The Antarctic Seafloor Annotated Imagery Database Authors: Jan Jansen, Victor Shelamoff, Charley Gros, Thomas Windsor, Nicole A Hill, David K Barnes, David A Bowden, Julian Gutt, Narissa Bax, Rachel V Downey, Marc P Eléaume, Alexandra L Post, Huw J Griffiths, Katrin Linse, Dieter Piepenburg, Autun Purser, Craig R Smith, Amanda Ziegler, Craig R Johnson Published: 2026 ABSTRACT AbstractMarine imagery can be a comparatively cost-effective way to collect data on seafloor organisms, biodiversity and habitat morphology. However, annotating these images to extract detailed biological information is time-consuming and expensive, and reference libraries of consistently annotated seafloor images are rarely publicly available. Here, we present the Antarctic Seafloor Annotated Imagery Database (AS-AID), a result of a multinational collaboration to collate and annotate seafloor imagery datasets from 21 Antarctic research campaigns between 1985 and 2019. AS-AID is comprised of 52,491 georeferenced downward facing seafloor images of which 3,599 have been labelled with a total of 632,252 expert annotations. Annotations are based on the Collaborative and Automated Tools for Analysis of Marine Imagery (CATAMI) classification scheme and have been reviewed by experts. In addition, because the pixel location of each annotation within each image is available, annotations can be viewed easily and customised to suit individual research priorities. This dataset can be used to investigate species distributions, community patterns, provide a reference to assess change through time, and can be used to train algorithms to automatically detect and annotate marine fauna. Background & Summary Marine imagery can be a cost-effective and non-destructive way to gather information about the seafloor1. Extracting detailed and high-quality biological information from these images, however, is a key bottleneck because image annotation is time-consuming and often expensive. Perhaps because these data are difficult to produce, they have historically rarely been shared on public repositories. There has been a recent societal and scientific shift towards sharing data, greater transparency and reproducibility of data (FAIR data principles)2,3, and increasing capabilities of machine learning methods that require large numbers of annotated images has encouraged more scientists and research institutions to make datasets of expert annotated marine images publicly available (e.g., FathomNet4, BenthOZ-20155). Nonetheless, for most regions around the world, including the Southern Ocean, expert annotated seafloor images are rarely publicly available. The sparsity of biological data has impeded efforts to benchmark the status of marine ecosystems, and therefore also prevents monitoring of ecological change. This is particularly evident for the highly biodiverse Antarctic seafloor6–8, which still lacks a comprehensive analysis of the distribution of its biodiversity on a continental scale. A database of the distribution and abundances of a broad range of organisms covering the entire Antarctic continental shelf is invaluable to help identify key conservation areas and manage these unique ecosystems. In this data descriptor, we introduce a database of expert annotated seafloor imagery from the Antarctic continental shelf and slope, the Antarctic Seafloor Annotated Imagery Database (AS-AID, Fig. 1). AS-AID is comprised of 52,491 images of which 3,599 are annotated with a total of 632,252 annotation-points and 174 faunal and substrate classifications. Images have been collected from research campaigns between 1985 and 2019 for which data was either publicly available or could be sourced by directly contacting data curators. This represents the first international effort to collate images from different regions of the Southern Ocean at a pan-Antarctic scale into a single database and annotate these images using a single classification scheme. Simple overview of the steps involved in building AS-AID, including the methodology used to generate the database. See methods section for more details. Seabed photo sourced from pangea.de. Repeatability is key to making data useful beyond the scope of a single project, which is why annotations in AS-AID are based on an internationally recognised standard for labelling marine images (CATAMI9), and all annotations have exact reference points (x and y coordinates) to their respective images. Users can view AS-AID annotations on Squidle+ (www.squidle.org) and review all annotations for a single label and customise annotation labels to suit their own research objective. AS-AID can be used to train deep learning algorithms for automated classification of marine fauna. Combining AS-AID with environmental data and statistical models can allow insights into the species distributions and ecological processes that shape Antarctic seafloor ecosystems10–13. Methods The Antarctic Seafloor Annotated Image Database (AS-AID) includes