Searching for seadragons: predicting micro-habitat use for the common (weedy) seadragon (Phyllopteryx taeniolatus) based on habitat and prey.

Habitat associations can be critical predictors of larger-scale organism distributions and range shifts. Here the authors consider how a critical habitat, kelp (Ecklonia radiata) and prey (mysid crustacean swarms), can influence small- and large-scale distribution on the iconic common (weedy) seadragon (Phyllopteryx taeniolatus:Syngnathidae). P. taeniolatus are charismatic fish endemic to the temperate reefs of southern Australia, reported to range from Geraldton, Western Australia (28.7667°S, 114.6167°E) around southern Australia to Port Stephens, New South Wales (32.614369°S, 152.325676°E). The authors test a previously developed model of seadragon habitat preferences to predict P. taeniolatus occurrence within four sites from Sydney to the northern limit of their range in eastern Australia. They determined that P. taeniolatus associations with Ecklonia and mysid shrimp can be extrapolated across multiple sites to predict the occurrence of individual P. taeniolatus within a location/site. For instance, the authors demonstrated a significant positive relationship between the density of mysid swarms and the density of P. taeniolatus, evident across all sites despite large differences in the density of mysid swarms among sites. The findings are the first to model P. taeniolatus habitat associations across multiple sites to the northern limit of their range and have applications in protecting P. taeniolatus populations and how they may respond under climate change scenarios, such as poleward kelp retractions.

INTRODUCTION

Predictive modelling of species distributions has long been of interest to ecologists and natural resource managers, and is a highly useful tool in ecology and conservation (Guisan & Thuiller, 2005). Species distribution models (SDMs) have become increasingly popular in recent years to monitor the impacts of climate change (Davis et al., 2021), biological invasions (Mainali et al., 2015), predicting threatened species occurrence (Poulos et al., 2015), and have also been used to inform IUCN Red List species assessments (Syfert et al., 2014). SDMs are informed by modelling complex species‐environment relationships (Fukuda et al., 2012). Rapid advancements in mapping the physical environment and geographical information systems (GIS) have increased the capabilities of SDMs, making it easier to obtain environmental data (Elith & Leathwick, 2009). Nonetheless, in areas where GIS data may not be readily available, or where key variables that affect a species cannot be mapped in this way (e.g., prey availability), more traditional field‐base surveys and statistical methods, such as generalised linear models (GLMs), are highly useful in modelling species occurrence and habitat associations and informing other current methods such as resource selection probability functions (RSPFs) (Elith & Leathwick, 2009; Keating & Cherry, 2004; Manly et al., 2007).

Collecting environmental data and biotic interactions to use as predictor variables must be carefully selected in relation to the target species biology (Elith & Leathwick, 2009; Mainali et al., 2015), as species distributions are driven by both biotic and abiotic interactions (Davis et al., 2021). For example, potential climate refugia for the kelp, Ecklonia radiata (hereafter Ecklonia), were mapped over 10 degrees of latitude on the east coast of Australia using SDMs (Davis et al., 2021). Abiotic and biotic variables, such as sea urchin density (Centrostephanus rodgersii), summer bottom temperature and photosynthetically available radiation, significantly predicted Ecklonia cover. Another example of modelling was the application of generalised linear mixed effects modelling (GLMMs) to model the occurrence and habitat preferences of the syngnathid, Hippocampus reidi, across three estuaries in Brazil (Aylesworth et al., 2015), showing that seahorse occurrence was strongly linked with shallow depths, warm temperatures and increased availability of holdfast structures.

Syngnathids (seadragons, pipefish, seahorses and pipehorses) are charismatic species defined by unique morphology, low densities and small home ranges (Harasti et al., 2012; Sanchez‐Camara et al., 2006; Sanchez‐Camara & Booth, 2004; Vincent et al., 2011). The unique biology of syngnathids renders them particularly vulnerable to the effects of habitat loss and climate change (Vincent et al., 2011). Nonetheless, there are only few published studies modelling syngnathid habitat preferences and species occurrence (Aylesworth et al., 2015; Curtis and Vincent 2005; Harasti et al., 2014; Hernández‐Urcera et al., 2021). As syngnathids have been shown to be effective flagship icons for conservation and under threat worldwide (Pollom et al., 2021; Shokri et al., 2009; Vincent et al., 2011), studies that accurately detail the distribution, occurrence and habitat associations of syngnathids are critical for effective management (Vincent et al., 2011).

