Accurate abundance estimation of cliff-breeding Bounty Island shags using drone-based 2D and 3D photogrammetry

. Effective seabird management strategies rely on accurate population estimates, with previous methods typically employing ground counts of a target species. However, difficult and often inaccessible breeding habitats are now able to be explored due to recent technological advancements in Unoccupied Aerial Vehicles (UAVs). This study tested a novel approach by combining high-resolution orthomosaics and 3D models to provide population estimates of the remote cliff-breeding Bounty Island shag ( Leucocarbo ranfurlyi ) on the sub-Antarctic Bounty Islands in November 2022. Our results report 573 breeding pairs, estimating a total population of approximately 1733 birds, breeding on 13 of the 14 main islands. Given the topographical constraints of surveying the islands by boat, the most comparable assessment in 1978 shows a similar count of breeding pairs, proposing the Bounty Island shag population is stable. However, long-term monitoring and additional research surrounding foraging strategies is crucial for developing conservation efforts for one of the rarest and spatially restricted shag species in the world. Our study demonstrates a reproducible method for estimating elusive wildlife populations that can be used across species with wider applications


INTRODUCTION
Population and community biology is a core principle of conservation science (Simberloff 1988).The two main pillars are population dynamics over time, either described as observations of population size from surveys or modeled using demographic parameters (Berryman 2020) and population genetics (Hamilton 2021).Although the latter approach can provide profound insights into the current state of wild populations (e.g., Romiguier et al. 2014) as well as the historical trajectory a species followed over millennia (e.g., Cole et al. 2019), it is dependent on the accessibility of genetic samples and requires an overall more complex scientific approach (Gurdebeke and Maelfait 2002).For this reason, the determination of effective population size over time is generally the starting point for the assessment of the state of wild populations.Effective population size and its temporal trajectory are the key parameters for the ranking system employed by the International Union for the Conservation of Nature's Red List of Threatened Species (IUCN Standards and Petitions Committee 2022).
The accuracy of a species' threat ranking based on population sizes depends on the quality of the data reported, which in turn, depends on the methods employed.The determination of http://www.ace-eco.org/vol18/iss2/art6/population size in seabirds is a case in point.Although seabirds often breed in discrete locations at predictable times of the year, which facilitates population assessments greatly, their breeding habits may nevertheless make it difficult to obtain reliable population counts.For example, burrow breeding birds are generally not readily observable, making it virtually impossible to reliably count them (McKechnie et al. 2007).Species breeding in the open may still be inaccessible, be it because they breed at remote and inaccessible and therefore rarely visited locations or because the topographic properties of their breeding habitat make observations difficult.The Bounty Island shag (Leucocarbo ranfurlyi) provides a good example.
The Bounty Island shag is endemic to the sub-Antarctic Bounty Islands (47.7602°S, 179.0392°E), a sub-Antarctic offshore archipelago located ca. 700 km southeast of the New Zealand mainland.Due to their exposed setting and lack of suitable anchorage, the Bounty Islands are rarely visited, with only a handful of scientific expeditions having made landfall since the late 1970s (Taylor 2006).Accordingly, there is only limited knowledge about this species, except that it is one of the rarest shag species worldwide, with a population ranging between 368 and 1138 adults as estimated by a few population counts conducted opportunistically since the 1990s (Michaux 2017).Bounty Island shags breed in small colonies on cliff-side ledges or on skyline ridges and are distributed across all main islands of the archipelago (Robertson and van Tets 1982).The inaccessibility of their breeding habitat makes it difficult to obtain full population counts despite the species' small population size.
The only full census of the Bounty Island shag's population was conducted in 1978 when the distribution and number of shag nests were mapped using a combination of telescope observations and photographs taken by helicopter (Robertson and van Tets 1982).Since then, shag nest numbers have only been estimated in 1997, 2004, and 2011 using observations made from boats, which limits reliability of those counts (Michaux 2017) as many shag nests located higher up in the cliffs likely remained out of view of the observers operating at sea level.In 2013, a post-breeding count of Bounty Island shags was conducted (Hiscock et al. 2014) that serves as the baseline for the species' current Red List assessment (Westrip 2018).Ultimately, the trajectory of the Bounty Island shags' population developments remains uncertain.
Methods of conducting seabird surveys has advanced significantly, particularly in the past few years, especially thanks to the emergence of Unoccupied Aerial Vehicles (UAV or drones) as a consumer product (Vas et al. 2015).Although initially drones were used to take aerial photographs of the study site that either were analyzed individually (e.g., McClelland et al. 2016, Sardà-Palomera et al. 2017) or stitched to single multi-image panoramas (e.g., Ratcliffe et al. 2015), photogrammetry has become the standard method to process drone-based images into high-resolution orthomosaics that allow counting of target species (e.g., Oosthuizen et al. 2020, Dunn et al. 2021, Fudala and Bialik 2022).Orthomosaics are generated by point-wise analysis of a sequence of overlapping images that cover the target survey area; the process accounts for the camera lens properties and flight altitude and incorporates GPS coordinates to generate georeferenced high-resolution images (Barazzetti et al. 2010, Remondino et al. 2012).Furthermore, aerial triangulation of the image data allows us to obtain fine-scale information about how individual pixels matched in image sequences are oriented in 3D-space, so that digital elevation models (DEM) and textured 3D models can also be reconstructed from the drone images (Remondino et al. 2012).The topographical aspect of drone-based imagery also starts to be used in an ecological context, for example to examine the interplay of geomorphology and animal habitat use (e.g., McDowall andLynch 2017, Oosthuizen et al. 2020).
Here, we demonstrate that beyond offering further analytical parameters, 3D models also provide the means to revisit survey areas virtually and thereby gain visual access to habitat pockets that may remain obscured when viewed from above (drone topdown view) or below (e.g., boat-based observations).This is particularly relevant for cliff-breeding seabirds like the Bounty Island shag.We used a combination of animal counts on highresolution orthomosaics and a review of the identified sites of occurrence in 3D models to determine the exact number of shag nests on the Bounty Islands.Our aim was to develop a novel approach to obtaining accurate population estimates, particularly for fauna (and flora) occupying steep and difficult to access areas to better inform conservation management.

