Effectiveness of movement-based animal signals is a function of display structure and habitat characteristics: simulations of Australian dragons

Habitat-specic characteristics can affect signal transmission such that different habitats dictate the optimal signal. One way to examine how the environment inuences signals is by comparing changes in signal ecacy in different habitats. Examinations of signal ecacy between different habitats has helped to explain signal divergence/convergence between populations and species utilising acoustic and colour signals. Although previous research has provided evidence for local adaptations and signal divergence in many species of lizards, comparative studies in movement-based signals are rare due to technical diculties in quantifying movements in nature and ethical restrictions in translocating animals between habitats. We demonstrate herein that these issues can be addressed using 3D animations, and compared the relative performance of the displays of four Australian lizard species in the habitats of each species under varying environmental conditions. Our simulations show that habitats differentially affect signal performance, and an interaction between display and habitat structure. Interestingly, the signal adapted to the noisier environment did not show an advantage in signal ecacy, but the noisy habitat was detrimental to the performance of all displays. Our study is one of the rst studies for movement-based signals that directly compares signal performance in multiple habitats, and our approach has laid the foundation for future investigations in motion ecology that have been intractable to conventional research methods.


Introduction
Animal signals evolve to be effective in the environment in which they are emitted [1][2][3], and therefore the diversity we see in animal communication systems is partially driven by the environment in which the animal lives [4][5]. An effective signal requires reliable detection by an intended receiver following transmission through ecologically complex environments [1]. It is evident that environmental structures and habitat-speci c characteristics can affect signal transmission [1,4,[6][7], such that different habitats dictate the optimal signal for reliable detection [8]. Dynamically changing abiotic and biotic noise in the environment also interferes with effective communication between individuals, and animals are required to produce signals that could compete with these irrelevant sensory inputs [3,[9][10]. Many animals modify their signalling behaviour in response to changing conditions [11][12][13][14][15], such that these moment-tomoment adjustments to signalling strategies represent a form of behavioural plasticity to maintain signal e cacy [16][17]. Clearly, habitat-speci c transmission properties are important in determining the optimal structure of a signal, and it is therefore imperative that we quantify signal e cacy in the context of the relevant noise landscape to understand in more detail the evolution of animal communication strategies.
One way to examine how the environment in uences signals is by comparing changes in signal e cacy in different habitats. Earlier research into quantifying acoustic characteristics of different habitat types provided the foundation for understanding diversity in the physical structure of avian songs [6,18]. Crosshabitat examination of signal e cacy between different habitats also helped to explain intraspeci c divergence of calling behaviours in anuran species [19]. Similarly, by contrasting signals in different habitats, signal divergence was described for lizard populations utilising colour-based signals [20], as well as closely related insect species utilising seismic communication strategies [21]. Structural differences in movement-based signals have also been observed in many lizard species occupying both similar and distinctive habitats [17,[22][23][24]. Bloch and Irschick [25] found temporal differences in the display sequences of the green anole, Anolis carolinensis, suggesting population divergence due to population density and habitat use. Similarly, structural differences in the core display of the Jacky dragon, Amphibolurus muricatus, between three different populations might be a consequence of behavioural plasticity in response to variation in habitat structure [26]. Clearly, movement-based signals also show variation consistent with local adaptation explanations, but detailed cross-habitat comparisons of signal structure has rarely been demonstrated. Aside from the technical di culties associated with quantifying the environmental conditions and animal displays [27], there are important legislative and ethical restrictions that prevent translocating animals between habitats that they do not naturally inhabit.
Plant movements are the primary source of motion noise that affects movement-based signal communication [14,[28][29][30][31]. However, the motion noise landscape is di cult to examine in a controlled, systematic way, as it differs vastly from one microhabitat to another as a result of the interaction between variable plant characteristics, habitat topography and microhabitat structure [32]. To our knowledge, the recent work by Ramos and Peters [33][34] represents the only attempts to consider the relative effectiveness of movement-based signals of lizards in multiple habitats. In the rst of these studies, evidence is presented for local adaptation, showing that two sympatric lizard species who utilise different movements in their displays, nonetheless produced similar motion speeds [33]. In the second paper, the authors showed structural differences between allopatric populations of the same species [34].
In both studies, habitat structure was considered to be the guiding force leading to convergence or divergence in structure respectively. The approach taken in both studies, however, was to lm displays in local environments, and then the movement of plants at a separate time, in a controlled manner, and to then contrast resultant motion speed data [35]. Although these are novel and informative insights, we have developed new techniques based around sophisticated three-dimensional (3D) animations that affords us the opportunity to consider movement-based signals embedded in noise [27].
