We would like to acknowledge and express our deep gratitude to members of the Ngāti Manuhiri Settlement Trust (NMST) and Ngāi Tai ki Tāmaki (NTkT) who have shared their knowledge and wisdom with us in the development of this research project. The land and water upon which we conduct our research is the traditional territory of these communities, and their cultural and spiritual connections to these entities are fundamental to their well-being and identity. As part of our commitment to ethical research practices, we have involved both groups in a process of engagement and collaboration to ensure that their perspectives, values, and priorities are reflected in our research design and implementation. Indigenous knowledge is multifaceted, diverse, and deeply rooted in cultural and historical contexts, and we are committed to treating this knowledge with the respect and sensitivity that it deserves. Since our research project has the potential to impact the communities with whom we work, we are committed to ongoing communication and engagement with them throughout the project lifecycle. A collaborative and inclusive approach to research is essential for building trust, respect, and positive relationships with all of our stakeholders, so we will continue to prioritize this approach in all of our work.

The reintroduction of habitat-forming species is rapidly gaining popularity as an active intervention tool to restore degraded marine ecosystems (Swan et al., 2016; Bayraktarov et al., 2020; Fitzsimons et al., 2020; Sievers et al., 2022). A number of these projects currently rely on the translocation of individuals from growth sites (e.g., aquaculture, nurseries, wild populations) to recipient sites with a need for restoration (e.g., degraded areas, De Paoli et al., 2015; Barton et al., 2017; Wilcox et al., 2018; Benjamin et al., 2023). The selection of restoration sites in degraded areas involves a careful evaluation of the risks associated with reintroduction activities and the suitability of the restoration site for translocated populations (Seddon et al., 2007; Elsäßer et al., 2013; Section 5, IUCN/SSC, 2013; Fitzsimons et al., 2020; Robinson et al., 2020). However, there are often many environmental unknowns that can hinder the success of translocation efforts, even after careful consideration of potential risks. As a result, translocations may fail due to unexpected factors, such as unknown criteria for suitable habitat, especially if the species is reintroduced outside of their historic range (Fischer and Lindenmayer, 2000; Stadtmann and Seddon, 2018). This is especially apparent at the beginning of restoration initiatives that work with species and sites with unknown risks (e.g., see outcomes of mussel translocations reported in De Paoli et al., 2015; Wilcox et al., 2018; Alder et al., 2020). There is mounting evidence that small (≥1.5 m²) experimental plots can be used to assess the details of site suitability of recipient environments over an extended period (e.g., 18 months, Benjamin et al., 2023). However, since this process takes time and monitoring resources, it is useful to consider additional tools that can further refine the site selection process ahead of this commitment of resources, across a shorter timeframe, and a larger number of candidate sites.

Predators in the recipient environment often pose one of the biggest limitations to the success of species translocations by consuming newly translocated individuals (hereafter founders; Fischer and Lindenmayer, 2000; Moseby et al., 2015; Robinson et al., 2020). Managing the risk of predation to founders can be difficult, but addressing this early in the translocation process can yield successful outcomes (e.g., see approaches to predation risk management in Robinson et al., 2020). To minimize founder losses and improve the success of reintroductions, translocations are often planned in areas where predator populations are naturally low or absent (e.g., Robinson et al., 2020). In terrestrial biomes, this can be achieved by reintroducing species to predator-free areas such as peninsulas (e.g., Tawharanui Regional Park, Aotearoa/New Zealand [hereafter New Zealand], Jahn et al., 2022) or islands (e.g., Tiritiri Matangi Island, New Zealand, Towns and Ballantine, 1993; Seddon et al., 2014; Moseby et al., 2015; Van Heezik and Seddon, 2018). However, in open marine environments with no control over potential predator movement or water flow, this can be difficult. This is especially evident when working with sessile, biogenic marine species that cannot easily relocate themselves, unlike what has been seen with translocated fish species (e.g., Cochran-Biederman et al., 2015). Furthermore, assessments of marine predator populations can be a complicated process due to the difficulties of obtaining in-situ observations of mobile predator species, which is also an issue for fisheries management (Cochrane, 1999).

