Ethics statement

The study was conducted in accordance with EC Directive 86/609/EEC for animal experiments, and with the current Spanish legislation involving aviation safety. Non-invasive methods were used during experiments. No animals were sacrificed, handled nor sampled and thus, animal ethic evaluation was not required. Field technicians had the required licenses to operate in the frequencies used for this work. All the field procedures were reviewed and accepted by the authority yielding the permit (Doñana National Park authorities, Junta de Andalucía; permit reference: 3236/FQH/mdcg).

Nest inspections

Field work was conducted in Doñana National Park (Spain: 37° 6’ N, 6° 28’ E) at the beginning of May 2014, when tree foliage is consistent with the average situation experienced by kites during the peak of territorial intrusions (trespassing) by non-breeding individuals attempting territory takeovers. By this date, most individuals were in the pre-laying or incubation stage of the breeding cycle (the mean laying date of the population is 15 April, but may range from March to mid-June), the periods with the highest decoration intensity and trespassing rates in the population [10]. A total of 15 nests were randomly selected for inspection within the Park, covering the most representative areas, tree types and range of nest-exposures (from very exposed to hidden in the canopy). Nests with eggs or chicks were excluded from the study to avoid disturbance, except for one nest that contained cold, already abandoned eggs, which were removed for contaminant-analysis before the flights and thus did not interfere with the experiment.

During each UAS flight-session, the target nest was climbed, cleaned of any decoration material if already present and then assessed through UAS-flights twice, i.e. with and without being decorated by us with a white plastic bag. A white plastic bag was chosen because this is by far the most preferred (in relation to other colors and materials) and commonly used item (90% occurrence) by black kites in Doñana to decorate their nests [11]. The tree-climber, experienced with checking hundreds of kite nests in previous years, was instructed to place the white plastic bag so as to simulate a high level of nest decoration, as observed in real kite nests (e.g. see photo in Fig 3b of [11]). Thus, when decorated, the white bag covered 60–80 percent of the nest platform corresponding to a high-level of nest decoration [10]. This value was chosen based on the following reasons: (1) Financial and logistical constraints limited the sample size to a maximum of 15 nests and precluded the possibility to examine how visibility varied along a gradient of decoration intensity. Therefore, we decided to focus on a restricted range of decoration-levels corresponding to individuals that make the active decision to decorate their nest and benefit from it (see below). (2) In this sense, the basic question of the study was: “for a kite that decorates the nest, does this increase its potential visibility to other kites?”. Based on previous experiments, the portion of the population that most clearly, decorate the nest is composed by individuals in prime age [10]. These are the ones that most frequently collected experimental decorations when offered by the researchers [10], and, among those that collected these experimental items, more than half had decoration levels of 60–80%. (3) Individuals that decorate the nest profusely (i.e. at the end-tail of the decoration range) are those that will most benefit by the signal in terms of lowered intrusion rates and fights against challengers, as shown in [10]. Thus, they would be the ones most interested in “designing” a decoration configuration that is visible from as far as possible (otherwise, by the time a conspecific sees the decoration, it may be too late to avoid the aggressive encounters due to the intrusion). (4) Finally, by focusing on high-decoration levels, the analysis of the basic question outlined above was conservative: if nest decoration, as simulated here, does not increase nest visibility and detection, then it would be even less likely to do so for lower levels of decoration.

In addition to the manipulation, each nest was characterized by its dimensions (length and width), measured in cm with a ruler, and “opening angles”, defined as the three angles of unobstructed view of the sky, measured with a compass from the center of the nest-cup outwards: (1) to the right in horizontal, compared to the North; (2) to the left in horizontal, compared to the North; and (3) vertically, compared to the level of the platform. This characterized the potential breadth of visibility of the nest from the outside of the tree both vertically and horizontally.