The aim of the present study was to consider directly the relative effectiveness of the movement-based displays of multiple lizard species in multiple habitats using 3D animation. We reused the subject of earlier work [27,36], the Jacky lizard, A. muricatus, and created three additional model lizards and their respective habitats: the long-nosed dragon, Gowidon longirostris, the Mallee dragon, Ctenophorus fordi, and the tawny dragon, C. decresii ( Figure S1,S2). In addition to quite distinct signals, these four species also occupy different habitat types, from densely vegetated coastal heath environments to sparsely vegetated rocky outcrops and semi-arid Mallee woodland with spini ex understories ( Figure S2). The associated noise landscape in these habitats also varies as a result of vegetation density and plant species present. Signals adapted to a noisier environment tend to perform better when placed in a less noisy environment [33][34]. Therefore, we predicted that the signal of A. muricatus would perform better in the other three habitats as it occupies a densely vegetated and noisy environment. Similarly, we would also expect the habitat of A. muricatus to have the greatest in uence on signal performance. The habitats of G. longirostris and C. decresii are very similar, so we predicted that these will in uence signal performance of all species in a similar manner, and that their displays would perform equivalently. Ctenophorus fordi, on the other hand, would produce relatively weak signals as this species utilises the simplest motor pattern compared to the other three species.
Habitat structure is a crucial determinant of signal structure across species and modalities and here we extend this literature to include the movement-displays of lizards from different habitats. The diverse habitats are each inhabited by only one of the focal species, and experimental translocation of absent species would not be permissible. While we have simulated these circumstances, 3D animation technology is su ciently advanced that these are excellent proxies for lming the same circumstances in nature (if it were possible). Our animations are detailed and founded on the natural systems they represent, while our analytical approach is quantitative but descriptive, and shows clearly that habitat structure is an important contributing factor for movement-based signalling systems.

Study species
Amphibolurus muricatus (snout-vent length, SVL, 121 mm) is a well-known species that has featured in many physiology, ecology and visual communication studies [14,[37][38][39][40]. The display of A. muricatus begins with tail-icking, followed by limb movements and a push-up, sometimes accompanied by throat extensions [41][42] (Fig. 1a). There are two parts to the tail-ick sequence: an initial intermittent movement at the tip of the tail followed by a continuous and more rigorous movement of the whole tail. The tail-icks serve as an introductory component to the rest of the display and are used to attract the attention of the intruder [31,43]. Amphibolurus muricatus inhabits woodlands and coastal heath of southeast Australia and are often seen perched on fallen timber [44]. The simulated A. muricatus habitat was based on sites around the Croajingolong National Park, Victoria, where the habitat is densely vegetated with a complex understory of tall grasses and shrubs ( Figure S3a).
In contrast to A. muricatus, the displays of Ctenophorus fordi, are much reduced in complexity, and almost exclusively comprised of head bobbing (Fig. 3c). Ctenophorus fordi are found in south-east Western Australia to western Victoria and New South Wales [44]. The species is smaller in size (SVL 58 mm) and lack obvious sexual dimorphism [45][46]; although Garcia et al. [47] suggests they are sexually dimorphic in appearance in the ultraviolet region of the spectrum. Displays are performed by both sexes and triggered when another individual (even other species of lizards) appears in their eld of vision [46]. The 3D model used herein was based on morphology and display footage taken from a population that inhabit dense clumps of spini ex grass (Triodia sp.) and Mallee eucalyptus (Eucalyptus sp.) in the Murray-Sunset National Park, Victoria ( Figure S3c). Both G. longirostris and C. decresii are highly territorial species that inhabit semi-arid rocky outcrops with low vegetation density. Gowidon longirostris are relatively large lizards (SVL 114 mm) mainly due to their exceedingly long tail compared to the other three species. The species is widely distributed in arid zones across western and central Australia [44]. The complex signalling sequence comprises limb waving and head bobbing, with push-ups followed by a forward body thrust/rise usually associated with tail lashing motions [48]. Throat and crest extensions are also common modi ers of the display (Fig. 1b). The model lizard used herein was based on morphology and display footage taken from a population located within and around Ormiston Gorge, within West-MacDonnell National Park, Western Australia. The habitat around Ormiston Gorge features large rocky outcrops with sparsely distributed vegetation including Eucalyptus trees, low shrubs (Acacias sp.) and a variety of native grasses and grass trees (Xanthorrhoea sp. Figure S3b).
