Marine habitat mapping of green turtle (Chelonia mydas) foraging grounds in the Northern Territory, Australia
Natalie Robson
A * , Carol Palmer A , Sam Banks A , Michele Thums B , Alana Grech C , Joanna Day D E , G , Dylan Cooper G and Garnet Hooper H A
B
C
D
E
F
G
H
Abstract
Green turtles (Chelonia mydas) are listed as vulnerable in Australia and are culturally significant to Indigenous Traditional Owners. However, their foraging habitats remain poorly understood, particularly in northern Australia.
To map green turtle foraging habitats in the Northern Territory, through collaboration with Traditional Owners and ranger groups to support Sea Country management.
Visual classifications of towed video transect data were used in a Support Vector Machine Learning Model to predict habitat across 379 km2 of remotely sensed satellite imagery, encompassing two green turtle foraging grounds within jointly managed parks, namely, Trepang Bay (Garig Gunak Barlu Marine Park) and Field Island (Kakadu National Park).
Foraging turtle habitat; algae and seagrass made up 30% of the Trepang Bay and 18.05% of the Field Island foraging areas. The classification accuracy of the model showed a high level of agreement at both sites (0.63 and 0.75 respectively).
These habitats provide good foraging grounds for green turtles and support different age classes for various behaviours, including resting and predator avoidance.
The simple and repeatable field methods used in this study allow for ongoing monitoring by ranger groups. The findings will support conservation planning and management in the Northern Territory.
Keywords: collaborative, ecology, feeding, management, rangers, remote sensing, seagrass, towed-video.
Introduction
The management of marine ecosystems has been shifting from a traditional single-species approach (Department of Agriculture, Water and the Environment 2022) towards an ecosystem-based management approach (Olsson et al. 2008; Fletcher et al. 2011; Smale et al. 2012; Smith et al. 2017). Ecosystem-based management aims to maintain the health and resilience of marine ecosystems by considering the interactions among species, habitats, and human activities. Habitat mapping is a crucial first step towards ecosystem-based and species management (Cogan et al. 2009; Guarinello et al. 2010).
Globally, many species of migratory marine megafauna are in decline, with their survival increasingly threatened by habitat loss and environmental changes (Bearzi et al. 2006; Sequeira et al. 2019; Lin et al. 2023). Sea turtles have suffered some of the most severe declines in abundance because of anthropogenic threats (Fuentes and Cinner 2010; Fuentes et al. 2023). However, global conservation efforts have led to an increase in sea turtle numbers in many populations (Hays et al. 2025). Providing adequate protection for these species requires an understanding of their habitat use, which is often provided by satellite telemetry studies (Schofield et al. 2013; Hays et al. 2018; Ferreira et al. 2021, 2023; Fossette et al. 2021; Peel et al. 2024). However, information on what constitutes the underlying habitat is largely lacking, presenting a significant challenge in generating evidence-based solutions.
Management and recovery plans for marine turtles tend to focus primarily on nesting grounds. However, marine turtles spend most of their lives at their foraging areas, only migrating to nesting grounds every 2–5 years on reaching maturity (30–50 years old) (Department of the Environment and Energy 2017). Turtles also display strong site fidelity to these foraging areas (Avens et al. 2003; González Carman et al. 2016; Shimada et al. 2016, 2020).
Green turtles (Chelonia mydas) are considered ‘endangered’ worldwide by the IUCN Red List (Seminoff 2023); however, there has been an increase in many local populations because of conservation efforts (Hays et al. 2025). Juvenile and adult green turtles are primarily herbivorous, foraging on algae, seagrass and occasionally mangroves (Mortimer 1981; Limpus and Limpus 2000; André et al. 2005; Holloway-Adkins and Dennis Hanisak 2017; Hays et al. 2018; Díaz-Abad et al. 2022). However, there is considerable variation in green turtle diets globally and within foraging grounds (Esteban et al. 2020). This diet means they depend on coastal habitats for their foraging grounds, often preferring to forage in shallow waters of ≤5 m (Mortimer 1981; André et al. 2005; Hazel et al. 2009; Reisser et al. 2013; Madeira et al. 2022).
In Australia, green turtles are listed as ‘vulnerable’ and ‘migratory’ under the Commonwealth Environment Protection and Biodiversity Conservation Act 1999 (EPBC 1999). However, their population status in the Northern Territory is unknown, and they are not listed under the Territory Parks and Wildlife Conservation Act 1976 (TPWC 1976). Green turtles are also a species of high cultural importance to coastal Traditional Owners throughout northern Australia. They appear as totems and in Dreamtime stories (often represented in Aboriginal art) (Butler et al. 2012; Department of the Environment and Energy 2017) and the meat of turtles and eggs from all species are an important food source (Kennett et al. 2004; Department of the Environment and Energy 2017; Delisle et al. 2018). Indigenous ranger groups are traditional custodians of the extensive Sea Country in the Northern Territory (Kennett et al. 2004; Rist et al. 2019). The term ‘Sea Country’ is used by Indigenous Australians to refer to any environment within their broader traditional estate that is associated with the sea or saltwater, including coastal areas, estuaries, beaches, marine areas, and islands (Ens et al. 2012; Dam Lam et al. 2019). Since 2008, Traditional Owners in the Northern Territory have been afforded legal rights to the intertidal zone of Aboriginal Land under the Aboriginal Land Rights (Northern Territory) Act 1976.
The northern coast of Australia presents a significant gap in our contemporary understanding of marine benthic habitats. Much of the coastline is very remote, and environmental factors such as high turbidity, very strong and variable tidal currents, and dangerous animals such as saltwater crocodiles (Crocodylus porosus) and box jellyfish (Cubozoa), have made marine research in this region difficult. In the Northern Territory, very little mapping has been undertaken outside of Darwin Harbour (Department of Environment and Natural Resources 2005; Geo Oceans 2011; Galaiduk et al. 2019; O2 Marine 2019). Given these restraints, remote sensing presents a valuable tool for mapping and monitoring the distribution of coastal marine habitats in the region (Monk et al. 2008; Wicaksono et al. 2019; Schill et al. 2021; Kuhwald et al. 2022; Wilson et al. 2022; AECOM, see www.aecom.com).
The integration of remote sensing with ground-truthed underwater video and photos is a widely used approach for identifying and mapping marine habitats (Hedley et al. 2012; Kuhwald et al. 2022; Wilson et al. 2022; AECOM, see www.aecom.com). Although remote sensing cannot provide the accuracy and level of detail that field surveys deliver, because it does not provide habitat classifications, it can be used to assess large-scale spatial patterns in habitat cover. Ground-truthing is used to provide that higher level of detail and is used to train a supervised image classification. The ground truth data can also then be used for determining the accuracy of the classification (Lyons et al. 2013; Komatsu et al. 2020; Kutser et al. 2020). This method is also invaluable for monitoring changes to marine habitats caused by climate change and other anthropogenic threats (Andréfouët et al. 2002; Hedley et al. 2012; Hovey and Fraser 2018).
Many different classification schemes exist for analysing marine habitat data globally, such as the European Nature Information System (EUNIS) and the Coordination of Information on the Environment and the Barcelona Convention classification (Montefalcone et al. 2021). In Australia, local examples include the News South Wales National Intertidal–Subtidal Benthic Habitat Map (NISB) ( Department of Climate Change, Energy, the Environment and Water 2023), and the Combined Biotope Classification Scheme (CBiCS) (Flynn 2018). A key feature of the CBiCS approach is that it enables the mapping and monitoring of benthic assemblages in conjunction with associated environmental and habitat features, which are classified as biotopes. This approach is commonly used internationally for marine habitats (Olenin and Ducrotoy 2006; Hooper et al. 2009; Monteiro et al. 2021). To support habitat classification from underwater imagery, the Commonwealth Scientific and Industrial Research Organisation (CSIRO) published the Collaborative and Automated Tools for Analysis of Marine Imagery (CATAMI), a standardised vocabulary for annotating benthic characteristics (biota and substrata) that can then be used to classify benthic habitats (Althaus et al. 2015). CATAMI combines coarse-level taxonomy and morphology into a hierarchical classification.