The common (weedy) seadragon, Phyllopteryx taeniolatus (Syngnathidae) Lacepede, 1804, is a distinctive fish endemic to the Great Southern Reef (GSR) of Australia. P. taeniolatus are reported to range from Geraldton, Western Australia (WA), to Port Stephens, New South Wales (NSW) (Edgar, 2008). Throughout their geographical range, P. taeniolatus are reported to occupy different habitats such as kelps and seagrass (Kuiter, 2009; Klanten et al., 2020). Nonetheless, in Sydney, towards the northern extent of their distribution, P. taeniolatus are strongly associated with Ecklonia habitat and mysid prey (Allan et al., 2021; Sanchez‐Camara et al., 2006). Mysid shrimp (Mysida spp.) make up >80% of the diet of P. taeniolatus (Kendrick & Hyndes, 2005). It is currently unknown if changes in habitat may affect vulnerable P. taeniolatus populations, such as those at the northern extent of their range. Modelling has shown that under more extreme climate change scenarios, kelp declines on the east coast of Australia may be exacerbated (Castro et al., 2020; Martínez et al., 2018). Declines of other critical marine habitats have been linked to declines in abundance of other syngnathid species (Correia et al., 2015; Harasti, 2016). Extensive destruction of seagrass habitat was linked to 73% decline of Hippocampus hippocampus and a 94% decline of Hippocampus guttulatus in the Rio Formosa, Portugal (Correia et al., 2015). Furthermore, seagrass loss driven by urbanisation in Malaysia caused significant declines in Hippocampus kuda abundance (Vincent et al., 2011), while declines in the endangered Hippocampus whitei were directly linked to declines in soft coral and sponge habitat (Harasti, 2016).

Modelling habitat requirements of P. taeniolatus across multiple sites throughout their northern range has not been done and is essential to assist with conservation and management. To date, studies assessing habitat use of P. taeniolatus have only been conducted in Sydney, primarily at Kurnell, NSW (Sanchez‐Camara et al., 2006; Allan et al., 2021). In the present study, the authors use GLMs to analyse how well previous knowledge (Allan et al., 2021) of P. taeniolatus habitat preferences, mysid density and Ecklonia cover can be applied to predict P. taeniolatus occurrence within a specific location. They compare data, including habitat composition, water depth and mysid densities, from four study sites across c. 200 km of coastline in NSW, including Broughton Island, which is the northern most recorded site for P. taeniolatus on the east coast of Australia, and three other sites within Sydney. They also assess the relationship of P. taeniolatus density and mysid shrimp density at a site and among locations. Based on previous studies (Allan et al., 2021; Sanchez‐Camara et al., 2006), it was expected that mysids and Ecklonia cover would predict P. taeniolatus occurrence, and that there would be a strong positive relationship between mysid density and P. taeniolatus density across sites.

MATERIALS AND METHODS

Ethics statement

This project was conducted in accordance with animal ethics permit UTS ACEC ETH17‐1707 and NSW DPI Permit F94/696. Any handling of animals complied with Australian animal welfare laws.

Study sites

All observations for this study were collected using scuba, from March 2020 to September 2021, and 48 dives for a total of 36.1 h were performed across all sites, with an average dive time of 44 min (±6.5 min). This study was conducted at four different sites from Sydney to Port Stephens, NSW, Australia (Figure 1): Kurnell, North Bondi (NB), Cabbage Tree Bay (CTB) and Broughton Island (BI). Due to the rarity and cryptic nature of P. taeniolatus, study sites were selected as they had known populations of P. taeniolatus. Kurnell and NB are previously studied locations for P. taeniolatus (Sanchez‐Camara et al., 2005; Sanchez‐Camara et al., 2006; Sanchez‐Camara et al., 2011; Sanchez‐Camara & Booth, 2004), whereas no research on P. taeniolatus had been undertaken at CTB and BI. CTB was selected as a site based on P. taeniolatus occurrence data (iNaturalist, 2021) and discussions with local divers, whereas BI was selected because there were reported sightings of P. taeniolatus by regional experts (Harasti pers. comm.). BI is located approximately 15 km to the northeast of Port Stephens, NSW, Australia, and is considered the northern limit of P. taeniolatus on the east coast of Australia; subsequently this is the northernmost studied population of this species. All four sites had similar benthic composition, being rocky reefs where Ecklonia is the dominate macroalgae, interspersed with sand flats and urchin barrens. Nonetheless, sites differed topographically. Kurnell and NB have sloping rocky reefs (depths: 3–13 m, and 5–21 m, respectively) characterised by a linear Ecklonia–sand interface spanning the length of the site. In contrast, CTB and BI are more three‐dimensional sand flats interspersed with rock and Ecklonia dominating the macrophyte community. Depth range of surveys across all sites varied: 9–13 m at Kurnell; 16–21 m at NB; 9–16 m at CTB; and 14–21 m at BI. Locations of each site were accurately determined using a diver‐towed GPS (Garmin eTrex10®) attached to a dive flag.