Ctenophorus decresii are averaged sized lizards (SVL 82 mm) found across eastern South Australia, including Kangaroo Island [49]. The species exhibits diverse colouration within and between populations [50] and is known to show individual variation in their agonistic displays [51]. Other than throat/crest extension, limb waving and head bobbing, C. decresii also displays a distinctive hind-leg push up with the tail coiled in a horizontal direction [49,52] (Fig. 1d). The hind-leg push up is utilised by males to re ect endurance and aggressiveness during territorial interactions [51]. The model lizard used herein was based on morphology and display footages of C. decresii lizards inhabiting the rocky outcrops of the Flinders Ranges National Park, South Australia. The habitat is prone to re, so we have incorporated charred fallen branches and sparse vegetation coverage, featuring mainly saltbush (Atriplex nummularia) and spinifex grass (Triodia sp.), as well as sparsely distributed Eucalyptus sp. and black oaks (Casurina sp.; Figure  S3d). Creating and animating the habitats The microhabitats were modelled using Autodesk Maya©2015 based on data collected from habitat surveys. As emphasised in Chouinard-Thuly et al. [53], the simulations created herein are based on data from real environments. We utilised the habitat of A. muricatus from earlier work [27,36], while the three new habitats were created as close replicates of the actual habitats. Brie y, the habitat models were constructed in accordance with a library of reference images photographed at each site. These included 360o panoramic photographs to capture the general layout of the habitat, while microhabitat detail was captured by photographing plants and substrates within close range of a known signalling site. The position of plants surrounding the signalling point were noted and spatial layout information was incorporated into each scene.
As Maya does not provide ready-made plant models, major plant species from our study sites, such as spinifex grass and eucalyptus trees, were individually modelled and integrated with precise wind controllers for setting the speed, intensity and frequency of plant movements. The wind controllers are set as a sliding bar to vary in the range [0, 1], resulting in increasing movement intensity. Deformation of plant models in response to simulated wind utilised built-in scripts within Maya, which are based on physically accurate equations [54]. By manipulating the numbers on the wind controller, we set the movements of modelled plants to visually match video recordings of actual plant movements in response to moderate wind conditions at each site. Video recordings of plants at each site were taken during periods with natural wind speed close to 0 m/s and plant movements were generated using a leaf blower positioned 2 and 3 m from the plant, with wind speeds measured at the plant of 2.5 and 4 m/s respectively. The camera was positioned 2 m from the base of the target plant and angled perpendicular to the direction of the wind emanating from the leaf blower. 3D lizard models and display sequences Detailed steps on how 3D models of lizards were created are described elsewhere [27]. Lizard display sequences were created using video recordings of each species performing movement-based displays in nature. As described above, the displays of these four species are quite distinct. The temporal structure of the display is presented in Fig. 1, computed as the extent of body movement over time from motion analysis (see below). The head-bob display of C. fordi generates the least movement (Fig. 1b), although relatively continuous throughout the sequence. In contrast, the display of A. muricatus is characterised by intermittent movement (Fig. 1a), beginning with small movements at the tip of the tail, then utilising the whole tail before the push-up component of the display. The temporal structure of both G. longirostris and C. decresii involve relatively constant movement once the display begins (Fig. 1c, d respectively). The display of C. decresii involves a gradual increase in the amount of body movement before it ends, while G. longirostris incorporates several rapid changes in body movement (rapid increase and decrease). All lizard and habitat models were built in proportion and to scale.

Scene composition and analysis
The models of all lizard species were placed into the habitat of all species (Fig. 2a,b; Figure S4). For a given habitat, the lighting conditions, location and orientation of the lizard in the scene and the position of the camera were matched for all lizard species. We used a single directional light to illuminate the entire scene with a uniform lighting effect. Lizard models were positioned in a given habitat according to where the local species would typically be found and orientated side-on to the camera. The animated signalling sequences of all lizard displays started at the same frame in the animation timeline, and were similar in duration (6.4 s or 160 frames, with approximately 5 to 10 frames differences between the set). All sequences were 8 s duration and exported as images at 25 fps and 720 × 576 resolution. We manipulated wind conditions by systematically increasing the extent of plant motion at all sites by manually adjusting plant controllers, con rming incremental (but non-linear) increases in noise using computational motion analysis as described in [36]. We repeated this to achieve 10 versions of each scene varying in environmental noise conditions. All four species were placed in the habitats of each species under all 10 noise conditions resulting in a nal set of 160 sequences. For each species-habitat combination, we also created a baseline sequence featuring only lizard movements that is required for our analytical approach.