The goal of this study was to enhance the understanding and management of green turtle foraging habitats in the Northern Territory through a collaborative partnership with Traditional Owners and ranger groups. The specific aims of the study were:
To identify and map the marine habitats at two remote green turtle foraging areas in the Northern Territory by using remote sensing and towed-video transects, and additionally, to investigate the similarities and differences in habitats between the two foraging areas and which benthic species are driving those differences. This detailed benthic habitat mapping can be used to inform ecosystem-based management plans that support the conservation of green turtles, a species that is both vulnerable in Australia and culturally significant.
To provide training and capacity building for Indigenous ranger groups, empowering them with the skills and knowledge necessary to collect, analyse, and utilise habitat data. This will enhance their ability to protect green turtles and their habitats and support Traditional Owner-led management.
Materials and methods
Study sites
The Northern Territory is characterised by a tropical monsoonal climate, with high temperatures, heavy seasonal rainfall and cyclones, alternated with extended rain-free periods. There are also complex tidal regimes, with two tides in some areas, one tide in other areas, huge tidal ranges (up to 8 m) in some areas and almost no tidal range in others (Department of Sustainability, Environment, Water, Population and Communities 2012). The Northern Territory has an average sea-surface temperature of 30.6°C in the wet season and 24.5°C in the dry season (URS 2008).
Garig Gunak Barlu Marine Park and Kakadu National Park are critical conservation areas in the Northern Territory, recognised for their ecological, cultural, and biodiversity significance (Dethmers et al. 2010; Northern Territory Government 2011; Groom et al. 2017; Department of Climate Change, Energy, the Environment and Water 2019). Within these parks, two green turtle foraging grounds were selected for this study (Fig. 1) on the basis of traditional knowledge, local knowledge, and existing scientific knowledge of foraging green turtles in the Northern Territory (Dethmers et al. 2010; Ferreira et al. 2021).
A map of the two study sites, (a) Trepang Bay and (b) Field Island, in the Northern Territory, Australia, with right-hand side maps showing spatial context.
Trepang Bay (Fig. 1a), on the Cobourg Peninsula, is part of Garig Gunak Barlu Marine Park and is jointly managed by the Northern Territory Government and Traditional Owners. The area experiences minimal freshwater and sediment input (Department of Environment, Water, Heritage and the Arts 2007). The bay has a smaller tidal range than do many areas of the Northern Territory (2–2.5 m). Coral reefs have been recorded along the Cobourg Peninsula; the taxonomic similarities of these reefs are like those found in the Torres Strait (Veron 2004). A coral bleaching event recorded between November 2002 and January 2003 caused significant destruction to corals and surrounding habitats in Coral Bay, within Port Essington, and the area has not been surveyed since (Department of Environment, Water, Heritage and the Arts 2007).
Field Island (Fig. 1b), managed by (Commonwealth) Kakadu National Park, is located within the Van Diemen Gulf. The gulf is a semi-enclosed, poorly flushed embayment with significant input from the East and South Alligator, Adelaide and Mary rivers. These large catchments contribute significantly to the inflow of sediment, freshwater, and nutrients into the system during the Northern Territory’s wet season. The area has a large tidal range (3–5 m), causing strong currents that scour the benthic habitats (Department of Environment, Water, Heritage and the Arts 2007).
The reefs surrounding Field Island are classified as a biologically important area (BIA) for foraging green turtles, although this delineation is predominantly based on expert opinion, rather than quantitative analysis of turtle data, and is currently being updated (Department of the Environment and Energy 2017). A marine benthic survey of the south-eastern Van Diemen Gulf in 2004 used a combination of aerial and vessel-based surveys, the latter including beam trawl, benthic grab, box dredge, gillnet and underwater video, to characterise the biota and substrates at each site (Russell and Smit 2007).
Of the areas surveyed, the intertidal habitats around Field Island had the most seagrass cover. Halophila decipiens was the dominant seagrass species and often occurred with Halodule uninervis, Halophila minor and Syringodium isoetifolium. Seagrass meadows studied were all intertidal, except for a small area just off the north-western tip of Field Island (Russell and Smit 2007). The extent of the seagrass meadows from 2004 is available from SeaMap Australia (see https://seamapaustralia.org/).
Data collection
Field surveys were supported by a Ranger Exchange Program facilitated by Larrakia Nation Rangers with six participating ranger groups. An additional exchange was conducted between the Garig Gunak Barlu National Park Rangers and the Garngi Rangers from Croker Island. In all, the Kakadu National Park Rangers, Garig Gunak Barlu National Park Rangers, Larrakia Nation Rangers, Gumurr-Marthakal Rangers, Kenbi Rangers, Garngi Rangers, Tiwi Rangers and Mardbalk Rangers all contributed to this data collection.
For each of the selected sites (Trepang Bay and Field Island), the extent of the turtle foraging grounds, and detailed mapping area were identified on the basis of the mean foraging turtle kernel utilisation density (KUD) from satellite-tracking data. This foraging movement data were collected from 13 green turtles (seven at Trepang Bay and six at Field Island) by Charles Darwin University between 2021 and 2023 (N. Robson, unpubl. data). The average tag duration was 117 days (range of 52–224 days). The mean 50% KUD was used to identify areas of high turtle use (Supplementary material Fig. S2a, b).
The habitat map of each green turtle foraging ground was created using a combination of remote-sensing tools. In this study, Sentinel-2 satellite imagery was used to create the habitat map; this imagery has been successfully used in marine habitat mapping worldwide (Kuhwald et al. 2022; Wilson et al. 2022). Sentinel-2 imagery is free to download through the Sentinel Australasia Regional Access (SARA) and is available at a spatial resolution of 10 m2 (pixel size). On the basis of field observations (visibility of the camera), the late dry season had the best water clarity at both sites. Therefore, images from the dry season (August–September) were selected for analysis on the basis of cloud cover (0%) and water clarity.
To create an accurate habitat map, it is important to ‘ground truth’ the remote-sensing mapping area and characterise benthic habitat types and associated assemblages. For this study, the ground-truthing was undertaken by running transects using a towed-video system during neap tides to optimise the quality of the seabed imagery (Carroll et al. 2020; Foster et al. 2020). The subsea video system used was a high-definition GoPro Hero 10 (set to ‘wide’ view, with a field of view of 122°) mounted onto a bespoke lightweight ballasted PVC pipe frame (Fig. S1). A wide-angle dive torch was attached to the frame above the camera, such that the light from the torch would illuminate the seabed at an oblique angle to the angle of view of the GoPro to mitigate backscatter. A hand-held Garmin GPSMap 64 (accuracy of <5 m) created a tracklog of each transect. The start and end depths were recorded using the vessel depth sounder. This system was chosen because of its compact nature and ease of use. Rangers participating in the survey were trained on the use of this video system, with the aim that it could be readily used for future habitat monitoring.