Predation remains a pervasive factor limiting a number of shellfish restoration initiatives and can impact the resilience of reefs to future disturbances (e.g., for mussels, De Paoli et al., 2015, 2017; Wilcox and Jeffs, 2019; Alder et al., 2020; for oysters, Tedford and Castorani, 2022; for clams, Whitlow et al., 2003). Part of this issue may be related to the inevitable loss of founders during the translocation process. For example, in the days to weeks following the translocation of mussels to the seafloor, it is common to incur losses due to the stresses encountered during and immediately after the translocation process (Wilcox et al., 2018). Some of these stresses occur:

Previous work has suggested that these losses can generate an odour plume that attracts nearby predators, which may further exacerbate mussel loss (e.g., in the lab with the Asian green mussel, Perna viridis, Shin et al., 2002; as suggested for in-situ losses of reintroduced green-lipped mussel, Perna canaliculus, Wilcox and Jeffs, 2019; at in-situ experimental plots of green-lipped mussels, Benjamin et al., 2023). Some of the mobile predator species that have so far limited shellfish reintroductions include:

Currently, the most effective predator deterrent in marine systems are physical barriers that inhibit predator access to founders (e.g., Capelle et al., 2019; Alder et al., 2020). However, the addition of barriers incurs substantial additional costs, may not be permitted by local legislation, and is often impractical at restoration scales that are ecologically meaningful (Alder et al., 2020; Schotanus et al., 2020; Alder et al., 2021). Therefore, there is a need to refine methods for understanding the presence and/or movement of mobile predator species that may limit the establishment of founders ahead of expensive, large-scale restoration efforts. There is some indication that methods and timing of translocations may limit predation and improve initial translocated mussel survival (Bertolini et al., 2018; Alder et al., 2022), however less is known about methods to assess the spatial variability of predators at desired restoration sites prior to restoration efforts. There is mounting evidence that the strategic placement of bivalve species can limit predation, such as the installation of euryhaline oyster reefs at less saline locations to limit the impact of freshwater intolerant oyster drills (Johnson and Smee, 2014; Pusack et al., 2019). These observations suggest that preliminary assessments of predator abundances could be a useful tool during initial site selection to reduce the possibility of subsequent losses. Reintroductions of small numbers of individuals can reduce costs and further clarify whether certain sites are suitable for survival of the reintroduced species and if it is worth continuing reintroduction efforts (e.g., as seen with freshwater fish, George et al., 2009).

In the south-eastern Hauraki Gulf of the North Island of New Zealand, endemic green-lipped mussel (Perna canaliculus) reefs once covered ≥ 600 km² of the seafloor, with the densest reefs inhabiting large stretches of the Firth of Thames (Paul, 2012). Throughout the majority of the 20^th century (1900-1969), these reefs were dredged to near functional extinction leading to the collapse of the dredge industry. Despite the collapse and subsequent lack of commercial bottom dredging in these areas, these reefs have failed to reestablish naturally (Paul, 2012). Over the past decade, community groups have started to restore these reefs throughout their native range. At present, mussel reef restoration currently focuses efforts at subtidal, soft-sediment areas devoid of biogenic structure and have seen variable rates of success with only a small number of pilot reefs surviving for more than 2 years. This means that it is important to gain an understanding of local conditions and the potential predators that might be attracted to the sudden introduction of both hard-bottom habitat and a novel food source. Understanding how these factors interact with mussel survival can then be used to determine which sites should be prioritised for future, longer-term site suitability trials, and eventually, large-scale restoration efforts. This study seeks to address whether rapid (week-long), small-scale (< 100 mussels m^-2) experimental translocations can be used as a preliminary screening process to improve site selection in degraded areas inside and outside the historic range of green-lipped mussels throughout Tīkapa Moana/the Hauraki Gulf (hereafter the Hauraki Gulf) of the North Island of New Zealand. This study will identify whether a remote assessment of predator abundance, environmental parameters (water velocity, wind speed, wind direction, rainfall, moon phase), and site characteristics (depth, substrate type, effective fetch) correlate with short-term translocated mussel survival following placement on the seafloor.