Study site and species
The Bounty Island group ("Bounties") consists of 14 islands as well as 11 islets and rocks that range in size from 0.4 to 12.7 ha with a combined landmass of 49.5 ha (erroneously reported as 135 ha by Land Information New Zealand (LINZ) 2022).This excludes four rocks completely awash in high seas (Fig. 1).The archipelago was first recorded on 19 September 1788 by Captain William Bligh and named after his ship the HMS Bounty a few months before the infamous mutiny (Bligh 1792).The archipelago consists principally of granite islands and rock stacks that rise steeply from sea level to heights of 40-60 m.Despite their barrenness and small size, the islands are inhabited by a recovering population of New Zealand fur seals (Arctocephalus forsteri) and are an important breeding hot spot for seabirds.
The Bounties host most of the world's breeding population of Salvin's albatross (Thalassarche salvini) (Sagar et al. 2015), nearly half of the global population of Erect-crested penguin (Eudyptes sclateri) (Wilson and Mattern 2019) as well as Fulmar prions (Pachyptila crassirostris) that breed on one other sub-Antarctic island outside the Bounties (Shepherd et al. 2022).The Bounty Island shag is the only endemic bird species on the archipelago.
The Bounty Island shag was first described in 1901 and is closely related to the Auckland Island shag (L.colensoi) (Rawlence et al. 2022).The first study to examine the biology and behavior was conducted in 1978 (Robertson and van Tets 1982) and to date provides the bulk of information about the species.The shags construct their nests primarily on cliff ledges using brown seaweed (Marginariella sp.); this nesting material is arranged in a circular fashion creating a nest bowl ca.35 cm in diameter (Robertson and van Tets 1982).The breeding season commences in October, with chicks hatching in late November and early December (Michaux 2017).