The number of transects at each site was defined on the basis of the relative size of the foraging ground and the complexity of the habitat (identified from bathymetry and satellite imagery). The location of each transect was chosen using a spatially balanced design (Foster et al. 2020). Spatially balanced designs are more efficient than other randomised designs because they tend to increase the balance on many environmental variables (in this case, depth). The R package MBHdesign (ver. 2.3.15, see https://cran.r-project.org/package=MBHdesign; Foster 2021) was used to create the survey design at each site, specially balanced by depth (using GEBCO_2021 bathymetry) (Foster et al. 2020). The survey points were added to ArcGIS Pro (see https://esriaustralia.com.au/products/arcgis-pro) to view any gaps and remove points, falling in waters too murky for accurate video transects. Additional points were manually added or adjusted using the Sentinal-2 imagery as a guide to ensure that each potential habitat type had been ground-truthed (Fig. 1, S2a, b). The camera system was deployed from the side of the vessel and towed for 5 min. The distance and direction of each transect were determined by the tides. The average transect distance was 81.11 m ± 51.20 at Trepang Bay and 151.80 m ± 76.50 at Field Island.
Eighty transects were undertaken at Trepang Bay, with two surveys per year (July and October in 2022 and May and November in 2023). Transect depths during the survey ranged from 0.5 to 17 m. Four surveys were also conducted at Field Island over those 2 years (April and August in 2022, and July and November in 2023), with 42 transects undertaken there. Most Field Island transects were conducted close to low tide. The depth of these transects during the survey ranged from 0.4 to 6.9 m.
Data analysis
The video files associated with the towed-video system were viewed in SMPlayer (ver. 24.5.0, see https://www.smplayer.info/). Screenshots were used to extract images throughout the video when the image met the acceptable criteria, such as being well-lit, having good visibility of features, and being within sufficient distance of the seafloor to obtain clear imagery of seabed features. A random selection of at least five images was selected from these screenshots from each video. If a distinct habitat change was observed along the transect (i.e. bare sand to coral reef), the transect was split and renamed (i.e. K04 and K04-1), and an additional five images were extracted from the new habitat type. The timing of this habitat change was then recorded on the tracklog.
Images were analysed in the GNU Image Manipulation Program, GIMP (ver. 2.10.6, see https://www.gimp.org/). Each image was overlayed with a 10 × 10 square grid and then analysed for percentage cover of biota and substrate types. Any sessile organism or substrate (sand, shell, mud, pebbles and rock) taking up over half of a grid square (≥0.5%) was identified and the percentage cover was recorded. Organisms taking up less than half of one grid square were identified and recorded as ‘present’. Motile organisms, such as fish, were identified and counted as individuals. The relative profile of each habitat (low, medium, or high) was also recorded. For consistency and comparability, the classification vocabulary followed the Collaborative and Automated Tools for Analysis of Marine Imagery (CATAMI) (Althaus et al. 2015). Each transect was classified by its depth range as either ‘Intertidal’ (<3 m) or ‘Subtidal’ (3–20 m), on the basis of the depth recorded on the vessel depth sounder and the tide at the time of the survey.
The species percentage cover data were analysed in PRIMER with PERMANOVA+ (ver. 7, see https://www.primer-e.com/). The initial analysis looked at these data for each image to assess the within-transect similarities in biological assemblage. A fourth-root transformation was used to downweigh the influence of more abundant habitat components (e.g. sand). A permutational ANOVA (PERMANOVA+) test was then used to test the null hypothesis (H0) that there is no difference in the benthic assemblages among transects and survey sites. A similarity percentages (SIMPER) breakdown procedure was undertaken to assess the average percentage similarity within transects and to calculate the contribution of each species (%) to the similarity within each transect.
To compare biological assemblages among transects, the raw percentage cover data from both sites were averaged by transect and a fourth-root transformation, and Bray–Curtis resemblance was then applied. A cluster analysis using similarity profiles (SIMPROF) was then undertaken on the habitat data. SIMPROF is a series of permutation tests run on biotic data that looks for statistically significant evidence of genuine clusters of sites that are a priori unstructured (Clarke et al. 2008). A two-dimensional non-metric multi-dimensional scaling (MDS) ordinations plot was created from a resemblance matrix based on Bray–Curtis similarity, with a minimum stress of 0.01 and the Kruskal fit scheme 1 (Primer 2024). A PERMANOVA+ test was then used to compare the groups identified by the cluster analysis and identify any significance amog groupings, with the null hypothesis that there is no difference among SIMPROF groupings. A SIMPER breakdown procedure was undertaken to assess the average percentage contribution of individual variables (biota and substrate) to the dissimilarity among objects in a Bray–Curtis dissimilarity matrix (Clarke 1993) to calculate the contribution of each species (%) to the dissimilarity between each SIMPROF group. The data for both study sites (Trepang Bay and Field Island) were combined, and the same set of analyses was performed among sites to identify any significant differences and similarities in the benthic assemblages of the two sites. These analyses were used to assign a biotope habitat classification to each video transect. The cluster with SIMPROF defined the transects representing a particular biotope, and the outcomes of the SIMPER analysis subsequently defined the biotope characteristics.
The Sentinel-2 imagery comes with 13 spectral bands (varying by wavelength and bandwidth). However, for identifying marine habitats we worked with only four bands with the best resolution (10 m2), including Band 2 (blue, central wavelength of 490 nm), Band 3 (green, central wavelength of 560 nm), Band 4 (red, central wavelength of 665 nm) and Band 8 (near infrared (NIR), central wavelength of 832.8 nm). The proportions of each band within a pixel can indicate different habitat types.
Geometric corrections (resampling and subsetting) of the Sentinel-2 imagery for both sites were undertaken in the Sentinel Application Platform (SNAP) (Fig. 2). The Sen2Coral plug-in was then used to de-glint (remove sun glare) and remove any white caps (Serco Italia SPA 2019).
A depth mask was applied to the imagery by using 2021 GEBCO bathymetry (GEBCO Compilation Group 2021) in R, by using the raster package (ver. 3.6-32, see https://cran.r-project.org/package=raster). The GEBCO bathymetry was originally compared with the depths recorded on the vessel depth sounder. However, the tidal charts for these two sites are not accurate (varying from the closest tide gauge), and the depths could not be accurately adjusted for the tides. This filtered out any habitats too deep to be picked up by remote sensing, generally waters deeper than 20 m, but this can depend on water clarity, turbidity etc. (Kutser et al. 2020). The depth mask differed among sites on the basis of the water clarity and the green turtle foraging areas (18 m for Trepang Bay and 6 m for Field Island) (Fig. 2, S2a, b).
The normalised difference vegetation index (NDVI) was then calculated using the raster band data by the following formula (Blaschke et al. 2014):
The NDVI always ranges between −1 and 0, with zero usually representing bare habitat and negative values indicating water. Raster cell statistics were used to filter out the positive values (land) from the NDVI raster, which created a land mask. The resulting image contained only pixels that could be used for the marine habitat classification.
The biotope classifications we identified from the analysis of each transect were mapped over the pre-processed Sentinel-2 imagery in ArcGIS Pro (Fig. 2), by using the GPS positions generated by the tracklog from each transect. To obtain the training data (training sample file) for the habitat classification model, we delineated a group of pixels from the satellite image underlying the biotope classification GPS point into a polygon representing features or potential habitats, such as a seagrass meadow or coral bommie, described from the image. This was undertaken using supervised pixel-based classification in the Training Samples Manager tool in ArcGIS Pro. All the pixels in the image were then statistically compared by their colour, with the pixels defined by the training samples by using a Support Vector Machine Learning (SVML) model. The SVML model was chosen as it is less susceptible to noise, correlated bands, and an unbalanced number or size of training sites within each class (ESRI 2022). Once trained, this model predicted the class membership (biotope) of each pixel across the whole mapping area (ESRI 2022). The output of this tool was a classified raster of the habitats (biotopes) in the mapping area (Fig. 2).