FIGURE 1

Location of Phyllopteryx taeniolatus survey sites from Sydney to Port Stephens, NSW, Australia. (a) Location of Broughton Island and Sydney on the east coast of Australia. (b) Location of study sites (Kurnell, North Bondi, Cabbage Tree Bay) within Sydney. (c) Location of study site on the northern side of Broughton Island

Habitat composition and environmental variables

Habitat composition at each site was determined using a point‐transect method (Choat & Bellwood, 1985). Eight transects, each 50 m in length, were placed haphazardly within each site. Due to differences in the site topography across sites, methodology was adjusted accordingly to avoid resampling over the same habitat. At Kurnell and NB, both having more linear reef formations, an initial transect was placed along the kelp–sand interface (c. 12 and c. 18 m depth, respectively) with three subsequent transects placed consecutively, spanning the length of the study site. The four remaining transects were placed at a parallel distance of 15 m towards the reef to ensuring all habitat was sampled. CTB and BI are more three‐dimensional kelp beds, where placing four 50 m transects would run out of the survey area. Therefore, two 50 m transects were placed successively along the reef flat starting from a random point and running towards the other end of the study site, totalling eight transects. Each group of two transects was placed at a parallel distance of 15 m. Transects were filmed facing directly downwards using a GoPro 5 camera (www.gopro.com). Screenshots were taken every 1.5 m on the transect, totalling 264 images per site. Benthic percent cover was determined using Coral Point Count with Excel (CPCe) (Kohler & Gill, 2006). Benthic categories followed Allan et al. (2021), i.e., rock, sand, sponge, Ecklonia, rubble, Sargassum spp., Caulerpa spp.

Additionally, depth was measured using a Suunto Zoop dive computer (www.suunto.com). If depth varied from start to endpoint of the transect, depth for that transect was recorded as an average of the two values.

P. taeniolatus and mysid swarm density

Mysid shrimp swarm density and P. taeniolatus were determined by calculating total number of mysid swarms and P. taeniolatus individuals encountered along the 50 m transects (4 m wide) at each site. The total area of 8 × 50 m transects was 1600 m2. Density estimates of P. taeniolatus were expressed as the number of individual seadragons per 100 m2, and number of mysid swarms per 100 m2. A mysid swarm was defined as a cohesive group of mysids swimming in regular spatial arrangement (Wittmann, 1977; Ohtsuka et al., 1995). Several attempts were made to quantify mysid swarm size with measurements and netting techniques (Ohtsuka et al., 1995). Nonetheless, due to the strong avoidance behaviours and patchy distribution of mysids (Gal et al., 1999; Kaltenberg & Benoit‐Bird, 2013) it was determined that this was not providing reliable and repeatable measurements for size of mysid swarms. Therefore, swarm size was not estimated, and a simple metric of swarm number/abundance was used to estimate densities.

Predicting P. taeniolatus occurrence from microhabitat usage

Data on microhabitat use of P. taeniolatus were collected by two divers actively searching the study area at each site for P. taeniolatus individuals. These data were collected to test whether previous knowledge on P. taeniolatus microhabitat use could predict individual's occurrence within a site, across the four study sites. When each seadragon was encountered, habitat use data were collected using the methods of Allan et al. (2021). Five‐metre‐long (three‐meters wide) belt transects were placed at the point of initial sighting of P. taeniolatus individuals and run in a haphazardly selected direction. Transects (where P. taeniolatus was absent) were also started haphazardly throughout the site to sample and to compare with habitat in transects with P. taeniolatus occurred.

Benthic cover was analysed in CPCe using the benthic categories stated above. From each 5 m transect video, three screenshots, each 5 m2, were taken and analysed to determine percent cover of different benthic groups in each image, resulting in an area of 15 m2 per transect. Depth and temperature were also collected at each P. taeniolatus sighting. Mysid swarms (defined above) were recorded as either present or absent within each transect.