We performed two flights per nest (one with and one without artificial decoration) using a hexacopter to photograph them (with a Sony Nex5 camera, 16 Megapixels resolution) at different distances and angles (Fig 1 and Figure A in S1 File). All nests were assessed following the same sequential procedure: (1) first, we took one zenithal image (“zenithal snapshot” hereafter), hovering at 50 m above ground level (AGL hereafter). (2) Then, the hexacopter flew at 10 m AGL around the tree, designing a circle of 15 m radius 360° and making a series of eight regularly-spaced photos of the treetop (hereafter “lateral snapshots”). Finally, the hexacopter flew away from the nest and then approached it making a series of photos (hereafter “approaching snapshots”) at distances of 115, 95, 75, 55, 35 and 15 m (at 30, 26, 22, 18, 14 and 10 m AGL, respectively; Fig 1). This sequence of declining distances and heights was chosen to simulate the progressive approach of a trespassing, prospecting black kite that intends to inspect the content of a nest. In Doñana, kites breed at high density in loose aggregations with small inter-nest distances (e.g. [12]), so that, on average, a prospecting trespasser would start to enter the defended area and elicit the aggressive response by the owner between 30–120 m from a nest. Thus, the sequence of distances of the approaching snapshots covered the gradient ranging from approaching the border of a territory from its outside to a full intrusion and mimicked the actual behavior observed during trespassing, although simplified into a straight-line flight.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Illustration of the flight plan used during the experiments.

The hexacopter first ascended to 50 m above ground level and took one zenithal image (zenithal snapshot), then flew at 15 m radius 360° around the tree and at 10 m AGL, taking regularly photos (lateral snapshots). Finally, the hexacopter approached the nest making a series of photos (approaching snapshots) at distances of 115, 95, 75, 55, 35, 15 m (30, 26, 22, 18, 14, 10 m AGL, respectively), simulating the aerial perspective of an approaching black kite that intends to inspect the nest.

https://doi.org/10.1371/journal.pone.0157440.g001

We randomly alternated the sequence of decoration treatments within nests to minimize the potential effect of light conditions on the detectability of the nests. All flights were performed in daylight between 7:30 and 21:00h (local time) by a professional team of operators and field coordinators from Flying-Cam (http://www.flying-cam.com). Each nest-assessment required around 1.5 hours of overall work (approximately 15–20 min to climb the tree and measure the nest characteristics, 45–50 min of active flight and 20–30 min of material download, assemblage, testing and reload on vehicles) to complete all recordings (of the decorated and non-decorated treatments), while moving from one nest to the next required ~ 20–60 min. Flights were performed on autonomous mode guided by the on-board autopilot that followed a GPS waypoints-trajectory (pre-programmed in the previous days based on the GPS position of the nest location), but keeping the capability for the human pilot to take control of the aircraft in the eventuality of danger. The take-off and landing were performed in manual mode. A second operator using real time video from the UAS adjusted the gimbal and remotely controlled the onboard camera to perform the snapshots of the nests. All the images were associated to their GPS coordinates and barometric altitude extracted from the flight tracks.

Preparation of nest images for visibility trials

We took 662 images from the 15 nests inspected with and without decoration (Figs 1 and 2). Preliminary trials by DC and MM showed that finding nests in the pictures taken at distances of 95 and 115m (26 and 30m AGL, respectively) during the approaching flights had a very low detection rate and thus, those images were removed for further analyses. Before performing the definitive trials, all images were revised and, if necessary, treated with Photoshop CS6 (Adobe, San Jose, CA, USA) to remove features potentially facilitating the detection of nest location (for example, the car in the nest proximity).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Images taken by the hexacopter of black kite nests experimentally decorated by the researchers (left side) and of the same nest without decoration (“self-control nest”, right side).