The accuracy of habitat classification models (such as SVML) decreases with depth and water clarity. An uncertainty layer was therefore added to the final habitat map for depths greater than 5 m in Trepang Bay and for 4 m at Field Island. Finally, a confusion matrix was run in ArcGIS Pro (using the Image Analyst toolbox) to look at the accuracy between the classified map and the ground-truth data points (ESRI 2023). The assessment included calculating user accuracy, producer accuracy, and the kappa index of agreement. These accuracy metrics range between 0 and 1, where 1 represents 100% accuracy. The Kappa index of agreement can be interpreted as follows: values of ≤0 as indicating no agreement, 0.01–0.20 as none to slight, 0.21–0.40 as fair, 0.41–0.60 as moderate, 0.61–0.80 as substantial and 0.81–1.00 as almost perfect agreement (Landis and Koch 1977).
Results
Analysis of biological data
A significant difference in the benthic assemblages among the transects at Trepang Bay was found from the PERMANOVA run on the detailed transect data (P = 0.001). The average similarity in the benthic assemblage within each transect ranged from 42.95 to 100%. The three transects with 100% average similarity all contained 100% mud and silt. The transect with the lowest average similarity had a mixed benthic assemblage, with the highest contributions coming from rock (38.85%), coarse sand (31.10%) and encrusting sponges (11.35%).
The cluster analysis identified 10 groupings (SIMPROF groups labelled a–j) for the averaged Trepang Bay transects, with the majority of transects falling into either group ‘j’ or ‘h’ (Fig. 3, S3a). A significant (P = 0.001) difference was found among the 10 SIMPROF groups. A two-dimensional, non-metric MDS ordination illustrates the differences among transects, with transects displayed as closer together being more similar in their benthic characteristics (Fig. 3).
Two-dimensional non-metric multi-dimensional scaling (MDS) ordinations of all transects within each site for (a) Trepang Bay and (b) Field Island, Northern Territory, Australia. The colours and symbols represent SIMPROF groups.
The average similarity within the two biggest SIMPROF groups (in terms of number of transects per group) was 57.81 and 44.17% for groups ‘h’ and ‘j’ respectively. This similarity within group ‘h’ was characterised by the presence of coral rubble, stony corals (massive and submassive) and coarse sand. The average similarity within group ‘j’ was characterised by the presence of coarse sand, biogenic material (coral rubble and shells) and brown algae (laminate and erect-fine branching). The highest recorded within-group similarity was 92.43% for group ‘d’, with this similarity being driven by the presence of mud and silt (Fig. 3).
A significant difference was found between intertidal and subtidal assemblages at Trepang Bay, on the basis of the PERMANOVA (P = 0.001). Plotting SIMPROF clusters by depth better clarified the groupings and was used to divide some of the SIMPROF groups into subgroups when assigning biotopes.
A significant difference in the benthic assemblages among the transects at Field Island was found from the PERMANOVA run on the detailed transect data (P = 0.001). The average similarity in the benthic assemblage within each transect ranged from 24.74 to 100%. The two transects with 100% average similarity both contained 100% coarse sand. The transect with the lowest average similarity was dominated by mud and silt (73.54% contribution), but also had a few scattered coral bommies, which accounts for the low average similarity.
The cluster analysis identified seven SIMPROF groupings for the averaged Field Island transects, with the majority of transects falling into groups ‘c’ or ‘d’ (Fig. 3, S3c). A significant (P = 0.001) difference was found among the seven groups, on the basis of PERMANOVA. A two-dimensional non-metric MDS ordination displays the distance between transects as Bray–Curtis similarity, with transects being displayed as closer together having a closer similarity in the benthic assemblage (Fig. 3). The Field Island MDS ordination had a high stress of 0.23, which meant that the plot did not adequately represent the dissimilarities in two-dimensional space (Fig. 3).
The average similarity within the two largest SIMPROF groups was 53.92 and 42.40% for groups ‘c’ and ‘d’ respectively. This similarity within group ‘c’ was characterised by the consistency in coverage of coarse sand. The average similarity within group ‘d’ was characterised by the consistent occurrence of coarse sand, biogenic material (coral rubble), brown algae (filamentous) and encrusting sponges and stony corals. The highest within-group similarity group ‘g’ (77.18%), which was found to be characterised by the contribution of mud and silt, seagrass (strap-like and elliptical leaves), biogenic material (coral rubble and shells) and pebbles (Fig. 3). A significant difference was found between intertidal and subtidal assemblages at Field Island, on the basis of the PERMANOVA (P = 0.001). Plotting the SIMPROF clusters by depth better clarified the groupings and was used to divide some of the SIMPROF further when assigning biotopes.
Running a SIMPROF cluster analysis for the data from both sites showed that some transects from Trepang Bay and Field Island were more similar to each other than to other transects within their respective sites. However, there was still a significant (P = 0.001) difference between the benthic assemblages at Trepang Bay and Field Island. Whereas the two-dimensional non-metric MDS ordination had a high stress (two-dimensional stress of 0.21, indicating poor representation), visualisation on a three-dimensional non-metric MDS ordination exhibited a clear divide among sites, although strong similarities between many transects from Trepang Bay and Field Island were still evident (a relevant perspective of the three-dimensional ordination is presented in Fig. 4). A SIMPER analysis of the SIMPROF groups showed that the biggest mixed site group (group ‘k’) had an average similarity of 51.66%, with this similarity being driven by the presence of coarse sand.
Biotope classification
Biotope classifications were characterised on the basis of SIMPER analysis of SIMPROF groups and considering depth ranges (intertidal vs subtidal). Some SIMPROF groups were combined into the same biotope for mapping simplicity or split into two groups on the basis of depth and species presence. For example, the Field Island Transect K030 was assigned SIMPROF group ‘c’ and was classified as ‘Subtidal coarse sand’, whereas Transect K010, also in SIMPROF group ‘c’, was classified as ‘Intertidal coarse sand and seagrass’ (Fig. S3). This additional aspect was undertaken with the consideration that statistical significance does not necessarily directly translate into biotope classifications. For example, depauperate and abundant examples of the same basic biotope may be separated by statistical analysis but may still display the same basic characteristics and fulfil similar ecological functions and services to a lesser or greater degree. The Trepang Bay analysis results (SIMPROF groups, depth range, SIMPER) indicated that eight different biotopes occurred along the transects at this site, whereas seven biotopes were characterised from Field Island transects (Table 1).