Statistical analysis

Habitat composition and environmental variables

The hypothesis that habitat composition would differ across the four study sites (Kurnell, NB, CTB and BI) was tested using a one‐factor permutational multivariate analysis of variance (PERMANOVA) conducted in PERMANOVA+ Version 1.0.5 within PRIMER‐E 6 (Plymouth Routines in Multivariate Ecological Research; www.primer-e.com). Analysis was done on the Bray–Curtis similarity matrix with significance determined from n = 9999 permutations, with site analysed as a fixed factor. Depth data across all sites did not pass normality testing and therefore was analysed using a non‐parametric independent samples Kruskal‐Wallis test.

Additionally, habitat composition in areas where P. taeniolatus were present vs. absent was compared across all sites pooled and analysed with a one‐factor ANOVA with a Tukey's HSD post hoc to discriminate variation between sites. Differences in each habitat category were then assessed among sites with ANOVA to discriminate differences in P. taeniolatus habitat use within each of the four study sites.

P. taeniolatus and mysid swarm density

Mysid swarm density (number of swarms recorded on 50 m belt transect) across the four study sites was compared using a one‐factor ANOVA with a Tukey's HSD post hoc to discriminate differences between sites. Estimates of P. taeniolatus density were similarly determined from the number of individuals found on each 50 m transect. The relationship of P. taeniolatus density and mysid availability is understudied, so this was analysed using linear regression to better understand the impact of food requirements on P. taeniolatus habitat selection across multiple sites. Densities were log10 (x + 1) transformed to ensure homoscedasticity of variance.

Predicting P. taeniolatus occurrence from habitat preferences

GLMs were used to determine the most important variables in predicting P. taeniolatus occurrence within an area. A stepwise binary logistic regression model was developed in SPSS (IBM Corp 2020) to predict P. taeniolatus presence within a site. P. taeniolatus presence/absence (coded 1/0) within 5 m microhabitat transects was fitted as the binary response variable. The environmental variables included in the model as predictors were based on previous results of habitat preferences of P. taeniolatus (Allan et al., 2021) and added to potentially increase the model's predictive value across the four study sites. Variables included in the full model were percent cover of Ecklonia, sand and rock, mysid presence/absence, site and depth (m) analysed as fixed effects. The stepwise procedure progressively simplified the model based on χ2 tests. Non‐significant terms were removed if they did not affect the deviance of the model (Aylesworth et al., 2015). Data from all sightings were pooled for analysis, to increase statistical power of the study (Allan et al., 2021; Sanchez‐Camara et al., 2006); nonetheless, there were too few sightings of P. taeniolatus to compare differences between sexes, life stages or reproductive phases.

RESULTS

Habitat composition across sites

Habitat composition did not significantly differ across sites (PERMANOVA, F1,31 =2.914, P >0.05; Figure 2). Average percent cover of the main benthic categories across all sites was 27% rock, 45% sand and 24% Ecklonia. All habitat categories were included in the analysis, but only the dominant three habitats are represented in Figure 2. Furthermore, average depth differed across all sites (Kruskal‐Wallis, H3 = 68.962, P <0.05). Mean depths at all sites were as follows: Kurnell, 11.2 m (±0.3 m); NB, 18.3 m (±0.6 m); CTB, 12.8 m (±0.3 m); and BI, 18.6 m (±0.3 m). Pair‐wise comparisons showed depth did not differ between Kurnell and CTB (H3 = −6.375, P >0.05), and between NB and BI (H 3 =2.000, P >0.05).

FIGURE 2

Mean (±s.e.) percent cover of major benthic categories, estimated from eight 50 m belt transects, at four sites from Sydney to Port Stephens, NSW, Australia (K, Kurnell; NB, North Bondi; CTB, Cabbage Tree Bay; BI, Broughton Island)

Habitat composition where P. taeniolatus were “present” vs. “absent” differed significantly across four sites, with the former having lower rock (present = 19.51%, absent = 24.80%, T204 = −2.106, P < 0.05; Figure 3), and higher Ecklonia (present = 31.38%, absent = 20.80%, T204 =4.740, P <0.05; Figure 3). Sand did not differ significantly between areas where P. taeniolatus were present and absent (present = 46.37%, absent = 50.58%, T204 = − 1.276, P > 0.05; Figure 3).