Our analytical approach to determining salience of lizard displays was the same as previous work [27,36]. Brie y, animation sequences featuring plant movements were analysed frame by frame using saliency analysis to generate a relative salience score in the range [0, 1] for all areas of the frame (Fig. 2ce). The scores of each sequence were then multiplied by its relevant binary mask, which featured only the lizard movement, to produce a relative salience score for the lizard movements in the environmental context imposed. This was repeated for all frames in the sequence to generate the relative salience over time (Fig. 2f). As in our earlier work, we then computed the proportion of the sequence that exceeded a given threshold as a function of changing receiver thresholds (1000 values between 0 and 1 varied at increments of 0.01; Fig. 2g). A single vector of values is obtained for each species in each wind condition of each habitat.

Data analysis
Our objective was to compare and contrast the performance of lizard displays in different habitats under varying environmental conditions. We have simulated a single exemplar of each species and habitat to achieve this, so inferential statistics are not appropriate. Nevertheless, our simulations provide novel and informative data that we have considered quantitatively. We began by running non-metric multidimensional scaling (MDS) analysis, and used the metaMDS function from the Vegan package [55] in the R Statistical Environment [57] to quantify similarities/dissimilarities in the resultant vectors for each combination of species, habitat and wind condition (160 vectors). All data were considered together; however, to facilitate visual inspection of results, we group the outcome by wind condition and present the same results twice, labelled by the signalling species and then by habitat. Our second approach was to consider performance not as a single vector, but after segmenting individual datapoints (Fig. 2g inset) and grouping into three receiver threshold categories: low (0-0.3), moderate (0.3-0.8), high (0.8-1.0). We rst con rmed the detrimental effect of wind for A. muricatus in its own habitat, before focusing on moderate receiver thresholds to compare species in all habitats during moderate and strong wind conditions.

Results
The effectiveness of displays by four Australian agamid lizard species were assessed in their own habitats and the habitats of each of the other species under different wind conditions. In each case, we obtained the proportion of the sequence that exceeded a speci ed threshold for response as a function of varying threshold values. Figure 3 provides the result of non-metric MDS analysis comparing similarities/dissimilarities between all 160 resultant vectors. The XY coordinate data are presented separately for calm (levels 1-3), moderate (levels 4-7) and strong (levels 8-10) wind conditions. With datapoints labelled by the species displaying (Fig. 3a), we observe some grouping by species, though these appear in clusters of a few points (i.e., a single habitat) rather than as a single species cluster that would imply similar performance across all habitats and conditions. This is consistent with the notion that performance of a given display is dependent on location. Another observation we can make here is that the performance of three species (A. muricatus, G. longirostris and C. decresii) are comparable in certain habitats (Fig. 3c, blue lines), while one species, C. fordi, is consistently on the outer and performs relatively poorly (Fig. 3c, red lines). In Fig. 3b we label data by the habitat in which displays are performed. What is most striking here is that displays by all species within the habitat of A. muricatus cluster together, and this gets tighter as wind conditions worsen. Performance in this habitat is strong to begin with, but declines as receivers become more discriminating (Fig. 3c, black lines), suggesting that the task of distinguishing signal from noise has become more di cult.
We next considered the data as individual performance scores grouped into low, moderate and high receiver threshold categories. Figure 4 provides the outcome for A. muricatus in its own habitat. When receivers are non-discriminating, display performance is strong even as wind conditions increase. However, the masking effect of wind becomes apparent when receivers start to be more selective in what will attract attention. Given highly selective receivers are predicted to rarely detect a signal in strong wind conditions, we chose to focus on moderate receiver response thresholds to compare species across all habitats. The performance of each species' display in each habitat is shown in Fig. 5 for moderate and strong wind conditions (Fig. 5). These data con rm our earlier observations that the habitat of A. muricatus is the most detrimental to display performance, while the display of C. fordi is consistently the worst performing. We also see that stronger wind reduces performance across all sites (scores reduced in Fig. 5b compared with Fig. 5a), although to varying degrees (greater spread in data in strong winds).

Discussion
Technical di culties in quantifying environmental conditions and movement-based displays in a systematic way has made examination of movement-based signals a challenging task. Legislative restrictions have made in situ cross-habitat comparisons improbable. Recent studies have attempted to contrast multiple signals in multiple habitats by quantifying signal and noise separately and then contrasting resultant motion speed data [33][34]. For the rst time, sophisticated 3D animations gave us the opportunity to closely examine movement-based signals in multiple habitats. Speci cally, we were able to quantify how varying environmental conditions, receiver behaviour and differences in display and microhabitat structures effected signal e cacy. Using four Australian agamid lizard species, we compared and contrasted how species-speci c movement-based signals would perform in structurally distinct environments. Our simulations suggested that signal e cacy varies as a function of receiver response thresholds, where at medium to high thresholds, the changes within and between species became clear. In line with our predictions, all signals performed the worst in A. muricatus habitat, the performance of G. longirostris and C. decresii were very similar and the display of C. fordi was the worst across all habitats. However, contrary to our prediction, the performance of A. muricatus was not more effective than G. longirostris and C. decresii, so the notion that signals expected to be adapted to a noisier environment will perform better than the ones adapted to less noisy environments was not supported.