| Site | SIMPROF group | Description and similarities | Biotope name | |
|---|---|---|---|---|
| Trepang Bay | h, e, a | Coral rubble, massive, submassive, and encrusting hard coral structures, coarse sand, and brown algae (laminate and erect fine branching) | Intertidal mixed coral reef habitat | |
| j | Coarse sand, coral rubble and brown algae (laminate and erect fine branching) | Intertidal algal reef | ||
| j2, i2 | Coarse sand, green algae and seagrass (mostly Halodule uninervis, but areas mixed with Halophila ovalis) | Intertidal mixed algae and seagrass | ||
| g | Coral rubble, coarse sand, encrusting, submassive, foliose and corymbose hard coral structures | Subtidal mixed coral reef habitat and algal reef | ||
| b, d2 | Mud and silt and brown filamentous algae (red and brown) | Intertidal mud and silt and algae | ||
| f | Coarse sand, hydroids, rock, encrusting sponges, soft whip corals, seasonal seagrass (Halophila ovalis) | Deeper subtidal mixed soft coral reef and seasonal seagrass | ||
| i | Coarse sand | Intertidal coarse sand | ||
| c, d | Mud and silt and brown erect branching red algae, with high levels of bioturbation | Subtidal mud and silt | ||
| Field Island | c | Coarse sand | Subtidal coarse sand | |
| c2, a | Coarse sand and seagrass (mixed Halophila ovalis, Thalassia hemprichii, Syringodium Isoetifolium and Halodule uninervis) | Intertidal coarse sand and seagrass | ||
| c3 | Coarse sand – mobile sandbanks | Intertidal coarse sand | ||
| d | Coarse sand, coral rubble, filamentous brown algae, encrusting sponges, and hard corals (encrusting) with patchy seagrass (Halodule uninervis) | Intertidal mixed coral reef habitat and algal reef | ||
| d2 | Coarse sand, coral rubble, filamentous brown algae, encrusting sponges, hard (foliose) and soft corals (branching and whip) | Subtidal mixed coral reef habitat and algal reef | ||
| b, e | Mud and silt, some red algae | Subtidal mud and silt | ||
| f, g | Mud and silt, seagrass (mixed Halophila ovalis and Thalassia hemprichii) and pebbles | Intertidal mud and silt and seagrass |
Habitat mapping
Using the Support Vector Machine Learning Model, the total area for which habitat was predicted at Trepang Bay was 246.3 km2, with the largest percentage of this area classified as ‘Subtidal mud and silt’ (40.44%) (Fig. 5). The habitat ‘Intertidal algal reef’ was the second most predicted biotope at 12.99%, followed by ‘Intertidal mud and silt’ (9.55%), ‘Intertidal mixed algae and seagrass’ (9.15%), ‘Intertidal coarse sand’ (8.12%), ‘Deeper subtidal mixed coral reef and season seagrass’ (7.92%), ‘Subtidal mixed coral reef habitat and algal reef’ (5.93%) and ‘Intertidal mixed coral reef habitat’ (5.88%).
Habitat map of Trepang Bay with the predicted biotope classifications from the Support Vector Machine Learning model trained using towed-video survey data. Depth contours are noted on the map and areas that are too deep for accurate remote sensing have been marked as ‘Uncertain’.
The total area for which habitat was predicted at Field Island was 132.3 km2; similar to Trepang Bay, the largest percentage of this area was classified as ‘Subtidal mud and silt’ (54.45%) (Fig. 6). The habitat ‘Intertidal coarse sand’ was the second-most predicted biotope at Field Island at 16.05%, followed by ‘Subtidal coarse sand’ (11.52%), ‘Intertidal coarse sand and seagrass’ (5.60%), ‘Intertidal mixed coral reef habitat and algal reef’ (5.33%), ‘Subtidal mixed coral reef habitat and algal reef’ (4.47%) and ‘Intertidal mud and silt and seagrass’ (2.65%).
Habitat map of the waters around Field Island with the predicted biotope classifications from the Support Vector Machine Learning model trained using towed-video survey data. Depth contours are noted on the map and areas that are too deep for accurate remote sensing have been marked as ‘Uncertain’.
The results of the accuracy assessment found that the kappa index for Trepang Bay was 0.63, indicating a substantial level of agreement between predicted and observed habitat classifications. The biotope ‘Subtidal mud and silt’ exhibited the highest user accuracy at 0.98 (almost perfect agreement), indicating the reliability of the classification for this habitat type. The lowest accuracy (0.28) was found for the more complex deeper habitat, ‘Subtidal mixed coral reef habitat and algal reef’, meaning that there was minimal agreement between observed and predicted for this habitat.
Similarly, for Field Island, the kappa index was found to be 0.75, signifying a higher, but still substantial, level of agreement between predicted and observed classifications. The biotope ‘Intertidal coarse sand’ demonstrated the highest user accuracy at 0.96 (almost perfect agreement). The lowest accuracy, of 0.53 (moderate agreement), was found for the habitat ‘subtidal coarse sand’, with this habitat type having the fewest ground-truthed transects.
Discussion
Through a collaborative partnership with Traditional Owners and ranger groups, progress has been made in enhancing the understanding and management of green turtle foraging habitats in the Northern Territory. This collaboration was essential in creating these habitat maps, as the expertise and knowledge of Traditional Owners and ranger groups provided valuable insights into the local marine ecosystems. This study marks the first application of ground-truthed remote-sensing techniques for mapping subtidal habitats in the Northern Territory. Sentinel-2 imagery was used to detect habitats such as seagrass meadows, coral reefs and algal reefs, as well as distinguishing among mud, coarse sand and rocky habitats with a high level of agreement at both sites (0.63 for Trepang Bay and 0.75 for Field Island). The successful utilisation of Sentinel-2 imagery has demonstrated the potential of remote sensing, when paired with ground-truthed transects, as a valuable tool for assessing and monitoring marine habitats in remote and challenging environments. Even with the environmental challenges of low water clarity and high tidal range (Kuhwald et al. 2022), this study has shown that remote sensing can be effectively applied to the Northern Territory’s shallow coastal waters. However, further research is required into both the seasonal and long-term changes in benthic assemblage, spatial extent and habitat health at these sites (Hedley et al. 2012; Lyons et al. 2013; Hovey and Fraser 2018; Krause et al. 2021).
The habitat classification model generally had a lower accuracy rating for deeper and more complex habitat types. Fine-scale habitat changes are more difficult to detect from satellite imagery with an increasing depth, and complex habitats can be difficult to distinguish at the spatial scale of 10 m2. However, this lower accuracy can be mitigated by increased ground-truthing (Hedley et al. 2012; Wilson et al. 2022).
The significant differences between the benthic assemblages at Trepang Bay and Field Island could be attributed to varying levels of freshwater input. Trepang Bay, characterised by smaller rivers and catchments, experiences less freshwater and sediment input than does Field Island, which is located within Van Diemen’s Gulf. The Gulf receives significant inflow from large river systems during the wet season (Department of Environment, Water, Heritage and the Arts 2007). This difference in hydrological dynamics, and associated differences in, for example, levels of sedimentation and turbidity, are likely to influence habitat composition and distribution, highlighting the importance of considering local environmental factors in habitat mapping and management efforts.
Although juvenile and adult green turtles are primarily herbivorous, foraging on algae, seagrass and occasionally mangroves (Mortimer 1981; André et al. 2005; Díaz-Abad et al. 2022), considerable variability in green turtle diet exists around the world, both among and within foraging grounds (Limpus and Limpus 2000; Palmer et al. 2021; Clyde-Brockway et al. 2022). The benthic assemblage results from Trepang Bay and Field Island both showed a diverse array of marine habitats, including mixed coral reef habitats, algal reefs and seagrass meadows. Algae and seagrass habitats made up a combined 30% of the mapped habitats in Trepang Bay and 18.05% at Field Island. The seagrass species found at Field Island during this study were similar to those found in 2004, with the new addition of Thalassia hemprichii (Russell and Smit 2007). These seagrass and algae habitats provide good foraging grounds for green turtles, with the diversity of marine habitats allowing for a range of different age classes of green turtles to forage at both sites. The more complex coral reef habitats also provide areas for rest and predator avoidance. This diversity of habitats also plays a crucial role in supporting other marine species such as hawksbill turtles (coral reef) and dugongs (seagrass) (André et al. 2005; Clyde-Brockway et al. 2022).
The collaborative partnership between ranger groups and scientists in field surveys not only facilitated the collection of ground-truthing data in very remote areas, but also demonstrated the potential for repeatability and effectiveness of these field methods for habitat monitoring. The towed-video system developed was highly cost-efficient, and lightweight, allowing it to be hand-haulable from small, open survey vessels and is easily constructed from materials that are commonly available in hardware stores. Although a live video feed was not used for this project’, this camera set-up allows for a wifi extender cable to connect personnel on the vessel to the live video feed at an extra cost. A live video feed would allow for more control of the camera system (distance from the bottom, avoiding collision with large rocks), particularly at deeper or low-visibility sites. This approach has demonstrated the potential value to ranger groups for increasing opportunities for data collection in remote areas where other systems were either too bulky or failed.