downward facing images of the seafloor at a depth range between 0–3,307 m from Antarctic research campaigns for which we were able to obtain access to both images and detailed metadata records. We focussed annotations in the first iteration to images deeper than 100 m because at these depths environmental conditions are more stable (e.g., relatively less impact on the fauna from iceberg scour) and more comparable between regions, which suits the continental scale biodiversity analysis that this database has primarily been developed for. Further, and also in its first iteration, AS-AID focusses on downward facing seafloor imagery rather than oblique facing imagery (where cameras can be mounted at different angles) because initial work indicated that downward facing images are more comparable between different research campaigns, with greater uniformity of illumination and resolution. Additionally, the area of seabed imaged with downward facing cameras can be more accurately estimated than from data collected with obliquely mounted cameras. In total, we identified 21 research campaigns from 2019 and prior that collected downward facing images of suitable quality with accessible metadata records (Figs. 1, 2 & Table 1). The only research campaigns that we identified as potentially suitable, but for which we could not access all relevant data were LMG1703 and ANA08D, which are therefore excluded from AS-AID at time of writing. More relevant research campaigns potentially exist for this time-period, and will be included in future iterations if they are identified as suitable. Overview of the circum-Antarctic distribution of sampling locations, including which research campaign (“Survey ID”) they originate from. More details on the individual research campaigns can be found in Tables 1–4. Table 1Research campaign (survey) details, including links and references to source data repositories where available (all images available on https://data.imas.utas.edu.au/imagery/IMAS_Antarctic/).Survey IDYearRegionContactReference for dataset (if available)Source dataset link (if available)License (if available)PS06 (ANT-III/3)1984/85Weddell SeaJ. GuttGutt21, Gutt22www.pangaea.de/?q = ps06 + benthos&f.author%5B%5D = Gutt%2C + Julian&f.pubyear%5B%5D = 1985CC-BY-3.0PS14 (ANT-VII/4)1989Weddell SeaJ. GuttGutt23, Gutt24, Gutt25, Gutt26,Gutt27, Gutt28, Gutt29, Gutt30,Gutt31, Gutt32, Gutt33, Gutt34,Gutt35, Gutt36, Gutt37, Gutt38,Gutt39, Gutt40www.pangaea.de/?q = ps14 + benthos&f.author%5B%5D = Gutt%2C + Julian&f.pubyear%5B%5D = 2010&f.pubyear%5B%5D = 1989CC-BY-3.0PS18 (ANT-IX/3)1990/91Lazarev SeaJ. GuttGutt41, Gutt42, Gutt43, Gutt44,Gutt45, Gutt46, Gutt47, Gutt48,Gutt49, Gutt50, Gutt51, Gutt52,Gutt53, Gutt54, Gutt55, Gutt56,Gutt57, Gutt58, Gutt59, Gutt60,Gutt61, Gutt62www.pangaea.de/?q = ps18 + benthos&f.author%5B%5D = Gutt%2C + JulianCC-BY-3.0PS61 (ANT-XIX/5)2002Antarctic PeninsulaJ. GuttArntz and Gutt63, Arntz and Gutt64https://doi.org/10.1594/PANGAEA.220747https://doi.org/10.1594/PANGAEA.220746CC-BY-3.0PS81 (ANT-XXIX/3)2013Antarctic PeninsulaD. PiepenburgPiepenburg, et al.65https://doi.org/10.1594/PANGAEA.872719CC-BY-3.0PS96 (ANT-XXXI/2)2015/16Weddell SeaD. PiepenburgPiepenburg66https://doi.org/10.1594/PANGAEA.862097CC-BY-3.0PS1182019Antarctic PeninsulaA. PurserPurser, et al.67https://doi.org/10.1594/PANGAEA.911904CC-BY-4.0TAN08022008Ross SeaD. Bowdenhttps://atlas.niwa.co.nz/CC-BY-3.0TAN18022018Ross SeaD. Bowdenhttps://atlas.niwa.co.nz/CC-BY-3.0TAN19012019Ross SeaD. Bowdenhttps://atlas.niwa.co.nz/CC-BY-3.0NBP08082008Antarctic PeninsulaC. Smith A. ZieglerGrange and Smith68LMG09022009Antarctic PeninsulaC. Smith A. ZieglerGrange and Smith68NBP10012010Antarctic PeninsulaC. Smith A. ZieglerGrange and Smith68NBP14022014East AntarcticaA. PostLeventer, et al.69, Post, et al.70https://www.usap-dc.org/view/dataset/601310CC BY-NC 4.0AA20112011East AntarcticaJ. SmithSmith71data.aad.gov.au/metadata/records/VMS_Benthic_PhotographyCC-BY-4.0NBP15022015Ross SeaP. BartBart72https://doi.org/10.15784/601182CC BY-NC 4.0LMG13112013Antarctic PeninsulaA. LeventerDomack73https://doi.org/10.15784/601311CC BY-NC 4.0JR2622011South OrkneyD. BarnesBarnes and Sands74, Barnes, et al.75JR150052015South OrkneyH. GriffithsBrasier, et al.76JR170012017Antarctic PeninsulaD. BarnesBarnes, et al.77, Zwerschke, et al.78JR170032018Antarctic PeninsulaK. LinseLinse, et al.79https://doi.org/10.5285/48dcef16-6719-45e5-a335-3a97f099e451http://www.nationalarchives.gov.uk/doc/open-government-licence/version/3/ Research campaign (survey) details, including links and references to source data repositories where available (all images available on https://data.imas.utas.edu.au/imagery/IMAS_Antarctic/). https://doi.org/10.1594/PANGAEA.220747 https://doi.org/10.1594/PANGAEA.220746 Sampling locations and image collection AS-AID images were collected during 21 research campaigns between 1985 and 2019 (Tables 1–4), using a range of different camera platforms (Table 4). Figure 2 