from a camouflage perspective, the appendage region would provide an optimal 205 transition zone when its mean luminance is exactly the mean of the object and background region's 206 luminance. 207 Once the first chick had been found, the second observer joined the predator and took chick 224 photographs. We used a Nikon D7000 camera converted to full spectrum including the UV range (Optic 225 Makario GmbH, Germany) and a Nikkor macro 105 mm lens that allows transmission of light at low 226 wavebands. The equipment was chosen because calibration data were available for this combination 227 (Troscianko and Stevens 2015). Each hiding background was photographed with and without the chick 228 using a UV pass filter for the UV spectrum and a UV/IR blocking filter ("IR -Neutralisationsfilter NG", 229 Optic Makario GmbH, Germany) for the visible spectrum. The camera was set to an aperture of f/8, 230 ISO 400 and the pictures were stored in "RAW" file format. We used exposure bracketing to produce 231 three images to ensure that at least one picture was not over or underexposed. A 25 % reflectance 232 standard (Zenith Polymer TM) placed in the corner of each picture enabled a subsequent standardizing 233 of light conditions. 234 In total, we took pictures of 32 chicks from 15 families. For 21 chicks we obtained photographs suitable 235 for further analyses with an unobstructed view to the entire chick and only one chick per photograph. 236 Of these, we randomly selected pictures of 15 chicks. Unfortunately, it was not possible to obtain 237 proper alignment of visual and UV pictures in ImageJ as either chick or camera moved slightly in the 238 break between changing filters for the two settings. Therefore, we restricted our analyses to human 239 colour vision and discarded the UV pictures for further analysis. 240 In each picture, we manually selected the chick outline and the feather-boundary as a basis for the 241 ROIs (Figure 4a-c). The chick outline included bill, legs, rings and all areas densely covered by feathers 242 without background shining through. We then marked the feather-boundary, i.e., the smoothened line 243 created by the protruding neoptile feather tips. In the next step, we transferred images of chicks with 244 or without protruding feathers, i.e. cropped at feather-boundary or chick outline, respectively, and 9 inserted them into a uniform or the natural background. First, we cropped the chick without protruding 246 feathers and transferred it into a uniform black background. Second, we cropped the chick including 247 all feathers and inserted it into exactly the same hiding spot on the picture of the natural background 248 ( Figure 4b). Third, we cropped the chick excluding the protruding feathers and transferred it into the 249 natural background (Figure 4c). 250

251
We then proceeded with LEIA following the protocol of experiment 1 with the following changes. 252 Again, the selected ROI was the contour region ranging from the chick outline extended by 30 pixels 253 towards the chick inside to the feather-boundary extended by 30 pixels towards the outside. We 254 excluded all areas of the ROI that showed a shadow of the chick as the chicks' shadow was missing on 255 the empty natural background images to which the cropped chicks were transferred to (Figure 4a-c). 256 We used the images of the cropped chicks on the black background to determine the threshold of the 257 HEI pixels according to the protocol of experiment 1 for each chick separately. For each cropped chick 258 that was transferred to the picture with the natural background, we compared the mean edge intensity 259 of the HEI pixels provided by LEIA with and without protruding feathers (Figure 4b-c) using a two-sided 260 paired t-test. 261

262
We also calculated mean luminance differences for each chick using the same cropped photographs 263 as for the LEIA. Similar to the artificial object experiment, the chick region included everything inside 264 the chick outline, the background region included everything outside the feather-boundary up to a 265 1500x1500 pixel-wide rectangle surrounding the chick and the feather region (FR) was between chick 266 outline and feather-boundary. Note that the FR is different from the contour region, which additionally 267 includes a small part of chick and background region. We reduced the FR by excluding all areas that 268 were shaded by the chick since this shadow was missing on the empty background images. 269 Additionally, we excluded the buffer zone ( Figure 3a, the area between the coloured regions) to cover 270 the whole variation in feather density in the FR (Figure 5a-b). Close to the chick outline, the feathers 271 were still relatively dense thinning more and more towards the feather-boundary as they were very 272 variable in length. 273 For each chick, we measured the mean luminance of all three regions in the luminance channel of the 274 image containing the chick without feathers (Figure 5a). The FR we measured again in the image 275 containing the chick with feathers (Figure 5b). 276 In theory, the best transition zone between chick and background that reduces the outline of the chick 277 against the background the most should have an exactly intermediate luminance between chick and 278 background region. In a first analysis, we checked whether the absolute distance of mean luminance 279 of the FR with feathers was closer to those optimal values than without feathers. Because the 280 luminance data were not normally distributed according to the Shapiro-Wilk normality test we 281 conducted a Wilcoxon paired signed rank test. To compare the data graphically in an intuitive way, we 282 transformed the values so that the chick region always was the reference with a value of 0, the 283 background region became 1. The two values measured in the FR stayed in their initial relative distance 284 to chick and background value. 285 The FR generally was quite narrow compared to chick and background region and its effect probably 286 acts predominantly from close proximity. Therefore, we focussed the next analysis only on chick and 287 FR. We assumed that the chick to a certain extent differs in luminance from its immediate background 288 in the FR and that including the feathers decreases this difference and thus possibly improves the 289 camouflage. Therefore, we compared the absolute distances between the mean luminance of chick 290 region and FR with and without feathers. As the data were normally distributed according to the 291