The experimental decoration covered ca 80% of the nest platform, corresponding to a high-level of decoration in real nests, typical of birds in prime age (see [10]). Black arrows indicate the position of the nests. The images have been cropped and enlarged for illustrative purposes.

https://doi.org/10.1371/journal.pone.0157440.g002

To estimate the detectability of black kite nests to trespassers, we conducted trials of nest detectability using 25 volunteers as "experimental conspecifics”. The assumption behind such approach is that, if decoration makes a nest more visible to a human observer, it will make it even more visible to the keener vision of a raptorial bird [13]. To make the assessment by volunteers more rapid and standardized, we developed a Java program that projected the images on a laptop computer in a predetermined order (see below) and asked each observer to try to detect one nest per image and click on its presumed location. The program allowed 15 seconds to the observer for locating the nest, recorded the latency to locate it and assessed if the “clicked area” matched the real nest location. Each time an observer clicked on an image, the program switched to the next one, regardless of whether the volunteer succeeded or not in correctly detecting the nest and without feedback about their success in locating the nests. Also, if the observer did not click on the image within 15 seconds, the nest was recorded as non-detected and the program switched to the next photograph. Previous to the trials, we conducted a training session with 10 pictures that portrayed real kite nests and provided feedback to the participants, so as to accustom them to the general appearance of the target object of their search and standardize as much as possible the search image across participants. The images shown during the training session (15 seconds per image) were the same for all the participants and none of them were used in the subsequent trials. The observers had no previous knowledge of the aims of the study and none participated more than once in the experiment.

Each observer visualized 150 images of the 15 experimental nests divided in two sets. In the first set, consisting of 60 images, the observer was exposed to: (i) a pair of zenithal images of each nest (with and without decoration) taken at 50 m AGL, and (ii) a pair of lateral images of each nest (with and without decoration) taken circularly around the nest at 15 m radius and 10 m AGL. In the second set of images, formed by a total of 90 photographs, the volunteers visualized three pairs of approaching snapshots per nest (for the 15 decorated and 15 non-decorated nests), ordered from farther to closer (75 m, 55 m and 35 m distance), thus simulating the progressive approach of a trespassing black kite that intends to inspect the content of the nest.

Cumulatively, each volunteer dedicated approximately 25 minutes to visualize all the images. The images projected and their relative order varied between trials as, for example, there was more than one lateral image available for the 10 m AGL snapshot. To minimize memory-biases (e.g. nest location and characteristics of the surrounding landscape), the pictures from the same nest (regardless of its decoration treatment) were separated from each other by at least four photographs. Similarly, in the case of the approaching flights, the series of approaching snapshots for the same nest (i.e. from the decorated and non-decorated treatment) were separated from each other by at least four series of photographs. Finally, we randomized the position of the nest within the images by cropping each photograph to the size of 4175 x 2774 pixels (a 15% of reduction from the initial resolution of 4912 x 3264 pixels) in Photoshop CS6. In this way, we reduced the habitat similarity between images of the same nest. Based on pre-trials, we are confident that such procedures prevented any substantial effect of memory-biases, which would be homogenous across treatments and observers anyway. All trials were conducted using the same screen (HP ZR22w, 22 inches, 1920 x 1080 resolution) positioned at a distance of 50 cm from the observer.

Statistical analyses

We used generalized linear mixed models (GLMM) to test whether (i) the probability of nest detection (0/1; binomial distribution, logit link function) and (ii) the latency to nest detection (time elapsed between initial image projection and nest detection, more details below; Gaussian distribution) was related to the decoration treatment (decorated vs non-decorated), the distance of the hexacopter to the nest and/or the interaction between nest decoration and distance. “Nest identity” and the “identity of the volunteers” were included in all the models as random factors. In the models on latency, we only included images in which the nests were detected by the participant in both treatments. Previously to running these models, latency to nest detection (max = 15 seconds) was log-transformed to satisfy the assumptions of normality. We built separate models to analyse zenithal, lateral and approaching snapshots. Additionally, in the case of the approaching snapshots, we ran models including only the first ‘correct detection’ of each nest (i.e. 2 images per nest, one per treatment) instead of the 3 consecutive images per treatment. For example, if Volunteer A detected nest X in the 75 m distance photo, then all subsequent detections of nest X by A (regardless of the decoration treatment) were not included in the analyses. Results remained unchanged (see Table A in S1 File) for both the probability of nest detection and latency to nest detection.