Sites were selected to cover a range of locations in areas outside (western area, sites W1-W6) and inside (eastern area, sites E1-E5) the historic range of dense, green-lipped mussel reefs in the Hauraki Gulf ( Figure 1 ). Prior to initial site surveys, a shortlist of sites was selected from a list vetted by local indigenous kaitiaki (i.e., Māori environmental stewards, Bennett et al., 2021) who provided guidance on suitable locations for the study (Western area, Ngāti Manuhiri; Eastern area, Ngāi Tai ki Tāmaki). Following the site vetting process and prior to the deployment of timelapse cameras, sites were surveyed using a remotely operated vehicle (BlueROV2) to qualify site characteristics and identify any existing habitat (e.g., shellfish or rocky reef) that may influence the outcome of the study.

To account for differences in site characteristics that can influence mussel loss and predator abundance at different sites during the study period, six site parameters were included ( Supplementary Table 1 ):

To account for differences in environmental variables that may influence mussel loss and predator abundance at sites during the study period (Perry et al., 2018; Rueda et al., 2019), four environmental parameters (based on daily totals and means) were compared among deployments:

Prior to each experimental translocation, adult mussels (shell length (SL) 91.5 ± 9.7 mm SD) sourced from long-line aquaculture farms were completely detached from growth substrate (similar to how they would be harvested for restoration) and separated into groups of 100 mussels each. All groups were taken from seawater, held in tanks circulating unfiltered seawater, and deployed to the seafloor within 48 hours of harvest. At each of the eleven experimental sites, three groups of mussels were placed at individual patches with 50 m spacings in between (300 mussels per site). This distance was considered sufficient to maintain similarities in site characteristics (e.g., depth, substrate type, predominant current directions) while minimising the potential for autocorrelation in fish communities among experimental groups within sites (Mellin et al., 2010; Figure 2 ).

A programmable timelapse camera was attached to a metal stake and placed 0.5 m away from each experimental mussel group (three per site) and 0.25 m above the seafloor (Kaiser Baas X450 4K Action Camera, 14 MP; based on the design specifications of CoralCam, Greene et al., 2020). This field of view allowed cameras to capture images of predator and non-predator species in the vicinity of mussel groups. This arrangement also helped to standardise depth-of-field and minimise any potential hydrodynamic disturbance to the mussels. All cameras were programmed to take one image every 10 minutes during peak daylight hours (06:30 – 18:30) for the first four days following placement on the seafloor. Due to the logistical constraints of overnight timelapse photography, images were only gathered during daylight hours and did not account for nocturnal predator activity.

Following each translocation period, all remaining mussels were recovered. Mussel survival was obtained for each group by dividing the number of live mussels at recovery by the total number of mussels at deployment (n = 100). Due to inclement weather limiting site access, deployment lengths ranged between one – two weeks. To account for differences in deployment lengths, mussel survival was divided by the number of days to obtain mussel loss per day (similar to Alder et al., 2020).

Image analysis followed a similar protocol to Alder et al., 2022. All captured digital images were assessed chronologically and categorised based on depth-of-field by the same reviewer. For both turbidity-obscured and clear images, all visible mobile species were counted and identified to the lowest taxonomic level. Counts followed a MaxN approach (similar to Priede and Merrett, 1996) where all identifiable species present within 1 m of the camera (i.e., just beyond the extent of experimental mussel groups) were recorded. For digital timelapse images, daily species counts were calculated for each site by summing the total counts of each species captured across the 228 images taken each day (i.e., one picture per camera every 10 minutes between 06:30 and 18:30, three cameras per site). Mobile species were classified as potential predators or non-predators of mussels based on previous knowledge of feeding preferences. Previous work (Alder et al., 2020, 2022) considered the following as potential predators: Australasian snapper (Pagrus auratus), spotted estuary smooth-hound (a.k.a rig shark, Mustelus lenticulatus), and the Sydney Octopus (Octopus tetricus). Non-predators included: White trevally (Pseduocaranx dentex), Parore (Girella tricuspidata), Porcupine fish (Tragulichthys jaculiferus), reef squid (Sepioteuthis lessoniana), velvet leatherjacket (Meuschenia scaber), yellow-eyed mullet (Aldrichetta forsteri), and spotties (Notolabrus celidotus) (Russell, 1983; Anderson, 1987; Michael, 1993; Anderson, 1997; Sazima, 1998; Raubenheimer et al., 2005; Grober-Dunsmore et al., 2007; Kendrick and Francis, 2010; King and Clark, 2010; Hauraki Gulf Forum, 2023).