Drone surveys
Between 10 and 15 November 2022, we conducted drone surveys of the 14 main islands as well as seven rocks (Fig. 1).For each island, aerial imagery was collected with DJI Mavic 2 Pro drones (Shenzhen DJI Sciences and Technologies, Nanshan, Shenzen, China).This drone model contains a L1D-20c Hasselblad camera with a fixed 35 mm lens with an approximate field of view (FOV) of 77° that can record still images at a resolution of 5473x3648 pixels (20 Megapixels).The flight time of the drone was approximately 25-30 min.and varied with wind conditions.We used mission planning software DroneDeploy (https://www.dronedeploy.com),which also features a mobile app that enables autonomous flight of the drones along precalculated flight paths.Paths depend on flight altitude as well as desired front and side overlap of images in the mission sequence.Mission flight altitudes were chosen so that the drone would survey each island 40-60 m above the island's highest point.Higher flight altitudes were chosen for islands located >1.5 km from the drone launch site (Funnel and Prion Islands) as well as Depot Island, as flying missions at higher altitudes helps reduce flight times.A minimum altitude of 40 m reduced potential disturbance effects and ensured the drone would operate above the Salvin's albatrosses circling over most of the islands at any given time of the day (Rexer-Huber and Parker 2020).Depending on the relative altitude the drone operated over the respective islands, the ground sampling distance (GSD; i.e., the area on the ground a single pixel represents in the imagery) varied between 0.34 and 0.52 cm.A total of 14 missions were flown, some of which covered several islands and rocks at the same time.Missions flown at speeds of 3-4 m/s lasted between 9.8 and 54.2 min.and resulted in 188-1022 images per mission.
Except for the two drone missions flown over the Eastern Group, which launched from the deck of the research vessel, all other missions launched from a rock platform on the highest point of Proclamation Island.During take-off and landing, it took the drone about 5-10 sec.to transition between launch and mission altitude, quickly clearing the air space used by circling albatross.This reduced collision risk substantially; no collision or other drone-seabird interactions occurred.During all missions, direct line of sight was maintained with the drone by a dedicated spotter.Binoculars were used for missions flown at greater distances from the launch sites.No evasive manoeuvers were required during any of the missions.

Impact of drone surveys on shags
The impact of drone surveys on wildlife has been examined in some detail over the past decade (e.g., Brisson-Curadeau et al. 2017, Egan et al. 2020).Disturbance impacts of drones depend to a large degree on the chosen flight altitude (Weimerskirch et al. 2018).Studies examining the effects of drones flying over shag colonies found that a flight altitude of 20 m is tolerated (Parker and Rexer-Huber 2022).The altitudes of the missions flown in this study were two to three times higher, further reducing the potential for disturbance.Observations of shag behavior during the flight mission using the drone's live video feed did not show any behavioral reaction in the shags or any other wildlife.Observed behavior may not accurately reflect disturbance effects on a physiological level (e.g., Ellenberg et al. 2013).However, given that the ambient noise level generated by albatross and penguins completely masked the sound generated by the drone, and that the drone flew above a layer of volant seabirds circling the islands at any given time of the day, makes it likely that the shags remained unaware of the drone's presence.

Image processing and bird counts
Images recorded during the missions were used to generate georeferenced as well as textured 3D models of the islands (for details on the principle of photogrammetry refer to Barazzetti et al. 2010, Remondino et al. 2012).Data were uploaded to the cloud-based DroneDeploy service (https://www.dronedeploy.com/), which processed image data into orthomosaics, models, and DEM.Orthomosaics were subsequently used to conduct bird counts in the image annotation software DotDotGoose (version 1.6.0,https://github.com/persts/DotDotGoose).For each island, various object classes were identified, namely the species that were visible in the respective orthomosaics (i.e., shags, penguins, albatross, seals).DotDotGoose overlays a grid of definable dimensions over the image, and objects can be marked with a color-coded dot denoting the different classes (Fig. 2).Distinguishing shags and penguins is possible using several characteristics, such as the spindle-shaped body contour as well as presence of crests on the head of penguins, or the wedge-shaped tail and long slender neck of shags.During analysis, all identifiable wildlife (i.e., Bounty Island shags, Erect-crested penguins, Salvin's albatross, Fulmar prions, NZ fur seals) were counted; here, we report only on the Bounty Island shag portion of the data.Once all objects were marked, the point data were exported as pixel coordinates in the orthomosaics.These pixel coordinates were back converted to geographic coordinates using a tool provided by the DotDotGoose developers (https://github.com/persts/DotDotGoose/tree/main/extra_tools).Each orthomosaic was annotated fully by three different observers, and resulting point data were imported for further analysis into ArcGis Pro (version 3.0.3,Esri Inc., Redlands, California, USA).