shows an overview of sampling locations for all image transects included in AS-AID. Images and metadata were downloaded and stored on external hard-drives, and image filenames were standardised for processing. Table 4Details on image collection, image characteristics and annotation.Survey IDCamera-platformDepth-range (IBCSO V2)Distance from seafloorSizeLaser-ImageColour-Annotationsfor imagesof imagedPoint-gridExhaustive searchincludedannotatedseabed areapointscropcorrectedall species & substratemobile speciesVME speciesPS06Photo sledge (FTS)349–1115sameConstant (weight)Unknown—NoYes97224130PS14Photo sledge (FTS)148–1347sameConstant (weight)0.56 m2—NoYes128521110471PS18Photo sledge (FTS)144–956176–956Constant (weight)1 m2—NoYes111241889589PS61Photo sledge (FTS)330–348sameConstant (weight)~1 m2—NoYes129621813PS81OFOS52–783sameVariableCalculated50 cm80%Yes1124288388816756PS96OFOS198–731sameVariableProvided + calculated50 cm80%No2343676024548PS118OFOBS82–2169sameVariableCalculated50 cm80%Yes280802490017272TAN0802DTIS287–2258sameVariableCalculated20 cmNoYes186846237911TAN1802DTIS664–1213sameVariableCalculated20 cmNoYes36288122796689TAN1901DTIS447–1373sameVariableCalculated20 cmNoYes402841683710894NBP0808YOYO573–642sameConstant (weight, 2.5 m)~3 m2 + calculated10 cmNoYes442827224LMG0902YOYO578–678sameConstant (weight, 2.5 m)~3 m2 + calculated10 cmNoYes885631836NBP1001YOYO436–725sameConstant (weight, 2.5 m)~3 m2 + calculated10 cmNoYes212765260650NBP1402YOYO281–899sameConstant (weight, 2.5 m)~4.8 m2 + calculated10 cm90%Yes84242474351AA2011CTD198–1453sameVariableCalculated50 cm90%Yes76685322569NBP1502YOYO175–597sameConstant (weight, 2.5 m)Calculated10 cmNoYes2646062421802LMG1311YOYO349–654sameConstant (weight, 2.5 m)Calculated10 cmNoYes64801720309JR262SUCS219–280sameConstant (frame)0.51 m2—NoYes183611229JR15005SUCS473–1017sameConstant (frame)0.51 m2—NoYes9504875666JR17001SUCS67–507sameConstant (frame)0.51 m2—NoYes15127448JR17003SUCS184–865sameConstant (frame)0.51 m2—NoYes68042702331TOTAL67–225838869218057262988 Details on image collection, image characteristics and annotation. Image pre-processing We inspected all 72,560 images visually and removed images/transects from AS-AID that were out of focus, too far or too close to the seafloor to allow identification of animals, had poor lighting, or had unreliable coordinates (Tables 2 & 3), resulting in 52,491 usable images. We then corrected images for tone, colour and contrast using the automated batch function in Adobe Photoshop Version 21 (all research campaigns apart from PS96, see Table 4). Further, because the corners of images from research campaigns PS81, PS96, PS118, AA2011 and NBP1402 are out of focus or badly illuminated we cropped these images to either 80% or 90% of their original size (Table 4). Table 2Data characteristics for each research campaign (survey) and comments on which data were excluded from annotation efforts. Metadata quality: * single longitude/latitude position for each transect; ** longitude/latitude available for start/end locations of each transect Locations of individual images interpolated from start/end locations; *** longitude/latitude available for each image.Survey IDMetadataTransectsCommentsqualityincludedPS06*2/10Excluded transect 288: located near/under iceshelf (no available environmental data) Excluded transects 301-303, 307, 309-311: Blurry imagesPS14**18/24Image quality lowExcluded transects 248, 250, 259, 261: Blurry/discoloured imagesExcluded transect 274: Coordinates unreliable (transect on land)Excluded transect 312: Located in the South AtlanticPS18**22/25Excluded transects 162-1, 173-2 & 207-1: located near/under iceshelf (no available environmental data)PS61**2/8Only included transects 235-1 & 249-1: all other transects are on the Scotia Arc which is not part of this databasePS81***29/30Excluded transect 159: Bad visibility making it difficult to identify organismsPS96***13/13—PS118***10/11Excluded transect 38: Bad visibility making it difficult to identify organismsTransects 39 & 69: Reduced transect starts and ends to form straight transect linesTAN0802***17/53Excluded transects 181-310 which are located on seamounts off the Antarctic continental shelf.TAN1802***31/31Transects 160-213 are located on seamounts off the Antarctic continental shelf.TAN1901***34/34Transect 209 is located on a seamount off the Antarctic continental shelf.NBP0808**5/6Excluded transect 1103: Bad illumination/image qualityLMG0902**6/6—NBP1001**21/22Excluded transect 1325: Uncertain reliability of coordinates (transect very long with few images only)NBP1402**10/10—AA2011***71/86Excluded transects CTD53, CTD55, CTD57-58, CTD61-64, CTD85, CTD90, CTD93, CTD98, CTD107, CTD115 & CTD138: no or only blurry/unusable imagesNBP1502**5/7Excluded transects KC16 & KC17: no metadata available/accessibleLMG1311**2/2—JR262*4/23Image quality very low4 transects on the South Orkney shelf included, all other transects are off the Antarctic continental