A future goal for the collaborative partnership is to continue the capability building of ranger groups and community members to collect long-term marine habitat data, such as tracking coral bleaching and seagrass extent. By training ranger groups in field data collection and data storage methods, a ranger database can be created for towed-video data, which can then be used by Indigenous natural and cultural resource management agencies to support their decision-making on Country. The collaborative approach used in this study highlights the possibility for future seasonal and annual monitoring of marine habitats by remote ranger groups, as well as increasing the amount of available ground-truthing data. Combining these data with information on the species that these habitats support, including green turtles, is extremely useful for management planning, including the development of marine turtle management plans and Indigenous protected areas in the Northern Territory (Parks and Wildlife Service of the Northern Territory 2011; Department of the Environment and Energy 2017). By combining remote-sensing technologies with field observations, we have gained valuable knowledge about the availability, diversity and quality of the marine habitats used by foraging green turtles in the Northern Territory.
Conflicts of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Declaration of funding
This project was funded by ARC Linkage Grant LP200100222, partnered with Kakadu National Park, Taronga Conservation Society Australia, Northern Territory Parks and Wildlife Service, Larrakia Nation Rangers, Gumurr-Marthakal Rangers and Sea Darwin. The Ranger Exchange and Sea Ranger operational costs were funded through the INPEX Coastal Offsets Aboriginal Ranger Grants Program and managed by Larrakia Nation.
Author contributions
Natalie Robson conceived and designed the study, developed the methodology, conducted the data collection, and performed the data analysis, mapping and writing. Carol Palmer secured funding through the grant application, conducted fieldwork, and supervised the project. Garnet Hooper contributed to the equipment (video system), survey design, data analysis, visualisation and statistical interpretation, and provided critical feedback on writing. Sam Banks contributed to the grant application, survey design and feedback. Michele Thums provided critical feedback and contributed to writing. Alana Grech and Joanna Day contributed to the grant application, survey design, and provided feedback. Robert Risk, Dylan Cooper, and the Kakadu Rangers made significant contributions to data collection. All authors discussed the results, reviewed, and approved the final paper.
Acknowledgements
We thank the ranger groups who assisted in the collection of data for this project: Kakadu National Park Rangers (South Alligator Branch), Garig Gunak Barlu National Park Rangers, Larrakia Nation Rangers, Gumurr-Marthakal Rangers, Kenbi Rangers, Garngi Rangers, Tiwi Rangers and Mardbalk Rangers. We also thank Kaline De Mello for reviewing an early draft of this paper.
References
Althaus F, Hill N, Ferrari R, et al. (2015) A standardised vocabulary for identifying benthic biota and substrata from underwater imagery: the CATAMI classification scheme. PLoS ONE 10(10), e0141039.
| Crossref | Google Scholar |
André J, Gyuris E, Lawler IR (2005) Comparison of the diets of sympatric dugongs and green turtles on the Orman Reefs, Torres Strait, Australia. Wildlife Research 32(1), 53-62.
| Crossref | Google Scholar |
Andréfouët S, Berkelmans R, Odriozola L, et al. (2002) Choosing the appropriate spatial resolution for monitoring coral bleaching events using remote sensing. Coral Reefs 21(2), 147-154.
| Crossref | Google Scholar |
Avens L, Braun-McNeill J, Epperly S, et al. (2003) Site fidelity and homing behavior in juvenile loggerhead sea turtles (Caretta caretta). Marine Biology 143(2), 211-220.
| Crossref | Google Scholar |
Bearzi G, Politi E, Agazzi S, et al. (2006) Prey depletion caused by overfishing and the decline of marine megafauna in eastern Ionian Sea coastal waters (central Mediterranean). Biological Conservation 127(4), 373-382.
| Crossref | Google Scholar |
Blaschke T, Hay GJ, Kelly M, et al. (2014) Geographic object-based image analysis – towards a new paradigm. ISPRS Journal of Photogrammetry and Remote Sensing 87, 180-191.
| Crossref | Google Scholar | PubMed |
Butler JRA, Tawake A, Skewes T, et al. (2012) Integrating traditional ecological knowledge and fisheries management in the Torres Strait, Australia: the catalytic role of turtles and dugong as cultural keystone species. Ecology and Society 17(4), 34.
| Crossref | Google Scholar |
Carroll A, Althaus F, Beaman R, Friedman A, Ierodiaconou D, Ingleton T, Jordan A, Linklater M, Monk J, Post A, Przeslawski R, Smith J, Stowar M, Tran M, Tyndall A (2020) 7. Marine sampling field manual for towed underwater camera systems. In ‘Field Manuals for Marine Sampling in Australian Waters, Version 2’. (Eds R Przeslawski, S Foster) pp. 212–236. (Geoscience Australia and CSIRO) Available at https://www.nespmarine.edu.au/system/files/Przeslawski_2020_NESP%20field%20manuals%20V2_all.pdf
Clarke KR (1993) Non-parametric multivariate analyses of changes in community structure. Australian Journal of Ecology 18, 117-143.
| Crossref | Google Scholar |
Clarke KR, Somerfield PJ, Gorley RN (2008) Testing of null hypotheses in exploratory community analyses: similarity profiles and biota-environment linkage. Journal of Experimental Marine Biology and Ecology 366(1–2), 56-69.
| Crossref | Google Scholar |
Clyde-Brockway CE, Heidemeyer M, Paladino FV, et al. (2022) Diet and foraging niche flexibility in green and hawksbill turtles. Marine Biology 169(8), 108.
| Crossref | Google Scholar |
Cogan CB, Todd BJ, Lawton P, Noji TT (2009) The role of marine habitat mapping in ecosystem-based management. ICES Journal of Marine Science 66(9), 2033-2042.
| Crossref | Google Scholar |
Dam Lam R, Gasparatos A, Chakraborty S, et al. (2019) Multiple values and knowledge integration in indigenous coastal and marine social-ecological systems research: a systematic review. Ecosystem Services 37, 100910.
| Crossref | Google Scholar |
Delisle A, Kiatkoski Kim M, Stoeckl N, et al. (2018) The socio-cultural benefits and costs of the traditional hunting of dugongs Dugong dugon and green turtles Chelonia mydas in Torres Strait, Australia. Oryx 52(2), 250-261.
| Crossref | Google Scholar |
Department of Agriculture, Water and the Environment (2022) Chelonia mydas – Green Turtle. In ‘Species profile and threats database’. (Commonwealth of Australia) Available at https://www.environment.gov.au/cgi-bin/sprat/public/publicspecies.pl?taxon_id=1765
Department of Climate Change, Energy, the Environment and Water (2019) Chapter one. World heritage values and attributes of Kakadu National Park. In ‘Australia’s Kakadu: protecting world heritage. Response by the Government of Australia to the UNESCO World Heritage Committee regarding Kakadu National Park. April 1999’. pp. 1–12. (Commonwealth of Australia) Available at https://www.dcceew.gov.au/sites/default/files/env/pages/3f3a19ff-9007-4ce6-8d4f-cd8ade380804/files/chap01.pdf
Department of Climate Change, Energy, the Environment and Water (2023) New South Wales – National intertidal-subtidal benthic NISB habitat map. (Commonwealth of Australia) Available at https://researchdata.edu.au/new-south-wales-habitat-map/1356978 [Dataset]
Department of Environment and Natural Resources (2005) Seagrass meadows of Arnhem Land, Kakadu and Gulf of Carpentaria. SeaMap Australia collaborative project. (DENR, Northern Territory Government) Available at http://metadata.imas.utas.edu.au/geonetwork/srv/en/metadata.show?uuid=b110ebcb-7584-413f-bf12-71651d9b3b18 [Dataset]
Department of the Environment and Energy (2017) Recovery plan for marine turtles in Australia 2017–2027. (Commonwealth of Australia) Available at https://www.dcceew.gov.au/sites/default/files/documents/recovery-plan-marine-turtles-2017.pdf
Department of Sustainability, Environment, Water, Population and Communities (2012) Marine bioregional plan for the North Marine Region: prepared under the Environment Protection and Biodiversity Conservation Act 1999. (Commonwealth of Australia) Available at https://www.dcceew.gov.au/sites/default/files/env/pages/0fcb6106-b4e3-4f9f-8d06-f6f94bea196b/files/north-marine-plan.pdf
Dethmers KEM, Jensen MP, FitzSimmons NN, et al. (2010) Migration of green turtles (Chelonia mydas) from Australasian feeding grounds inferred from genetic analyses. Marine and Freshwater Research 61(12), 1376-1387.