Selection of the minimum adequate models was done by sequentially dropping non-significant terms from fully saturated models (containing all main effects and interactions) in a hierarchical way, starting with the least significant terms. We performed likelihood ratio tests to confirm whether the inclusion of a predictor was significantly informative. In these tests, the full model, including the focal predictor, is compared with its restricted counterpart without the same predictor, and the significance of the predictor is obtained by a chi-square distribution.

We systematically performed model diagnostics statistics while modelling to avoid misleading conclusions based on statistical artifacts. To this end, we checked assumptions about the distribution of residuals through diagnostics plots and examined collinearity. These analyses did not show any obvious deviation from GLMM assumptions or any collinearity problems. Statistical analyses were implemented in R 3.1.2 (R Development Core Team 2015) with the package lmerTest [14].

Potential limitations of the study

Although the use of humans observers is a popular approach in evolutionary studies to measure crypticity or conspicuousness (e.g. [22–24]), we recognize that the use of humans to rate pictures is an oversimplification of field conditions. However, it offers significant advantages by permitting image manipulation and controlled conditions, in terms of time, distance, and perspective, to assess detection of both decorated and undecorated nests. Furthermore, some experiments on crypsis have reported similar detection rates between avian and human predators (e.g. [22,24–26]). Humans and birds, including black kites, share a visually oriented sensory system and an excellent color and detail discrimination. If anything, diurnal raptors possess a higher visual acumen [27,28]. A main concern using human observers as black kite surrogates is that humans, unlike birds, have no ultra-violet (UV) sensitivity [29,30]. In the context of the present study, the use of humans to assess nest visibility may have led to underestimate the detection capacity of actual kites due to the UV component of the white decorations. Thus, given such potential differences in visual acuity and UV-perception, our results are likely to represent a conservative estimate of the detection advantage offered by nest ornamentation.

A second, potential limitation of the study was represented by the fact that the experimental material used to decorate the nests in our experiments was a standardized, new plastic bag. In our experience, kites tend to place new, shiny materials in the nest when first decorating it. For example, many kites readily collected new pieces of white plastic when these were experimentally offered in a previous study [10]. However, these materials will unavoidably become dirtier and less bright with time through physical wear, exposure to elements and contact with the nest contents. Thus, even if kites regularly refill their nests (F. Sergio, pers. obs.), our manipulation may have included brighter materials than experienced by an average intruder. In this sense, our measures may reflect the “ideal” decoration-decision by a kite at the moment of placing the item, later altered by environmental constraints.

Finally, we imposed high levels of nest decoration during the flights, so as to ensure an over-focus on individuals that intend to signal and most benefit from it. Due to financial and logistical constraints, we were unable to examine the level of visibility-enhancement offered by intermediate to low levels of decoration, associated with individuals that may strike a balance between an attempt to threaten conspecifics and to pass sufficiently unnoticed to avoid excessive intruder pressure.

In contrast with the above limitations, note that, given the proverbially extreme latero-vertical maneuverability of kite flight, a real kite intruder would be much more adept at finding the distance, flight altitude and angle that maximizes the visibility of an observed nest than our simplified, pre-planned, straight trajectories. This likely made our assessment unavoidably highly conservative and could have amply counter-balanced our over-decoration of kite nests. Thus, if on one hand we could have over-decorated our nests, on the other hand, we used a simulated intruder with very limited capacity of opportunistic aerial relocation for nest visibility-maximization from a distance. Because of the above uncontrollable factors, difficult to incorporate in an assessment of this kind, this study should be best seen as a clear, unavoidable simplification of an ideal situation, focused on physical signal transmission and only considered in conjunction and through its congruence with previous analyses of the same communication system [10].