shelf/slopeJR15005**21/21—JR17001***7/7—JR17003***12/12— Data characteristics for each research campaign (survey) and comments on which data were excluded from annotation efforts. Metadata quality: * single longitude/latitude position for each transect; ** longitude/latitude available for start/end locations of each transect Locations of individual images interpolated from start/end locations; *** longitude/latitude available for each image. Image quality low Excluded transects 248, 250, 259, 261: Blurry/discoloured images Excluded transect 274: Coordinates unreliable (transect on land) Excluded transect 312: Located in the South Atlantic Excluded transect 38: Bad visibility making it difficult to identify organisms Transects 39 & 69: Reduced transect starts and ends to form straight transect lines Image quality very low 4 transects on the South Orkney shelf included, all other transects are off the Antarctic continental shelf/slope Table 3Overview of the number of images per research campaign (survey) and how subsets were created.Survey IDNumber of imagesImage subsettingTotalIncludedSubset 1Subset 2PS065001695Transect 289: selected 5 random imagesTransect 292: selected all good quality images (4)PS141523109811965Spatially-balanced-randomHowever: Start/end coordinates not trustworthy (ship speed would be > 1 kt) for transects 245, 250, 256, 259, 260, 261, 275, 278-3, 280 & 293: Selected 5 random images out of the first 25 images from these transects, using starting GPS coordinates as image position.PS181705145610359Spatially-balanced-randomHowever: Start/end coordinates not trustworthy (ship speed would be > 1 kt) for transects 171-1, 173-1, 206-2, 211-2 & 222-1: Selected 5 random images out of the first 25 images from these transects, using starting GPS coordinates as image position.PS61139129127Spatially-balanced-randomPS8114543145101041528Spatially-balanced-randomPS9627362727217112Spatially-balanced-randomPS118126296056260133Spatially-balanced-randomTransect 11-2: GPS-coordinates wiggle a lot between images - > subset based on distance from start rather than distance along transectTransect 39-1: removed the first 800 and the last 487 images to result in a straight lined transectTransect 69-1: removed the first 29 and the last 50 images to result in a straight lined transectTAN0802130461088717391Spatially-balanced-randomTAN180274705566336175Spatially-balanced-randomTAN190183015952373195Spatially-balanced-randomNBP0808300242 (41)4122Spatially-balanced-randomLMG0902288260 (82)8243Spatially-balanced-randomNBP100118081480 (207)197104Spatially-balanced-randomNBP14027614537843Spatially-balanced-randomAA201118534357171Selected 1 random image per transect due to low number of usable imagesNBP150228691719245124Spatially-balanced-randomLMG13112632226031Spatially-balanced-randomJR26261311710Selected 5 random images per transectJR150057735058853Spatially-balanced-randomJR17001142120148Spatially-balanced-randomJR170038502796333Spatially-balanced-randomTOTAL7256054143 (52491)35991912“Total” is the number of images downloaded from the repository; “Included” is the number of images that show the seafloor, are in focus and are neither blurry nor too bright/dark, and are included in our database. “Subset 1” and “Subset 2” are the number of images that were selected for annotation for the point-scoring method and the exhaustive search respectively. Note that only annotated images are publicly available for research campaigns NBP0808 (41 images), LMG0902 (82 images) and NBP1001 (207 images). Overview of the number of images per research campaign (survey) and how subsets were created. Transect 289: selected 5 random images Transect 292: selected all good quality images (4) Spatially-balanced-random However: Start/end coordinates not trustworthy (ship speed would be > 1 kt) for transects 245, 250, 256, 259, 260, 261, 275, 278-3, 280 & 293: Selected 5 random images out of the first 25 images from these transects, using starting GPS coordinates as image position. Spatially-balanced-random However: Start/end coordinates not trustworthy (ship speed would be > 1 kt) for transects 171-1, 173-1, 206-2, 211-2 & 222-1: Selected 5 random images out of the first 25 images from these transects, using starting GPS coordinates as image position. Spatially-balanced-random Transect 11-2: GPS-coordinates wiggle a lot between images - > subset based on distance from start rather than distance along transect Transect 39-1: removed the first 800 and the last 487 images to result in a straight lined transect Transect 69-1: removed the first 29 and the last 50 images to result in a straight lined transect “Total” is the number of images downloaded from the repository; “Included” is the number of images that show the seafloor, are in focus and are neither blurry nor too bright/dark, and are included in our database. “Subset 1” and “Subset 2” are the number of images that were selected for annotation for the point-scoring method and the