| Crossref | Google Scholar |
Díaz-Abad L, Bacco-Mannina N, Madeira FM, et al. (2022) eDNA metabarcoding for diet analyses of green sea turtles (Chelonia mydas). Marine Biology 169, 18.
| Crossref | Google Scholar |
Ens EJ, Finlayson M, Preuss K, et al. (2012) Australian approaches for managing ‘country’ using Indigenous and non-Indigenous knowledge. Ecological Management & Restoration 13(1), 100-107.
| Crossref | Google Scholar |
ESRI (2022) Deep learning using the ArcGIS Image Analyst extension. (ESRI) Available at https://pro.arcgis.com/en/pro-app/2.8/help/analysis/image-analyst/deep-learning-in-arcgis-pro.htm [Verified 7 June 2022]
ESRI (2023) Compute confusion matrix (Image Analyst). (ESRI) Available at https://pro.arcgis.com/en/pro-app/3.3/tool-reference/image-analyst/compute-confusion-matrix.htm
Esteban N, Mortimer JA, Stokes HJ, et al. (2020) A global review of green turtle diet: sea surface temperature as a potential driver of omnivory levels. Marine Biology 167, 183.
| Crossref | Google Scholar |
Ferreira LC, Thums M, Fossette S, et al. (2021) Multiple satellite tracking datasets inform green turtle conservation at a regional scale. Diversity and Distributions 27(2), 249-266.
| Crossref | Google Scholar |
Ferreira LC, Thums M, Whiting S, et al. (2023) Exposure of marine megafauna to cumulative anthropogenic threats in north-west Australia. Frontiers in Ecology and Evolution 11, 1229803.
| Crossref | Google Scholar |
Flynn A (2018) CBiCS classification of Victorian biotopes. Collaboration with Fathom Pacific Version 1. Report to Australian Marine Ecology. Available at https://www.researchgate.net/publication/342260470
Fossette S, Ferreira LC, Whiting SD, et al. (2021) Movements and distribution of hawksbill turtles in the Eastern Indian Ocean. Global Ecology and Conservation 29, e01713.
| Crossref | Google Scholar |
Foster S (2021) MBHdesign: an R package for efficient spatial survey designs. Methods in Ecology and Evolution 12, 415-420.
| Google Scholar |
Foster SD, Monk J, Lawrence E, Hayes KR, Hosack GR, T. Langlois, Hooper G, Przeslawski R (2020) Statistical considerations for monitoring and sampling. In ‘Field Manuals for Marine Sampling to Monitor Australian Waters, Version 2’. (Eds R Przeslawski, S Foster) pp. 26–50. (Geoscience Australia and CSIRO) Available at https://survey-design-field-manual.github.io/files/Chapter%202_Survey%20Design%20Field%20Manual_V2.pdf
Fuentes MMPB, Cinner JE (2010) Using expert opinion to prioritize impacts of climate change on sea turtles’ nesting grounds. Journal of Environmental Management 91(12), 2511-2518.
| Crossref | Google Scholar | PubMed |
Fuentes MMPB, McMichael E, Kot CY, et al. (2023) Key issues in assessing threats to sea turtles: knowledge gaps and future directions. Endangered Species Research 52, 303-341.
| Crossref | Google Scholar |
GEBCO Compilation Group (2021) The GEBCO_2021 grid. General bathymetric charts of the OCEEAN. (GEBCO) Available at https://www.gebco.net/data-products/gridded-bathymetry-data/gebco-2021
González Carman V, Bruno I, Maxwell S, et al. (2016) Habitat use, site fidelity and conservation opportunities for juvenile loggerhead sea turtles in the Río de la Plata, Argentina. Marine Biology 163(1), 1-13.
| Crossref | Google Scholar |
Groom RA, Griffiths AD, Chaloupka M (2017) Estimating long-term trends in abundance and survival for nesting flatback turtles in Kakadu National Park, Australia. Endangered Species Research 32(1), 203-211.
| Crossref | Google Scholar |
Guarinello ML, Shumchenia EJ, King JW (2010) Marine habitat classification for ecosystem-based management: a proposed hierarchical framework. Environmental Management 45, 793-806.
| Crossref | Google Scholar |
Hays GC, Alcoverro T, Christianen MJA, et al. (2018) New tools to identify the location of seagrass meadows: marine grazers as habitat indicators. Frontiers in Marine Science 5, 9.
| Crossref | Google Scholar |
Hays GC, Laloë J-O, Seminoff JA (2025) Status, trends and conservation of global sea turtle populations. Nature Reviews Biodiversity 1(2), 119-133.
| Crossref | Google Scholar |
Hazel J, Lawler IR, Hamann M (2009) Diving at the shallow end: green turtle behaviour in near-shore foraging habitat. Journal of Experimental Marine Biology and Ecology 371(1), 84-92.
| Crossref | Google Scholar |
Hedley J, Roelfsema C, Koetz B, et al. (2012) Capability of the Sentinel 2 mission for tropical coral reef mapping and coral bleaching detection. Remote Sensing of Environment 120, 145-155.
| Crossref | Google Scholar |
Holloway-Adkins KG, Dennis Hanisak M (2017) Macroalgal foraging preferences of juvenile green turtles (Chelonia mydas) in a warm temperate/subtropical transition zone. Marine Biology 164(8), 161.
| Crossref | Google Scholar |
Kennett R, Robinson CJ, Kiessling I, et al. (2004) Indigenous initiatives for co-management of Miyapunu/Sea turtle. Ecological Management & Restoration 5, 159-166.
| Crossref | Google Scholar |
Komatsu T, Hashim M, Nurdin N, Noiraksar T, Prathep A, Stankovic M, Phuoc T, Son H, Thu M, Van Luong C, Wouthyzen S, Phauk S, Muslim AM, Yahya NN, Terauchi G, Sagawa T, Hayashizaki K (2020) Practical mapping methods of seagrass beds by satellite remote sensing and ground truthing. Coastal Marine Science 43(1), 1-25.
| Crossref | Google Scholar |
Krause JR, Hinojosa-Corona A, Gray AB, et al. (2021) Emerging sensor platforms allow for seagrass extent mapping in a turbid estuary and from the meadow to ecosystem scale. Remote Sensing 13(18), 3681.
| Crossref | Google Scholar |
Kuhwald K, Schneider von Deimling J, Schubert P, et al. (2022) How can Sentinel-2 contribute to seagrass mapping in shallow, turbid Baltic Sea waters? Remote Sensing in Ecology and Conservation 8, 328-346.
| Crossref | Google Scholar |
Kutser T, Hedley J, Giardino C, et al. (2020) Remote sensing of shallow waters – a 50 year retrospective and future directions. Remote Sensing of Environment 240, 111619.
| Crossref | Google Scholar |
Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics 33, 159-174.