Mapping of shag nests
Although orthomosaics are of very high detail, resulting point data occasionally contained misclassifications (e.g., shags classified as penguins) or false positives (e.g., shadows or rocks mistakenly tagged as shags).Therefore, count data were verified by using a two-pronged approach.Firstly, points identifying the same animal class placed by the three observers within 15 cm were combined using the three points' centroid coordinate as the final object location and accepted as correct.Secondly, points placed only by one or two observers were isolated and manually validated by visually confirming or rejecting that the marked object was a shag.Rejected points were removed from the data set.
In the second step of the analysis, all accepted points were manually reviewed to determine the number of shag nests.The circular extent of the nest halo is quite visible on the orthomosaics due to the strong contrast of the brown nesting material on the light gray or white rocks (see Fig. 2).Any point marking a bird in such circumstances was counted as a nest.
Orthomosaics represent a top-down view of the scenery below.As a result, any areas or objects obscured by overhangs remain invisible on the orthomosaics.This is particularly relevant in the case of the Bounty Island shag as many nests are tucked away under rock overhangs and alcoves that remain invisible when seen directly from above.This problem was circumvented by inspecting textured 3D models of the study area (Fig. 3); DroneDeploy offers a web browser-based 3D viewer for this purpose.Each island was inspected in 3D view along its entire circumference.Any nests and shags not apparent on the orthomosaics were mapped manually in ArcGIS Pro by adding points at the approximate location as determined by rotating the 3D view to top-down view and identifying details of the topography directly over the nest or bird.After review of all island models, the resulting point feature representing all locations of nests and individuals (i.e., visible and obscured) provided the final count for each island.

Spatial analysis
The point feature representing shag nests was used to conduct spatial analyses of both their vertical and horizontal distributions.The elevation of each nest above sea level was derived from the island's respective DEM rasters via the "Extract Value By Point" function in ArcGIS Pro.Clusters of nests occurred mostly on distinctive ledges with physical boundaries (i.e., cliff edge, back wall) that were readily identifiable in the orthomosaics.Shag nests were lined up either single file or in a zigzag pattern along the respective ledge's back wall.A line feature was generated that followed the back wall's trajectory, allowing it to determine its length.In a few instances, shag nests located on broader platforms appeared to be distributed more evenly (e.g., Fig. 2).However, upon inspection in the 3D model it became apparent that nests were still lined up along vertical rock features providing a back wall.Every such "back wall" was traced in ArcGIS as a line feature to determine its length.For each shag nest, the distance to the nearest nest was determined by running the "Near" function in ArcGIS Pro.For each island, "nest spacing" was defined as the median nest distance.The sum of back wall lengths was defined as "total ledge space."The theoretical nest capacity for each island was then calculated as the quotient of total ledge space and nest spacing.