exhaustive search respectively. Note that only annotated images are publicly available for research campaigns NBP0808 (41 images), LMG0902 (82 images) and NBP1001 (207 images). Image sub-setting Annotating all images from all research campaigns is neither feasible nor desirable due to the large differences in sampling effort among research campaigns (i.e., the density of images along the transect lines is highly variable). We therefore created a subset of the full dataset by selecting a spatially balanced random subset of the images using their coordinates and the distance along the transect line with the R package MBHdesign version 2.2.2. Where only one coordinate for the transect was available, or where recorded end-coordinates of the transects did not match with expected values given the duration of the deployment and the normal ship-speed during deployment (max 1 kn), we randomly selected 5 of the first 25 images in those transects and assigned these images to the starting position of the transect (Table 2). The primary image subset we created contains approximately 1 image for every 100 m of transect (3,599 images, Table 3). Further, we created a smaller secondary subset of images for a more detailed exhaustive search of each image, containing approximately 1 image for every 200 m of transect (1912 images, Table 3). Image annotation We used both a point-grid and an exhaustive search for annotating the images using the web-based annotation toolkits CoralNet (coralnet.ucsd.edu/) and Biigle BIIGLE 2.0 (www.biigle.de/). Point-grid annotation allows estimating percentage cover of all substrates and organisms present in the image, but the number of points that make up the grid influence how well these percentages are estimated, and how well rare and/or small organisms are detected14. In contrast, labelling organisms individually in an exhaustive search reduces detectability issues for rare species. However, individual labels present a challenge for many colonial organisms because individual colony sizes can vary greatly and it is frequently ambiguous how many colonies there are. We chose to first label all images in the primary subset using a point-grid, and then label the secondary subset using an exhaustive search. In the exhaustive search we focussed only on two groups of organisms: Mobile species, and small or rare species comprising Vulnerable Marine Ecosystem taxa (VME taxa as defined by the Commission for the Conservation of Antarctic Marine Living Resources, CCAMLR15) that are not colonial. For point-grid annotation we overlaid all images in the primary subset with a regular grid of 108 points (9 × 12) to closely match the 3 × 4 height to width ratio of the images. We chose 108 points as a good compromise between sampling effort and time constraints. For each point we only identified what substrate (e.g. fine substrates, Fig. 3) or organisms (such as different types of suspension feeders, Fig. 4) could be found within the single pixel in the centre of this point. While this required us to regularly zoom in and out of the image to determine the exact position of the point, it reduced potential bias towards labelling a point as an organism instead of substrate. We used CoralNet16 to manually annotate all images and used the inbuilt deep-learning algorithm, which automatically suggests the five most likely annotation labels for each point, to more quickly select the correct label. Example output from AS-AID: Percentage of grid-points identified as fine substrate (mud or sand) in the images from each sampling location. Example output from AS-AID: Percentage cover of grid-points identified as suspension feeders (combining ascidians, bryozoans, cnidarians, sponges) in images from each sampling location. For individually labelling organisms in the secondary subset via an exhaustive search, we used the online toolkit BIIGLE 2.0 and their “lawnmower mode” to systematically search through an image17. We identified each organism from the respective morphospecies list and drew a close-fitting circle around the organism if they were mainly (>50% of their body) within the image. For all images in the primary subset, we used BIIGLE to annotate visible laser-points that allow calculating the imaged seabed area. After completing annotations and quality control (detailed further below), we uploaded all images and annotations onto Squidle+ for making the data publicly available. Image classification scheme We classified all organisms based on CATAMI9 (Collaborative and Annotation Tools for Analysis of Marine Imagery), and added labels to the classification tree when we could confidently distinguish organisms further into separate morphospecies. CATAMI is a hierarchical classification scheme that uses taxonomy at the higher levels of the classification tree and morphology at the lower levels of the classification tree. This approach accounts for the fact that it is often difficult to identify seafloor fauna to a fine