The poor performance of the display by C. fordi relative to other species was anticipated and might be attributable to differences in signal function. Unlike the other three species in the study, C. fordi are not territorial and their displays are not considered aggressive responses to intruders. It has been suggested that the likely function of the simple head bob display is for species recognition [46]. Displays are performed at close range when lizards encounter other lizards, so the necessity to distinguish itself from surrounding plant motion is removed. The other three species in this study are all assumed to be territorial species [42,52,57], responding to intruders at greater distances and when the intruder is possibly unaware of the presence of the territory owner. In these circumstances, reliable detection and processing of territorial displays are essential for establishing social dominance [58][59].
The potential masking effect for movement-based signals is dependent on prevailing wind and the distribution of plants in the environment [32]. Complex image motion backgrounds are expected in dense vegetation, which would likely require animals to generate certain types of movements for effective communication [3]. The habitat of A. muricatus is densely vegetated with tall grasses and small shrubs ( Figure S1a). The plants are often positioned close to the signalling animal, with short signaller-plant distances predicted to reduce detectability of signals [60]. In comparison, the habitats of G. longirostris and C. decresii feature sparsely vegetated environments, with plants typically separated from the signalling animals. However, the C. decresii habitat we modelled herein was based on a recently burnt area which would feature a reduced level of motion noise in the environment, this information could change for C. decresii as the vegetation recovers from the re over the next few years. The signaller-plant distances in the habitat of C. fordi were also very close, but in this case the dominant plant species, spinifex grass (Triodia sp.), does not generate considerable movement in response to wind at the base of the plant where the lizards were positioned. As such, the potential masking effect of plant environments is attributable to the interaction between wind conditions, the spatial position of plants in the environment and the structure of the plant itself that will dictate how it responds to wind.
Habitats with common structural properties, regardless of geographic location, are predicted to result in signals with similar structural characteristics [61][62]. Both G. longirostris and C. decresii inhabit sparsely vegetated rocky outcrops, and the similarities in the structure of their signals and relative signal performance in noise could be attributed to structural similarities in the environments they inhabit.
Interestingly, the performance of their displays was just as effective as the signal of A. muricatus in the habitat of A. muricatus suggesting different movements might be equally effective. However, it is in the other habitats that we start to see subtle interactions between display and habitat structure. The displays of G. longirostris perform best in two of these habitats, while C. decresii displays perform better in their own habitat. Ramos and Peters [48] found no evidence for broad scale habitat characteristics in uencing signal use in Australian dragons, but it is intriguing to consider the possibility that the use of different motor patterns is partially driven by ne-scale habitat characteristics that in uence the motion noise environment. Both G. longirostris and C. decresii signals involve almost continuous movement featuring whole body push-ups/forward thrusts. In contrast, the push-up component of the display of A. muricatus is brief and follows an occasionally lengthy period of tail icking [14]. Tail-icks represent the introductory component to the rest of the display, reportedly to capture the attention of a receiver [14,43].
It is also this component of the signal that exhibits plasticity in response to changing environmental conditions [14], whereby the duration of tail icking is lengthened in noisy conditions and the movement becomes more intermittent. Tail-icks are observed in C. decresii, although it is not a core component of the display [35], and there are no reports for tail icking by G. longirostris [48]. Further research is needed to determine whether either of these species possess a component of their display that exhibits the kind of plasticity needed for variable environments. Alternatively, perhaps the environments that they inhabit have not selected for such plasticity as the rocky gorges provide a clear background against which the signal can be reliably detected regardless of plant movement speeds.
This study has added to a growing literature showing that habitat characteristics and the associated noise landscape has a signi cant evolutionary effect on signal design and likely promotes changes in signalling strategies under certain conditions [14,[63][64]. How signals function in social interactions is another important consideration for driving variation in signal structure [46,58,[65][66]. Cross-habitat examination of signal performance has been bene cial for other communication systems [19,20,21,23], but the present study is one of the rst comparable studies for movement-based signals. Our innovative approach allowed for examination of different movement-based signals embedded in noise across different habitats. We acknowledge that the results herein are generated from a single representative display per species, in a single representative habitat per species. Nevertheless, our method and analytical approach has provided quantitative data showing that the effectiveness of movement-based animal signals will be affected by the motion noise environment, and might indeed direct signalling behaviour in much the same way that it has for other modes of signalling.