RESULTS
Bounty Island shags were found to be present on 13 of the 14 main islands (Table 1; Fig. 4).The only island not to feature shags was the southernmost island in the Center Group, Castle Island; no other species of bird or seals were found to be present on that island either.A total of 910 shags were identified on orthomosaics and 3D models.Of these, 573 birds were sitting on a nest; the remaining 337 individuals were either in attendance of their incubating partner or roosting by themselves.
Eighty-three nests, 15% of the entire nest count, were only visible when islands were inspected in 3D view.On Prion Island, nearly half (44%) of the shag nests became apparent only when the 3D model of the island was examined.On Coronet Island, an entire ledge containing 39% of the island's shag nests was completely obscured by overhanging ledges also occupied by shags (Fig. 3).Graph is organized to reflect the three main island groups, Main Group (blue), Center Group (green), and East Group (yellow).
The only islands where all shag nests were visible on orthomosaics were Depot, Penguin, Proclamation, Ruatara, and Tunnel Islands, which also happen to be the islands with the lowest shag nest counts of the entire archipelago (Table 1).
North Rock (138 nests) and Lion Island (112 nests) host nearly half of the species' breeding population (43%).Other important breeding islands are Molly Cap, Prion, Coronet and Ruatara Island (42-56 nests; see Table 1).Except for Molly Cap and Ruatara, these islands host substantially lower populations of other wildlife, with Bounty Island shags being the sole occupants of Coronet Island (Fig. 3).
Bounty Island shag nest aggregations principally occur with northern, eastern, or southern orientations.Except for two nests in the southwest of Funnel Island, Bounty Island shags did not occupy any west-facing cliffs (Fig. 4).The ledges occupied by the birds for breeding were exclusively found at elevations >20 m (Fig. 5).Distances between shag nests range between 0.8 and 1.2 m (Table 1; Fig. 6), the exception being Proclamation Island, where limited ledge space results in low number of nests that occur fragmented across the northern face of the island (Fig. 4).
Considering nest spacing in relation to the available ledge space, there would be room for a total of 812 Bounty Island shags nests.This would mean that the current breeding population occupies around 70% of the available Bounty Island breeding habitat (Table 1).

DISCUSSION
Bounty Island shag population assessment Using counts from a combination of high-resolution orthomosaics and textured 3D models, this study reliably determined the total population of the endemic Bounty Island shag to consist of 573 breeding pairs.The current estimate for the number of mature individuals listed by the Fig. 6.Nest spacing of Bounty Island shags on the different islands across the archipelago.Graph is organized to reflect the three main island groups, Main Group (blue), Center Group (green), and East Group (yellow).
IUCN Red List ranges between 874-975 mature individuals (Westrip 2018).As each nest represents a pair of shags, our survey indicates that there were at least 1144 mature individuals on the Bounty Islands.Using the Red List estimation that mature individuals usually comprise about two-thirds of a total population (Westrip 2018), the 2022 count would result in an estimate for the total Bounty Island population of around 1733 birds.The current Red List estimate (1386.5±75.5 birds) is primarily based on counts of birds in February 2013 that were conducted from a rigid-hull inflatable boat circumnavigating the islands of the archipelago over the course of 2.5 h (Hiscock et al. 2014).As such, our study provides a substantially more reliable assessment of the Bounty Island shag population.The only estimate of the breeding population comparable to our study dates back to 1978.Using a combination of photographs taken by helicopter and telescope observations made from Proclamation Island, Robertson and van Tets (1982) estimated a total of 569 breeding pairs of shags to be present that year.This estimate is remarkably close to the one derived in our study, which would suggest no significant change in the Bounty Islands shag population size over the past 45 years.Other attempts of estimating nest numbers were conducted in 1997 (120 pairs; Clarke et al. 1998) and2004 (359 pairs;De Roy and Amey 2004).
Although these figures at a first glance seem to suggest wild fluctuations in the Bounty Island shag breeding population, these likely stem from the fact that those estimates were based on opportunistic nest counts made from small boats.Given the birds' tendency to establish nests that are well tucked away toward the ledge backing, it seems safe to assume that a substantial number of nests were missed because they remained obscured when seen from sea level.
The general distribution of shags determined in this study closely resembles distribution maps provided in Robertson and van Tets (1982) and Clarke et al. (1998).However, there appear to be some differences in shag nest numbers when breeding locations are compared.No breeding shags were reported on either Ruatara or Penguin Island in 1978, both islands now contain a significant number of shag nests (n = 64; see Table 1), which represent about 11% of the entire Bounty Island shag population.In 1978, Lion Island was the single greatest stronghold in terms of nest numbers for the species, with 165 nests (Robertson and van Tets 1982), whereas we only counted 112 nests.North Rock, while already an important breeding site 45 years ago with 71 shag nests, had nearly twice as many shags breeding on it in 2022 (n = 138).Overall, it appears as if shag nests numbers have decreased predominantly on the Main (1978 vs. 2022: 304 vs. 255 nests, net difference: -16%) and Center groups (154 vs. 124 nests, -19%), whereas nest numbers increased on the East Group (111 vs. 194 nests, +75%; Fig. 7).Given the nearly identical population estimates in 1978 and 2022, displacement by other wildlife appears to be possible factor behind the inter-island changes in numbers.