taxonomic resolution without physical specimens and that the morphology of seafloor fauna is often reflective of function. Where further discrimination is known or needed these can be added at the end of the respective CATAMI branch. We then integrated our label tree with the Australian Morphospecies Catalog (AMC) on Squidle+, meaning all labels we used can be found in AMC. For the scoring of VMEs, we combined CATAMI with the VME classification guide provided by CCAMLR15, and these labels are also integrated into the AMC on Squidle+. A list of the 20 most commonly used labels for each of the three datasets can be found in Tables 5–7. Table 6The top 20 most common labels in the exhaustive search for mobile species.Label names (all within Australian Morphospecies Catalog category “Biota”)FrequencyEchinoderms > Ophiuroids > Brittle / snake stars78798Worms46017Echinoderms > Ophiuroids > Brittle / snake stars > Suspension feeders > Five arms7667Crustacea > Prawns / shrimps / mysids6574Sea spiders5278Crustacea > Amphipoda2698Worms > Polychaetes > Tube worms2601Echinoderms > Sea cucumbers > Benthic2450General Unknown Biology2427Echinoderms > Ophiuroids > Brittle / snake stars > Suspension feeders > Six arms > Red Six arm Suspension feeder2403Echinoderms > Ophiuroids > Brittle / snake stars > Suspension feeders > Five arms > White Five arm Suspension feeder2195Brachiopods1982Echinoderms > Feather stars > Unstalked crinoids1841Echinoderms > Ophiuroids > Brittle / snake stars > Suspension feeders1711Echinoderms > Ophiuroids > Brittle / snake stars > Curly arms1648Echinoderms > Sea stars1644Echinoderms > Ophiuroids > Brittle / snake stars > Suspension feeders > Six arms > White Six arm Suspension feeder1581Echinoderms > Sea cucumbers > Benthic > Suspension feeder1432Echinoderms > Sea cucumbers > Benthic > Suspension feeder > Buried > Red tentacles1053Echinoderms > Ophiuroids > Brittle / snake stars > Suspension feeders > Seven arms843More information on the frequency of each annotation label (including for less commonly used labels) can be found on squidle.org. The top 20 most common labels in the exhaustive search for mobile species. More information on the frequency of each annotation label (including for less commonly used labels) can be found on squidle.org. Table 5The top 20 most common labels in the point-scoring analysis.Label nameFrequency2 Physical > Substrate > Unconsolidated (soft) > Sand / mud ( < 2 mm)2104842 Physical > Substrate > Unconsolidated (soft) > Pebble / gravel820712 Physical > Substrate > Unconsolidated (soft) > Pebble / gravel > Biologenic19912Unscorable146291.1 Biota > Matrix > Turf/Sediment/Silt matrix123521.1 Biota > Matrix > Bryozoa / Cnidaria Matrix89202 Physical > Substrate > Consolidated (hard) > Cobbles50722 Physical > Substrate > Consolidated (hard) > Rock44601.1 Biota > Bryozoa > Hard > Branching > Antler34812 Physical > Substrate > Consolidated (hard) > Boulders23341.1 Biota > Matrix17961.1 Biota > Echinoderms > Ophiuroids > Brittle / snake stars16601.1 Biota > Bryozoa > Soft > Dendroid16121.1 Biota > Bryozoa > Hard > Fenestrate > Bryozoa Lace15071.1 Biota > Cnidaria > Hydrocorals > Branching12871.1 Biota > Bryozoa > Hard > Branching > Leafy12471.1 Biota > General Unknown Biology12411.1 Biota > Echinoderms > Feather stars > Unstalked crinoids10131.1 Biota > Ascidians > Unstalked > Colonial8251.1 Biota > Sponges > Massive forms > Simple823More information on the frequency of each annotation label (including for less commonly used labels) can be found on squidle.org. The top 20 most common labels in the point-scoring analysis. More information on the frequency of each annotation label (including for less commonly used labels) can be found on squidle.org. Table 7The top 20 most common labels in the exhaustive search for selected VME species.Label names (all within Australian Morphospecies Catalog category “Biota”)FrequencyAscidians > Unstalked > Colonial27185Cnidaria > Corals > Stony corals > Solitary8258Ascidians > Stalked > Colonial7242Cnidaria > Hydroids > Solitary3583Cnidaria > Corals > Black & Octocorals > Whip > Sea Whip > Gorgonacea3331Cnidaria > Colonial anemones > Zoanthids2399Cnidaria > Corals > Black & Octocorals > Branching (3D) > Non-fleshy > Bottle-brush > Simple > Gorgonacea2379Cnidaria > True anemones1828Ascidians > Stalked > Solitary1607Ascidians > Unstalked > Solitary1413Brachiopods982Cnidaria > Corals > Black & Octocorals > Fan (2D) > Fern-frond > Simple615Cnidaria > Hydroids > Colonial436Cnidaria > Corals > Black & Octocorals > Branching (3D) > Fleshy > Mushroom364Cnidaria > Corals > Black & Octocorals > Branching (3D) > Non-fleshy > Arborescent > Gorgonacea253Cnidaria > Corals > Black & Octocorals > Branching (3D) > Non-fleshy > Bushy205Cnidaria > Corals > Black & Octocorals > Branching (3D) > Non-fleshy > Bottle-brush > Complex > Gorgonacea177Cnidaria > Corals > Black & Octocorals > Quill (seapen)165Cnidaria > Corals > Black & Octocorals > Whip > Sea