An automated work-flow for pinniped surveys: A new tool for monitoring population dynamics

Abstract

Detecting changes in population trends depends on the accuracy of estimated mean population growth rates and thus the quality of input data. However, monitoring wildlife populations poses economic and logistic challenges especially in complex and remote habitats. Declines in wildlife populations can remain undetected for years unless effective monitoring techniques are developed, guiding appropriate management actions. We developed an automated survey workflow using unmanned aerial vehicles (drones) to quantify the number and size of individual animals, using the well-studied Scandinavian harbour seal (Phoca vitulina) as a model species. We compared ground-based counts using telescopes with manual flights, using a zoom photo/video, and pre-programmed flights producing orthomosaic photo maps. We used machine learning to identify and count both pups and older seals and we present a new method for measuring body size automatically. We evaluate the population’s reproductive success using drone data, historical counts and predictions from a Leslie matrix population model. The most accurate and time-efficient results were achieved by performing pre-programmed flights where individual seals are identified by machine learning and their body sizes are measured automatically. The accuracy of the machine learning detector was 95–97% and the classification error was 4.6 ± 2.9 for pups and 3.1 ± 2.1 for older seals during good light conditions. There was a clear distinction between the body sizes of pups and older seals during breeding time. We estimated 320 pups in the breeding season 2021 with the drone, which is well beyond the expected number, based on historical data on pup production. The new high quality data from the drone survey confirms earlier indications of a deteriorating reproductive rate in this important harbour seal colony. We show that aerial drones and machine learning are powerful tools for monitoring wildlife in inaccessible areas which can be used to assess annual recruitment and seasonal variations in body condition.

120‐years of ecological monitoring data shows that the risk of overhunting is increased by environmental degradation for an isolated marine mammal population: The Baltic grey seal

“…The study of other indicators of population development, such as pup production and body condition in seal colonies, could provide valuable insight into the best application of resource limitation in population modelling (Bowen et al, 2020;Infantes et al, 2022).…”

Growth and opportunities for drone surveillance in pinniped research

“…With visible‐light aerial imagery, deep learning techniques have already been applied to estimate aggregate pinniped counts (Hoekendijk et al 2021), detect individual pinnipeds (Dujon et al 2021), and classify pinnipeds by age class (Salberg 2015, Infantes et al 2022), though success and generalisability vary widely between examples. Upcoming applications also include deep learning for photogrammetry, as has already been demonstrated with drone‐based photography of cetaceans (Gray et al 2019) and more recently with harbour seals (Infantes et al 2022), and deep learning for individual identification, as has been demonstrated with ground‐based photography of harbour seals (Nepovinnykh et al 2018, 2022, Birenbaum et al 2022). In this early stage of its technological deployment, deep learning for computer vision remains an experimental technique in pinniped research, and still few examples characterise its error and generalisability across large‐scale applications.…”

“…Some current drone applications with pinnipeds leverage thermal or multispectral imagery to facilitate detection by high contrast in drone imagery (Seymour et al 2017, Sweeney et al 2019, Larsen et al 2022b), but many more studies rely exclusively on visible‐light photography to detect pinnipeds. With visible‐light aerial imagery, deep learning techniques have already been applied to estimate aggregate pinniped counts (Hoekendijk et al 2021), detect individual pinnipeds (Dujon et al 2021), and classify pinnipeds by age class (Salberg 2015, Infantes et al 2022), though success and generalisability vary widely between examples. Upcoming applications also include deep learning for photogrammetry, as has already been demonstrated with drone‐based photography of cetaceans (Gray et al 2019) and more recently with harbour seals (Infantes et al 2022), and deep learning for individual identification, as has been demonstrated with ground‐based photography of harbour seals (Nepovinnykh et al 2018, 2022, Birenbaum et al 2022).…”

“…2D and 3D measurements (reviewed in Hodgson et al 2020), but drone-specific applications remain few and experimental. First attempts have demonstrated success with 2D measurements from single photographs(Krause et al 2017, Alvarado et al 2020), 2D measurements from orthomosaics (Fig.4;Allan et al 2019, Hodgson et al 2020, Infantes et al 2022, and 3D measurements from structure-from-motion models(Hodgson et al 2020, Shero et al 2021. Such studies generally require validation against conventional ground-truth measurements with captured animals to confirm the veracity of photogrammetric methods(Krause et al 2017, Alvarado et al 2020) -though, at sufficient sample sizes, UAS-derived measurements have been validated against archival ground-truth measurements of a comparable sample(Allan et al 2019).…”

An unmanned aerial vehicle pipeline to estimate body volume at scale for ecological monitoring

“…Using UAVs is cheaper and quicker than traditional monitoring methods (Jackson et al, 2022) and can be applied over large spatial scales (Bogdan et al, 2021). To date, though, UAVs have been used for 2D size estimates of individual animals, including at the population scale (Gray et al, 2019; de Kock et al, 2021; Infantes et al, 2022). However, 3D body size (volume) estimates can also be estimated using UAVs by structure-from-motion photogrammetry (Westoby et al, 2012), taking a series of overlapping images from a known altitude that are stitched to create a geometrically accurate 3D model (Hodgson et al, 2020; Shero et al, 2021).…”