Shapiro-Wilk normality test we conducted a two-sided paired t-test. 292
For an easier comparison of the measurements, we transformed the luminance values in this analysis. 293 The chick region again was the reference with a value of 0. As the background region was excluded, 294 we scaled the FR without feathers to 1. The value measured in the FR with feathers stayed in its initial 295 relative distance to the other two values. 296 The analysis aimed to check if the FR meets the basic requirement of a transition zone having 297 intermediate luminance. Thus, we checked whether the mean luminance value of the FR with feathers 298 fell between the one of chick region (mean luminance = 0) and FR without feathers (mean 299 luminance = 1) constituting the immediate surrounding background to account for the local scale. We 300 calculated the probability for the FR with feathers of having a value between 0 and 1 when randomly 301 distributed. For this, we drew a random sample (n = 10,000) from a normal distribution with the mean 302 and standard deviation in the transformed data. Then, we ran an exact binomial test to determine 303 whether the observed intermediate luminance value was different from the expected value. 304

305
Experiment 1: Artificial object 306 Local Edge Intensity Analysis 307 All images showed multimodal density distributions of pixels (Figure 2a). Pixels showing the highest 308 edge intensities were found at the object outline. These HEI pixels showed prominent modal peaks in 309 all multimodal density distributions (Figure 2a). For the object without appendages, 1.59% of pixels 310 made up the distinct modal area with a mean edge intensity of 2.7 (Figure 2a, '0'). Consequently, we 311 used a threshold of 1.59% to define HEI pixels for all images. Adding appendages reduced the mean 312 edge intensities of the HEI pixels with the lowest mean edge intensity reached in the image with 256 313 appendages (Figure 2a-b). 314

Appendage characteristics 315
Increasing appendage thickness (Scenario 1) resulted in overall higher mean edge intensities 316 suggesting higher detectability than in the basic scenario. With thicker appendages, the lowest mean 317 edge intensity of the HEI pixels was reached already with 128 appendages. Images with more than 128 318 appendages had higher mean edge intensity values implying a deterioration of camouflage (Figure 2b). 319 Increasing appendage transparency (Scenario 2) yielded overall slightly higher mean edge intensities 320 than observed in the basic scenario. The lowest mean edge intensities were reached with more 321 appendages than in the basic scenario (Figure 2c  Background complexity and spatial acuity 329 Introducing background complexity (Scenario 4) resulted in similar mean edge intensities of the HEI 330 pixels for 256 appendages as in the basic scenario for large squares. The ROI on the background with 331 small squares showed slightly higher mean edge intensities for the HEI pixels than for the background 332 with large squares. More appendages did not lead to such a pronounced increase of mean edge 333 intensities as in the basic scenario (Figure 2e). Lowering the spatial acuity of the perceiver (Scenario 5) 334 decreased the mean edge intensity severely. At a spatial acuity of 10 cpd, the minimum mean edge 335 intensity of the HEI pixels in the image with 256 appendages was only half of the value obtained in the 336 basic scenario (Figure 2f). 337

ROI Size 338
Changing the ROI size and examining a larger part of background and object ( Figure S2a  For eight of the 15 analysed chicks, the empty background image was slightly shifted because of a 387 camera movement. Therefore, we corrected their position manually to place the chicks exactly at the 388 same spot in the empty background. 389 After removing the areas of the ROIs where the chick shaded the background, we were able to analyse 390 on average 72 % of the contour region with LEIA. Across the ROIs of the 15 chicks, the mean threshold 391 for the HEI pixels was 0.9826 (Table S3). Consequently, we compared on average 1.74 % of the pixels 392 between photographs of cropped chicks with and without the protruding neoptile feathers. 393 For 13 of 15 chicks (87 %), the mean edge intensities of HEI pixels were lower for the cropped image 394 of each chick with protruding neoptile feathers (e.g. Figure 4b) than for the corresponding images 395 without protruding neoptile feathers (e.g. Figure 4c). Accordingly, images including the protruding 396 feathers showed lower mean edge intensities of HEI pixels than those excluding them (Figure 4d, 397 paired t-test: t = 4.365, df = 14, p-value < 0.001). The mean edge intensity difference of HEI pixels 398 between measurements with and without feathers was 0.178 (95 %CI: 0.091, 0.265). 399 amphibians (Rauhaus et al. 2012) and reptiles (Buxton 1923). A striking example is provided by many 511 insect larvae such as hairy caterpillars which, as chicks, have typically reduced mobility in comparison 512 with the adult form. Birds have a strong influence on caterpillar mortality (Campbell and Sloan 1977), 513 but hairy caterpillars are less preferred prey for avian predators than non-hairy caterpillars (Whelan et 514 al. 1989). We suggest that concealing the outline might be one currently underappreciated function of 515 hairy appendages contributing to improved camouflage. 516

Conclusion 517
The 'irregular marginal form' as a camouflage strategy has inspired early researchers on camouflage 518 (Cott 1940) but evidence for this mechanism so far has been limited. Our results suggest that body 519 appendages such as feathers or hairs can help to create an 'irregular marginal form' that serves to 520 diffuse the object outline. Appendages with the characteristics of protruding neoptile feathers reduced 521 the edge intensity in a proof of principle analysis and on images of precocial chicks taken in their 522 natural environment. Appendages also served to reduce mean luminance differences when both 523 object and background were uniformly coloured but this mechanism failed to contribute to outline 524 diffusion when we analysed images of chicks in their natural backgrounds. Author contributions -VAR, CK and DM conceptualised the study. TV carried out field work. 539 VAR and CK analysed, interpreted the data and wrote the manuscript. All authors revised the 540 manuscript. 541 Permits -Fieldwork permits to collect the data were granted by the Secretaría de Medio 542 Ambiente y Recursos Naturales (SEMARNAT). All field activities were performed in accordance 543 with the approved ethical guidelines outlined by SEMARNAT. 544