Whip > Alcyonacea121Molluscs > Bivalves > Scallop > Adamussium colbecki110More information on the frequency of each annotation label (including for less commonly used labels) can be found on squidle.org. The top 20 most common labels in the exhaustive search for selected VME species. More information on the frequency of each annotation label (including for less commonly used labels) can be found on squidle.org. Quality control & expert review All annotations went through rigorous quality control comprising three stages: Labelling review, internal QA/QC, and expert review. Labelling review concerns the exhaustive search only: After all images of a research campaign had been fully annotated in BIIGLE, a second scorer systematically searched through all images again to label any organisms the first scorer had missed. The use of two scorers to search though each image gave us high confidence that few individuals have been missed. Internal QA/QC: We conducted a two-stage internal QA/QC both individually for each research campaign during annotation and also for all research campaigns combined. In the internal QA/QC we reviewed each label, corrected any misclassification and discussed with the team if necessary. For reviewing labels from the grid annotation, we cropped 50 × 50 pixel thumbnails around each grid-point and sorted the thumbnails by their label in a morphospecies library. For reviewing the individual annotations from the exhaustive search using BIIGLE we used BIIGLE’s inbuilt “Largo” review function. After this first review, we then created a second library containing only thumbnails of morphospecies whose labels had changed during the review. This second review stage was to ensure all changes made to the original scoring were correct. Expert review: In the third stage of our review process, co-authors contributed as needed with their taxonomic expertise to ensure the morphospecies grouping and labelling was consistent, and to help identify some of the unknown organisms encountered. Data Records All seafloor images, a static version of the annotations and the associated metadata are deposited at the IMAS data portal18,19 and can be accessed under the following: https://doi.org/10.25959/2R0E-PB18 and https://doi.org/10.25959/SXGA-V355. The image record contains the campaign names for each scientific survey, and within each of these folders are the individual deployments containing the images as well as metadata files (“navdata.csv”) with the following key variables: “key” = Image name “pose.lon” = Longitude “pose.lat” = Latitude “pose.depth” = Depth at the seafloor in m “pose.alt” = distance of the camera to the seafloor in m “timestamp.start” = time in UTC “data.transect_ID” = deployment identifier “data.image.area” = imaged seabed area in m2 if available “data.image_area_source” = how estimates of imaged seabed area were obtained “data.license” = image license “data.source_link” = DOI link to original data source where available “data.equipment” = camera system used to obtain the imagery The static version of the annotations at the time of publication18 contains three csv files (Points, Mobiles and VMEs) for each of the 21 research campaigns. The csv files contain all data variables available to download from Squidle+ , the most important ones are described in more detail below: “point.media.key” = image name “point.media.path_best_thm” = persistent file path to image “point.y” = annotation point location along the y-axis “point.x” = annotation point location along the x-axis “point.bbx” = bounding box coordinates containing the annotated fauna “point.pose.lon” = Longitude of the image “point.pose.lat” = Latitude of the image “label.lineage_names” = full AMC annotation label Classification catalogues with example annotations for each label used in AS-AID are publicly available20 and can be access under the following: https://doi.org/10.25959/J4EE-9A56. All images, annotations and the classification scheme can also be accessed on Squidle+ by filtering for “IMAS Collated Antarctic Imagery”. Technical Validation Annotations in AS-AID have gone through a rigorous review process involving experienced scientists, trained students and taxonomists. During the annotation process, the annotation team (consisting of two scientists, one postgraduate student and one undergraduate student) met regularly (daily at the beginning, weekly at a later stage and every time a research campaign was fully annotated) to discuss uncertain annotations, ensure consistent annotation between the scorers and create new labels where necessary. All annotations were checked for correct labelling once a research campaign was fully annotated and the entire database was checked again after all research campaigns had been fully annotated. Further, expert taxonomists were consulted in particular for key taxa such as sponges, cnidarians and bryozoans. The largest uncertainties in this database were distinguishing whether an object was