FIGURE 3

Mean percent cover of Phyllopteryx taeniolatus microhabitats (±s.e.) for the main three habitat groups (Ecklonia, sand and rock) across all study sites in NSW, Australia (K, Kurnell; NB, North Bondi; CTB, Cabbage Tree Bay; BI, Broughton Island). White bars represent mean habitat cover with P. taeniolatus present and grey bars where P. taeniolatus were absent (* indicates significant difference within sites)

P. taeniolatus vs. mysid swarm density

Pooling 50 m transects within a site (8 × 50 m), there were 28 mysid swarms observed at Kurnell, 9 at NB, 7 at CTB and 16 at BI, whereas the density of P. taeniolatus along the 50 m transects was highest at Kurnell, BI, CTB and NB, respectively. There was a significant positive relationship between P. taeniolatus and mysid swarm densities across all sites (F1,31 = 18.218, R 2 =0.378, P <0.001; Figure 4). Furthermore, the density of mysid swarms was significantly higher at Kurnell than all other sites (Tukey's HSD post hoc, P <0.05), whereas there was no significant difference in mysid density among the other three sites.

FIGURE 4

Density of Phyllopteryx taeniolatus individuals and mysid swarm densities along 50 m transects at four sites from Sydney to Port Stephens, NSW, Australia (K, Kurnell, blue circles; NB, North Bondi, orange squares; CTB, Cabbage Tree Bay, grey triangles; BI, Broughton Island, white circle)

Predicting P. taeniolatus occurrence from habitat preferences

During this study, there was a total of 114 sightings of 62 individuals of P. taeniolatus across all sites. GLMs determined that mysid presence/absence within a microhabitat and percent cover of Ecklonia were the most important variables in predicting P. taeniolatus presence within multiple sites (Table 1). In the exploratory stepwise analysis, the variables’ sand percent cover, rock percent cover and depth were removed from the final model (Table 1). Increased percent cover of Ecklonia and presence of mysid shrimp both significantly increased the probability of finding an individual seadragon within a random microhabitat. Furthermore, there was no significant relationship between mysid presence and Ecklonia cover in the analysis (β =0.778, P >0.05). Rock was a non‐significant predictor; the negative coefficient is indicative of previous findings that P. taeniolatus prefer areas with higher kelp and lower rock cover. Site was retained in the final model, as it did not affect the overall deviance of the model; nonetheless, it was a non‐significant predictor; that is, there was no difference in habitat preferences across site. The final model predicted 71% of the variation in P. taeniolatus occurrence across all sites.

TABLE 1

Final outcome of stepwise binary logistic regression model with Phyllopteryx taeniolatus presence/absence fitted as the response variable across four sites in NSW Australia

Effects in the model	β	Exp (β)	Exp (β) 95% c.i.		P
Mysid P/A*	1.568	4.797	2.394	9.613	0.0001*
Ecklonia % cover*	0.054	1.055	1.032	1.079	0.0001*
Site	0.306	1.358	0.946	1.951	0.10

Note: Ecklonia % cover, mysid P/A and site were retained in the final model as the most important effects (* indicates significance in the final model, P/A = presence/absence). Refer to Supporting Information Table for complete stepwise model.

Significance indicated in bold value.

DISCUSSION

The results show that previously developed knowledge of P. taeniolatus preference for Ecklonia habitat and mysid prey (Allan et al., 2021) can be applied to predict the occurrence of P. taeniolatus within multiple sites over c. 200 km of coastline. Furthermore, the authors identify a significant positive relationship between the densities of mysid shrimp and P. taeniolatus across all study sites, despite large differences in mysid occurrence among sites.

The population of P. taeniolatus at Broughton Island is the reported northern limit of their range and is believed to be the most northern established population of P. taeniolatus on the east coast of Australia. This is the first study to assess P. taeniolatus at their northern limit in NSW. As such, studying the habitat associations of range edge populations may provide critical baseline information for assessing the impacts of climate change and habitat loss on P. taeniolatus. The authors were expecting, given that the site is at the northern range edge, that a key predictor, such as Ecklonia cover or mysid density, may be low here, but not so (Davis et al., 2021; Williams et al., 2020). Responses of syngnathids to alterations in habitat quality and quantity, prey availability or physiological stressors may affect survival, abundance growth and reproduction of these unique fishes (Vincent et al., 2011). Previously, population declines in an endangered seahorse (H. whitei) have been directly linked to the decline in critical habitats such as soft corals and sponges (Harasti, 2016). Therefore, the strong dependence of P. taeniolatus on kelp habitat and predicted shifts and declines in Ecklonia habitat (Castro et al., 2020; Martínez et al., 2018) may directly impact the persistence of P. taeniolatus, particularly in range edge populations. Therefore, having accurate baseline data on the range limits of P. taeniolatus is crucial for assessing the subsequent impacts of climate change and range shifts of associated organisms such as Ecklonia or mysids (Sorte et al., 2010).