At the end of the translocation period, corrected mussel loss was significantly different among sites (Test statistic (10) = 25.921, p = 0.004). Pairwise comparisons revealed a two-fold higher daily loss among most eastern sites compared with western sites ( Figure 3 ).

A total of 9665 digital timelapse images were captured across three separate deployments and 11 sites between October and November 2022. Of the images gathered, 4993 were classified as clear and 4672 were classified as turbidity-obscured images. From all of the clear images with identifiable mobile species, a total of 1212 individuals from 10 different species were counted ( Supplementary Table 2 ). For all the camera observations of mobile species, 1142 were of potential predator species and 70 were of non-predator species, with Australasian snapper accounting for 93% of the individuals observed ( Supplementary Figure 2 ). Comparisons of community compositions based on hourly counts detected significant differences among sites (square root transformation, pseudo-F_10,1065 = 15.134, p = 0.0001). Pairwise tests detected similar patterns in species assemblages across most sites, with generally higher predator abundances detected at eastern sites compared to western sites ( Figure 4A ).

There were no clear daily patterns that emerged in the total daily counts of all predator species among sites. The proportion of turbidity-obscured images were significantly different among sites (square-root transformation, F_10,148 = 62.6, p = 0.001), days (F_4,148 = 34.5, p = 0.001), and the interaction between the two (F_40,148 = 2.0, p = 0.004; Figure 4B ). Despite a high variation in the number of turbidity-obscured images among sites and areas, a positive correlation was detected when the number of turbid images were compared against summed daily species counts, suggesting a positive relationship between turbid sites and predator presence (Pearson’s r = 0.412, p = 0.001; Figure 4B ). Post-hoc analyses revealed a significantly higher proportion of turbidity-obscured images were recorded for eastern sites compared with western sites.

Comparisons of environmental parameters among sites demonstrated that differences in the number of turbid images, depth, and effective fetch contributed to most of the variation across both areas ( Figure 5A ). Comparisons among normalised daily environmental parameters and square-root transformed species counts revealed little correlation (RELATE, rho = 0.067). Of the environmental parameters examined, total rainfall, moon phase, the proportion of turbid images, and water velocity were most closely related to patterns in species abundance (BEST correlation = 0.239). Overall, environmental parameters explained 41.3% of the total variation observed in patterns of species counts across all sites (DistLM). A distance-based redundancy analysis (dbRDA) indicated that the turbid images (Pseudo-F = 5.86, p = 0.005), moon phase (Pseudo-F = 8.94, p < 0.001), and water velocity (Pseudo-F = 7. 76, p < 0.001) helped drive 32.5% (dbRDA1) and 8.8% (dbRDA2) and of the total variation in daily community compositions ( Figure 5B ).

An assessment of the site-specific factors used to predict mussel survival (i.e., daily mussel loss per group) following transfer to the seafloor revealed that mussel survival was positively correlated with water velocity (β = 0.19, 95% CI [0.15, 0.24], t₍₁₅₆₎ = 8.15, p < 0.001), and negatively correlated with the total number of predators (β = -0.09, 95% CI [-0.16, -0.01], t₍₁₅₆₎ = -2.28, p = 0.024), and the proportion of turbidity-obscured images (-0.07, 95% CI [-0.10, -0.03], t₍₁₅₆₎ = -3.50, p < 0.001).

An assessment of the deployment-specific factors used to predict mussel survival (i.e., mean daily mussel loss per site) following transfer to the seafloor revealed that mussel survival was negatively correlated with total rainfall (β = -0.17, 95% CI [-0.19, -0.05], t₍₄₃₎ = -3.96, p << 0.001), moon phase (β = -0.95, 95% CI [-1.08, 0.27], t₍₄₃₎ = -4.01, p << 0.001), and tidal range (β = -10.83, 95% CI [-7.40, 2.29], t₍₄₃₎ = -3.23, p = 0.001).