an organism or substrate, and whether an organism was dead or alive, and we created a number of categories to label these cases accordingly, such as e.g., “Unknown Biological Matrix” or “Unidentifiable”. In AS-AID, we have opted to use experts to consecutively perform tasks on the same set of images. In particular, this meant one expert labelled the images while a second expert checked the annotations in the review process and flagged any misclassifications for discussion. While this process didn’t allow label accuracy to be quantified, it did allow organisms to be labelled more consistently throughout the entire database. Researchers are invited to assess the validity of AS-AID for their own research purposes. Every annotation in AS-AID is linked to an image with a unique filename and contains the x- and y- coordinates where the organisms is located on that image as well as other metadata such as imaged area and imaging gear. Therefore, every annotation is fully traceable, allowing researchers to customise and assess the validity of AS-AID for their needs. Acknowledgements We thank the crews and onboard scientific parties of the expeditions which collected the data within this study for providing the images and metadata for analysis in this study, either directly or via online raw data archives. We thank Amy Leventer for helping with data acquisition and Fanny Hermand for her help during the early stages of this project. This research was supported by the Australian Research Council Discovery Project DP190101858 and by the Australian Research Council Special Research Initiative, Australian Centre for Excellence in Antarctic Science (Project Number SR200100008). NZ involvement is supported by Ministry of Business Innovation and Employment programme C01X1710, Ross Sea Research and Monitoring Programme; is the world’s largest MPA effective? C.G. has been supported by a CSIRO-UTAS Quantitative Marine Science Scholarship. N.B. would like to thank the John Ellerman Foundation as part of the Marine and Coastal Program Coordinator role at the South Atlantic Environmental Research Institute (SAERI). Author contributions J.J.: Conceptualization, Methodology, Software, Validation, Investigation, Data Curation, Writing - Original Draft, Writing – Review & Editing, Visualization, Supervision, Project Administration, Funding acquisition. V.S.: Methodology, Validation, Investigation, Data Curation, Writing – Review & Editing. C.G.: Methodology, Validation, Investigation, Data Curation, Writing – Review & Editing. T.W.: Methodology, Validation, Investigation, Writing – Review & Editing. N.H.: Conceptualization, Methodology, Writing – Review & Editing, Supervision, Project Administration, Funding acquisition. D.K.B.: Investigation, Resources, Writing – Review & Editing. D.A.B.: Investigation, Resources, Writing – Review & Editing. J.G.: Investigation, Resources, Writing – Review & Editing. N.B.: Investigation, Writing – Review & Editing. R.D.: Investigation, Writing – Review & Editing. M.E.: Investigation, Writing – Review & Editing. A.L.P.: Investigation, Writing – Review & Editing. H.G.: Resources, Writing – Review & Editing. K.L.: Resources, Writing – Review & Editing. A.P.: Resources, Writing – Review & Editing. D.P.: Investigation, Resources, Writing – Review & Editing. C.S.: Resources, Writing – Review & Editing. A.Z.: Resources, Writing – Review & Editing. C.J.: Conceptualization, Methodology, Writing – Review & Editing, Supervision, Project Administration, Funding acquisition. Funding This research has been funded as part of Australian Research Council Discovery Project 190101858. Data availability All images, annotations and the classification scheme can be openly accessed online via the following links: Images18: https://doi.org/10.25959/SXGA-V355 Annotations19: https://doi.org/10.25959/2R0E-PB18. Classification catalogs20: https://doi.org/10.25959/J4EE-9A56. All data can also be accessed on Squidle+ by filtering for “IMAS Collated Antarctic Imagery”. Code availability Image annotations and metadata can also be directly accessed using the software packages “SQAPI” for R (available from https://github.com/sajessop/SQAPI) and “sqapi” for python (available using “pip install sqapi”). Competing interests The authors declare no competing financial interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Citations Data Availability Statement All images, annotations and the classification scheme can be openly accessed online via the following links: Images18: https://doi.org/10.25959/SXGA-V355 Annotations19: https://doi.org/10.25959/2R0E-PB18. Classification catalogs20: https://doi.org/10.25959/J4EE-9A56. All data can also be accessed on Squidle+ by filtering for “IMAS Collated Antarctic Imagery”. Image annotations and metadata can also be directly accessed using the software packages “SQAPI” for R (available from https://github.com/sajessop/SQAPI) and “sqapi” for python (available using “pip install sqapi”).