Redistribution of shag nests across archipelago
In the late 18th century, the Bounty Islands were host to a substantial population of around 52,000 New Zealand fur seals (Taylor 1982).Sealing efforts, particularly in the year 1808-1809 when >50,000 seals were killed, almost extirpated the species from the archipelago in the early 19th century.Sealing on the Bounties ceased completely in 1891, and the fur seal population has been recovering slowly ever since, with most recent estimates ranging around 21,500 individuals (Taylor 1996).The greatest concentrations of seals occur on the Main Group as well as on Prion Island, and to a lesser extent, on Funnel Island of the Center Group (Taylor 1982(Taylor , 1996)).On Molly Cap and North Rock, we found very few seals present (one and six individuals, respectively, Mattern 2023).Fur seals are capable climbers, and especially younger animals often seek out narrow ledges in steep cliffs that allow them to rest away from dominant bulls guarding their harems on flatter areas of the islands (e.g., Taylor 1982, Goldsworthy andShaughnessy 1994).The change in nest numbers between the different islands may, therefore, be related to increasing numbers of seals occupying accessible shag breeding habitat.Similarly, it was assumed that increasing seal presence caused a redistribution of penguin nests on Proclamation Island (Mattern et al. 2021).However, compared with penguins, which may face reduced breeding habitat on some islands if seal numbers continue to increase, our data indicate that there still is ledge habitat available for shags to occupy.Bounty Island shags' preference for ledge breeding provides them with exclusive breeding habitat across the Bounty archipelago.
Many of the ledges are only accessible from the air which excludes penguins and seals from competing with shags for the space.Moreover, the narrowness of these ledges makes them difficult for large birds like albatross to land.Therefore, competition for nest space will only be intra-specific.If nest spacing is an appropriate proxy for available nesting habitat, there would be room for a breeding population of up to 800 pairs of Bounty Island shag (Table 1).Although there are probably other factors besides ledge space determining actual nest capacity such as local microclimate or subtle changes in ledge slope (e.g.Oosthuizen et al. 2020), it is still safe to assume that available nesting habitat is currently not a limiting factor for the Bounty Island shag population.
Judging by the sparse data available, the population of the Bounty Island shag appears to be stable.This, in turn, could indicate that the species is operating at the Bounty Islands marine environment's carrying capacity.Very little is known about the species' foraging requirements.Robertson and van Tets report stomach contents of nine shags that, besides fish and cephalopods, consisted predominantly of prey found at the seafloor (e.g., snails, sea urchins, hermit crabs), suggesting at least a partly benthic foraging strategy.No information is available on the species' diving behavior or that of any other related shag species in New Zealand (Miskelly 2022).However, similar-sized Antarctic shags (Phalacrocorax atriceps) have been recorded to dive to depths of up to 115 m (e.g., Croxall et al. 1991, Casaux et al. 2001).Even assuming that Bounty Island shags can forage at similar depths, the available benthic foraging habitat is fairly restricted as the area inside the 100 m depth contour around the islands is limited to approximately 32 km² (Mitchell et al. 2012).In this light, available foraging habitat may be a limiting factor for the Bounty Island shag population.
Although our study does not indicate significant changes in the Bounty Island shag population in the past 45 years, it is nevertheless vital to keep a close eye on future developments of the species.With an estimate around 1,700 birds, the Bounty Island shag is one of the rarest and spatially most restricted shag species in the world.Any change in the population numbers could be indicative of significant problems that may be alleviated through conservation efforts.With such a small population size, it is vital to adopt monitoring practices that provide the most accurate estimates possible, while using methods that are clear, transparent, and precisely reproducible.The methods we employed in this study address both these requirements.