In this study, mean habitat composition of the three main benthic groups, sand, rock and Ecklonia, did not differ significantly across all sites from Sydney to Port Stephens; nonetheless, depth differed across sites. In this study, mean depth at Kurnell and CTB was shallower than that at NB and BI; nonetheless, depth was a non‐significant effect in the predictive model of P. taeniolatus occurrence. Habitat preferences of another syngnathid, Hippocampus reidi, have been associated with deeper depths in the presence of increased wave action (Aylesworth et al., 2015). Wave and current exposure were not included in this analysis, as this model focused on microhabitat selection, where depth and exposure were relatively uniform within a site. Nonetheless, future species distribution models for P. taeniolatus may benefit from including wave and current exposure of sites. As P. taeniolatus are poor swimmers and do not possess prehensile tails, wave and current exposure may be important factors in future models of habitat occupancy to determine new locations where P. taeniolatus may be present.

The distribution of Ecklonia extends to c. 27°S on the east coast of Australia (Wernberg et al., 2019), with areas adjacent to the north of Port Stephens having comparable Ecklonia cover to BI (Williams et al., 2020), the current range limit of P. taeniolatus. Therefore, it is likely that Ecklonia cover, while highly important for P. taeniolatus populations, may not be the primary limiting factor for the establishment of populations extending north from Port Stephens. Nonetheless, local declines in Ecklonia habitat from increased herbivory or urbanisation (Williams et al., 2020) would possibly result in P. taeniolatus declines, as has been shown in H. whitei populations (Harasti, 2016). Additionally, the strong relationship between mysid density and P. taeniolatus density in this study was evident, despite differences between sites in mysid density. Surveying sites of similar habitat composition to those in this study, and assessing mysid availability, may provide valuable insight into mysid density being a potential limiting factor in P. taeniolatus latitudinal distribution, as sites with few mysid swarms in this study had similarly lower number of P. taeniolatus. It is also possible that P. taeniolatus distribution may be limited by geographical barriers, such as large areas of sand devoid of kelp north of BI, certain species traits, as P. taeniolatus have low larval dispersal and exhibit high site‐fidelity (Klanten et al., 2020; Sanchez‐Camara & Booth, 2004). There is currently no large‐scale mapping of Ecklonia habitat within the observed depth range of P. taeniolatus, only mapping in smaller areas and predictive modelling (Davis et al., 2021; Jordan, 2010; Williams et al., 2020). Rigorous data on Ecklonia cover throughout the range of P. taeniolatus would allow for the development of more sophisticated species distribution models or habitat occupancy models.

This is the first study to model P. taeniolatus habitat associations towards the northern extent of their range over 200 km of coastline, and the first such study for any seadragon species. Studies monitoring habitat associations across large geographic areas are rare in syngnathids (Aylesworth et al., 2015; Hernández‐Urcera et al., 2021), but may be extremely useful in the development of conservation and management strategies for vulnerable syngnathids or other species (Vincent et al., 2011), especially given the high spatial differences in P. taeniolatus population genetics (Klanten et al., 2020). Nonetheless, habitat mapping and more readily available environmental GIS data of Ecklonia habitat could lead to the development of more sophisticated SDMs detailing the range of P. taeniolatus throughout Australia. Given the highly charismatic nature, and fully protected status of P. taeniolatus, ensuring the protection of Ecklonia habitat and areas of high mysid density may be crucial to the persistence of seadragon populations in the future.

AUTHOR CONTRIBUTIONS

S.J.A. (corresponding author) contributed to conceptualisation of the study and study design, data acquisition, analysis and drafting of the original manuscript. M.J.O. contributed to data acquisition, analysis and critically revising the work. O.S.K contributed to study design and conceptualisation, data analysis and critical review of the manuscript. D.H. contributed to data collection, analysis and interpretation, and critical revisions of the manuscript. D.J.B. contributed to conceptualising the study design and concept, data collection, analysis and interpretation and manuscript revisions.

TABLE S1 Supporting information

Click here for additional data file.