| Crossref | Google Scholar | PubMed |
Limpus CJ, Limpus DJ (2000) Mangroves in the diet of Chelonia mydas in Queensland, Australia. Marine Turtle Newsletter 89, 13-15.
| Google Scholar |
Lin M, Liu M, Turvey ST, et al. (2023) An interview-based investigation of marine megafauna bycatch in the northern South China Sea. Biological Conservation 286, 110297.
| Crossref | Google Scholar |
Lyons MB, Roelfsema CM, Phinn SR (2013) Towards understanding temporal and spatial dynamics of seagrass landscapes using time-series remote sensing. Estuarine, Coastal and Shelf Science 120, 42-53.
| Crossref | Google Scholar |
Madeira FM, Rebelo R, Catry P, et al. (2022) Fine-scale foraging segregation in a green turtle (Chelonia mydas) feeding ground in the Bijagós archipelago, Guinea Bissau. Frontiers in Marine Science 9, 984219.
| Crossref | Google Scholar |
Monk J, Ierodiaconou D, Bellgrove A, et al. (2008) Using community-based monitoring with GIS to create habitat maps for a marine protected area in Australia. Journal of the Marine Biological Association of the United Kingdom 88(5), 865-871.
| Crossref | Google Scholar |
Montefalcone M, Tunesi L, Ouerghi A (2021) A review of the classification systems for marine benthic habitats and the new updated Barcelona Convention classification for the Mediterranean. Marine Environmental Research 169, 105387.
| Crossref | Google Scholar |
Monteiro JG, Jiménez JL, Gizzi F, et al. (2021) Novel approach to enhance coastal habitat and biotope mapping with drone aerial imagery analysis. Scientific Reports 11(1), 574.
| Crossref | Google Scholar |
Mortimer JA (1981) The feeding ecology of the West Caribbean Green Turtle (Chelonia mydas) in Nicaragua. Biotropica 13, 49-58.
| Crossref | Google Scholar |
Northern Territory Government (2011) ‘Cobourg marine park: plan of management.’ (Eds Cobourg Peninsula Sanctuary and Marine Park Board, Parks and Wildlife Service of the Northern Territory, Department of Natural Resources, Environment, The Arts and Sport) (Parks and Wildlife Service Northern Territory) Available at https://rsis.ramsar.org/RISapp/files/49345442/documents/AU1_mgt210423.pdf
O2 Marine (2019) Appendix S: Technical Report For Benthic Habitats and Communities. Darwin industrial processing facility: benthic habitat and communities. (O2 Marine) Available at https://ntepa.nt.gov.au/__data/assets/pdf_file/0005/761495/draft_eis_darwin_processing_facility_appendixS_technical_report_benthic_habitats_communties.pdf
Olenin S, Ducrotoy J-P (2006) The concept of biotope in marine ecology and coastal management. Marine Pollution Bulletin 53(1-4), 20-29.
| Crossref | Google Scholar | PubMed |
Olsson P, Folke C, Hughes TP (2008) Navigating the transition to ecosystem-based management of the Great Barrier Reef, Australia. Proceedings of the National Academy of Sciences 105, 9489-9494.
| Crossref | Google Scholar |
Palmer JL, Beton D, Çiçek BA, et al. (2021) Dietary analysis of two sympatric marine turtle species in the eastern Mediterranean. Marine Biology 168(6), 94.
| Crossref | Google Scholar |
Peel LR, Whiting SD, Pendoley K, et al. (2024) I still call Australia home: satellite telemetry informs the protection of flatback turtles in Western Australian waters. Ecosphere 15(5), e4847.
| Crossref | Google Scholar |
Primer (2024) PRIMER v7: user manual/tutorial: accuracy & fit scheme. Available at https://learninghub.primer-e.com/books/primer-v7-user-manual-tutorial/page/accuracy-fit-scheme [Verified 19 March 2023]
Reisser J, Proietti M, Sazima I, et al. (2013) Feeding ecology of the green turtle (Chelonia mydas) at rocky reefs in western South Atlantic. Marine Biology 160(12), 3169-3179.
| Crossref | Google Scholar |
Rist P, Rassip W, Yunupingu D, et al. (2019) Indigenous protected areas in Sea Country: Indigenous-driven collaborative marine protected areas in Australia. Aquatic Conservation: Marine and Freshwater Ecosystems 29(S2), 138-151.
| Crossref | Google Scholar |
Russell BC, Smit N (2007) Report of a marine biodiversity survey of inshore soft bottom benthos of the SE Van Diemen Gulf and NW Arnhem Land between the Goulburn Islands and Castlereagh Bay, Northern Territory. Report to the Marine & Biodiversity Division, Department of the Environment, Water, Heritage and the Arts, and Parks Australia North.
Schill SR, McNulty VP, Pollock FJ, et al. (2021) Regional high-resolution benthic habitat data from planet dove imagery for conservation decision-making and marine planning. Remote Sensing 13(21), 4215.
| Crossref | Google Scholar |
Schofield G, Dimadi A, Fossette S, et al. (2013) Satellite tracking large numbers of individuals to infer population level dispersal and core areas for the protection of an endangered species. Diversity and Distributions 19(7), 834-844.
| Crossref | Google Scholar |
Seminoff JA (2023) Green turtle Chelonia mydas (amended version of 2004 assessment). In ‘The IUCN Red List of Threatened Species 2023’. e.T4615A247654386. (International Union for Conservation of Nature and Natural Resources) Available at https://www.iucnredlist.org/species/4615/247654386
Sequeira AMM, Hays GC, Sims DW, et al. (2019) Overhauling ocean spatial planning to improve marine megafauna conservation. Frontiers in Marine Science 6, 639.
| Crossref | Google Scholar |
Serco Italia SPA (2019) Sen2Coral toolbox for coral reef monitoring, Great Barrier Reef (version 1.1). Available at https://rus-copernicus.eu/portal/the-rus-library/learn-by-yourself/
Shimada T, Limpus C, Jones R, et al. (2016) Sea turtles return home after intentional displacement from coastal foraging areas. Marine Biology 163(1), 8.
| Crossref | Google Scholar |
Shimada T, Limpus CJ, Hamann M, et al. (2020) Fidelity to foraging sites after long migrations. Journal of Animal Ecology 89(4), 1008-1016.
| Crossref | Google Scholar | PubMed |
Smale DA, Kendrick GA, Harvey ES, et al. (2012) Regional-scale benthic monitoring for ecosystem-based fisheries management (EBFM) using an autonomous underwater vehicle (AUV). ICES Journal of Marine Science 69, 1108-1118.
| Crossref | Google Scholar |
Smith DC, Fulton EA, Apfel P, et al. (2017) Implementing marine ecosystem-based management: lessons from Australia. ICES Journal of Marine Science 74, 1990-2003.
| Crossref | Google Scholar |
URS Australia Pty Ltd (2008) Appendix 9. Nearshore marine water quality and sediment study. Ichthys gas field development project nearshore marine water quality and sediment study. Report prepared for INPEX Browse, Ltd. (URS: Perth, WA, Australia) Available at https://www.inpex.com.au/media/ufvddwdg/draft-eis-technical-appendices-appendix-9-nearshore-marine-water-quality-and-sediment-study.pdf
Wicaksono P, Aryaguna PA, Lazuardi W (2019) Benthic habitat mapping model and cross validation using machine-learning classification algorithms. Remote Sensing 11(11), 1279.
| Crossref | Google Scholar |
Wilson KL, Wong MC, Devred E (2022) Comparing Sentinel-2 and WorldView-3 imagery for coastal bottom habitat mapping in Atlantic Canada. Remote Sensing 14(5), 1254.
| Crossref | Google Scholar |