2D and 3D photogrammetry improve survey quality
Our study is the first to use a combination of high-resolution orthomosaics and textured 3D models to count seabirds.The use of drone-based surveys has seen a marked increase in recent years (Hodgson et al. 2018, Pfeifer et al. 2019, Bishop et al. 2022).The general approach to such drone surveys is to record sequences of top-down (nadir) photographs and subsequently generate composite, high-resolution images of the survey area either by stitching images (Mattern et al. 2021, Dunn et al. 2021, Parker and Rexer-Huber 2022) or using dedicated photogrammetry solutions (e.g., Oosthuizen et al. 2020, Hayes et al. 2021, Fudala and Bialik 2022).The latter approach, although computationally more demanding, provides significantly better results than image stitching as the resulting orthomosaics are orthorectified and georeferenced, allowing accurate spatial analysis (Hayes et al. 2021).Moreover, photogrammetry provides significant ancillary data such as digital elevation maps, which in turn allow the threedimensional reconstruction of a survey area's topography, and ultimately, textured 3D models that combine orthomosaic with a topographic mesh (Barazzetti et al. 2010, Remondino et al. 2012).
3D models effectively provide a post hoc tool and verification method for field-based surveys.Study sites can be revisited virtually and without the constraints of a fixed vantage point, which is a significant limitation, particularly in boat-based surveys (Bishop et al. 2022).As our study shows, the commonly used top-down view of the study area can considerably underestimate occupation in steep cliffs and overhangs.We encourage using the described 3D model approach when estimating animal population in topographically challenging settings.Barazzetti, L., M. Scaioni, and F. Remondino. 2010.Orientation and 3D modelling from markerless terrestrial images: combining accuracy with automation.The Photogrammetric Record 25:356-381.https://doi.org/10.1111/j.1477-9730.2010.00599.x

Fig. 1 .
Fig. 1.Overview of New Zealand sub-Antarctic region (top panel) and the main study site, the Bounty Islands (bottom panel).

Fig. 2 .
Fig. 2. Orthomosaic of North Rock, Bounty Islands, generated from drone images (top panel) and working view of wildlife counts using the software DotDotGoose.Point annotation depicted in the lower panel is incomplete to provide clear views on not yet counted target objects.White rectangle in top panel indicates location of detail visible in software window.

Fig. 3 .
Fig. 3. 2D nadir (top panel) and 3D (bottom panel) views of the northeast end of Coronet Island, Bounty Islands, New Zealand (see also Fig. 1) showing Bounty Island shag nest on rock ledges.Yellow lines in the 3D view indicate areas with shag nests not visible in nadir view due to overhanging rock ledges.Twenty-five out of a total of 65 shag nests only became apparent when inspecting the 3D model of the island.The 3D model can be interactively explored at https://skfb.ly/oDSnw.

Fig. 4 .
Fig. 4. Distribution of Bounty Island shags across the archipelago.Red dots mark nests, yellow dots roosting birds.Outline and topography derived from Digital Elevation Models created via photogrammetry.Isolines represent 10 m height contour intervals.Note the different scales of the individual panels.An interactive map can be explored at https://www.tawaki-project.org/bounty-island-shags/.

Fig. 5 .
Fig. 5. Vertical distribution of Bounty Island shags on the different islands across the archipelago.Graph is organized to reflect the three main island groups, Main Group (blue), Center Group (green), and East Group (yellow).

Fig. 7 .
Fig. 7. Changes in Bounty Island shag nest numbers across the archipelago between 1978 and 2022.Green bars indicate increases, red bars decreases in nest numbers in November 2022 when compared with data from 1978 (Robertson and van Tets 1982).

Table 1 .
Bounty Island shag nest characteristics on all surveyed islands of the Bounty Island archipelago, New Zealand in November 2022.Bottom row provides a summary of the main table rows.Theoretical nest capacity is derived from median observed nest spacing and available ledge space for each island.Ranges are provided in parenthesis for median values.